fix(release): harden embedding store and plugins for ReMe 0.4.1.9 (#503)

* chore(release): prepare ReMe 0.4.1.9

* refactor(config): remove daily_cookbook and streamline plugin configs

- Delete the entire daily_cookbook.yaml standalone application config
- Remove qwenpaw dependencies verification and related CI workflow steps
- Simplify release workflows by removing qwenpaw verification and enforcing reme-ai >=0.4.1.9
- Update plugin start commands and examples to use 'default' or 'demo' configs instead of daily_cookbook
- Adjust imports and tests related to daily_cookbook removal and injected_job_kwargs enhancements
- Refactor agent wrapper to support injected_job_kwargs for job parameter injection in auto-fin and daily-paper
- Improve daily_paper digest prompt to include configured daily directory and correct historical search constraints
- Update dependency versions in pyproject.toml files to require reme-ai >=0.4.1.9 and remove qwenpaw optional dependencies
- Clean up unused environment variables and obsolete test cases related to daily_cookbook and verification steps

* fix(local_embedding_store): retry batch computation on vector space changes

- Add up to 3 attempts to recompute embedding batch if vector space changes during processing
- Log warnings when maximum retries reached and discard stale results
- Prevent caching results from outdated vector spaces to maintain consistency
- Add tests to verify retry behavior and abort after continuous vector space churn

fix(daily_paper): update digest search logic and tests

- Change search to query existing memory, not only previous articles in daily_dir
- Allow multiple searches outside daily_dir but limit links to dated markdown in daily_dir before today
- Update test assertions to reflect revised search and linking rules

* fix(embedding): retry vector space changes per request

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,97 @@
+name: Bug report
+description: Report reproducible incorrect or unexpected ReMe behavior
+title: "[Bug]: "
+labels: [bug]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for helping improve ReMe. Please remove secrets, API keys, and private memory content before submitting.
+
+  - type: textarea
+    id: description
+    attributes:
+      label: Description
+      description: What happened, and what did you expect instead?
+      placeholder: Describe the observed and expected behavior.
+    validations:
+      required: true
+
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: Steps to reproduce
+      description: Provide the smallest configuration and command sequence that reproduces the problem.
+      placeholder: |
+        1. Configure ...
+        2. Run ...
+        3. Observe ...
+    validations:
+      required: true
+
+  - type: textarea
+    id: config
+    attributes:
+      label: Relevant configuration
+      description: Include only relevant values and redact credentials, tokens, endpoints, and private paths.
+      render: yaml
+
+  - type: textarea
+    id: logs
+    attributes:
+      label: Logs or traceback
+      description: Paste relevant output after removing secrets and private workspace content.
+      render: shell
+
+  - type: input
+    id: reme-version
+    attributes:
+      label: ReMe version
+      placeholder: e.g. 0.4.1.8 or a commit SHA
+    validations:
+      required: true
+
+  - type: input
+    id: python-version
+    attributes:
+      label: Python version
+      placeholder: e.g. 3.11.9
+    validations:
+      required: true
+
+  - type: dropdown
+    id: os
+    attributes:
+      label: Operating system
+      options:
+        - Linux
+        - macOS
+        - Windows
+        - Other
+    validations:
+      required: true
+
+  - type: dropdown
+    id: area
+    attributes:
+      label: Affected area
+      options:
+        - CLI or configuration
+        - HTTP, MCP, or local service
+        - Memory or workspace files
+        - Search, catalog, graph, or index
+        - Model or agent integration
+        - ReMe Studio
+        - Plugin or external integration
+        - Packaging or installation
+        - Other
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: safety
+    attributes:
+      label: Data safety
+      options:
+        - label: I removed credentials and private memory content from this report.
+          required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: false
+contact_links:
+  - name: ReMe documentation
+    url: https://reme.agentscope.io
+    about: Read the installation, configuration, and usage guides.
+  - name: Existing issues
+    url: https://github.com/agentscope-ai/ReMe/issues
+    about: Search for existing reports and discussions before opening a new issue.

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,64 @@
+name: Feature request
+description: Propose a focused enhancement to ReMe
+title: "[Feature]: "
+labels: [enhancement]
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem
+      description: What user problem or limitation should this change address?
+    validations:
+      required: true
+
+  - type: textarea
+    id: proposal
+    attributes:
+      label: Proposed behavior
+      description: Describe the desired behavior and its user-visible contract.
+    validations:
+      required: true
+
+  - type: dropdown
+    id: area
+    attributes:
+      label: Area
+      options:
+        - CLI or configuration
+        - Jobs or steps
+        - Memory or workspace files
+        - Search, catalog, graph, or index
+        - Service or client
+        - Model or agent integration
+        - ReMe Studio
+        - Plugin or external integration
+        - Documentation
+        - Other
+    validations:
+      required: true
+
+  - type: textarea
+    id: ownership
+    attributes:
+      label: Local-first and compatibility considerations
+      description: Explain any effect on user-owned files, rebuildable state, configuration, schemas, or service interfaces.
+
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: Alternatives considered
+      description: Describe workarounds or alternative designs you considered.
+
+  - type: textarea
+    id: examples
+    attributes:
+      label: Example usage
+      description: Show the proposed CLI, configuration, API, or UI behavior when useful.
+      render: shell
+
+  - type: checkboxes
+    id: contribution
+    attributes:
+      label: Contribution
+      options:
+        - label: I am willing to help implement or test this feature.

.github/ISSUE_TEMPLATE/question.yml

@@ -0,0 +1,53 @@
+name: Usage question
+description: Ask for help using or configuring ReMe
+title: "[Question]: "
+labels: [question]
+body:
+  - type: markdown
+    attributes:
+      value: Please check the documentation and existing issues before asking a new question.
+
+  - type: textarea
+    id: goal
+    attributes:
+      label: What are you trying to achieve?
+    validations:
+      required: true
+
+  - type: textarea
+    id: attempted
+    attributes:
+      label: What have you tried?
+      description: Include relevant commands or configuration, with secrets and private memory content removed.
+    validations:
+      required: true
+
+  - type: input
+    id: reme-version
+    attributes:
+      label: ReMe version
+      placeholder: e.g. 0.4.1.8 or a commit SHA
+
+  - type: dropdown
+    id: area
+    attributes:
+      label: Area
+      options:
+        - Installation
+        - Configuration
+        - CLI or service usage
+        - Memory and workspace management
+        - Search and retrieval
+        - ReMe Studio
+        - Plugin or integration
+        - Other
+
+  - type: checkboxes
+    id: checked
+    attributes:
+      label: Before submitting
+      options:
+        - label: I checked the [ReMe documentation](https://reme.agentscope.io) and searched existing issues.
+          required: true
+        - label: I removed credentials and private memory content.
+          required: true

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,35 @@
+## Summary
+
+<!-- Explain the problem and the smallest coherent change that addresses it. -->
+
+## Related issue
+
+<!-- Use "Fixes #123" when applicable. -->
+
+## Contract and data impact
+
+- [ ] No public configuration, schema, CLI, endpoint, streaming, or workspace-layout contract changes
+- [ ] No user-owned memory files are deleted or rewritten
+- [ ] Derived indexes, catalogs, graphs, caches, and metadata remain rebuildable
+
+<!-- If any item is unchecked, describe the impact and migration or recovery path. -->
+
+## Validation
+
+<!-- List the exact checks run and their results. Explain relevant checks that were not run. -->
+
+- [ ] Focused tests pass
+- [ ] Unit tests pass, or omitted tests are explained below
+- [ ] `pre-commit run --all-files` passes, or omitted checks are explained below
+- [ ] Frontend checks were run when `reme_studio/` changed
+
+## Checklist
+
+- [ ] I reviewed the diff for unrelated changes and sensitive data
+- [ ] Tests cover intentional behavior changes
+- [ ] Defaults, schemas, and concise documentation were updated together when required
+- [ ] Long-lived clients, tasks, services, and executors follow the application lifecycle
+
+## Screenshots or additional notes
+
+<!-- Include UI screenshots, compatibility notes, or follow-up work when relevant. -->

.github/workflows/_build-docs.yml

@@ -0,0 +1,58 @@
+name: _Build documentation
+
+on:
+  workflow_call:
+    inputs:
+      run_tests:
+        description: Run the documentation test suite before building
+        required: false
+        default: true
+        type: boolean
+      upload_pages_artifact:
+        description: Upload the build for a later GitHub Pages deployment job
+        required: false
+        default: false
+        type: boolean
+
+permissions:
+  contents: read
+
+jobs:
+  build:
+    name: Build documentation
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: github-pages
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Node
+        uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: '22.22.3'
+          cache: npm
+          cache-dependency-path: github-pages/package-lock.json
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Run tests
+        if: inputs.run_tests
+        run: npm test
+
+      - name: Build documentation
+        run: npm run build
+
+      - name: Configure Pages
+        if: inputs.upload_pages_artifact
+        uses: actions/configure-pages@45bfe0192ca1faeb007ade9deae92b16b8254a0d # v6
+
+      - name: Upload Pages artifact
+        if: inputs.upload_pages_artifact
+        uses: actions/upload-pages-artifact@7b1f4a764d45c48632c6b24a0339c27f5614fb0b # v4
+        with:
+          path: github-pages/dist

.github/workflows/_build-python-packages.yml

@@ -0,0 +1,88 @@
+name: _Build Python packages
+
+on:
+  workflow_call:
+    inputs:
+      expected_version:
+        description: Expected release version; omit for a consistency-only check
+        required: false
+        default: ''
+        type: string
+      upload_artifacts:
+        description: Upload distributions for later publish jobs
+        required: false
+        default: false
+        type: boolean
+
+permissions:
+  contents: read
+
+jobs:
+  distributions:
+    name: Build Python distributions
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: '3.11'
+
+      - name: Install build dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install build packaging pytest twine
+
+      - name: Validate package versions
+        if: inputs.expected_version == ''
+        run: python scripts/bump_version.py --check
+
+      - name: Validate release version
+        if: inputs.expected_version != ''
+        env:
+          EXPECTED_VERSION: ${{ inputs.expected_version }}
+        run: python scripts/bump_version.py --check --expected-version "${EXPECTED_VERSION}"
+
+      - name: Run package tests
+        run: PYTHONPATH=. python -m pytest tests/unit/test_package_versions.py -q
+
+      - name: Build and check distributions
+        run: |
+          mkdir -p dist/reme
+          python -m build --outdir dist/reme
+          python -m twine check dist/reme/*
+
+      - name: Verify distributions and isolated installation
+        run: |
+          REME_WHEEL="$(pwd)/$(ls dist/reme/reme_ai-[0-9]*.whl)"
+          python -m zipfile -l "${REME_WHEEL}" | (! grep 'reme/web/')
+          python -m zipfile -l "${REME_WHEEL}" | (! grep 'reme_studio/')
+          python -m venv "${RUNNER_TEMP}/reme-package-smoke"
+          "${RUNNER_TEMP}/reme-package-smoke/bin/python" -m pip install "${REME_WHEEL}[as]"
+          cd "${RUNNER_TEMP}"
+          "${RUNNER_TEMP}/reme-package-smoke/bin/python" -c "import reme"
+
+      - name: Verify released core dependencies
+        if: inputs.expected_version != ''
+        run: |
+          REME_WHEEL="$(pwd)/$(ls dist/reme/reme_ai-[0-9]*.whl)"
+          python -m venv "${RUNNER_TEMP}/reme-core-package-smoke"
+          "${RUNNER_TEMP}/reme-core-package-smoke/bin/python" -m pip install "${REME_WHEEL}[core]"
+          cd "${RUNNER_TEMP}"
+          "${RUNNER_TEMP}/reme-core-package-smoke/bin/python" - <<'PY'
+          from reme_studio import static_dir
+
+          assert (static_dir() / "index.html").is_file()
+          PY
+
+      - name: Upload ReMe distributions
+        if: inputs.upload_artifacts
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: reme-distributions
+          path: dist/reme/
+          if-no-files-found: error

.github/workflows/ci-docs.yml

@@ -0,0 +1,48 @@
+name: CI / Documentation
+
+on:
+  push:
+    branches: [main, master, dev, develop]
+    paths:
+      - '.github/workflows/ci-docs.yml'
+      - '.github/workflows/_build-docs.yml'
+      - 'AGENTS.md'
+      - 'README.md'
+      - 'README_ZH.md'
+      - 'docs/**'
+      - 'github-pages/**'
+      - 'reme_studio/README*.md'
+      - 'reme_studio/public/og.jpg'
+      - 'typescript/README*.md'
+      - 'plugins/*/README*.md'
+      - 'benchmark/*/README*.md'
+  pull_request:
+    branches: [main, master, dev, develop]
+    paths:
+      - '.github/workflows/ci-docs.yml'
+      - '.github/workflows/_build-docs.yml'
+      - 'AGENTS.md'
+      - 'README.md'
+      - 'README_ZH.md'
+      - 'docs/**'
+      - 'github-pages/**'
+      - 'reme_studio/README*.md'
+      - 'reme_studio/public/og.jpg'
+      - 'typescript/README*.md'
+      - 'plugins/*/README*.md'
+      - 'benchmark/*/README*.md'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  documentation:
+    name: Test and build documentation
+    uses: ./.github/workflows/_build-docs.yml
+    with:
+      run_tests: true

.github/workflows/ci-packages.yml

@@ -0,0 +1,40 @@
+name: CI / Python packages
+
+on:
+  push:
+    branches: [main, master, dev, develop]
+    paths:
+      - '.github/workflows/ci-packages.yml'
+      - '.github/workflows/_build-python-packages.yml'
+      - '.github/workflows/release-python.yml'
+      - 'pyproject.toml'
+      - 'README.md'
+      - 'reme/**'
+      - 'scripts/bump_version.py'
+      - 'tests/unit/test_package_versions.py'
+      - 'LICENSE'
+  pull_request:
+    branches: [main, master, dev, develop]
+    paths:
+      - '.github/workflows/ci-packages.yml'
+      - '.github/workflows/_build-python-packages.yml'
+      - '.github/workflows/release-python.yml'
+      - 'pyproject.toml'
+      - 'README.md'
+      - 'reme/**'
+      - 'scripts/bump_version.py'
+      - 'tests/unit/test_package_versions.py'
+      - 'LICENSE'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  distributions:
+    name: Build and verify distributions
+    uses: ./.github/workflows/_build-python-packages.yml

.github/workflows/ci-python-quality.yml

@@ -0,0 +1,40 @@
+name: CI / Python quality
+
+on:
+  push:
+  pull_request:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  pre-commit:
+    name: Pre-commit
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Setup Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: '3.11'
+          cache: pip
+
+      - name: Update setuptools
+        run: |
+          pip install -U setuptools wheel
+
+      - name: Install
+        run: |
+          pip install -q -e reme_studio -e ".[dev,core]"
+          pip install -q --no-deps -e plugins/auto-fin -e plugins/daily_paper
+
+      - name: Pre-commit starts
+        run: pre-commit run --all-files

.github/workflows/ci-python-tests.yml

@@ -0,0 +1,54 @@
+name: CI / Python tests
+
+on:
+  push:
+    branches: [main, master, dev, develop]
+  pull_request:
+    branches: [main, master, dev, develop]
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  unit-tests:
+    name: Unit Tests - py${{ matrix.python-version }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.11", "3.12", "3.13"]
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: 'pip'
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip setuptools wheel
+          pip install -e reme_studio -e ".[dev,core]"
+          pip install --no-deps -e plugins/auto-fin
+          pip install -e plugins/daily_paper
+          pip install coverage
+
+      - name: Run unit tests
+        run: |
+          coverage run -m pytest tests/unit plugins/auto-fin plugins/daily_paper \
+            -v \
+            --tb=long \
+            -s \
+            --log-cli-level=WARNING
+
+      - name: Generate coverage report
+        run: coverage report -m

.github/workflows/ci-reme-studio.yml

@@ -0,0 +1,90 @@
+name: CI / ReMe Studio
+
+on:
+  push:
+    paths:
+      - "reme_studio/**"
+      - ".github/workflows/ci-reme-studio.yml"
+      - ".github/workflows/release-reme-studio.yml"
+      - "scripts/package_studio.py"
+      - "tests/unit/test_package_versions.py"
+      - "pyproject.toml"
+      - "LICENSE"
+  pull_request:
+    paths:
+      - "reme_studio/**"
+      - ".github/workflows/ci-reme-studio.yml"
+      - ".github/workflows/release-reme-studio.yml"
+      - "scripts/package_studio.py"
+      - "tests/unit/test_package_versions.py"
+      - "pyproject.toml"
+      - "LICENSE"
+
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  studio:
+    name: Studio checks
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: reme_studio
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Setup Node
+        uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: "22.22.3"
+          cache: npm
+          cache-dependency-path: reme_studio/package-lock.json
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Run format check
+        run: npm run format:check
+
+      - name: Run lint
+        run: npm run lint
+
+      - name: Run tests
+        run: npm test
+
+      - name: Verify npm package
+        run: |
+          npm pack --pack-destination "${RUNNER_TEMP}"
+          tar -tzf "${RUNNER_TEMP}"/agentscope-ai-reme_studio-*.tgz | grep '^package/dist-static/index.html$'
+
+      - name: Set up Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: "3.11"
+
+      - name: Build and verify Python package
+        working-directory: .
+        run: |
+          python -m pip install build packaging pytest twine
+          PYTHONPATH=. python -m pytest tests/unit/test_package_versions.py -q
+          python scripts/package_studio.py
+          python -m build reme_studio --outdir dist/studio
+          python -m twine check dist/studio/*
+          STUDIO_WHEEL="$(pwd)/$(ls dist/studio/reme_studio-*.whl)"
+          python -m venv "${RUNNER_TEMP}/reme-studio-package-smoke"
+          "${RUNNER_TEMP}/reme-studio-package-smoke/bin/python" -m pip install "${STUDIO_WHEEL}"
+          cd "${RUNNER_TEMP}"
+          "${RUNNER_TEMP}/reme-studio-package-smoke/bin/python" - <<'PY'
+          from reme_studio import static_dir
+
+          assert (static_dir() / "index.html").is_file()
+          PY

.github/workflows/ci-typescript.yml

@@ -0,0 +1,51 @@
+name: CI / TypeScript integrations
+
+on:
+  push:
+    branches: [main, master, dev, develop]
+    paths:
+      - '.github/workflows/ci-typescript.yml'
+      - '.github/workflows/release-typescript.yml'
+      - 'typescript/**'
+  pull_request:
+    branches: [main, master, dev, develop]
+    paths:
+      - '.github/workflows/ci-typescript.yml'
+      - '.github/workflows/release-typescript.yml'
+      - 'typescript/**'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  package:
+    name: Type-check, test, and pack
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: typescript
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: '22.22.3'
+          cache: npm
+          cache-dependency-path: typescript/package-lock.json
+
+      - run: npm ci
+      - run: npm run format:check
+      - run: npm run lint
+      - run: npm run typecheck
+      - run: npm test
+      - run: npm run test:package
+      - name: Validate OpenClaw package contract
+        run: npx --yes clawhub@0.23.3 package validate . --json

.github/workflows/ci-windows.yml

@@ -0,0 +1,51 @@
+name: CI / Windows
+
+on:
+  push:
+    branches: [main, master, dev, develop]
+  pull_request:
+    branches: [main, master, dev, develop]
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  cli-smoke:
+    name: CLI smoke - py${{ matrix.python-version }}
+    runs-on: windows-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.11"]
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: 'pip'
+
+      - name: Install package
+        run: |
+          python -m pip install --upgrade pip setuptools wheel
+          pip install -e ".[dev,as]"
+
+      - name: Run version job
+        run: reme start config=tests/fixtures/config/version-smoke.yaml job=version
+
+      - name: Run Windows path tests
+        run: |
+          python -m pytest `
+            tests/unit/test_auto_dream.py::test_scan_day_files_includes_nested_md_and_excludes_interests `
+            tests/unit/test_auto_dream.py::test_dream_extract_matches_posix_catalog_paths `
+            tests/unit/test_read_with_neighbors.py::test_read_with_neighbors_uses_posix_nested_path `
+            -v

.github/workflows/deploy-docs.yml

@@ -0,0 +1,52 @@
+name: Deploy / Documentation
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - "github-pages/**"
+      - "docs/**"
+      - "README.md"
+      - "README_ZH.md"
+      - "reme_studio/README*.md"
+      - "reme_studio/public/og.jpg"
+      - "typescript/README*.md"
+      - "plugins/*/README*.md"
+      - "benchmark/*/README*.md"
+      - "AGENTS.md"
+      - ".github/workflows/deploy-docs.yml"
+      - ".github/workflows/_build-docs.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: pages
+  cancel-in-progress: true
+
+jobs:
+  build:
+    name: Build documentation
+    uses: ./.github/workflows/_build-docs.yml
+    with:
+      run_tests: true
+      upload_pages_artifact: true
+    permissions:
+      contents: read
+      pages: write
+      id-token: write
+
+  deploy:
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    runs-on: ubuntu-latest
+    needs: build
+    permissions:
+      pages: write
+      id-token: write
+    steps:
+      - name: Deploy
+        id: deployment
+        uses: actions/deploy-pages@cd2ce8fcbc39b97be8ca5fce6e763baed58fa128 # v5

.github/workflows/policy-pr-title.yml

@@ -0,0 +1,40 @@
+name: Policy / PR title
+
+on:
+  pull_request:
+    branches: [main, master, dev, develop]
+    types: [opened, edited, synchronize, reopened]
+
+permissions:
+  contents: read
+  pull-requests: read
+
+jobs:
+  check-pr-title:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check PR title format
+        uses: amannn/action-semantic-pull-request@48f256284bd46cdaab1048c3721360e808335d50 # v6.1.1
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        with:
+          types: |
+            feat
+            fix
+            docs
+            ci
+            refactor
+            test
+            chore
+            perf
+            style
+            build
+            revert
+          requireScope: false
+          scopePattern: ^[a-z0-9_-]+$
+          scopePatternError: |
+            The scope must contain only lowercase letters, numbers, hyphens, and underscores.
+            Example: "feat(memory): add redis cache support"
+          validateSingleCommit: false
+          ignoreLabels: |
+            ignore-semantic-pull-request

.github/workflows/pre-commit.yml

@@ -1,38 +0,0 @@
-name: Pre-commit
-
-on: [ push, pull_request ]
-
-jobs:
-  run:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: True
-      matrix:
-        os: [ ubuntu-latest ]
-    env:
-      OS: ${{ matrix.os }}
-      PYTHON: '3.10'
-    steps:
-      - uses: actions/checkout@master
-      - name: Setup Python
-        uses: actions/setup-python@master
-        with:
-          python-version: '3.10'
-      - name: Update setuptools
-        run: |
-          pip install -U setuptools wheel
-      - name: Install
-        run: |
-          pip install -q -e .[dev]
-      - name: Install pre-commit
-        run: |
-          pre-commit install
-      - name: Pre-commit starts
-        run: |
-          pre-commit run --all-files > pre-commit.log 2>&1 || true
-          cat pre-commit.log
-          if grep -q Failed pre-commit.log; then
-            echo -e "\e[41m  [**FAIL**] Please install pre-commit and format your code first. \e[0m"
-            exit 1
-          fi
-          echo -e "\e[46m  ********************************Passed******************************** \e[0m"

.github/workflows/python-publish.yml

@@ -1,40 +0,0 @@
-# This workflow will upload a Python Package using Twine when a release is created
-# For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python#publishing-to-package-registries
-
-# This workflow uses actions that are not certified by GitHub.
-# They are provided by a third-party and are governed by
-# separate terms of service, privacy policy, and support
-# documentation.
-
-name: Publish Python Package to Pypi
-
-on:
-  workflow_dispatch:
-  release:
-    types: [published]
-
-permissions:
-  contents: read
-
-jobs:
-  deploy:
-
-    runs-on: ubuntu-latest
-
-    steps:
-    - uses: actions/checkout@v4
-    - name: Set up Python
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.10'
-    - name: Install dependencies
-      run: |
-        python -m pip install --upgrade pip
-        pip install setuptools wheel build
-    - name: Build package
-      run: python -m build
-    - name: Publish package to PyPI
-      uses: pypa/gh-action-pypi-publish@release/v1
-      with:
-        user: __token__
-        password: ${{ secrets.PYPI_API_TOKEN }}

.github/workflows/release-auto-fin.yml

@@ -0,0 +1,157 @@
+# 发布操作手册：
+# 1. 先将 plugins/auto-fin/pyproject.toml 中的 project.version 更新为待发布版本并合入目标分支。
+# 2. 确认插件依赖的 reme-ai 版本已经发布到 PyPI；本工作流会在构建阶段验证该依赖可下载。
+# 3. 确认 PyPI Trusted Publisher 已绑定本仓库、此工作流和 pypi environment，且 PyPI 上不存在相同版本。
+# 4. 在 GitHub 仓库的 Actions 页面选择“Release / Auto Fin plugin”，点击“Run workflow”。
+# 5. 输入与 project.version 完全一致的版本号（例如 0.1.0）后运行；版本也可以带 v 前缀。
+#
+# 推荐发布顺序：reme-ai -> reme-auto-fin -> QwenPaw 更新依赖并通过 plugins: [auto-fin] 启用。
+# 当前仅支持 workflow_dispatch 手动触发，不会因 push、tag 或 release 自动发布。
+
+name: Release / Auto Fin plugin
+
+run-name: Publish reme-auto-fin ${{ inputs.version }}
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: Version from plugins/auto-fin/pyproject.toml (for example, 0.1.0)
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+concurrency:
+  group: publish-reme-auto-fin
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    env:
+      RELEASE_VERSION: ${{ inputs.version }}
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: '3.11'
+
+      - name: Install test and build dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install build packaging pytest pytest-asyncio twine
+          python -m pip install -e ".[core]"
+          python -m pip install --no-deps -e plugins/auto-fin
+
+      - name: Validate package name and release version
+        id: package
+        run: |
+          python - "${RELEASE_VERSION}" <<'PY'
+          import os
+          import sys
+          import tomllib
+          from pathlib import Path
+
+          from packaging.requirements import Requirement
+          from packaging.version import Version
+
+          project = tomllib.loads(Path("plugins/auto-fin/pyproject.toml").read_text(encoding="utf-8"))["project"]
+          expected = Version(sys.argv[1].removeprefix("v"))
+          actual = Version(project["version"])
+          if project["name"] != "reme-auto-fin":
+              raise SystemExit(f"Expected project name 'reme-auto-fin', found {project['name']!r}")
+          if actual != expected:
+              raise SystemExit(f"Package version is {actual}, but workflow input is {expected}")
+          requirements = [requirement for requirement in project["dependencies"] if requirement.startswith("reme-ai")]
+          if len(requirements) != 1:
+              raise SystemExit(f"Expected one reme-ai dependency, found {requirements!r}")
+          reme_requirement = Requirement(requirements[0])
+          if reme_requirement.name != "reme-ai" or reme_requirement.extras:
+              raise SystemExit(f"Expected a base reme-ai dependency, found {requirements[0]!r}")
+          if Version("0.4.1.8") in reme_requirement.specifier or Version("0.4.1.9") not in reme_requirement.specifier:
+              raise SystemExit(f"Expected reme-ai>=0.4.1.9, found {requirements[0]!r}")
+          with Path(os.environ["GITHUB_OUTPUT"]).open("a", encoding="utf-8") as output:
+              print(f"reme_requirement={reme_requirement}", file=output)
+          print(f"Publishing {project['name']} {actual}")
+          PY
+
+      - name: Run Auto Fin tests
+        run: python -m pytest plugins/auto-fin -q
+
+      - name: Require the plugin-enabled ReMe release on PyPI
+        env:
+          REME_REQUIREMENT: ${{ steps.package.outputs.reme_requirement }}
+        run: |
+          python -m pip download --no-deps \
+            --dest "${RUNNER_TEMP}/reme-auto-fin-base" \
+            "${REME_REQUIREMENT}"
+
+      - name: Build and check distributions
+        run: |
+          mkdir -p dist/auto-fin
+          python -m build plugins/auto-fin --outdir dist/auto-fin
+          python -m twine check dist/auto-fin/*
+
+      - name: Verify distributions and isolated installation
+        run: |
+          AUTO_FIN_WHEEL="$(pwd)/$(ls dist/auto-fin/reme_auto_fin-*.whl)"
+          AUTO_FIN_SDIST="$(pwd)/$(ls dist/auto-fin/reme_auto_fin-*.tar.gz)"
+          python -m zipfile -l "${AUTO_FIN_WHEEL}" | grep 'dist-info/licenses/LICENSE'
+          python -m tarfile -l "${AUTO_FIN_SDIST}" | grep '/LICENSE'
+          python -m venv "${RUNNER_TEMP}/reme-auto-fin-smoke"
+          "${RUNNER_TEMP}/reme-auto-fin-smoke/bin/python" -m pip install \
+            "agentscope[model-ollama]==2.0.7" "${AUTO_FIN_WHEEL}"
+          cd "${RUNNER_TEMP}"
+          "${RUNNER_TEMP}/reme-auto-fin-smoke/bin/python" - <<'PY'
+          from importlib.metadata import distribution
+
+          from reme.plugin_manifest import load_package_manifest
+
+          package = distribution("reme-auto-fin")
+          plugins = {entry.name: entry for entry in package.entry_points if entry.group == "reme.plugins"}
+          assert plugins["auto-fin"].value == "reme_auto_fin"
+          manifest = load_package_manifest("reme_auto_fin", plugin_name="auto-fin")
+          assert set(manifest.backends) == {
+              "auto_fin_data_step",
+              "auto_fin_topic_step",
+              "auto_fin_merge_step",
+          }
+          assert set(manifest.application_defaults["jobs"]) == {
+              "auto_fin",
+              "auto_fin_cron",
+          }
+          PY
+
+      - name: Upload distributions
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: reme-auto-fin-${{ inputs.version }}
+          path: dist/auto-fin/
+          if-no-files-found: error
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+
+    steps:
+      - name: Download distributions
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7
+        with:
+          name: reme-auto-fin-${{ inputs.version }}
+          path: dist/auto-fin
+
+      - name: Publish reme-auto-fin
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+        with:
+          packages-dir: dist/auto-fin

.github/workflows/release-daily-paper.yml

@@ -0,0 +1,157 @@
+# Release checklist:
+# 1. Update project.version in plugins/daily_paper/pyproject.toml and merge it into the target branch.
+# 2. Publish the required reme-ai version before this plugin; the build verifies that dependency on PyPI.
+# 3. Configure PyPI Trusted Publishing for this repository/workflow and its pypi environment.
+# 4. Run "Release / Daily Paper plugin" from GitHub Actions with the exact project version (a v prefix is accepted).
+#
+# Recommended order: reme-ai -> reme-daily-paper -> downstream applications enabling plugins: [daily-paper].
+# This workflow is intentionally manual and never publishes from a push, tag, or GitHub release event.
+
+name: Release / Daily Paper plugin
+
+run-name: Publish reme-daily-paper ${{ inputs.version }}
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: Version from plugins/daily_paper/pyproject.toml (for example, 0.1.0)
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+concurrency:
+  group: publish-reme-daily-paper
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    env:
+      RELEASE_VERSION: ${{ inputs.version }}
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: '3.11'
+
+      - name: Install test and build dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install build packaging pytest pytest-asyncio twine
+          python -m pip install -e ".[core]"
+          python -m pip install -e plugins/daily_paper
+
+      - name: Validate package name, dependencies, and release version
+        id: package
+        run: |
+          python - "${RELEASE_VERSION}" <<'PY'
+          import os
+          import sys
+          import tomllib
+          from pathlib import Path
+
+          from packaging.requirements import Requirement
+          from packaging.version import Version
+
+          project = tomllib.loads(Path("plugins/daily_paper/pyproject.toml").read_text(encoding="utf-8"))["project"]
+          expected = Version(sys.argv[1].removeprefix("v"))
+          actual = Version(project["version"])
+          if project["name"] != "reme-daily-paper":
+              raise SystemExit(f"Expected project name 'reme-daily-paper', found {project['name']!r}")
+          if actual != expected:
+              raise SystemExit(f"Package version is {actual}, but workflow input is {expected}")
+          requirements = [Requirement(value) for value in project["dependencies"]]
+          reme_requirements = [requirement for requirement in requirements if requirement.name == "reme-ai"]
+          if len(reme_requirements) != 1 or reme_requirements[0].extras:
+              raise SystemExit(f"Expected one base reme-ai dependency, found {reme_requirements!r}")
+          if Version("0.4.1.8") in reme_requirements[0].specifier or Version("0.4.1.9") not in reme_requirements[0].specifier:
+              raise SystemExit(f"Expected reme-ai>=0.4.1.9, found {reme_requirements!r}")
+          if sum(requirement.name == "pypdf" for requirement in requirements) != 1:
+              raise SystemExit("Expected exactly one pypdf dependency")
+          with Path(os.environ["GITHUB_OUTPUT"]).open("a", encoding="utf-8") as output:
+              print(f"reme_requirement={reme_requirements[0]}", file=output)
+          print(f"Publishing {project['name']} {actual}")
+          PY
+
+      - name: Run Daily Paper tests
+        run: python -m pytest plugins/daily_paper -q
+
+      - name: Require the plugin-enabled ReMe release on PyPI
+        env:
+          REME_REQUIREMENT: ${{ steps.package.outputs.reme_requirement }}
+        run: |
+          python -m pip download --no-deps \
+            --dest "${RUNNER_TEMP}/reme-daily-paper-base" \
+            "${REME_REQUIREMENT}"
+
+      - name: Build and check distributions
+        run: |
+          mkdir -p dist/daily-paper
+          python -m build plugins/daily_paper --outdir dist/daily-paper
+          python -m twine check dist/daily-paper/*
+
+      - name: Verify distributions and isolated installation
+        run: |
+          DAILY_PAPER_WHEEL="$(pwd)/$(ls dist/daily-paper/reme_daily_paper-*.whl)"
+          DAILY_PAPER_SDIST="$(pwd)/$(ls dist/daily-paper/reme_daily_paper-*.tar.gz)"
+          python -m zipfile -l "${DAILY_PAPER_WHEEL}" | grep 'reme_daily_paper/plugin.yaml'
+          python -m zipfile -l "${DAILY_PAPER_WHEEL}" | grep 'reme_daily_paper/analyze.yaml'
+          python -m zipfile -l "${DAILY_PAPER_WHEEL}" | grep 'dist-info/licenses/LICENSE'
+          python -m tarfile -l "${DAILY_PAPER_SDIST}" | grep '/LICENSE'
+          python -m venv "${RUNNER_TEMP}/reme-daily-paper-smoke"
+          "${RUNNER_TEMP}/reme-daily-paper-smoke/bin/python" -m pip install \
+            "agentscope[model-ollama]==2.0.7" "${DAILY_PAPER_WHEEL}"
+          cd "${RUNNER_TEMP}"
+          "${RUNNER_TEMP}/reme-daily-paper-smoke/bin/python" - <<'PY'
+          from importlib.metadata import distribution
+
+          from reme.plugin_manifest import load_package_manifest
+
+          package = distribution("reme-daily-paper")
+          plugins = {entry.name: entry for entry in package.entry_points if entry.group == "reme.plugins"}
+          assert plugins["daily-paper"].value == "reme_daily_paper"
+          manifest = load_package_manifest("reme_daily_paper", plugin_name="daily-paper")
+          assert set(manifest.backends) == {
+              "daily_paper_collect_step",
+              "daily_paper_rank_step",
+              "daily_paper_select_step",
+              "daily_paper_analyze_step",
+              "daily_paper_digest_step",
+          }
+          assert set(manifest.application_defaults["jobs"]) == {"daily_paper", "daily_paper_cron"}
+          PY
+
+      - name: Upload distributions
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: reme-daily-paper-${{ inputs.version }}
+          path: dist/daily-paper/
+          if-no-files-found: error
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+
+    steps:
+      - name: Download distributions
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7
+        with:
+          name: reme-daily-paper-${{ inputs.version }}
+          path: dist/daily-paper
+
+      - name: Publish reme-daily-paper
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+        with:
+          packages-dir: dist/daily-paper

.github/workflows/release-python.yml

@@ -0,0 +1,47 @@
+name: Release / Python packages
+
+# Configure a PyPI Trusted Publisher for this repository, workflow, and its
+# pypi environment before running the manual release.
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: Release version
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+concurrency:
+  group: publish-reme-ai
+  cancel-in-progress: false
+
+jobs:
+  build:
+    name: Build and verify distributions
+    uses: ./.github/workflows/_build-python-packages.yml
+    with:
+      expected_version: ${{ inputs.version }}
+      upload_artifacts: true
+
+  publish-reme:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - name: Download ReMe distributions
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7
+        with:
+          name: reme-distributions
+          path: dist/reme
+
+      - name: Publish ReMe
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+        with:
+          packages-dir: dist/reme
+          skip-existing: true

.github/workflows/release-reme-studio.yml

@@ -0,0 +1,158 @@
+# Release checklist:
+# 1. Update reme_studio/pyproject.toml, package.json, and package-lock.json to the same Studio version.
+# 2. Configure npm Trusted Publishing and PyPI Trusted Publishing with the pypi environment.
+# 3. Run this workflow manually with the exact Studio version.
+
+name: Release / ReMe Studio
+
+run-name: Publish ReMe Studio ${{ inputs.version }} (${{ inputs.npm_tag }})
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: Version from the Studio Python and npm manifests
+        required: true
+        type: string
+      npm_tag:
+        description: npm distribution tag
+        required: true
+        default: latest
+        type: choice
+        options:
+          - next
+          - latest
+
+permissions:
+  contents: read
+
+concurrency:
+  group: publish-reme-studio
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    env:
+      RELEASE_VERSION: ${{ inputs.version }}
+      NPM_TAG: ${{ inputs.npm_tag }}
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: "22.22.3"
+          cache: npm
+          cache-dependency-path: reme_studio/package-lock.json
+
+      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: "3.11"
+
+      - name: Validate Studio package names and version
+        run: |
+          python - <<'PY'
+          import json
+          import os
+          import tomllib
+          from pathlib import Path
+
+          studio = Path("reme_studio")
+          python_manifest = tomllib.loads((studio / "pyproject.toml").read_text(encoding="utf-8"))["project"]
+          npm_manifest = json.loads((studio / "package.json").read_text(encoding="utf-8"))
+          expected = os.environ["RELEASE_VERSION"].removeprefix("v")
+          if python_manifest["name"] != "reme_studio":
+              raise SystemExit(f"Unexpected Python package name: {python_manifest['name']}")
+          if npm_manifest["name"] != "@agentscope-ai/reme_studio":
+              raise SystemExit(f"Unexpected npm package name: {npm_manifest['name']}")
+          if python_manifest["version"] != expected or npm_manifest["version"] != expected:
+              raise SystemExit(
+                  f"Studio manifests are {python_manifest['version']} and {npm_manifest['version']}; "
+                  f"workflow input is {expected}",
+              )
+          prerelease = "-" in expected
+          if prerelease != (os.environ["NPM_TAG"] == "next"):
+              raise SystemExit("Prereleases must use next; stable releases must use latest")
+          PY
+
+      - name: Install dependencies and run checks
+        working-directory: reme_studio
+        run: |
+          npm ci
+          npm run format:check
+          npm run lint
+          npm test
+
+      - name: Build Studio distributions
+        run: |
+          python -m pip install build twine
+          mkdir -p dist/studio-python dist/studio-npm
+          npm pack ./reme_studio --pack-destination dist/studio-npm
+          python scripts/package_studio.py
+          python -m build reme_studio --outdir dist/studio-python
+          python -m twine check dist/studio-python/*
+
+      - name: Verify Studio distributions and isolated installation
+        run: |
+          STUDIO_WHEEL="$(pwd)/$(ls dist/studio-python/reme_studio-*.whl)"
+          tar -tzf dist/studio-npm/*.tgz | grep '^package/dist-static/index.html$'
+          python -m venv "${RUNNER_TEMP}/reme-studio-package-smoke"
+          "${RUNNER_TEMP}/reme-studio-package-smoke/bin/python" -m pip install "${STUDIO_WHEEL}"
+          cd "${RUNNER_TEMP}"
+          "${RUNNER_TEMP}/reme-studio-package-smoke/bin/python" - <<'PY'
+          from reme_studio import static_dir
+
+          assert (static_dir() / "index.html").is_file()
+          PY
+
+      - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: reme-studio-${{ inputs.version }}
+          path: |
+            dist/studio-python/*
+            dist/studio-npm/*
+          if-no-files-found: error
+
+  publish-python:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7
+        with:
+          name: reme-studio-${{ inputs.version }}
+          path: dist
+
+      - name: Publish ReMe Studio to PyPI
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+        with:
+          packages-dir: dist/studio-python
+          skip-existing: true
+
+  publish-npm:
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: "24"
+          registry-url: https://registry.npmjs.org
+
+      - uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7
+        with:
+          name: reme-studio-${{ inputs.version }}
+          path: dist
+
+      - name: Publish ReMe Studio to npm
+        env:
+          NPM_TAG: ${{ inputs.npm_tag }}
+        run: npm publish dist/studio-npm/*.tgz --access public --tag "${NPM_TAG}" --provenance

.github/workflows/release-typescript.yml

@@ -0,0 +1,167 @@
+# Release checklist:
+# 1. Update typescript/package.json and package-lock.json to the release version and merge them.
+# 2. Configure npm Trusted Publishing for agentscope-ai/ReMe and this workflow file.
+# 3. Run this workflow manually with the exact package version (an optional v prefix is accepted).
+# 4. Configure ClawHub Trusted Publishing or CLAWHUB_TOKEN before enabling ClawHub publication.
+# 5. Use the `next` tag for prereleases and `latest` only for stable releases.
+
+name: Release / TypeScript integrations
+
+run-name: Publish @agentscope-ai/reme ${{ inputs.version }} (${{ inputs.npm_tag }})
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: Version from typescript/package.json (for example, 0.1.0)
+        required: true
+        type: string
+      npm_tag:
+        description: npm distribution tag
+        required: true
+        default: latest
+        type: choice
+        options:
+          - next
+          - latest
+      publish_clawhub:
+        description: Also publish the verified tarball to ClawHub
+        required: true
+        default: false
+        type: boolean
+
+permissions:
+  contents: read
+
+concurrency:
+  group: publish-agentscope-ai-reme
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    outputs:
+      version: ${{ steps.validate.outputs.version }}
+    env:
+      RELEASE_VERSION: ${{ inputs.version }}
+      NPM_TAG: ${{ inputs.npm_tag }}
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Set up Node
+        uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: '22.22.3'
+
+      - name: Validate package name and release version
+        id: validate
+        working-directory: typescript
+        run: |
+          node --input-type=module <<'JS'
+          import { appendFileSync, readFileSync } from 'node:fs';
+
+          const manifest = JSON.parse(readFileSync('package.json', 'utf8'));
+          const expected = process.env.RELEASE_VERSION.replace(/^v/, '');
+          if (manifest.name !== '@agentscope-ai/reme') {
+            throw new Error(`Unexpected package name: ${manifest.name}`);
+          }
+          if (manifest.version !== expected) {
+            throw new Error(`package.json is ${manifest.version}, workflow input is ${expected}`);
+          }
+          const prerelease = manifest.version.includes('-');
+          const npmTag = process.env.NPM_TAG;
+          if (prerelease !== (npmTag === 'next')) {
+            throw new Error(prerelease
+              ? 'Prerelease versions must use the next npm tag'
+              : 'Stable versions must use the latest npm tag');
+          }
+          console.log(`Preparing ${manifest.name}@${manifest.version}`);
+          appendFileSync(process.env.GITHUB_OUTPUT, `version=${manifest.version}\n`);
+          JS
+
+      - name: Install dependencies
+        working-directory: typescript
+        run: npm ci
+
+      - name: Type-check and test
+        working-directory: typescript
+        run: |
+          npm run format:check
+          npm run lint
+          npm run typecheck
+          npm test
+          npm run test:package
+          npx --yes clawhub@0.23.3 package validate . --json
+
+      - name: Pack npm tarball
+        working-directory: typescript
+        run: |
+          mkdir -p "${RUNNER_TEMP}/reme-typescript-package"
+          npm pack --pack-destination "${RUNNER_TEMP}/reme-typescript-package"
+
+      - name: Upload npm tarball
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: agentscope-ai-reme-${{ inputs.version }}
+          path: ${{ runner.temp }}/reme-typescript-package/*.tgz
+          if-no-files-found: error
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write
+
+    steps:
+      - name: Set up Node for npm
+        uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+        with:
+          node-version: '24'
+          registry-url: https://registry.npmjs.org
+
+      - name: Download npm tarball
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7
+        with:
+          name: agentscope-ai-reme-${{ inputs.version }}
+          path: dist/typescript
+
+      - name: Reject an existing package version
+        env:
+          PACKAGE_VERSION: ${{ inputs.version }}
+        run: |
+          PACKAGE_VERSION="${PACKAGE_VERSION#v}"
+          if npm view "@agentscope-ai/reme@${PACKAGE_VERSION}" version >/dev/null 2>&1; then
+            echo "@agentscope-ai/reme@${PACKAGE_VERSION} already exists" >&2
+            exit 1
+          fi
+
+      - name: Publish to npm
+        env:
+          NPM_TAG: ${{ inputs.npm_tag }}
+        run: npm publish dist/typescript/*.tgz --access public --tag "${NPM_TAG}" --provenance
+
+  publish-clawhub:
+    if: ${{ inputs.publish_clawhub }}
+    needs: build
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+    uses: openclaw/clawhub/.github/workflows/package-publish.yml@87ca030c30f3cfb78ab15c8e66b5ff1469c8f9c8 # v0.23.3
+    with:
+      owner: agentscope-ai
+      family: code-plugin
+      version: ${{ needs.build.outputs.version }}
+      tags: ${{ inputs.npm_tag }}
+      source_repo: ${{ github.repository }}
+      source_commit: ${{ github.sha }}
+      source_ref: ${{ github.ref }}
+      source_path: typescript
+      package_artifact_name: agentscope-ai-reme-${{ inputs.version }}
+      wait_for_publication: true
+    secrets:
+      clawhub_token: ${{ secrets.CLAWHUB_TOKEN }}

.github/workflows/security-codeql.yml

@@ -0,0 +1,46 @@
+name: Security / CodeQL
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+  schedule:
+    - cron: '0 1 * * 1'
+  workflow_dispatch:
+
+permissions:
+  actions: read
+  contents: read
+  packages: read
+  security-events: write
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  analyze:
+    name: Analyze ${{ matrix.language }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        language: [python, javascript-typescript]
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+
+      - name: Initialize CodeQL
+        uses: github/codeql-action/init@cdf488f595d80d6e07e03d4674febd5ab45fa938 # v4
+        with:
+          languages: ${{ matrix.language }}
+          build-mode: none
+
+      - name: Perform CodeQL analysis
+        uses: github/codeql-action/analyze@cdf488f595d80d6e07e03d4674febd5ab45fa938 # v4
+        with:
+          category: /language:${{ matrix.language }}

.gitignore

@@ -1,45 +1,82 @@
-.vscode
-.env*
+# OS / editor
 .DS_Store
-.idea
-venv/
-.ipynb_checkpoints
-.__pycache__
-__pycache__
-*.log
-tmp*
-temp*
-private*
-dist/
-nohup*
-cache
-log/
-.trash/
-runs
-logs
-rag_nodes_index.jsonl
-alfworld_data
-step_experiences/*
-build/*
-*.egg-info/*
-cookbook/appworld/data/*
-cookbook/appworld/experiments/*
-cookbook/appworld/exp_result/*
-file_vector_store/*
-cookbook/appworld/file_vector_store/*
-/.venv/
-site/*
-docs/_build/*
-test_compact_storage/*
-test_working_memory/*
+.idea/
+.vscode/
+.qoder/
 *.code-workspace
-local_vector_store/*
-reme_profile/*
-chroma_vector_store/*
-bench_results/*
-meta_memory/*
-*.sqlite3
-**/data/*.json
+
+# Local environment
+.env
+.env.*
+!.env.example
+!example.env
+.venv/
+venv/
+env/
+private*/
+
+# Python caches / test artifacts
+__pycache__/
+*.py[cod]
+*$py.class
+.ipynb_checkpoints/
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+coverage.xml
+htmlcov/
+
+# Packaging / build outputs
+build/
+dist/
+node_modules/
+*.egg-info/
+typescript/reports/
+
+# Logs / temporary files
+*.log
+nohup.out
+nohup*.out
+log/
+logs/
+runs/
+tmp*/
+temp*/
+.trash/
+
+# ReMe runtime data
+.reme/
+reme_workspace/
+reme_workspace_auto_fin_real_test*/
+vault/
 *.db
-memories/*
-.reme/*
+*.sqlite
+*.sqlite3
+
+# Documentation build outputs
+docs/_build/
+site/
+
+evaluation/
+# The pi-Bench suite ships its own trace-history render config, which must
+# stay in git even though it lives under an evaluation/ directory.
+!benchmark/pibench/config/bench/evaluation/
+!benchmark/pibench/config/bench/evaluation/**
+datasets/
+
+# Claude Code skills (local only)
+.claude/skills/
+
+# Benchmark memory workspaces (created on demand by run.py via mkdir)
+benchmark/*/workspaces/
+
+# Benchmark datasets (LongMemEval via download.py, BEAM via git clone)
+benchmark/*/dataset/
+
+# Benchmark outputs (created on demand by run.py via mkdir)
+benchmark/*/results/
+
+# integration tests outputs
+tests/integration/logs/
+daily/

.pre-commit-config.yaml

@@ -1,9 +1,10 @@
+exclude: ^skills/
+
 repos:
   - repo: https://github.com/pre-commit/pre-commit-hooks
     rev: v6.0.0
     hooks:
       - id: check-ast
-        exclude: ^(test/|cookbook/|reme_ai/)
       - id: check-yaml
       - id: check-xml
       - id: check-toml
@@ -14,42 +15,37 @@ repos:
     rev: v4.0.0
     hooks:
       - id: add-trailing-comma
-        exclude: ^(test/|cookbook/|reme_ai/)
   - repo: https://github.com/psf/black
-    rev: 25.9.0
+    rev: 26.5.1
     hooks:
     - id: black
-      exclude: ^(test/|cookbook/|reme_ai/)
-      args: [--line-length=120]
+      args: [--line-length=120, --target-version=py311]
   - repo: https://github.com/PyCQA/flake8
     rev: 7.3.0
     hooks:
       - id: flake8
-        exclude: ^(test/|cookbook/|reme_ai/)
         args: [
           "--extend-ignore=E203",
           "--max-line-length=120"
         ]
   - repo: https://github.com/pylint-dev/pylint
-    rev: v4.0.2
+    rev: v4.0.6
     hooks:
       - id: pylint
         exclude:
             (?x)(
                 ^docs
-                | ^test/
-                | ^cookbook/
                 | pb2\.py$
                 | grpc\.py$
                 | \.demo$
                 | \.md$
                 | \.html$
-                | reme_ai/
           )
         args: [
           --disable=W0511,
           --disable=W0718,
           --disable=W0122,
+          --disable=W1203,
           --disable=C0103,
           --disable=R0913,
           --disable=R0917,
@@ -82,7 +78,7 @@ repos:
           --max-module-lines=1500,
         ]
   - repo: https://github.com/regebro/pyroma
-    rev: "5.0"
+    rev: "5.0.1"
     hooks:
       - id: pyroma
         args: [--min=10, .]

AGENTS.md

@@ -0,0 +1,214 @@
+# AGENTS.md
+
+This file guides coding agents working in the ReMe repository. Keep changes small, testable, and consistent with the
+contracts expressed by the current code.
+
+## Project Principles
+
+ReMe is a local-first, file-native memory system for agents.
+
+- User-owned workspace files are the durable source of truth.
+- Indexes, catalogs, graphs, caches, and generated metadata must remain rebuildable.
+- Prefer transparent formats and predictable behavior over hidden state.
+- Preserve user control over workspace paths, configuration, and service boundaries.
+- Keep concepts focused on project intent; let code and schemas describe implementation.
+
+When convenience conflicts with these principles, favor data ownership, recoverability, and explicit behavior.
+
+## Sources of Truth
+
+Use this order when documentation and implementation disagree:
+
+1. Current code and public Pydantic schemas.
+2. Tests that describe supported behavior.
+3. CLI behavior and the built-in configuration.
+4. README files and other development documentation.
+
+Do not duplicate large implementation descriptions in documentation. Express the stable contract and link to the
+relevant module where useful. When behavior changes intentionally, update the implementation, schemas, tests, defaults,
+and concise documentation together.
+
+## Repository Map
+
+- `reme/reme.py`: CLI entry point; dispatches `start`, `find_reme`, and client calls.
+- `reme/application.py`: application assembly, dependency ordering, job execution, and lifecycle.
+- `reme/config/config_parser.py`: YAML/JSON loading, environment expansion, dot-notation parsing, and deep config
+  merging.
+- `reme/config/default.yaml`: default service, jobs, steps, and components. Other files in
+  `reme/config/` are named configuration variants.
+- `reme/schema/application_config.py`: typed application, component, and job configuration.
+- `reme/schema/`: request, response, streaming, memory, graph, and file contracts.
+- `reme/components/application_context.py`: application-wide wiring and in-memory shared state.
+- `reme/components/runtime_context.py`: request-scoped data, response, streaming queue, and stop event.
+- `reme/components/base_component.py`: component lifecycle, dependency binding, and workspace helpers.
+- `reme/components/component_registry.py`: the frozen built-in registry template and application-local registry factory.
+- `reme/components/job/`: base, stream, background, and cron job implementations.
+- `reme/components/service/`: local CLI, HTTP, and MCP service backends.
+- `reme/components/`: agent wrappers, model adapters, stores, catalogs, graphs, indexes, clients, tokenizers, and
+  outbound proxies.
+- `reme/steps/`: registered job steps grouped by common, file I/O, index, evolve, cookbook, benchmark, and transfer
+  concerns.
+- `reme/utils/`: shared utilities, including service discovery, logging, web-static resolution, session I/O, token
+  accounting, and wikilink handling.
+- `tests/unit/`: primary fast, isolated validation suite.
+- `tests/integration/`: service/model tests that may need credentials or external processes.
+- `reme_studio/`: ReMe Studio frontend source plus the independently published `reme_studio` Python package and
+  `@agentscope-ai/reme_studio` npm static distribution.
+- `typescript/`: the independently published `@agentscope-ai/reme` package, including the shared TypeScript client and
+  DeepSeek Harness and OpenClaw adapters.
+- `plugins/`: installable ReMe extensions, such as Auto Fin.
+- `integrations/`: adapters that connect ReMe to external agent hosts, such as Claude Code, DSH, and Hermes Agent.
+- `skills/`: standalone skills; `reme_memory` calls ReMe, while other skills may use separate tools or direct-file
+  conventions.
+- `benchmark/` and `cookbook/`: runnable evaluations and example workflows.
+- `docs/`: README-linked supporting pages and figures.
+
+## Development Setup
+
+ReMe requires Python 3.11 or newer. Install the editable development environment with:
+
+```bash
+pip install -e reme_studio -e ".[dev,core]"
+```
+
+Before changing behavior, inspect the adjacent implementation, schema, built-in config, and focused tests. Follow
+existing async and typing patterns unless the task explicitly requires a new contract.
+
+## Configuration and CLI Contracts
+
+- CLI syntax is `reme ACTION key=value ...`; leading `-` or `--` on arguments is accepted.
+- Nested overrides use dot notation. Values support null, booleans, numbers, JSON collections, and quoted JSON strings;
+  leading-zero numeric-looking values remain strings.
+- `config=<name-or-path>` loads a discovered config name or a `.yaml`, `.yml`, or `.json` file. With no explicit config
+  path, `default` is loaded when available.
+- Config files expand `${VAR}` and `${VAR:-default}` recursively. An undefined variable without a default is an error.
+- CLI/config overrides are deep-merged over the loaded file. Do not silently change this merge behavior or stable
+  configuration keys.
+- `ApplicationConfig` normalizes `workspace_dir` to an expanded absolute path. `session_dir`
+  must remain workspace-relative; standard transcripts live under `{session_dir}/dialog`.
+- `reme start` runs the configured service. `reme start job=<name> ...` switches to the one-shot CLI service and runs
+  the job through the normal application lifecycle.
+- Other actions use a client selected from the running service configuration when discoverable, otherwise from local
+  config. Client-selection arguments must not leak into the job payload.
+
+## Registration and Application Lifecycle
+
+Component and Step discovery is import-driven:
+
+- Implementations declare a non-`BASE` `component_type` and register with `@R.register("backend")`
+  or `R.register(Class, "backend")`.
+- Component packages must be imported through `reme/components/__init__.py`.
+- Step packages/modules must be reachable through their package `__init__.py` chain and ultimately
+  `reme/steps/__init__.py`.
+- Adding an implementation without its registration import leaves it undiscoverable at runtime. Treat implementation,
+  registration, import side effect, defaults, and tests as one change.
+
+`Application` validates config through `ApplicationContext`, creates workspace directories, instantiates the service,
+configured components, and jobs, and then manages lifecycle as follows:
+
+- Components start in topological dependency order. Missing required dependencies and cycles fail explicitly; optional
+  dependencies may resolve to `None`.
+- Jobs start after components in this order: base jobs, stream jobs, background jobs, then cron jobs.
+- Shutdown closes everything in reverse start order and then shuts down the optional thread pool.
+- If startup fails, already-started resources are closed.
+- `BaseComponent.start()` and `close()` are lock-protected and idempotent. Dependencies created by a standalone
+  `default_factory` are owned and closed by the parent component.
+
+Keep async clients, tasks, executors, and services under this lifecycle. Do not introduce an untracked long-lived
+resource.
+
+## Jobs, Steps, and State
+
+`BaseJob` resolves configured Step classes during job startup and constructs fresh Step instances for every invocation.
+Job-level kwargs are merged into each `RuntimeContext`, with call-time kwargs taking precedence. Sequential Steps in one
+invocation share the same `RuntimeContext` and `Response`.
+
+Treat Step instances as invocation-scoped:
+
+- Constructor fields and `self.kwargs` hold Step configuration and resolved dependencies. They may be cached or adjusted
+  during that one invocation, but must not be relied on across Job calls.
+- `self.context.data` holds request inputs and intermediate values shared by sequential Steps.
+- `self.context.response.answer`, `success`, and `metadata` are request-scoped output. Because the same response travels
+  through the Step chain, later Steps may consume metadata produced earlier, but it is not application-lifetime or
+  durable storage.
+- `self.app_context.metadata` holds in-memory state shared across Job/Step invocations for the life of one
+  `Application`, such as counters, tool-context state, session maps, or locks.
+- Workspace files or a dedicated Component/store hold durable state that must survive restart.
+
+Use narrow, namespaced keys in `app_context.metadata` and protect shared mutable values against concurrent access. The
+search/draft helpers intentionally mirror tool-context state into
+`self.kwargs` only when no `ApplicationContext` exists for standalone use and unit tests; do not generalize that
+compatibility fallback into persistent runtime state. If shared state becomes a stable service contract or needs
+dedicated lifecycle, locking, or persistence, promote it to a typed context field or Component.
+
+Additional Step contracts:
+
+- `Ref` dependencies resolve in this order: Step kwargs, current `RuntimeContext`, then the named application component.
+  The value is cached only on the current Step instance and cleared before each call.
+- `input_mapping` and `output_mapping` copy keys within `RuntimeContext.data`; missing sources are ignored.
+- Dispatched Steps receive the current `RuntimeContext`, so their data and response are shared.
+- Base jobs convert uncaught Step errors into `Response(success=False)`; stream jobs emit an error chunk and always a
+  terminal `DONE`; background jobs let errors reach their supervisor.
+- Background jobs are never service-exposed. MCP also skips stream jobs. Respect `enable_serve`
+  and any configured service job allowlist.
+
+## Workspace and File Safety
+
+- Application startup creates the workspace plus configured metadata, session, memory-session, resource, daily, and
+  digest directories.
+- File-operation paths are resolved against the workspace and must stay inside it. Home-relative paths are unsupported,
+  traversal escapes are rejected, and `_allowed_paths` restrictions fail closed when invalid.
+- Preserve per-path locking, encoding detection, byte limits, truncation behavior, and optimistic
+  `expected_mtime` checks when modifying file operations.
+- Do not bypass the existing file steps or stores in a way that weakens workspace containment.
+- Never write test state into the repository's `.reme/`; use `tmp_path` or another isolated workspace.
+- Do not delete or rewrite user memory to repair an index or make a test pass. Rebuild derived state from source files
+  instead.
+
+## Validation
+
+Use the narrowest useful check while iterating, then broaden it according to risk.
+
+Focused test:
+
+```bash
+pytest tests/unit/path/to/test_file.py -v
+```
+
+Main unit suite:
+
+```bash
+pytest tests/unit -v --tb=long -s --log-cli-level=WARNING
+```
+
+Repository formatting and lint checks:
+
+```bash
+pre-commit run --all-files
+```
+
+Black and Flake8 use a 120-character line limit and Python 3.11 formatting; Pylint is also run by pre-commit. If
+`reme_studio/` changes, use its Node 22.13+ scripts and run the proportionate checks from that directory, such as
+`npm run format:check`, `npm run lint`, or `npm test`.
+
+Integration tests may contact real model providers, services, or agent subprocesses and can require credentials. Do not
+run credentialed or externally mutating tests automatically; run them only when the task requires them and the necessary
+environment has been supplied or authorized. Mock network, model, and subprocess boundaries in unit tests.
+
+## Change Guardrails
+
+- Preserve unrelated user changes in a dirty working tree.
+- Make the smallest coherent change and avoid unrelated cleanup or broad refactors.
+- Do not edit generated output when the source can be changed instead. The publish workflow builds
+  `reme_studio/dist-static` and stages it under `reme_studio/src/reme_studio/static`; change `reme_studio/` source for
+  frontend work.
+- Do not silently change CLI flags, configuration keys, workspace layouts, serialized schemas, endpoint shapes,
+  streaming termination, or service interfaces. Preserve compatibility where practical and document intentional
+  migrations.
+- Do not introduce dependencies without a concrete repository-level need.
+- Do not commit `.env` files, credentials, runtime memory, logs, indexes, caches, benchmark outputs, or generated
+  Studio distributions.
+- State which validations passed and which relevant checks were not run in the final handoff.
+
+If a requirement is ambiguous, infer intent from nearby code, schemas, defaults, and tests. Ask the user only when the
+remaining choice would materially alter a public contract, user data, or an external system.

CLAUDE.md

@@ -0,0 +1 @@
+AGENTS.md

benchmark/appworld/appworld_react_agent.py

@@ -1,360 +0,0 @@
-# flake8: noqa: E402, E501
-# pylint: disable=E0611
-"""A minimal ReAct Agent for AppWorld tasks."""
-import os
-import re
-import time
-import json
-import datetime
-from typing import List, Any
-
-
-import ray
-import requests
-from tqdm import tqdm
-from loguru import logger
-from openai import OpenAI
-from jinja2 import Template
-from dotenv import load_dotenv
-
-from prompt import NEW_PROMPT_TEMPLATE
-from appworld import AppWorld, load_task_ids
-
-os.environ["APPWORLD_ROOT"] = "."
-
-load_dotenv("../../.env")
-
-
-@ray.remote
-class AppworldReactAgent:
-    """A minimal ReAct Agent for AppWorld tasks."""
-
-    def __init__(
-        self,
-        index: int,
-        task_ids: List[str],
-        experiment_name: str,
-        model_name: str = "qwen3-8b",
-        temperature: float = 0.9,
-        max_interactions: int = 30,
-        max_response_size: int = 129024,
-        num_trials: int = 1,
-        use_memory: bool = False,
-        memory_base_url: str = "http://0.0.0.0:8002/",
-        use_memory_addition: bool = False,
-        use_memory_deletion: bool = False,
-        delete_freq: int = 10,
-        freq_threshold: int = 5,
-        utility_threshold: float = 0.5,
-    ):
-
-        self.index: int = index
-        self.task_ids: List[str] = task_ids
-        self.experiment_name: str = experiment_name
-        self.model_name: str = model_name
-        self.temperature: float = temperature
-        self.max_interactions: int = max_interactions
-        self.max_response_size: int = max_response_size
-        self.num_trials: int = num_trials
-        self.use_memory: bool = use_memory
-        self.use_memory_addition: bool = use_memory_addition if use_memory else False
-        self.use_memory_deletion: bool = use_memory_deletion if use_memory else False
-        self.delete_freq: int = delete_freq
-        self.freq_threshold: int = freq_threshold
-        self.utility_threshold: float = utility_threshold
-
-        self.llm_client = OpenAI()
-        self.memory_base_url: str = memory_base_url
-
-        self.history: List[List[List[dict]]] = [[] for _ in range(num_trials)]
-        self.retrieved_memory_list: List[List[List[Any]]] = [[] for _ in range(num_trials)]
-
-        for run_id in range(num_trials):
-            for _ in range(len(task_ids)):
-                self.retrieved_memory_list[run_id].append([])
-                self.history[run_id].append([])
-
-    def call_llm(self, messages: list) -> str:
-        """Call the LLM to generate a response to the messages."""
-        for i in range(100):
-            try:
-                response = self.llm_client.chat.completions.create(
-                    model=self.model_name,
-                    messages=messages,
-                    temperature=self.temperature,
-                    extra_body={"enable_thinking": False},
-                    seed=0,
-                )
-
-                return response.choices[0].message.content
-
-            except Exception as e:
-                logger.exception(f"encounter error with {e.args}")
-                time.sleep(1 + i * 10)
-
-        return "call llm error"
-
-    def prompt_messages(self, run_id, task_index, previous_memories: None, world: AppWorld):
-        """Prompt the messages to the LLM."""
-        app_descriptions = json.dumps(
-            [{"name": k, "description": v} for (k, v) in world.task.app_descriptions.items()],
-            indent=1,
-        )
-        dictionary = {"supervisor": world.task.supervisor, "app_descriptions": app_descriptions}
-        sys_prompt = Template(NEW_PROMPT_TEMPLATE.lstrip()).render(dictionary)
-        query = world.task.instruction
-        if self.use_memory:
-            if len(previous_memories) == 0:
-                response = self.get_memory(world.task.instruction)
-                if response and "memory_list" in response["metadata"]:
-                    self.retrieved_memory_list[run_id][task_index] = response["metadata"]["memory_list"]
-                    task_memory = re.sub(r"\bMemory\s*(\d+)\s*[:]", r"Experience \1:", response["answer"])
-                    logger.info(f"loaded task_memory: {task_memory}")
-                    query = (
-                        "Task:\n"
-                        + query
-                        + "\n\nSome Related Experience to help you to complete the task:\n"
-                        + task_memory
-                    )
-            else:
-                formatted_memories = []
-                for i, memory in enumerate(previous_memories, 1):
-                    condition = memory["when_to_use"]
-                    memory_content = memory["content"]
-                    memory_text = f"Experience {i}:\n When to use: {condition}\n Content: {memory_content}\n"
-                    formatted_memories.append(memory_text)
-                query = (
-                    "Task:\n"
-                    + query
-                    + "\n\nSome Related Experience to help you to complete the task:\n"
-                    + "\n".join(formatted_memories)
-                )
-        messages = [
-            {"role": "system", "content": sys_prompt},
-            {"role": "user", "content": query},
-        ]
-        self.history[run_id][task_index] = messages
-
-    @staticmethod
-    def get_reward(world) -> float:
-        """Get the reward for the Appworld world."""
-        tracker = world.evaluate()
-        num_passes = len(tracker.passes)
-        num_failures = len(tracker.failures)
-        return num_passes / (num_passes + num_failures)
-
-    def extract_code_and_fix_content(
-        self,
-        text: str,
-        ignore_multiple_calls=True,
-    ) -> tuple[str, str]:
-        """Extract the code and fix the content."""
-        full_code_regex = r"```python\n(.*?)```"
-        partial_code_regex = r".*```python\n(.*)"
-
-        original_text = text
-        output_code = ""
-        match_end = 0
-        # Handle multiple calls
-        for re_match in re.finditer(full_code_regex, original_text, flags=re.DOTALL):
-            code = re_match.group(1).strip()
-            if ignore_multiple_calls:
-                text = original_text[: re_match.end()]
-                return code, text
-            output_code += code + "\n"
-            match_end = re_match.end()
-        # check for partial code match at end (no terminating ```)  following the last match
-        partial_match = re.match(
-            partial_code_regex,
-            original_text[match_end:],
-            flags=re.DOTALL,
-        )
-        if partial_match:
-            output_code += partial_match.group(1).strip()
-            # terminated due to stop condition. Add stop condition to output.
-            if not text.endswith("\n"):
-                text = text + "\n"
-            text = text + "```"
-        if len(output_code) == 0:
-            return text, text
-        else:
-            return output_code, text
-
-    def execute(self):
-        """Execute the Appworld tasks."""
-        result = []
-        counter = 0
-        for task_index, task_id in enumerate(tqdm(self.task_ids, desc=f"run_index={self.index}")):
-            t_result = None
-            previous_memories = []
-            # Run each task num_trials times
-            for run_id in range(self.num_trials):
-                start_time = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-                with AppWorld(task_id=task_id, experiment_name=f"{self.experiment_name}_run_{run_id}") as world:
-                    before_score = self.get_reward(world)
-                    for i in range(self.max_interactions):
-                        if i == 0:
-                            self.prompt_messages(
-                                run_id=run_id,
-                                task_index=task_index,
-                                previous_memories=previous_memories,
-                                world=world,
-                            )
-                        code_msg = self.call_llm(self.history[run_id][task_index])
-                        code, _ = self.extract_code_and_fix_content(code_msg)
-                        self.history[run_id][task_index].append({"role": "assistant", "content": code})
-
-                        output = world.execute(code)
-                        # if len(output) > self.max_response_size:
-                        #     # logger.warning(f"output exceed max size={len(output)}")
-                        #     output = output[: self.max_response_size]
-                        self.history[run_id][task_index].append(
-                            {"role": "user", "content": "Output:\n```\n" + output + "```\n\n"},
-                        )
-
-                        if world.task_completed():
-                            break
-
-                    after_score = self.get_reward(world)
-                    uplift_score = after_score - before_score
-
-                    if self.use_memory:
-                        if self.use_memory_addition:
-                            new_traj_list = [
-                                self.get_traj_from_task_history(task_id, self.history[run_id][task_index], after_score),
-                            ]
-                            previous_memories = self.summary_memory(new_traj_list)
-                            if after_score == 1:
-                                self.add_memory(previous_memories)
-
-                        # update the freq & utility attributes of retrieved memories
-                        update_utility: bool = after_score == 1
-                        self.update_memory_information(self.retrieved_memory_list[run_id][task_index], update_utility)
-
-                    counter += 1
-                    if self.use_memory_deletion:  # and counter % self.delete_freq == 0:
-                        self.delete_memory()
-
-                    t_result = {
-                        "task_id": world.task_id,
-                        "run_id": run_id,
-                        "experiment_name": self.experiment_name,
-                        "task_completed": world.task_completed(),
-                        "before_score": before_score,
-                        "after_score": after_score,
-                        "uplift_score": uplift_score,
-                        "task_history": self.history[run_id][task_index],
-                        "task_start_time": start_time,
-                    }
-                    if after_score == 1:
-                        break
-            result.append(t_result)
-
-        return result
-
-    def handle_api_response(self, response: requests.Response):
-        """Handle API response with proper error checking"""
-        if response.status_code != 200:
-            print(f"Error: {response.status_code}")
-            print(response.text)
-            return None
-
-        return response.json()
-
-    def get_memory(self, query: str):
-        """Retrieve relevant task memories based on a query"""
-        response = requests.post(
-            url=f"{self.memory_base_url}retrieve_task_memory",
-            json={
-                "query": query,
-                "enable_llm_rerank": False,
-                "enable_score_filter": False,
-                "top_k": 5,
-                "enable_llm_rewrite": False,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return None
-
-        logger.info(f"query: {query}, response: {result}")
-        return result
-
-    def get_traj_from_task_history(self, task_id: str, task_history: list, reward: float):
-        """Get the trajectory from the task history."""
-        pattern = r"\n\nSome Related Experience to help you to complete the task:.*"
-        task_history[1]["content"] = re.sub(pattern, "", task_history[1]["content"], flags=re.DOTALL)
-        return {
-            "task_id": task_id,
-            "messages": task_history,
-            "score": reward,
-        }
-
-    def summary_memory(self, trajectories):
-        """Generate a summary of conversation messages and create task memories"""
-
-        response = requests.post(
-            url=f"{self.memory_base_url}summary_task_memory",
-            json={
-                "trajectories": trajectories,
-                "success_threshold": 1.0,
-                "enable_soft_comparison": True,
-                "validation_threshold": 0.5,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return []
-
-        # Extract memory list from response
-        memory_list = result.get("metadata", {}).get("memory_list", [])
-        print(f"Task memory list created: {len(memory_list)} memories")
-        return memory_list
-
-    def add_memory(self, memory_list):
-        """Add the memory to the memory pool."""
-        response = requests.post(
-            url=f"{self.memory_base_url}add_task_memory",
-            json={
-                "memory_list": memory_list,
-            },
-        )
-        response.raise_for_status()
-
-    def update_memory_information(self, memory_list, update_utility: bool = False):
-        """Update the memory information."""
-        response = requests.post(
-            url=f"{self.memory_base_url}record_task_memory",
-            json={
-                "memory_list": memory_list,
-                "update_utility": update_utility,
-            },
-        )
-        response.raise_for_status()
-        logger.info(response.json())
-
-    def delete_memory(self):
-        """Delete the memory from the memory pool."""
-        response = requests.post(
-            url=f"{self.memory_base_url}delete_task_memory",
-            json={
-                "freq_threshold": self.freq_threshold,
-                "utility_threshold": self.utility_threshold,
-            },
-        )
-        response.raise_for_status()
-
-
-def main():
-    """Main function to run the Appworld React Agent."""
-    dataset_name = "train"
-    task_ids = load_task_ids(dataset_name)
-    agent = AppworldReactAgent(index=0, task_ids=task_ids[0:1], experiment_name=dataset_name, num_trials=1)
-    result = agent.execute()
-    logger.info(f"result={json.dumps(result)}")
-
-
-if __name__ == "__main__":
-    main()

benchmark/appworld/prompt.py

@@ -1,660 +0,0 @@
-# flake8: noqa: E402, E501
-# pylint: disable=C0114,C0301
-# This is a basic prompt template containing all the necessary onboarding information to solve AppWorld tasks. It explains the role of the agent and the supervisor, how to explore the API documentation, how to operate the interactive coding environment and call APIs via a simple task, and provides key instructions and disclaimers.
-
-# You can adapt it as needed by your agent. You can also choose to bypass API docs app and build your own API retrieval, e.g., for FullCodeRefl, IPFunCall, etc, we asked an LLM to predict relevant APIs separately and put its documentation directly in the prompt.
-PROMPT_TEMPLATE = """
-USER:
-I am your supervisor and you are a super intelligent AI Assistant whose job is to achieve my day-to-day tasks completely autonomously.
-
-To do this, you will need to interact with app/s (e.g., spotify, venmo, etc) using their associated APIs on my behalf. For this you will undertake a *multi-step conversation* using a python REPL environment. That is, you will write the python code and the environment will execute it and show you the result, based on which, you will write python code for the next step and so on, until you've achieved the goal. This environment will let you interact with app/s using their associated APIs on my behalf.
-
-Here are three key APIs that you need to know to get more information
-
-# To get a list of apps that are available to you.
-print(apis.api_docs.show_app_descriptions())
-
-# To get the list of apis under any app listed above, e.g. supervisor
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-# To get the specification of a particular api, e.g. supervisor app's show_account_passwords
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-Each code execution will produce an output that you can use in subsequent calls. Using these APIs, you can now generate code, that the environment will execute, to solve the task.
-
-For example, consider the task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-What is the password for my Spotify account?
-
-ASSISTANT:
-# Okay. Lets first find which apps are available to get the password by looking at the app descriptions.
-print(apis.api_docs.show_app_descriptions())
-
-USER:
-[
-  {
-    "name": "api_docs",
-    "description": "An app to search and explore API documentation."
-  },
-  {
-    "name": "supervisor",
-    "description": "An app to access supervisor's personal information, account credentials, addresses, payment cards, and manage the assigned task."
-  },
-  ...
-  {
-    "name": "spotify",
-    "description": "A music streaming app to stream songs and manage song, album and playlist libraries."
-  },
-  {
-    "name": "venmo",
-    "description": "A social payment app to send, receive and request money to and from others."
-  },
-  ...
-]
-
-
-ASSISTANT:
-# Looks like the supervisor app could help me with that. Lets see what apis are available under this app.
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-
-USER:
-[
-  ...
-  "show_account_passwords : Show your supervisor's account passwords."
-  ...
-]
-
-
-ASSISTANT:
-# I can use `show_account_passwords` to get the passwords. Let me see its detailed specification to understand its arguments and output structure.
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-USER:
-{
-  'app_name': 'supervisor',
-  'api_name': 'show_account_passwords',
-  'path': '/account_passwords',
-  'method': 'GET',
-  'description': "Show your supervisor's app account passwords.",
-  'parameters': [],
-  'response_schemas': {
-    'success': [{'account_name': 'string', 'password': 'string'}],
-    'failure': {'message': 'string'}
-  }
-}
-
-ASSISTANT:
-# Okay, it requires no arguments. So I can just call it directly.
-print(apis.supervisor.show_account_passwords())
-
-USER:
-[
-  {
-    "account_name": "spotify",
-    "password": "dummy_spotify_pass"
-  },
-  {
-    "account_name": "file_system",
-    "password": "dummy_fs_pass"
-  },
-  ...
-]
-
-
-ASSISTANT:
-# So the Spotify password is an entry in the `passwords` list with the account_name=spotify.
-spotify_password = [account_password["account_name"] == "spotify" for account_password in passwords][0]["password"]
-print(spotify_password)
-
-
-USER:
-dummy_spotify_pass
-
-ASSISTANT:
-# When the task is completed, I need to call apis.supervisor.complete_task(). If there is an answer, I need to pass it as an argument `answer`. I will pass the spotify_password as an answer.
-apis.supervisor.complete_task(answer=spotify_password)
-
-
-USER:
-Marked the active task complete.
-
-
-----------------------------------------------
-
-USER:
-**Key instructions and disclaimers**:
-
-1. The email addresses, access tokens and variables (e.g. spotify_password) in the example above were only for demonstration. Obtain the correct information by calling relevant APIs yourself.
-2. Only generate valid code blocks, i.e., do not put them in ```...``` or add any extra formatting. Any thoughts should be put as code comments.
-3. You can use the variables from the previous code blocks in the subsequent code blocks.
-4. Write small chunks of code and only one chunk of code in every step. Make sure everything is working correctly before making any irreversible change.
-5. The provided Python environment has access to its standard library. But modules and functions that have a risk of affecting the underlying OS, file system or process are disabled. You will get an error if do call them.
-6. Any reference to a file system in the task instructions means the file system *app*, operable via given APIs, and not the actual file system the code is running on. So do not write code making calls to os-level modules and functions.
-7. To interact with apps, only use the provided APIs, and not the corresponding Python packages. E.g., do NOT use `spotipy` for Spotify. Remember, the environment only has the standard library.
-8. The provided API documentation has both the input arguments and the output JSON schemas. All calls to APIs and parsing its outputs must be as per this documentation.
-9. For APIs that return results in "pages", make sure to consider all pages.
-10. To obtain current date or time, use Python functions like `datetime.now()` or obtain it from the phone app. Do not rely on your existing knowledge of what the current date or time is.
-11. For all temporal requests, use proper time boundaries, e.g., if I ask for something that happened yesterday, make sure to consider the time between 00:00:00 and 23:59:59. All requests are concerning a single, default (no) time zone.
-12. Any reference to my friends, family or any other person or relation refers to the people in my phone's contacts list.
-13. All my personal information, and information about my app account credentials, physical addresses and owned payment cards are stored in the "supervisor" app. You can access them via the APIs provided by the supervisor app.
-14. Once you have completed the task, call `apis.supervisor.complete_task()`. If the task asks for some information, return it as the answer argument, i.e. call `apis.supervisor.complete_task(answer=<answer>)`. For tasks that do not require an answer, just skip the answer argument or pass it as None.
-15. The answers, when given, should be just entity or number, not full sentences, e.g., `answer=10` for "How many songs are in the Spotify queue?". When an answer is a number, it should be in numbers, not in words, e.g., "10" and not "ten".
-16. You can also pass `status="fail"` in the complete_task API if you are sure you cannot solve it and want to exit.
-17. You must make all decisions completely autonomously and not ask for any clarifications or confirmations from me or anyone else.
-
-USER:
-Using these APIs, now generate code to solve the actual task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-{{ instruction }}
-"""
-
-PROMPT_TEMPLATE_WITH_EXPERIENCE = """
-USER:
-I am your supervisor and you are a super intelligent AI Assistant whose job is to achieve my day-to-day tasks completely autonomously.
-
-To do this, you will need to interact with app/s (e.g., spotify, venmo, etc) using their associated APIs on my behalf. For this you will undertake a *multi-step conversation* using a python REPL environment. That is, you will write the python code and the environment will execute it and show you the result, based on which, you will write python code for the next step and so on, until you've achieved the goal. This environment will let you interact with app/s using their associated APIs on my behalf.
-
-Here are three key APIs that you need to know to get more information
-
-# To get a list of apps that are available to you.
-print(apis.api_docs.show_app_descriptions())
-
-# To get the list of apis under any app listed above, e.g. supervisor
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-# To get the specification of a particular api, e.g. supervisor app's show_account_passwords
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-Each code execution will produce an output that you can use in subsequent calls. Using these APIs, you can now generate code, that the environment will execute, to solve the task.
-
-For example, consider the task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-What is the password for my Spotify account?
-
-ASSISTANT:
-# Okay. Lets first find which apps are available to get the password by looking at the app descriptions.
-print(apis.api_docs.show_app_descriptions())
-
-USER:
-[
-  {
-    "name": "api_docs",
-    "description": "An app to search and explore API documentation."
-  },
-  {
-    "name": "supervisor",
-    "description": "An app to access supervisor's personal information, account credentials, addresses, payment cards, and manage the assigned task."
-  },
-  ...
-  {
-    "name": "spotify",
-    "description": "A music streaming app to stream songs and manage song, album and playlist libraries."
-  },
-  {
-    "name": "venmo",
-    "description": "A social payment app to send, receive and request money to and from others."
-  },
-  ...
-]
-
-
-ASSISTANT:
-# Looks like the supervisor app could help me with that. Lets see what apis are available under this app.
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-
-USER:
-[
-  ...
-  "show_account_passwords : Show your supervisor's account passwords."
-  ...
-]
-
-
-ASSISTANT:
-# I can use `show_account_passwords` to get the passwords. Let me see its detailed specification to understand its arguments and output structure.
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-USER:
-{
-  'app_name': 'supervisor',
-  'api_name': 'show_account_passwords',
-  'path': '/account_passwords',
-  'method': 'GET',
-  'description': "Show your supervisor's app account passwords.",
-  'parameters': [],
-  'response_schemas': {
-    'success': [{'account_name': 'string', 'password': 'string'}],
-    'failure': {'message': 'string'}
-  }
-}
-
-ASSISTANT:
-# Okay, it requires no arguments. So I can just call it directly.
-print(apis.supervisor.show_account_passwords())
-
-USER:
-[
-  {
-    "account_name": "spotify",
-    "password": "dummy_spotify_pass"
-  },
-  {
-    "account_name": "file_system",
-    "password": "dummy_fs_pass"
-  },
-  ...
-]
-
-
-ASSISTANT:
-# So the Spotify password is an entry in the `passwords` list with the account_name=spotify.
-spotify_password = [account_password["account_name"] == "spotify" for account_password in passwords][0]["password"]
-print(spotify_password)
-
-
-USER:
-dummy_spotify_pass
-
-ASSISTANT:
-# When the task is completed, I need to call apis.supervisor.complete_task(). If there is an answer, I need to pass it as an argument `answer`. I will pass the spotify_password as an answer.
-apis.supervisor.complete_task(answer=spotify_password)
-
-
-USER:
-Marked the active task complete.
-
-
-----------------------------------------------
-
-USER:
-**Key instructions and disclaimers**:
-
-1. The email addresses, access tokens and variables (e.g. spotify_password) in the example above were only for demonstration. Obtain the correct information by calling relevant APIs yourself.
-2. Only generate valid code blocks, i.e., do not put them in ```...``` or add any extra formatting. Any thoughts should be put as code comments.
-3. You can use the variables from the previous code blocks in the subsequent code blocks.
-4. Write small chunks of code and only one chunk of code in every step. Make sure everything is working correctly before making any irreversible change.
-5. The provided Python environment has access to its standard library. But modules and functions that have a risk of affecting the underlying OS, file system or process are disabled. You will get an error if do call them.
-6. Any reference to a file system in the task instructions means the file system *app*, operable via given APIs, and not the actual file system the code is running on. So do not write code making calls to os-level modules and functions.
-7. To interact with apps, only use the provided APIs, and not the corresponding Python packages. E.g., do NOT use `spotipy` for Spotify. Remember, the environment only has the standard library.
-8. The provided API documentation has both the input arguments and the output JSON schemas. All calls to APIs and parsing its outputs must be as per this documentation.
-9. For APIs that return results in "pages", make sure to consider all pages.
-10. To obtain current date or time, use Python functions like `datetime.now()` or obtain it from the phone app. Do not rely on your existing knowledge of what the current date or time is.
-11. For all temporal requests, use proper time boundaries, e.g., if I ask for something that happened yesterday, make sure to consider the time between 00:00:00 and 23:59:59. All requests are concerning a single, default (no) time zone.
-12. Any reference to my friends, family or any other person or relation refers to the people in my phone's contacts list.
-13. All my personal information, and information about my app account credentials, physical addresses and owned payment cards are stored in the "supervisor" app. You can access them via the APIs provided by the supervisor app.
-14. Once you have completed the task, call `apis.supervisor.complete_task()`. If the task asks for some information, return it as the answer argument, i.e. call `apis.supervisor.complete_task(answer=<answer>)`. For tasks that do not require an answer, just skip the answer argument or pass it as None.
-15. The answers, when given, should be just entity or number, not full sentences, e.g., `answer=10` for "How many songs are in the Spotify queue?". When an answer is a number, it should be in numbers, not in words, e.g., "10" and not "ten".
-16. You can also pass `status="fail"` in the complete_task API if you are sure you cannot solve it and want to exit.
-17. You must make all decisions completely autonomously and not ask for any clarifications or confirmations from me or anyone else.
-18. Some Related Experience to help you to complete the task:
-{{experience}}
-
-USER:
-Using these APIs, now generate code to solve the actual task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-{{ instruction }}
-"""
-
-NEW_PROMPT_TEMPLATE = """
-USER:
-I am your supervisor and you are a super intelligent AI Assistant whose job is to achieve my day-to-day tasks completely autonomously.
-
-To do this, you will need to interact with app/s (e.g., spotify, venmo etc) using their associated APIs on my behalf. For this you will undertake a *multi-step conversation* using a python REPL environment. That is, you will write the python code and the environment will execute it and show you the result, based on which, you will write python code for the next step and so on, until you've achieved the goal. This environment will let you interact with app/s using their associated APIs on my behalf.
-
-Here are three key APIs that you need to know to get more information
-
-# To get a list of apps that are available to you.
-
-```python
-print(apis.api_docs.show_app_descriptions())
-```
-
-# To get the list of apis under any app listed above, e.g. spotify
-
-```python
-print(apis.api_docs.show_api_descriptions(app_name='spotify'))
-```
-
-# To get the specification of a particular api, e.g. spotify app's login api
-
-```python
-print(apis.api_docs.show_api_doc(app_name='spotify', api_name='login'))
-```
-
-Each code execution will produce an output that you can use in subsequent calls. Using these APIs, you can now generate code, that I will execute, to solve the task. Let's start with the task
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-Task: How many playlists do I have in Spotify?
-
-ASSISTANT:
-Okay. Lets first find which APIs are available to use in Spotify.
-Code:
-```python
-print(apis.api_docs.show_api_descriptions(app_name='spotify'))
-```
-
-USER:
-Output:
-```
-[
- ...
- "login : Login to your account.",
- "logout : Logout from your account.",
- ...
-]
-```
-
-
-ASSISTANT:
-Okay. Looks like I can use the `login` api. Lets find its specifications.
-
-Code:
-```python
-print(apis.api_docs.show_api_doc(app_name='spotify', api_name='login'))
-```
-
-
-USER:
-Output:
-```
-{
- "app_name": "spotify",
- "api_name": "login",
- "path": "/auth/token",
- "method": "POST",
- "description": "Login to your account.",
- "parameters": [
-  {
-   "name": "username",
-   "type": "string",
-   "required": true,
-   "description": "Your account email.",
-   "default": null,
-   "constraints": []
-  },
-  {
-   "name": "password",
-   "type": "string",
-   "required": true,
-   "description": "Your account password.",
-   "default": null,
-   "constraints": []
-  }
- ],
- "response_schemas": {
-  "success": {
-   "token_type": "string",
-   "access_token": "string"
-  },
-  "failure": {
-   "message": "string"
-  }
- }
-}
-```
-
-
-ASSISTANT:
-I need the supervisor's username and password. Lets see if any app can help me get that.
-
-Code:
-```python
-print(apis.api_docs.show_app_descriptions())
-```
-
-
-USER:
-Output:
-```
-{{ app_descriptions }}
-```
-
-
-ASSISTANT:
-Looks like the supervisor app could help me with that.
-
-Code:
-```python
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-```
-
-
-USER:
-Output:
-```
-[
-  ...
-  "show_account_passwords : Show your supervisor's account passwords."
-  ...
-]
-```
-
-
-ASSISTANT:
-I can use `show_account_passwords` to get the passwords. So I will call that and save it in a variable for future reference.
-Code:
-```python
-passwords=apis.supervisor.show_account_passwords()
-print(passwords)
-```
-
-
-USER:
-Output:
-```
-[
-  {
-    "account_name": "spotify",
-    "password": "dummy_spotify_pass"
-  },
-  {
-    "account_name": "file_system",
-    "password": "dummy_fs_pass"
-  }
-]
-```
-
-
-ASSISTANT:
-So the Spotify password is an entry in the `passwords` list with the account_name=spotify. I can use it to login now with the provided email address: {{ supervisor.email }}.
-
-Code:
- ```python
-spotify_password = [account_password["account_name"] == "spotify" for account_password in passwords][0]["password"]
-login_result = apis.spotify.login(username='{{ supervisor.email }}', password=spotify_password)
-print(login_result)
-```
-
-
-USER:
-Output:
-```
-{
- "token_type": "Bearer",
- "access_token": "fake_access_token"
-}
-```
-
-
-ASSISTANT:
-First, let's check the available APIs for the Spotify app.
-Code:
-```python
-print(apis.api_docs.show_api_descriptions(app_name='spotify'))
-```
-
-
-USER:
-Output:
-```
-[
-...
-{
-  "name": "show_playlist_library",
-  "description": "Get a list of playlists in the user's playlist library."
-},
-{
-  "name": "show_playlist",
-  "description": "Get detailed information about a specific playlist. You can view your own playlists or others' playlists if they are public."
- },
-...
-]
-```
-
-
-ASSISTANT:
-The `show_playlist_library` API seems to be the one we need. Let's check its specifications.
-Code:
-```python
-print(apis.api_docs.show_api_doc(app_name='spotify', api_name='show_playlist_library'))
-```
-
-
-USER:
-Output:
-```
-{
- "app_name": "spotify",
- "api_name": "show_playlist_library",
- "path": "/private_playlists",
- "method": "GET",
- "description": "Get a list of playlists in the user's playlist library.",
- "parameters": [
-  {
-   "name": "access_token",
-   "type": "string",
-   "required": true,
-   "description": "Access token obtained from spotify app login.",
-   "default": null,
-   "constraints": []
-  },
-  {
-   "name": "page_index",
-   "type": "integer",
-   "required": false,
-   "description": "The index of the page to retrieve.",
-   "default": 0,
-   "constraints": [
-    "value >= 0.0"
-   ]
-  },
-  {
-   "name": "page_limit",
-   "type": "integer",
-   "required": false,
-   "description": "The maximum number of results to return per page.",
-   "default": 5,
-   "constraints": [
-    "value >= 1.0, <= 20.0"
-   ]
-  },
-  {
-   "name": "is_public",
-   "type": "boolean",
-   "required": false,
-   "description": "Whether to show public playlists or private playlists.",
-   "default": null,
-   "constraints": []
-  }
- ],
- "response_schema": [
-  {
-   "title": "string",
-   "created_at": "2019-01-01T00:00:00",
-   "is_public": true,
-   "rating": 0.0,
-   "like_count": 1,
-   "owner_email": "user@example.com",
-   "playlist_id": 1,
-   "song_ids": [
-    1
-   ]
-  }
- ]
-}
-```
-
-
-ASSISTANT:
-I need to page through all the playlists to get the list of playlists and save it in `playlists`.
-Code:
-```python
-page_index = 0
-playlists = []
-while page_index < 10:
-  playlist_page = apis.spotify.show_playlist_library(access_token=spotify_access_token, page_index=page_index)
-  if playlist_page:
-    playlists.extend(playlist_page)
-    page_index += 1
-  else:
-    break
-num_playlists = len(playlists)
-print(num_playlists)
-
-```
-
-USER:
-Output:
-```
-23
-```
-
-
-ASSISTANT:
-Now that the task is completed, I can call apis.supervisor.complete_task(). Since this task has an answer to be returned, I will pass that as an argument.
-
-Code:
-```python
-apis.supervisor.complete_task(answer=num_playlists)
-```
-
-
-USER:
-Output:
-Marked the active task complete.
-
-
-----------------------------------------------
-
-USER:
-**Key instructions**:
-(1) Make sure to end code blocks with ``` followed by a newline(\n).
-
-(2) Remember you can use the variables in your code in subsequent code blocks.
-
-(3) Remember that the email addresses, access tokens and variables (e.g. spotify_password) in the example above are not valid anymore.
-
-(4) You can use the "supervisor" app to get information about my accounts and use the "phone" app to get information about friends and family.
-
-(5) Always look at API specifications (using apis.api_docs.show_api_doc) before calling an API.
-
-(6) Write small chunks of code and only one chunk of code in every step. Make sure everything is working correctly before making any irreversible change.
-
-(7) Many APIs return items in "pages". Make sure to run through all the pages by looping over `page_index`.
-
-(8) Once you have completed the task, make sure to call apis.supervisor.complete_task(). If the task asked for some information, return it as the answer argument, i.e. call apis.supervisor.complete_task(answer=<answer>). Many tasks do not require an answer, so in those cases, just call apis.supervisor.complete_task() i.e. do not pass any argument.
-
-USER:
-Using these APIs, now generate code to solve the actual task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-"""

benchmark/appworld/requirements.txt

@@ -1,7 +0,0 @@
-fastapi
-uvicorn
-uuid
-jinja2
-loguru
-openai
-pandas

benchmark/appworld/run_appworld.py

@@ -1,197 +0,0 @@
-# pylint: disable=E0611
-"""Run the Appworld React Agent."""
-
-import os
-import json
-import time
-from pathlib import Path
-
-import ray
-import requests
-from loguru import logger
-from dotenv import load_dotenv
-from appworld import load_task_ids
-from appworld_react_agent import AppworldReactAgent
-
-os.environ["APPWORLD_ROOT"] = "."
-
-load_dotenv("../../.env")
-
-
-def run_agent(
-    run_index: int,
-    max_workers: int,
-    model_name: str,
-    dataset_name: str,
-    experiment_suffix: str,
-    num_trials: int = 1,
-    use_memory: bool = False,
-    memory_base_url: str = "http://0.0.0.0:8002/",
-    use_memory_addition: bool = False,
-    use_memory_deletion: bool = False,
-    delete_freq: int = 10,
-    freq_threshold: int = 5,
-    utility_threshold: float = 0.5,
-    batch_size: int = 4,
-):
-    """Run the Appworld React Agent."""
-    experiment_name = dataset_name + "_" + experiment_suffix
-    path: Path = Path(f"./exp_result/{model_name}")
-    path.mkdir(parents=True, exist_ok=True)
-
-    task_ids = load_task_ids(dataset_name)
-
-    result: list = []
-
-    def dump_file():
-        with open(path / f"{experiment_name}.jsonl", "a", encoding="utf-8") as f:
-            for x in result:
-                f.write(json.dumps(x) + "\n")
-
-    if max_workers > 1:
-        # Process tasks in batches
-        total_tasks = len(task_ids)
-        num_batches = (total_tasks + batch_size - 1) // batch_size  # Ceiling division
-
-        logger.info(f"Total tasks: {total_tasks}, Batch size: {batch_size}, Number of batches: {num_batches}")
-
-        for batch_idx in range(num_batches):
-            # Initialize Ray for this batch
-            start_idx = batch_idx * batch_size
-            end_idx = min(start_idx + batch_size, total_tasks)
-            batch_task_ids = task_ids[start_idx:end_idx]
-
-            logger.info(f"Starting batch {batch_idx + 1}/{num_batches} with {len(batch_task_ids)} tasks")
-
-            # Initialize Ray with the number of CPUs needed for this batch
-            ray.init(num_cpus=len(batch_task_ids))
-
-            future_list: list = []
-            for i, task_id in enumerate(batch_task_ids):
-                actor = AppworldReactAgent.remote(
-                    index=start_idx + i,
-                    model_name=model_name,
-                    task_ids=[task_id],
-                    experiment_name=experiment_name,
-                    num_trials=num_trials,
-                    use_memory=use_memory,
-                    memory_base_url=memory_base_url,
-                    use_memory_addition=use_memory_addition,
-                    use_memory_deletion=use_memory_deletion,
-                    delete_freq=delete_freq,
-                    freq_threshold=freq_threshold,
-                    utility_threshold=utility_threshold,
-                )
-                future = actor.execute.remote()
-                future_list.append(future)
-                time.sleep(1)
-
-            logger.info(f"Batch {batch_idx + 1} submit complete, waiting for results...")
-
-            # Collect results from this batch
-            for i, (task_id, future) in enumerate(zip(batch_task_ids, future_list)):
-                try:
-                    t_result = ray.get(future)
-                    if t_result:
-                        if isinstance(t_result, list):
-                            result.extend(t_result)
-                        else:
-                            result.append(t_result)
-                except Exception:
-                    logger.exception(f"run ray error with task_id={task_id}")
-
-                logger.info(f"Batch {batch_idx + 1}: task {i + 1}/{len(batch_task_ids)} complete")
-
-            # Shutdown Ray to free resources before next batch
-            ray.shutdown()
-            logger.info(f"Batch {batch_idx + 1}/{num_batches} complete, Ray resources released")
-
-            # Optional: small delay between batches
-            if batch_idx < num_batches - 1:
-                time.sleep(2)
-
-        dump_file()
-
-    else:
-        agent = AppworldReactAgent(
-            index=run_index,
-            model_name=model_name,
-            task_ids=task_ids,
-            experiment_name=experiment_name,
-            num_trials=num_trials,
-            use_memory=use_memory,
-            memory_base_url=memory_base_url,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=delete_freq,
-            freq_threshold=freq_threshold,
-            utility_threshold=utility_threshold,
-        )
-        result = agent.execute()
-
-        dump_file()
-
-
-def handle_api_response(response: requests.Response):
-    """Handle API response with proper error checking"""
-    if response.status_code != 200:
-        print(f"Error: {response.status_code}")
-        print(response.text)
-        return None
-
-    return response.json()
-
-
-def load_memory(path: str = "docs/library", api_url: str = "http://0.0.0.0:8002/"):
-    """Load memories from disk into the vector store"""
-    response = requests.post(
-        url=f"{api_url}load_memory",
-        json={
-            "load_file_path": path,
-            "clear_existing": True,
-        },
-    )
-
-    result = handle_api_response(response)
-    if result:
-        print(f"Memory loaded from {path}")
-
-
-def main():
-    """Main function to run the Appworld React Agent."""
-    max_workers = 16
-    batch_size = 8
-
-    num_runs = 4  # Number of runs
-    num_trials = 1  # for self-reflection
-    model_name = "qwen3-8b"
-    use_memory = True
-    use_memory_addition = False
-    use_memory_deletion = False
-    memory_base_url = "http://0.0.0.0:8002/"
-
-    if use_memory:
-        load_file_path = "docs/library/paper_data/task/appworld_qwen3_8b.jsonl"
-        load_memory(load_file_path, memory_base_url)
-
-    for i in range(num_runs):
-        run_agent(
-            run_index=i,
-            max_workers=max_workers,
-            model_name=model_name,
-            dataset_name="test_normal",
-            experiment_suffix="with-fixed-memory",
-            num_trials=num_trials,
-            use_memory=use_memory,
-            memory_base_url=memory_base_url,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=5,
-            freq_threshold=5,
-            utility_threshold=0.5,
-            batch_size=batch_size,
-        )
-
-
-if __name__ == "__main__":
-    main()

benchmark/appworld/run_exp_statistic.py

@@ -1,164 +0,0 @@
-"""Run the experiment statistic."""
-
-import json
-from collections import defaultdict
-from pathlib import Path
-
-import pandas as pd
-from loguru import logger
-
-
-def calculate_best_at_k(scores: list, k: int) -> float:
-    """
-    Calculate best@k
-    Divide scores into groups of size k, take the maximum value in each group,
-    then average these maximum values
-
-    Args:
-        scores: List of after_score values for all runs of a task
-        k: Group size
-
-    Returns:
-        best@k value
-    """
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        group_maxs.append(max(group))
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def calculate_pass_at_k(scores: list, k: int) -> float:
-    """Calculate pass@k."""
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        is_pass = 1.0 if max(group) >= 1.0 else 0.0
-        group_maxs.append(is_pass)
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def get_possible_k_values(total_runs: int) -> list:
-    """
-    Get all possible k values (factors of total_runs)
-
-    Args:
-        total_runs: Total number of runs
-
-    Returns:
-        List of k values in descending order
-    """
-    k_values = []
-    for k in range(1, total_runs + 1):
-        if total_runs % k == 0:
-            k_values.append(k)
-    return sorted(k_values, reverse=True)  # Sort from large to small
-
-
-def run_exp_statistic():
-    """Run the experiment statistic."""
-    path: Path = Path("./exp_result/qwen3-8b")
-
-    # Store results for all experiments
-    all_results = {}
-
-    for file in path.glob("*.jsonl"):  # [f for f in path.glob("*.jsonl") if not f.stem[-1].isdigit()]
-        # Group results by task_id
-        task_results = defaultdict(list)
-
-        with open(file, "r", encoding="utf-8") as f:
-            for line in f:
-                if not line.strip():
-                    continue
-                data = json.loads(line)
-
-                if isinstance(data, list):
-                    for part_data in data:
-                        task_id = part_data["task_id"]
-                        after_score = part_data["after_score"]
-                        task_results[task_id].append(after_score)
-                else:
-                    task_id = data["task_id"]
-                    after_score = data["after_score"]
-                    task_results[task_id].append(after_score)
-
-        if not task_results:
-            logger.warning(f"No valid data found in file {file}")
-            continue
-
-        # Check if each task has consistent number of runs
-        run_counts = [len(scores) for scores in task_results.values()]
-        if len(set(run_counts)) > 1:
-            logger.warning(f"Inconsistent number of runs for different tasks in file {file}: {set(run_counts)}")
-            continue
-
-        num_runs = run_counts[0]
-        logger.info(f"File {file}: {len(task_results)} tasks, {num_runs} runs per task")
-
-        # Get all possible k values
-        k_values = get_possible_k_values(num_runs)
-        logger.info(f"Calculable best@k values: {k_values}")
-
-        # Calculate various best@k values
-        file_results = {"file": file.name}
-
-        for k in k_values:
-            best_at_k_scores = []
-            pass_at_k_scores = []
-            for task_id, scores in task_results.items():
-                try:
-                    best_k_score = calculate_best_at_k(scores, k)
-                    pass_at_k_score = calculate_pass_at_k(scores, k)
-                    pass_at_k_scores.append(pass_at_k_score)
-                    best_at_k_scores.append(best_k_score)
-                except ValueError as e:
-                    logger.error(f"Error calculating best@{k} for task {task_id}: {e}")
-                    continue
-
-            if best_at_k_scores:
-                avg_best_at_k = sum(best_at_k_scores) / len(best_at_k_scores)
-                file_results[f"best@{k}"] = avg_best_at_k
-                logger.info(f"file={file.name} best@{k}={avg_best_at_k:.4f}")
-
-            if pass_at_k_scores:
-                avg_pass_at_k = sum(pass_at_k_scores) / len(pass_at_k_scores)
-                file_results[f"pass@{k}"] = avg_pass_at_k
-                logger.info(f"file={file.name} pass@{k}={avg_pass_at_k:.4f}")
-
-        all_results[file.name] = file_results
-
-    # Create and display table
-    if all_results:
-        df = pd.DataFrame(list(all_results.values()))
-        df = df.set_index("file")
-
-        # Sort columns by the number in column name (best@8, best@4, best@2, best@1)
-        pass_columns = [col for col in df.columns if col.startswith("pass@")]
-        # best_columns = [col for col in df.columns]
-        pass_columns.sort(key=lambda x: x, reverse=False)
-        df = df[pass_columns]
-
-        print("\n" + "=" * 80)
-        print("Experiment Results Summary Table")
-        print("=" * 80)
-        print(df.round(4))
-        print("=" * 80)
-
-        # Save table to CSV
-        output_path = path / "experiment_summary.csv"
-        df.to_csv(output_path)
-        logger.info(f"Results table saved to: {output_path}")
-    else:
-        logger.warning("No valid experiment results found")
-
-
-if __name__ == "__main__":
-    run_exp_statistic()

benchmark/beam/README.md

@@ -0,0 +1,124 @@
+[中文版 / Chinese version](./README_ZH.md)
+
+# BEAM Benchmark
+
+BEAM is a benchmark for **memory capability over long-context chat cases**. Each
+case contains a very long chat history split into batches; ReMe converts each
+batch into a session, ingests them in chronological order, then answers probing
+questions via an agentic (ReAct) mode. Answers are scored with BEAM's
+rubric-based `answer_judge` job, which produces both a graded score and a binary
+verdict, and per-type averages are reported.
+
+BEAM ships dataset variants by chat size — `100K` / `500K` / `1M` / `10M` — so
+memory systems can be stressed at different context lengths. Question types
+include abstention, contradiction resolution, event ordering, information
+extraction, instruction following, knowledge update, multi-session reasoning,
+preference following, summarization, and temporal reasoning.
+
+> For the shared setup (dependencies, credentials, log conventions) see the
+> [top-level benchmark README](../README.md).
+
+## 1. Get the Dataset
+
+BEAM is a public repository, cloned into `benchmark/beam/dataset/`:
+
+```bash
+mkdir -p benchmark/beam/dataset
+cd benchmark/beam/dataset
+git clone https://github.com/mohammadtavakoli78/BEAM.git
+```
+
+After cloning, `benchmark/beam/dataset/BEAM/` should contain `chats/`, `src/`,
+`topics/` and other subdirectories.
+
+## 2. Run
+
+From the repository root:
+
+```bash
+python benchmark/beam/run.py
+python benchmark/beam/run.py --config benchmark/beam/config.yaml
+python benchmark/beam/run.py -q                        # quiet
+python benchmark/beam/run.py --eval_only               # reuse existing workspaces, query + judge only
+```
+
+## 3. Pipeline
+
+1. For each case, load `chat.json` and convert each batch into a ReMe session.
+2. Ingest sessions in chronological order into an isolated workspace, then `digest_update`.
+3. Answer each probing question via agentic (ReAct) mode.
+4. Score answers with BEAM's rubric-based `answer_judge` job and print per-type averages.
+
+## 4. Key config — `benchmark/beam/config.yaml`
+
+| Key | Meaning |
+| --- | --- |
+| `dataset.beam_root` | BEAM dataset root (`benchmark/beam/dataset/BEAM`). |
+| `dataset.chat_size` | Variant to run: `100K` / `500K` / `1M` / `10M`. |
+| `dataset.case_ids` | Specific cases (e.g. `["1","2"]`); empty = all cases. |
+| `dataset.start_index` / `num_items` | Case pagination (`num_items` `0` = all). |
+| `dataset.workspace_root` | Per-case workspace root (`benchmark/beam/workspaces/beam`). |
+| `evaluation.num_workers` | `0` = auto, `1` = sequential, `>1` = parallel. |
+| `reme.config` | ReMe config used (`beam.yaml`). |
+| `output.dir` | Results directory (`benchmark/beam/results`). |
+
+## 5. Outputs
+
+Results are JSON files written to `output.dir` as
+`results_<chat_size>_<timestamp>.json`, with a per-type score summary also
+printed to the console. Logging conventions are shared across benchmarks — see
+the [top-level README](../README.md#outputs--logs).
+
+## 6. Reference Results
+
+> The results below use the longmemeval-version prompt.
+
+### 100K
+
+agentscope==2.0.4.post1, conda reme env, 20 workers, eval-only (reusing prebuilt memory)
+(2026-08-05, 20 cases / 400 Qs, total 46.0 min)
+
+| Type | Agentic | Binary | input tok/q | output tok/q | total tok/q | tool calls/q |
+|---|---|---|---|---|---|---|
+| abstention | 0.550 | 0.550 | 96,031 | 1,070 | 97,101 | 4.58 |
+| contradiction_resolution | 0.438 | 0.412 | 32,263 | 872 | 33,135 | 2.48 |
+| event_ordering | 0.501 | 0.423 | 140,195 | 5,163 | 145,358 | 4.70 |
+| information_extraction | 0.873 | 0.832 | 50,245 | 883 | 51,128 | 3.15 |
+| instruction_following | 0.750 | 0.725 | 37,986 | 848 | 38,834 | 2.67 |
+| knowledge_update | 0.688 | 0.675 | 31,198 | 651 | 31,849 | 2.27 |
+| multi_session_reasoning | 0.626 | 0.584 | 85,038 | 4,563 | 89,601 | 4.28 |
+| preference_following | 0.925 | 0.912 | 34,281 | 989 | 35,270 | 2.50 |
+| summarization | 0.623 | 0.461 | 89,657 | 2,056 | 91,713 | 4.12 |
+| temporal_reasoning | 0.637 | 0.625 | 34,563 | 1,049 | 35,612 | 2.52 |
+| **OVERALL** | **0.661** | **0.620** | **63,146** | **1,814** | **64,960** | **3.33** |
+
+Memory Construction average token consumption (default agent, full build over 20 cases):
+
+| Agent | input tok/case | output tok/case | total tok/case |
+|---|---|---|---|
+| default | 2,172,316 | 136,697 | 2,309,013 |
+
+### 1M
+
+agentscope==2.0.4.post1, conda reme env, 20 workers, full memory build
+(2026-08-05, 35 cases / 700 Qs, total 459.2 min)
+
+| Type | Agentic | Binary | input tok/q | output tok/q | total tok/q | tool calls/q |
+|---|---|---|---|---|---|---|
+| abstention | 0.429 | 0.429 | 118,707 | 1,178 | 119,886 | 4.20 |
+| contradiction_resolution | 0.391 | 0.364 | 49,787 | 810 | 50,597 | 2.50 |
+| event_ordering | 0.558 | 0.456 | 201,514 | 3,889 | 205,403 | 4.79 |
+| information_extraction | 0.809 | 0.772 | 78,950 | 894 | 79,844 | 3.00 |
+| instruction_following | 0.852 | 0.832 | 55,757 | 924 | 56,681 | 2.81 |
+| knowledge_update | 0.779 | 0.771 | 45,981 | 665 | 46,646 | 2.37 |
+| multi_session_reasoning | 0.658 | 0.612 | 138,133 | 2,873 | 141,006 | 4.40 |
+| preference_following | 0.798 | 0.777 | 51,796 | 920 | 52,716 | 2.53 |
+| summarization | 0.693 | 0.537 | 158,794 | 2,905 | 161,700 | 4.44 |
+| temporal_reasoning | 0.536 | 0.536 | 100,176 | 3,148 | 103,324 | 3.90 |
+| **OVERALL** | **0.650** | **0.609** | **99,959** | **1,821** | **101,780** | **3.49** |
+
+Memory Construction average token consumption (default agent, full build over 35 cases):
+
+| Agent | input tok/case | output tok/case | total tok/case |
+|---|---|---|---|
+| default | 31,943,817 | 1,417,061 | 33,360,878 |

benchmark/beam/README_ZH.md

@@ -0,0 +1,119 @@
+# BEAM 评测
+
+[English version](./README.md)
+
+BEAM 是一个面向**长上下文对话场景**的记忆能力评测基准。每个 case 包含一段被切分为多个
+batch 的超长对话；ReMe 将每个 batch 转换为一个会话，按时间顺序摄入后，以 agentic（ReAct）
+模式回答探测问题。答案由 BEAM 基于 rubric 的 `answer_judge` 任务打分，同时给出分级分数与二元
+判定，并输出各类型平均分。
+
+BEAM 按对话规模提供多种数据变体 —— `100K` / `500K` / `1M` / `10M`，可在不同上下文长度下
+压测记忆系统。题型包括 abstention（拒答）、contradiction resolution（矛盾消解）、event
+ordering（事件排序）、information extraction（信息抽取）、instruction following（指令遵循）、
+knowledge update（知识更新）、multi-session reasoning（多会话推理）、preference following
+（偏好遵循）、summarization（摘要）与 temporal reasoning（时间推理）。
+
+> 公共设置（依赖、凭据、日志约定）见[总评测说明](../README_ZH.md)。
+
+## 1. 获取数据集
+
+BEAM 是公开仓库，clone 到 `benchmark/beam/dataset/` 下：
+
+```bash
+mkdir -p benchmark/beam/dataset
+cd benchmark/beam/dataset
+git clone https://github.com/mohammadtavakoli78/BEAM.git
+```
+
+clone 完成后，`benchmark/beam/dataset/BEAM/` 目录下应包含 `chats/`、`src/`、`topics/` 等子目录。
+
+## 2. 运行
+
+在仓库根目录执行：
+
+```bash
+python benchmark/beam/run.py
+python benchmark/beam/run.py --config benchmark/beam/config.yaml
+python benchmark/beam/run.py -q                        # 安静模式
+python benchmark/beam/run.py --eval_only               # 复用已有工作区，仅执行查询 + 评判
+```
+
+## 3. 流程
+
+1. 为每个 case 加载 `chat.json`，将每个 batch 转换为一个 ReMe 会话。
+2. 按时间顺序将会话摄入独立工作区，随后执行 `digest_update`。
+3. 以 agentic（ReAct）模式回答每个探测问题。
+4. 通过 BEAM 基于 rubric 的 `answer_judge` 任务打分，并输出各类型平均分。
+
+## 4. 关键配置 —— `benchmark/beam/config.yaml`
+
+| 配置项 | 含义 |
+| --- | --- |
+| `dataset.beam_root` | BEAM 数据集根目录（`benchmark/beam/dataset/BEAM`）。 |
+| `dataset.chat_size` | 运行的变体：`100K` / `500K` / `1M` / `10M`。 |
+| `dataset.case_ids` | 指定 case（如 `["1","2"]`），空表示全部。 |
+| `dataset.start_index` / `num_items` | case 分页（`num_items` 为 `0` 表示全部）。 |
+| `dataset.workspace_root` | case 工作区根目录（`benchmark/beam/workspaces/beam`）。 |
+| `evaluation.num_workers` | `0` = 自动，`1` = 串行，`>1` = 并行。 |
+| `reme.config` | 使用的 ReMe 配置（`beam.yaml`）。 |
+| `output.dir` | 结果目录（`benchmark/beam/results`）。 |
+
+## 5. 输出
+
+结果以 JSON 文件写入 `output.dir`，文件名为 `results_<chat_size>_<timestamp>.json`，
+同时控制台会打印含各类型分数的汇总。日志约定在各基准间通用，见
+[总说明](../README_ZH.md#输出与日志)。
+
+## 6. 参考结果
+
+> 以下结果使用 longmemeval 版本的 prompt。
+
+### 100K
+
+agentscope==2.0.4.post1，conda reme 环境，20 并发，eval-only（复用已构建 memory）
+（2026-08-05，20 cases / 400 Qs，总耗时 46.0 min）
+
+| 题型 | Agentic | Binary | input tok/q | output tok/q | total tok/q | tool calls/q |
+|---|---|---|---|---|---|---|
+| abstention | 0.550 | 0.550 | 96,031 | 1,070 | 97,101 | 4.58 |
+| contradiction_resolution | 0.438 | 0.412 | 32,263 | 872 | 33,135 | 2.48 |
+| event_ordering | 0.501 | 0.423 | 140,195 | 5,163 | 145,358 | 4.70 |
+| information_extraction | 0.873 | 0.832 | 50,245 | 883 | 51,128 | 3.15 |
+| instruction_following | 0.750 | 0.725 | 37,986 | 848 | 38,834 | 2.67 |
+| knowledge_update | 0.688 | 0.675 | 31,198 | 651 | 31,849 | 2.27 |
+| multi_session_reasoning | 0.626 | 0.584 | 85,038 | 4,563 | 89,601 | 4.28 |
+| preference_following | 0.925 | 0.912 | 34,281 | 989 | 35,270 | 2.50 |
+| summarization | 0.623 | 0.461 | 89,657 | 2,056 | 91,713 | 4.12 |
+| temporal_reasoning | 0.637 | 0.625 | 34,563 | 1,049 | 35,612 | 2.52 |
+| **OVERALL** | **0.661** | **0.620** | **63,146** | **1,814** | **64,960** | **3.33** |
+
+Memory Construction 平均 token 消耗（default agent，20 cases 全量构建）：
+
+| Agent | input tok/case | output tok/case | total tok/case |
+|---|---|---|---|
+| default | 2,172,316 | 136,697 | 2,309,013 |
+
+### 1M
+
+agentscope==2.0.4.post1，conda reme 环境，20 并发，全量构建 memory
+（2026-08-05，35 cases / 700 Qs，总耗时 459.2 min）
+
+| 题型 | Agentic | Binary | input tok/q | output tok/q | total tok/q | tool calls/q |
+|---|---|---|---|---|---|---|
+| abstention | 0.429 | 0.429 | 118,707 | 1,178 | 119,886 | 4.20 |
+| contradiction_resolution | 0.391 | 0.364 | 49,787 | 810 | 50,597 | 2.50 |
+| event_ordering | 0.558 | 0.456 | 201,514 | 3,889 | 205,403 | 4.79 |
+| information_extraction | 0.809 | 0.772 | 78,950 | 894 | 79,844 | 3.00 |
+| instruction_following | 0.852 | 0.832 | 55,757 | 924 | 56,681 | 2.81 |
+| knowledge_update | 0.779 | 0.771 | 45,981 | 665 | 46,646 | 2.37 |
+| multi_session_reasoning | 0.658 | 0.612 | 138,133 | 2,873 | 141,006 | 4.40 |
+| preference_following | 0.798 | 0.777 | 51,796 | 920 | 52,716 | 2.53 |
+| summarization | 0.693 | 0.537 | 158,794 | 2,905 | 161,700 | 4.44 |
+| temporal_reasoning | 0.536 | 0.536 | 100,176 | 3,148 | 103,324 | 3.90 |
+| **OVERALL** | **0.650** | **0.609** | **99,959** | **1,821** | **101,780** | **3.49** |
+
+Memory Construction 平均 token 消耗（default agent，35 cases 全量构建）：
+
+| Agent | input tok/case | output tok/case | total tok/case |
+|---|---|---|---|
+| default | 31,943,817 | 1,417,061 | 33,360,878 |

benchmark/beam/config.yaml

@@ -0,0 +1,24 @@
+# BEAM evaluation configuration
+# This file controls what/how to evaluate.
+
+dataset:
+  beam_root: "benchmark/beam/dataset/BEAM"   # BEAM dataset root
+  chat_size: "1M"                        # 100K | 500K | 1M | 10M (dataset variant)
+  case_ids: []                          # empty = all cases; or ["1", "2", "3"]
+  start_index: 0                        # first case index (for pagination)
+  num_items: 0                          # 0 = all cases; >0 = limit
+  workspace_root: "benchmark/beam/workspaces/beam"  # workspace root for case workspaces
+
+evaluation:
+  num_workers: 20                       # 0 = auto; 1 = sequential; >1 = parallel (per-case)
+  compress_session: false               # true = compress session chunks in search_v2 (query-aware); false = no compression
+
+reme:
+  config: "beam.yaml"                   # reme config (in reme/config/)
+
+output:
+  dir: "benchmark/beam/results"
+  log_dir: "logs"                       # log directory (relative to project root)
+  log_prefix: "beam"                    # benchmark name used in log filenames
+  log_to_console: true
+  log_to_file: true

benchmark/beam/kill.sh

@@ -0,0 +1,76 @@
+#!/bin/bash
+# 杀死指定进程及其所有子进程
+# Usage: bash kill.sh <PID>
+
+if [ -z "$1" ]; then
+    echo "Usage: bash kill.sh <PID>"
+    echo "  杀死指定进程及其所有子进程"
+    exit 1
+fi
+
+PID=$1
+
+# 检查进程是否存在
+if ! kill -0 "$PID" 2>/dev/null; then
+    echo "进程 $PID 不存在"
+    exit 1
+fi
+
+# 递归收集所有子进程（包括子进程的子进程）
+collect_children() {
+    local parent=$1
+    local children
+    children=$(ps -o pid= --ppid "$parent" 2>/dev/null | tr -d ' ')
+    for child in $children; do
+        collect_children "$child"
+    done
+    echo "$parent"
+}
+
+# 收集进程树（子进程在前，父进程在后，保证先杀子再杀父）
+PROCESS_TREE=$(collect_children "$PID")
+TOTAL=$(echo "$PROCESS_TREE" | wc -l | tr -d ' ')
+
+echo "进程树（共 $TOTAL 个进程）:"
+while read -r p; do
+    cmd=$(ps -o args= -p "$p" 2>/dev/null | head -c 80)
+    printf "  PID=%-8s %s\n" "$p" "$cmd"
+done <<< "$PROCESS_TREE"
+
+# 先 SIGTERM 优雅终止
+echo ""
+echo "发送 SIGTERM..."
+while read -r p; do
+    kill "$p" 2>/dev/null
+done <<< "$PROCESS_TREE"
+
+# 等待最多 5 秒
+for i in $(seq 1 5); do
+    alive=false
+    while read -r p; do
+        if kill -0 "$p" 2>/dev/null; then
+            alive=true
+        fi
+    done <<< "$PROCESS_TREE"
+    if [ "$alive" = false ]; then
+        break
+    fi
+    sleep 1
+done
+
+# 检查是否还有残留，强制 SIGKILL
+remaining=false
+while read -r p; do
+    if kill -0 "$p" 2>/dev/null; then
+        remaining=true
+    fi
+done <<< "$PROCESS_TREE"
+
+if [ "$remaining" = true ]; then
+    echo "部分进程未响应，发送 SIGKILL..."
+    while read -r p; do
+        kill -9 "$p" 2>/dev/null
+    done <<< "$PROCESS_TREE"
+fi
+
+echo "已终止进程树（根 PID=$PID，共 $TOTAL 个进程）"

benchmark/beam/run.py

@@ -0,0 +1,891 @@
+"""BEAM evaluation runner for ReMe.
+
+Evaluates ReMe's memory capability using the BEAM dataset.
+Each case gets an isolated workspace; chat.json batches are ingested as
+sessions in chronological order; finally probing questions are answered
+via an agentic (ReAct) approach, then
+judged by BEAM's rubric-based LLM-as-judge.
+
+Usage:
+    python benchmark/beam/run.py
+    python benchmark/beam/run.py --config benchmark/beam/config.yaml
+    python benchmark/beam/run.py -q                          # quiet: only eval-level logs
+    python benchmark/beam/run.py --log-level WARNING         # reduce eval runner logs
+    python benchmark/beam/run.py --reme-log-level WARNING    # reduce reme internal logs
+    python benchmark/beam/run.py --eval_only                 # query+judge only, reuse existing workspace
+"""
+
+import json
+import logging
+import os
+import re
+import shutil
+import time
+import threading
+from datetime import datetime
+from pathlib import Path
+
+import yaml
+from dotenv import load_dotenv
+
+# Load .env from project root
+_PROJECT_ROOT = Path(__file__).parent.parent.parent
+load_dotenv(_PROJECT_ROOT / ".env")
+
+# Workspace root — read from config.yaml (dataset.workspace_root)
+_WORKSPACE_ROOT_DEFAULT = "benchmark/beam/workspaces/beam"
+
+# ---------------------------------------------------------------------------
+# Logging
+# ---------------------------------------------------------------------------
+_DEFAULT_LOG_FORMAT = "%(asctime)s | %(levelname)s | %(message)s"
+
+logging.basicConfig(level=logging.INFO, format=_DEFAULT_LOG_FORMAT)
+logger = logging.getLogger("beam")
+
+# Noisy library loggers silenced by default
+_NOISY_LOGGERS = [
+    "httpx",
+    "httpcore",
+    "openai",
+    "uvicorn",
+    "multipart",
+    "asyncio",
+    "watchfiles",
+    "filelock",
+]
+
+
+def setup_logging(
+    log_level: str,
+    reme_log_level: str,
+    log_dir: str | None = None,
+):
+    """Configure logging for the eval runner and reme internals.
+
+    Args:
+        log_level: Level for the eval runner logger (DEBUG/INFO/WARNING/ERROR).
+        reme_log_level: Level for reme's internal loguru logger.
+        log_dir: Per-run log directory (absolute path). None = no file logging.
+    """
+    numeric = getattr(logging, log_level.upper(), logging.INFO)
+    # Eval runner logger
+    logging.getLogger().setLevel(numeric)
+    logger.setLevel(numeric)
+
+    # Suppress noisy library loggers when above DEBUG
+    if numeric > logging.DEBUG:
+        for name in _NOISY_LOGGERS:
+            lib_logger = logging.getLogger(name)
+            lib_logger.setLevel(max(numeric, logging.WARNING))
+
+    # Add file handler for eval runner if log_dir is specified
+    if log_dir:
+        os.makedirs(log_dir, exist_ok=True)
+        log_filepath = os.path.join(log_dir, "runner.log")
+        file_handler = logging.FileHandler(log_filepath, encoding="utf-8")
+        file_handler.setLevel(numeric)
+        file_handler.setFormatter(logging.Formatter(_DEFAULT_LOG_FORMAT))
+        logging.getLogger().addHandler(file_handler)
+        logger.info(f"Eval runner log file: {log_filepath}")
+
+    # Reme internal logger (loguru) — will be applied per-worker via _configure_worker
+    os.environ["REME_LOG_LEVEL"] = reme_log_level.upper()
+    if log_dir:
+        os.environ["REME_LOG_DIR"] = log_dir
+
+
+def _configure_worker(
+    log_level: str,
+    reme_log_level: str,
+    log_dir: str | None = None,
+):
+    """Set up logging inside a multiprocessing worker process.
+
+    Must be called at the top of each worker because child processes inherit
+    parent state but loguru sinks are NOT shared across fork/spawn.
+    """
+    numeric = getattr(logging, log_level.upper(), logging.INFO)
+    logging.basicConfig(level=numeric, format=_DEFAULT_LOG_FORMAT, force=True)
+    logging.getLogger("beam").setLevel(numeric)
+    if numeric > logging.DEBUG:
+        for name in _NOISY_LOGGERS:
+            logging.getLogger(name).setLevel(max(numeric, logging.WARNING))
+
+    # Add file handler for eval runner in worker process
+    if log_dir:
+        os.makedirs(log_dir, exist_ok=True)
+        pid = os.getpid()
+        log_filepath = os.path.join(log_dir, f"worker-{pid}.log")
+        file_handler = logging.FileHandler(log_filepath, encoding="utf-8")
+        file_handler.setLevel(numeric)
+        file_handler.setFormatter(logging.Formatter(_DEFAULT_LOG_FORMAT))
+        logging.getLogger().addHandler(file_handler)
+
+    # Re-initialize loguru for reme internals at the desired level
+    from reme.utils import get_logger
+
+    reme_log_dir = log_dir or "logs"
+    get_logger(log_dir=reme_log_dir, level=reme_log_level.upper(), force_init=True)
+
+
+# ---------------------------------------------------------------------------
+# Config loading
+# ---------------------------------------------------------------------------
+def load_eval_config(config_path: str | None = None) -> dict:
+    """Load evaluation config yaml with env-var expansion."""
+    if config_path is None:
+        config_path = str(Path(__file__).parent / "config.yaml")
+    with open(config_path, encoding="utf-8") as f:
+        raw = f.read()
+
+    # Expand ${VAR} and ${VAR:-default}
+    def _expand(m):
+        expr = m.group(1)
+        if ":-" in expr:
+            key, default = expr.split(":-", 1)
+            return os.environ.get(key, default)
+        return os.environ.get(expr, "")
+
+    raw = re.sub(r"\$\{([^}]+)\}", _expand, raw)
+    return yaml.safe_load(raw)
+
+
+# ---------------------------------------------------------------------------
+# BEAM data loading
+# ---------------------------------------------------------------------------
+def parse_beam_time_anchor(time_str: str) -> datetime:
+    """Parse BEAM time_anchor format: 'March-15-2024' -> datetime."""
+    for fmt in ("%B-%d-%Y", "%b-%d-%Y"):
+        try:
+            return datetime.strptime(time_str, fmt)
+        except ValueError:
+            continue
+    raise ValueError(f"Cannot parse time_anchor: {time_str!r}")
+
+
+def load_beam_chat(chat_path: Path, chat_size: str, case_id: str) -> list[dict]:
+    """Load BEAM chat.json and convert to ReMe session format.
+
+    Each batch becomes one session with all its turns flattened.
+    Each turn resolves its own time_anchor independently; turns without
+    an explicit time_anchor inherit from the most recent preceding turn.
+    Returns list of sessions, each with:
+      - session_id: str
+      - date: str (YYYY-MM-DD)  — derived from the *first* turn's time
+      - messages: list[dict] with name, role, content, created_at
+    """
+    with open(chat_path, encoding="utf-8") as f:
+        batches = json.load(f)
+
+    sessions = []
+    for batch in batches:
+        batch_num = batch["batch_number"]
+
+        # Resolve batch-level fallback (used when no turn has a time_anchor)
+        batch_anchor = batch.get("time_anchor")
+        if not batch_anchor:
+            batch_anchor = "January-1-2024"
+
+        # Flatten all turns, resolving time_anchor per turn
+        messages = []
+        prev_dt = None  # carries forward from previous turn
+        first_dt = None  # for session-level date
+
+        for turn in batch["turns"]:
+            # Find this turn's own time_anchor from its messages
+            turn_anchor = None
+            for msg in turn:
+                if msg.get("time_anchor"):
+                    turn_anchor = msg["time_anchor"]
+                    break
+
+            if turn_anchor:
+                dt = parse_beam_time_anchor(turn_anchor)
+            elif prev_dt is not None:
+                dt = prev_dt  # inherit from previous turn
+            else:
+                dt = parse_beam_time_anchor(batch_anchor)
+
+            if first_dt is None:
+                first_dt = dt
+            prev_dt = dt
+
+            for msg in turn:
+                role = msg["role"]
+                messages.append(
+                    {
+                        "name": role,
+                        "role": role,
+                        "content": msg["content"],
+                        "created_at": dt.strftime("%Y-%m-%dT%H:%M:%S"),
+                    },
+                )
+
+        sessions.append(
+            {
+                "session_id": f"beam_{chat_size}_{case_id}_batch{batch_num}",
+                "date": first_dt.strftime("%Y-%m-%d"),
+                "messages": messages,
+            },
+        )
+
+    return sessions
+
+
+def get_available_cases(beam_root: Path, chat_size: str) -> list[str]:
+    """Return sorted list of case IDs for a given chat size."""
+    chats_dir = beam_root / "chats" / chat_size
+    if not chats_dir.exists():
+        return []
+    return sorted(
+        [d.name for d in chats_dir.iterdir() if d.is_dir()],
+        key=int,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Answer generation
+# ---------------------------------------------------------------------------
+async def answer_question_agentic(app, question: str, compress_session: bool = False) -> tuple[str, dict]:
+    """Answer a probing question using ReMe's agentic_answer job.
+
+    Returns (answer, metadata)
+    """
+    from reme.utils.evaluation_interface import track_agent_token_usage, track_job_counts
+
+    with (
+        track_job_counts(["search"], app.context) as tool_counts,
+        track_agent_token_usage(
+            ["bench"],
+            app.context,
+        ) as token_usages,
+    ):
+        query_resp = await app.run_job(
+            "agentic_answer",
+            query=question,
+            compress_session=compress_session,
+        )
+    answer = (query_resp.answer or "").strip()
+
+    return answer, {
+        "mode": "agentic",
+        "tool_counts": tool_counts,
+        "token_usage": token_usages["bench"],
+    }
+
+
+# ---------------------------------------------------------------------------
+# BEAM rubric-based LLM-as-Judge
+# ---------------------------------------------------------------------------
+async def judge_answer(
+    app,
+    question: str,
+    llm_response: str,
+    rubric: list[str],
+    question_type: str = "",
+) -> dict:
+    """Judge an answer via the answer_judge job (beam_rubric_judge_step)."""
+    judge_resp = await app.run_job(
+        "answer_judge",
+        llm_response=llm_response,
+        rubric=rubric,
+        probing_question=question,
+        question_type=question_type,
+    )
+    result = {
+        "llm_judge_score": (judge_resp.metadata or {}).get("llm_judge_score", 0.0),
+        "llm_judge_responses": (judge_resp.metadata or {}).get("llm_judge_responses", []),
+    }
+    # Include event_ordering extra metrics if present
+    eo = (judge_resp.metadata or {}).get("event_ordering")
+    if eo:
+        result["event_ordering"] = eo
+    return result
+
+
+# ---------------------------------------------------------------------------
+# Main evaluation pipeline
+# ---------------------------------------------------------------------------
+async def evaluate_case(eval_config: dict, case_id: str, eval_only: bool = False) -> dict:
+    """Evaluate a single BEAM case end-to-end.
+
+    Args:
+        eval_config: The evaluation configuration dict.
+        case_id: The case directory name (e.g. "1").
+        eval_only: If True, skip ingestion and only run query+judge
+            using the existing workspace.
+
+    Returns:
+        A results dict with all questions, answers, and judgments.
+    """
+    from reme import Application
+    from reme.config import resolve_app_config
+
+    dataset_cfg = eval_config["dataset"]
+    chat_size = dataset_cfg["chat_size"]
+    compress_session = bool(eval_config["evaluation"].get("compress_session", False))
+    beam_root = _PROJECT_ROOT / dataset_cfg.get("beam_root", "benchmark/beam/dataset/BEAM")
+    chat_path = beam_root / "chats" / chat_size / case_id / "chat.json"
+    probing_questions_path = beam_root / "chats" / chat_size / case_id / "probing_questions" / "probing_questions.json"
+
+    if not chat_path.exists():
+        raise FileNotFoundError(f"Chat file not found: {chat_path}")
+    if not probing_questions_path.exists():
+        raise FileNotFoundError(f"Probing questions not found: {probing_questions_path}")
+
+    logger.info(
+        "[Case %s] size=%s%s",
+        case_id,
+        chat_size,
+        " [eval_only]" if eval_only else "",
+    )
+
+    # Workspace setup
+    workspace_root = _PROJECT_ROOT / dataset_cfg.get("workspace_root", _WORKSPACE_ROOT_DEFAULT)
+    case_dir = workspace_root / f"{chat_size}_{case_id}"
+    workspace_dir = str(case_dir / ".reme")
+
+    if eval_only:
+        if not case_dir.exists() or not Path(workspace_dir).exists():
+            raise FileNotFoundError(
+                f"[Case {case_id}] eval_only: workspace not found at {case_dir}. "
+                f"Run without --eval_only first to build the workspace.",
+            )
+    else:
+        if case_dir.exists():
+            shutil.rmtree(case_dir)
+            logger.info(f"[Case {case_id}] Cleaned existing workspace: {case_dir}")
+        else:
+            logger.info(f"[Case {case_id}] Workspace not found, creating: {case_dir}")
+        case_dir.mkdir(parents=True, exist_ok=True)
+
+    # Pre-initialize ReMe's loguru logger with the correct log_dir
+    output_cfg = eval_config.get("output", {})
+    if output_cfg.get("log_to_file", False):
+        reme_log_dir = os.environ.get("REME_LOG_DIR")
+        if reme_log_dir:
+            from reme.utils import get_logger
+
+            get_logger(
+                log_dir=reme_log_dir,
+                level=os.environ.get("REME_LOG_LEVEL", "INFO"),
+                log_to_console=output_cfg.get("log_to_console", True),
+                log_to_file=True,
+                force_init=True,
+            )
+
+    cfg = resolve_app_config(
+        config=eval_config["reme"]["config"],
+        workspace_dir=workspace_dir,
+        log_to_console=output_cfg.get("log_to_console", True),
+        log_to_file=output_cfg.get("log_to_file", False),
+        enable_logo=False,
+    )
+
+    app = Application(**cfg)
+    await app.start()
+
+    from reme.utils.evaluation_interface import check_agent_token_usage  # noqa: E402
+
+    _MEM_AGENT_NAMES = ("default", "bench")
+    sessions_ingested = 0
+    memory_token_usage: dict[str, dict[str, int | None]] = {}
+    try:
+        if not eval_only:
+            # ── Phase 1: Ingest sessions (with token tracking) ─────────
+            sessions = load_beam_chat(chat_path, chat_size, case_id)
+            logger.info(f"[Case {case_id}] Loaded {len(sessions)} sessions from chat.json")
+
+            # Snapshot token counters before memory construction
+            mem_token_start = {name: check_agent_token_usage(name, app.context) for name in _MEM_AGENT_NAMES}
+
+            for i, session in enumerate(sessions):
+                logger.info(
+                    f"[Case {case_id}] Ingesting session {i+1}/{len(sessions)}: "
+                    f"id={session['session_id']} date={session['date']} "
+                    f"msgs={len(session['messages'])}",
+                )
+                resp = await app.run_job(
+                    "auto_memory",
+                    messages=session["messages"],
+                    session_id=session["session_id"],
+                    date=session["date"],
+                )
+                if not resp.success:
+                    logger.warning(f"[Case {case_id}] auto_memory failed: {resp.answer}")
+                else:
+                    logger.info(
+                        f"[Case {case_id}] auto_memory success: " f"{resp.answer[:100] if resp.answer else ''}",
+                    )
+                await app.run_job("index_update")
+                sessions_ingested += 1
+
+            # Final digest update
+            logger.info(f"[Case {case_id}] Running digest_update...")
+            await app.run_job("digest_update")
+            logger.info(f"[Case {case_id}] Ingestion complete.")
+
+            # Compute memory construction token deltas
+            for name in _MEM_AGENT_NAMES:
+                end_usage = check_agent_token_usage(name, app.context)
+                delta: dict[str, int | None] = {}
+                for metric in _TOKEN_USAGE_METRICS:
+                    current = end_usage[metric]
+                    start = mem_token_start[name][metric]
+                    delta[metric] = None if current is None else current - (start or 0)
+                memory_token_usage[name] = delta
+            logger.info(f"[Case {case_id}] Memory construction token usage: {memory_token_usage}")
+
+        # ── Phase 2: Answer + Judge probing questions ───────────────
+        with open(probing_questions_path, encoding="utf-8") as f:
+            probing_questions = json.load(f)
+
+        total_questions = sum(len(v) for v in probing_questions.values())
+        logger.info(f"[Case {case_id}] Total probing questions: {total_questions}")
+
+        all_question_results = []
+        q_idx = 0
+
+        for q_type in probing_questions:
+            logger.info(
+                f"[Case {case_id}] Question type: {q_type} " f"({len(probing_questions[q_type])} questions)",
+            )
+
+            for i, q in enumerate(probing_questions[q_type]):
+                q_idx += 1
+                question = q["question"]
+                rubric = q.get("rubric", [])
+                logger.info(
+                    f"[Case {case_id}] [{q_idx}/{total_questions}] " f"{q_type} Q{i+1}: {question[:100]}...",
+                )
+
+                q_result = {
+                    "question_type": q_type,
+                    "question_index": i,
+                    "question": question,
+                    "rubric": rubric,
+                }
+
+                # Agentic answer
+                try:
+                    agentic_answer, agentic_meta = await answer_question_agentic(
+                        app,
+                        question,
+                        compress_session=compress_session,
+                    )
+                except Exception as e:
+                    logger.error(f"[Case {case_id}] Agentic answer failed: {e}")
+                    agentic_answer = f"(error: {e})"
+                    agentic_meta = {"error": str(e)}
+
+                if not agentic_answer:
+                    agentic_answer = "(no answer generated)"
+                logger.info(f"[Case {case_id}] Agentic answer: {agentic_answer[:200]}...")
+                logger.info(
+                    f"[Case {case_id}] Agentic tool calls: {agentic_meta.get('tool_counts', {})}",
+                )
+                logger.info(f"[Case {case_id}] Bench token usage: {agentic_meta.get('token_usage', {})}")
+
+                # Judge agentic answer
+                logger.info(f"[Case {case_id}] Judging agentic ({q_type})...")
+                agentic_judgment = await judge_answer(
+                    app,
+                    question,
+                    agentic_answer,
+                    rubric,
+                    question_type=q_type,
+                )
+                logger.info(
+                    f"[Case {case_id}] Agentic score: " f"{agentic_judgment['llm_judge_score']:.3f}",
+                )
+
+                q_result["agentic_response"] = agentic_answer
+                q_result["agentic_judgment"] = agentic_judgment
+                q_result["agentic_metadata"] = agentic_meta
+
+                all_question_results.append(q_result)
+
+    finally:
+        await app.close()
+
+    return {
+        "case_id": case_id,
+        "chat_size": chat_size,
+        "sessions_ingested": sessions_ingested,
+        "total_questions": len(all_question_results),
+        "questions": all_question_results,
+        "memory_token_usage": memory_token_usage,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Worker: runs a single case in its own process with its own event loop
+# ---------------------------------------------------------------------------
+def _evaluate_case_worker(task_input: tuple) -> dict:
+    """Worker function for multiprocessing. Each process gets its own event loop."""
+    eval_config, case_id, log_level, reme_log_level, eval_only, log_dir = task_input
+    import asyncio  # pylint: disable=import-outside-toplevel
+
+    _configure_worker(log_level, reme_log_level, log_dir=log_dir)
+
+    # Suppress httpx GC noise
+    logging.getLogger("asyncio").setLevel(logging.CRITICAL)
+
+    return asyncio.run(evaluate_case(eval_config, case_id, eval_only=eval_only))
+
+
+def _indexed_worker(indexed_input: tuple) -> tuple:
+    """Module-level wrapper for imap_unordered with index tracking."""
+    idx, task_input = indexed_input
+    return idx, _evaluate_case_worker(task_input)
+
+
+def _resolve_num_workers(configured: int) -> int:
+    """Resolve num_workers: 0=auto (cpu_count-2, min 1), 1=sequential, >1=parallel."""
+    if configured == 0:
+        return max(1, (os.cpu_count() or 4) - 2)
+    return max(1, configured)
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+def main(  # pylint: disable=too-many-statements
+    config_path: str | None = None,
+    log_level: str = "INFO",
+    reme_log_level: str = "INFO",
+    eval_only: bool = False,
+):
+    """Run the BEAM evaluation pipeline.
+
+    Args:
+        config_path: Path to the YAML config file.
+        log_level: Log level for the eval runner.
+        reme_log_level: Log level for reme internal logs.
+        eval_only: If True, skip ingestion and only run query+judge using
+            existing workspaces.
+    """
+    from multiprocessing import Pool  # pylint: disable=import-outside-toplevel
+
+    # Load config BEFORE logging setup so log_dir is available
+    eval_config = load_eval_config(config_path)
+
+    # Resolve per-run log directory from config
+    output_cfg = eval_config.get("output", {})
+    log_dir_abs = None
+    if output_cfg.get("log_to_file", False):
+        log_dir_raw = output_cfg.get("log_dir", "logs")
+        log_prefix = output_cfg.get("log_prefix", "beam")
+        run_ts = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+        log_dir_abs = str(_PROJECT_ROOT / log_dir_raw / f"{log_prefix}_{run_ts}")
+
+    setup_logging(log_level, reme_log_level, log_dir=log_dir_abs)
+    dataset_cfg = eval_config["dataset"]
+    chat_size = dataset_cfg["chat_size"]
+    beam_root = _PROJECT_ROOT / dataset_cfg.get("beam_root", "benchmark/beam/dataset/BEAM")
+
+    # Determine which cases to run
+    case_ids = dataset_cfg.get("case_ids") or []
+    if not case_ids:
+        case_ids = get_available_cases(beam_root, chat_size)
+
+    # Pagination
+    start = dataset_cfg.get("start_index", 0)
+    num_items = dataset_cfg.get("num_items", 0)
+    if num_items > 0:
+        case_ids = case_ids[start : start + num_items]
+    elif start > 0:
+        case_ids = case_ids[start:]
+
+    if not case_ids:
+        logger.error(f"No cases found for chat_size={chat_size}")
+        return
+
+    logger.info(
+        "Evaluating %d case(s) for chat_size=%s: %s%s",
+        len(case_ids),
+        chat_size,
+        case_ids,
+        " [eval_only: query+judge only]" if eval_only else "",
+    )
+
+    # Resolve parallelism
+    num_workers = _resolve_num_workers(eval_config["evaluation"].get("num_workers", 1))
+    logger.info(f"Using {num_workers} worker(s)")
+
+    # Create output directory
+    output_dir = _PROJECT_ROOT / output_cfg.get("dir", "benchmark/beam/results")
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    # Create workspace root directory
+    workspace_root = _PROJECT_ROOT / dataset_cfg.get("workspace_root", _WORKSPACE_ROOT_DEFAULT)
+    workspace_root.mkdir(parents=True, exist_ok=True)
+
+    # Pre-check: verify all workspaces exist in eval_only mode
+    if eval_only:
+        missing_cases = []
+        for case_id in case_ids:
+            case_dir = workspace_root / f"{chat_size}_{case_id}"
+            if not case_dir.exists() or not (case_dir / ".reme").exists():
+                missing_cases.append(case_id)
+        if missing_cases:
+            preview = missing_cases[:10]
+            suffix = "..." if len(missing_cases) > 10 else ""
+            raise FileNotFoundError(
+                f"eval_only: {len(missing_cases)} workspace(s) not found under {workspace_root}. "
+                f"Missing cases: {preview}{suffix}. "
+                f"Run without --eval_only first to build the workspaces.",
+            )
+
+    # Build task args
+    task_args = [(eval_config, case_id, log_level, reme_log_level, eval_only, log_dir_abs) for case_id in case_ids]
+
+    # Progress tracking
+    total_items = len(task_args)
+    completed_count = [0]
+    start_time = time.time()
+    progress_lock = threading.Lock()
+
+    def _print_progress(prefix: str = "PROGRESS"):
+        elapsed = time.time() - start_time
+        elapsed_min = elapsed / 60
+        done = completed_count[0]
+        pct = 100.0 * done / total_items if total_items else 0
+        eta_str = "N/A"
+        if done > 0:
+            eta_sec = elapsed / done * (total_items - done)
+            eta_str = f"{eta_sec/60:.1f}min"
+        print(
+            f"[{prefix}] {datetime.now().strftime('%Y-%m-%d %H:%M:%S')} | "
+            f"{done}/{total_items} ({pct:.1f}%) completed | "
+            f"elapsed={elapsed_min:.1f}min | ETA={eta_str}",
+            flush=True,
+        )
+
+    def _progress_timer():
+        """Background thread: print progress every 10 minutes."""
+        while not _timer_stop.is_set():
+            _timer_stop.wait(600)
+            if not _timer_stop.is_set():
+                with progress_lock:
+                    _print_progress()
+
+    _timer_stop = threading.Event()
+    timer_thread = threading.Thread(target=_progress_timer, daemon=True)
+    timer_thread.start()
+
+    # Run evaluation
+    if num_workers == 1:
+        results = []
+        for task_input in task_args:
+            result = _evaluate_case_worker(task_input)
+            results.append(result)
+            with progress_lock:
+                completed_count[0] += 1
+    else:
+        results = [None] * total_items
+        indexed_args = list(enumerate(task_args))
+
+        with Pool(processes=num_workers) as pool:
+            for idx, result in pool.imap_unordered(_indexed_worker, indexed_args):
+                results[idx] = result
+                with progress_lock:
+                    completed_count[0] += 1
+
+    # Stop progress timer
+    _timer_stop.set()
+    timer_thread.join(timeout=2)
+
+    # Save results
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    output_file = output_dir / f"results_{chat_size}_{timestamp}.json"
+    with open(output_file, "w", encoding="utf-8") as f:
+        json.dump(results, f, ensure_ascii=False, indent=2)
+    logger.info(f"Results saved to {output_file}")
+
+    # Final progress
+    _print_progress("FINAL")
+
+    # Print concise summary
+    print("\n" + "=" * 70)
+    print(f"  BEAM EVALUATION RESULTS  |  size={chat_size}  cases={len(results)}")
+    print("=" * 70)
+
+    # Per-type stats (agentic only)
+    type_scores: dict[str, list[float]] = {}
+    type_binary_scores: dict[str, list[float]] = {}
+    all_scores: list[float] = []
+    all_binary_scores: list[float] = []
+    all_tool_call_totals: list[int] = []
+    all_token_usages: list[dict[str, int | None]] = []
+    all_memory_token_usages: list[dict[str, dict[str, int | None]]] = []
+
+    for case_result in results:
+        if "error" in case_result:
+            continue
+        mem_usage = case_result.get("memory_token_usage", {})
+        if mem_usage:
+            all_memory_token_usages.append(mem_usage)
+        for q in case_result.get("questions", []):
+            judgment = q.get("agentic_judgment", {})
+            score = judgment.get("llm_judge_score", 0.0)
+            # Binary: convert each rubric item score to 0/1, then average
+            judge_responses = judgment.get("llm_judge_responses", [])
+            if judge_responses:
+                binary_scores_per_item = [1.0 if r.get("score", 0) >= 1.0 else 0.0 for r in judge_responses]
+                binary_score = sum(binary_scores_per_item) / len(binary_scores_per_item)
+            else:
+                binary_score = 1.0 if score > 0.99 else 0.0
+            qtype = q["question_type"]
+            if qtype not in type_scores:
+                type_scores[qtype] = []
+                type_binary_scores[qtype] = []
+            type_scores[qtype].append(score)
+            type_binary_scores[qtype].append(binary_score)
+            all_scores.append(score)
+            all_binary_scores.append(binary_score)
+            metadata = q.get("agentic_metadata", {})
+            all_tool_call_totals.append(sum(metadata.get("tool_counts", {}).values()))
+            all_token_usages.append(metadata.get("token_usage", {}))
+
+    # Memory construction token usage summary
+    if all_memory_token_usages:
+        print("\n  ── Memory Construction Token Usage ──")
+        for agent_name in ("default", "bench"):
+            for metric in _TOKEN_USAGE_METRICS:
+                values = [
+                    usage[agent_name][metric]
+                    for usage in all_memory_token_usages
+                    if usage.get(agent_name, {}).get(metric) is not None
+                ]
+                if values:
+                    total = sum(values)
+                    mean, std = _mean_and_std(values)
+                    print(
+                        f"    {agent_name}/{metric}: total={total} mean={mean:.2f} std={std:.2f} ({len(values)} cases)",
+                    )
+                else:
+                    print(f"    {agent_name}/{metric}: unavailable")
+        print()
+
+    print("\n  ── AGENTIC ──")
+    if all_scores:
+        for qtype in sorted(type_scores.keys()):
+            scores = type_scores[qtype]
+            avg = sum(scores) / len(scores) if scores else 0
+            bin_scores = type_binary_scores[qtype]
+            bin_avg = sum(bin_scores) / len(bin_scores) if bin_scores else 0
+            print(f"    {qtype:<40s}: {avg:.3f}  binary={bin_avg:.3f}  ({len(scores)} Qs)")
+        overall = sum(all_scores) / len(all_scores) if all_scores else 0
+        binary_overall = sum(all_binary_scores) / len(all_binary_scores) if all_binary_scores else 0
+        print(f"    {'-'*38}")
+        print(f"    {'OVERALL':<40s}: {overall:.3f}  binary={binary_overall:.3f}  ({len(all_scores)} Qs)")
+        tool_call_mean, tool_call_std = _mean_and_std(all_tool_call_totals)
+        print(f"    Tool calls/query: mean={tool_call_mean:.2f} std={tool_call_std:.2f}")
+        print("    Bench reported tokens/query:")
+        for metric in _TOKEN_USAGE_METRICS:
+            values = [usage[metric] for usage in all_token_usages if usage.get(metric) is not None]
+            if values:
+                mean, std = _mean_and_std(values)
+                print(f"      {metric}: mean={mean:.2f} std={std:.2f}")
+            else:
+                print(f"      {metric}: unavailable")
+    else:
+        print("    (no results)")
+
+    # Per-case summary
+    print("\n  ── Per-Case Summary ──")
+    for case_result in results:
+        case_id = case_result["case_id"]
+        if "error" in case_result:
+            print(f"    Case {case_id}: ERROR — {case_result['error']}")
+            continue
+        n_qs = case_result.get("total_questions", 0)
+        n_sessions = case_result.get("sessions_ingested", 0)
+        mem_usage = case_result.get("memory_token_usage", {})
+        parts = [f"Case {case_id}: {n_sessions} sessions, {n_qs} questions"]
+        # Append memory construction total tokens if available
+        for agent_name in ("default", "bench"):
+            agent_usage = mem_usage.get(agent_name, {})
+            total = agent_usage.get("total_tokens")
+            if total is not None:
+                parts.append(f"mem_{agent_name}_tokens={total}")
+        questions = case_result.get("questions", [])
+        scores = [q.get("agentic_judgment", {}).get("llm_judge_score", 0.0) for q in questions]
+        if scores:
+            avg = sum(scores) / len(scores)
+            # Binary: 0/1 per rubric item, average per question, then across questions
+            bin_scores = []
+            for q in questions:
+                judge_responses = q.get("agentic_judgment", {}).get("llm_judge_responses", [])
+                if judge_responses:
+                    item_bins = [1.0 if r.get("score", 0) >= 1.0 else 0.0 for r in judge_responses]
+                    bin_scores.append(sum(item_bins) / len(item_bins))
+                else:
+                    s = q.get("agentic_judgment", {}).get("llm_judge_score", 0.0)
+                    bin_scores.append(1.0 if s > 0.99 else 0.0)
+            bin_avg = sum(bin_scores) / len(bin_scores)
+            parts.append(f"agentic={avg:.3f} binary={bin_avg:.3f}")
+        print(f"    {' | '.join(parts)}")
+
+    print("=" * 70)
+    total_elapsed = time.time() - start_time
+    print(f"\n  Total time: {total_elapsed/60:.1f} min")
+    print("\n" + "=" * 70)
+    print("  [DONE] BEAM EVALUATION COMPLETED SUCCESSFULLY")
+    print("=" * 70 + "\n")
+
+
+_TOKEN_USAGE_METRICS = (
+    "input_tokens",
+    "output_tokens",
+    "total_tokens",
+)
+
+
+def _mean_and_std(values: list[int]) -> tuple[float, float]:
+    """Return population mean and standard deviation for one per-question metric."""
+    if not values:
+        return 0.0, 0.0
+    mean = sum(values) / len(values)
+    return mean, (sum((value - mean) ** 2 for value in values) / len(values)) ** 0.5
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser(description="BEAM evaluation runner")
+    parser.add_argument("--config", type=str, default=None, help="Path to config.yaml")
+    parser.add_argument(
+        "--log-level",
+        type=str,
+        default="INFO",
+        choices=["DEBUG", "INFO", "WARNING", "ERROR"],
+        help="Log level for the eval runner (default: INFO)",
+    )
+    parser.add_argument(
+        "--reme-log-level",
+        type=str,
+        default="INFO",
+        choices=["DEBUG", "INFO", "WARNING", "ERROR"],
+        help="Log level for reme internal logs — loguru (default: INFO)",
+    )
+    parser.add_argument(
+        "-q",
+        "--quiet",
+        action="store_true",
+        help="Shortcut for --log-level WARNING --reme-log-level WARNING",
+    )
+    parser.add_argument(
+        "--eval_only",
+        action="store_true",
+        help="Skip ingestion. Reuse existing workspaces and only run query+judge.",
+    )
+    args = parser.parse_args()
+
+    if args.quiet:
+        args.log_level = "WARNING"
+        args.reme_log_level = "WARNING"
+
+    main(args.config, args.log_level, args.reme_log_level, eval_only=args.eval_only)

benchmark/bfcl/bfcl_agent.py

@@ -1,726 +0,0 @@
-# flake8: noqa: E402
-# pylint: disable=too-many-return-statements
-"""A minimal ReAct Agent for BFCL-v3(multi-turn) tasks."""
-
-import re
-import os
-import time
-import json
-import warnings
-import tempfile
-import datetime
-from pathlib import Path
-from typing import Dict, List, Any
-
-import ray
-import requests
-from tqdm import tqdm
-from loguru import logger
-from openai import OpenAI
-from dotenv import load_dotenv
-
-from bfcl_utils import (
-    load_test_case,
-    handle_user_turn,
-    handle_tool_calls,
-    extract_tool_schema,
-    extract_single_turn_response,
-    extract_multi_turn_responses,
-    capture_and_print_score_files,
-    create_error_response,
-)
-from bfcl_eval.model_handler.api_inference.qwen import QwenAPIHandler
-from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_utils import (
-    is_empty_execute_response,
-)
-from bfcl_eval.eval_checker.eval_runner import (
-    multi_turn_runner,
-    ast_file_runner,
-)
-from bfcl_eval.eval_checker.eval_runner_helper import record_cost_latency
-from bfcl_eval.utils import (
-    is_multi_turn,
-    is_relevance_or_irrelevance,
-    find_file_with_suffix,
-    load_file,
-)
-
-os.environ["BFCL_DATA_PATH"] = "data/multiturn_data_base_val.jsonl"
-os.environ["BFCL_ANSWER_PATH"] = "data/possible_answer"
-load_dotenv("../../.env")
-
-
-@ray.remote
-class BFCLAgent:
-    """A minimal ReAct Agent for BFCL-v3(multi-turn) tasks."""
-
-    def __init__(
-        self,
-        index: int,
-        task_ids: List[str],
-        experiment_name: str,
-        data_path: str = os.getenv("BFCL_DATA_PATH"),
-        answer_path: Path = Path(os.getenv("BFCL_ANSWER_PATH")),
-        model_name: str = "qwen3-8b",
-        temperature: float = 0.9,
-        max_interactions: int = 30,
-        max_response_size: int = 2000,
-        num_trials: int = 1,
-        enable_thinking: bool = False,
-        use_memory: bool = False,
-        use_memory_addition: bool = False,
-        use_memory_deletion: bool = False,
-        delete_freq: int = 10,
-        freq_threshold: int = 5,
-        utility_threshold: float = 0.5,
-        memory_base_url: str = "http://0.0.0.0:8002/",
-    ):
-
-        self.index: int = index
-        self.task_ids: List[str] = task_ids
-        self.categories: List[str] = [task_id.rsplit("_", 1)[0] if "_" in task_id else task_id for task_id in task_ids]
-        self.experiment_name: str = experiment_name
-        self.data_path: str = data_path
-        self.answer_path: Path = answer_path
-        self.model_name: str = model_name
-        self.temperature: float = temperature
-        self.max_interactions: int = max_interactions
-        self.max_response_size: int = max_response_size
-        self.num_trials: int = num_trials
-        self.enable_thinking: bool = enable_thinking
-        self.use_memory: bool = use_memory
-        self.use_memory_addition: bool = use_memory_addition if use_memory else False
-        self.use_memory_deletion: bool = use_memory_deletion if use_memory else False
-        self.delete_freq: int = delete_freq
-        self.freq_threshold: int = freq_threshold
-        self.utility_threshold: float = utility_threshold
-        self.memory_base_url: str = memory_base_url
-
-        self.llm_client = OpenAI()
-
-        self.history: List[List[List[dict]]] = [[] for _ in range(num_trials)]
-        self.retrieved_memory_list: List[List[List[Any]]] = [[] for _ in range(num_trials)]
-        self.test_entry: List[List[Dict[str, Any]]] = [[] for _ in range(num_trials)]
-        self.original_test_entry: List[List[Dict[str, Any]]] = [[] for _ in range(num_trials)]
-        self.tool_schema: List[List[List[dict]]] = [[] for _ in range(num_trials)]
-        self.current_turn = [[0 for _ in range(len(task_ids))] for _ in range(num_trials)]
-
-        for run_id in range(num_trials):
-            for task_index in range(len(task_ids)):
-                self.init_state(run_id, task_index)
-
-    def init_state(self, run_id, i) -> Dict[str, Any]:
-        """Initialize the state of the agent."""
-        self.test_entry[run_id].append(load_test_case(self.data_path, self.task_ids[i]))
-        self.original_test_entry[run_id].append(self.test_entry[run_id][i].get("extra", {}))
-        self.tool_schema[run_id].append(extract_tool_schema(self.test_entry[run_id][i].get("tools", [{}])))
-
-        msg = self.test_entry[run_id][i].get("messages", [])
-        self.history[run_id].append(msg)
-        self.retrieved_memory_list[run_id].append([])
-        self.current_turn[run_id][i] = 1
-
-    def update_task_history_with_memory(self, run_id, task_index, previous_memories: None):
-        """Update the task history with memory."""
-        query = self.history[run_id][task_index][0]["content"]
-        if len(previous_memories) == 0:
-            response = self.get_memory(query)
-            if response and "memory_list" in response["metadata"]:
-                self.retrieved_memory_list[run_id][task_index] = response["metadata"]["memory_list"]
-                task_memory = re.sub(r"\bMemory\s*(\d+)\s*[:]", r"Experience \1 :", response["answer"])
-                logger.info(f"loaded task_memory: {task_memory}")
-                self.history[run_id][task_index][0] = self.get_query_with_memory(query, task_memory)
-        else:
-            formatted_memories = []
-            for i, memory in enumerate(previous_memories, 1):
-                condition = memory["when_to_use"]
-                memory_content = memory["content"]
-                memory_text = f"Experience {i} :\n When to use: {condition}\n Content: {memory_content}\n"
-                formatted_memories.append(memory_text)
-            self.history[run_id][task_index][0] = self.get_query_with_memory(query, "\n".join(formatted_memories))
-
-    def get_query_with_memory(self, query: str, memory: str):
-        """Get the query with memory."""
-        return {
-            "role": "user",
-            "content": "Task:\n" + query + "\n\nSome Related Experience to help you to complete the task:\n" + memory,
-        }
-
-    def get_query_without_experience(self, query: str):
-        """Get the query without experience."""
-        if "\n\nSome Related Experience" in query:
-            query = query.split("\n\nSome Related Experience")[0].split("Task:\n")[-1]
-        return query
-
-    def get_traj_from_task_history(self, task_id: str, task_history: list, reward: float):
-        """Get the trajectory from the task history."""
-        return {
-            "task_id": task_id,
-            "messages": task_history,
-            "score": reward,
-        }
-
-    def handle_api_response(self, response: requests.Response):
-        """Handle API response with proper error checking"""
-        if response.status_code != 200:
-            print(f"Error: {response.status_code}")
-            print(response.text)
-            return None
-
-        return response.json()
-
-    def get_memory(self, query: str):
-        """Retrieve relevant task memories based on a query"""
-        response = requests.post(
-            url=f"{self.memory_base_url}retrieve_task_memory",
-            json={
-                "query": query,
-                "enable_llm_rerank": False,
-                "enable_score_filter": False,
-                "top_k": 5,
-                "enable_llm_rewrite": False,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return None
-
-        logger.info(f"query: {query}, response: {result}")
-        return result
-
-    def summary_memory(self, trajectories):
-        """Generate a summary of conversation messages and create task memories"""
-        response = requests.post(
-            url=f"{self.memory_base_url}summary_task_memory",
-            json={
-                "trajectories": trajectories,
-                "success_threshold": 1.0,
-                "enable_soft_comparison": True,
-                "validation_threshold": 0.5,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return []
-
-        # Extract memory list from response
-        memory_list = result.get("metadata", {}).get("memory_list", [])
-        logger.info(f"add new memories: {memory_list}")
-        return memory_list
-
-    def add_memory(self, memory_list):
-        """Add the memory to the memory pool."""
-        response = requests.post(
-            url=f"{self.memory_base_url}add_task_memory",
-            json={
-                "memory_list": memory_list,
-            },
-        )
-        response.raise_for_status()
-
-    def update_memory_information(self, memory_list, update_utility: bool = False):
-        """Update the memory information."""
-        response = requests.post(
-            url=f"{self.memory_base_url}record_task_memory",
-            json={
-                "memory_list": memory_list,
-                "update_utility": update_utility,
-            },
-        )
-        response.raise_for_status()
-        logger.info(response.json())
-
-    def delete_memory(self):
-        """Delete the memory from the memory pool."""
-        response = requests.post(
-            url=f"{self.memory_base_url}delete_task_memory",
-            json={
-                "freq_threshold": self.freq_threshold,
-                "utility_threshold": self.utility_threshold,
-            },
-        )
-        response.raise_for_status()
-
-    def call_llm(self, messages: list, tool_schemas: list[dict]) -> str:
-        """Call the LLM."""
-        for i in range(100):
-            try:
-                response = self.llm_client.chat.completions.create(
-                    model=self.model_name,
-                    messages=messages,
-                    tools=tool_schemas,
-                    temperature=self.temperature,
-                    seed=0,
-                    extra_body={"enable_thinking": self.enable_thinking},
-                    stream=self.enable_thinking,
-                    parallel_tool_calls=True,
-                )
-                if not self.enable_thinking:
-                    out_msg = response.choices[0].message
-                    return out_msg.model_dump(exclude_unset=True, exclude_none=True)
-                else:
-                    reasoning_content = ""  # Complete reasoning process
-                    answer_content = ""  # Define complete response
-                    tool_info = []  # Store tool invocation information
-                    is_answering = (
-                        False  # Determine whether the reasoning process has finished and response has started
-                    )
-
-                    for chunk in response:
-                        if not chunk.choices:
-                            # Handle usage information
-                            continue
-
-                        delta = chunk.choices[0].delta
-                        # Handle AI's thought process (chain reasoning)
-                        if hasattr(delta, "reasoning_content") and delta.reasoning_content is not None:
-                            reasoning_content += delta.reasoning_content
-                        # Handle final response content
-                        else:
-                            if not is_answering:  # Print title when entering the response phase for the first time
-                                is_answering = True
-                            if delta.content is not None:
-                                answer_content += delta.content
-
-                            # Handle tool invocation information (support parallel tool calls)
-                            if delta.tool_calls is not None:
-                                for tool_call in delta.tool_calls:
-                                    index = tool_call.index  # Tool call index, used for parallel calls
-
-                                    # Dynamically expand tool information storage list
-                                    while len(tool_info) <= index:
-                                        tool_info.append(
-                                            {
-                                                "id": "",
-                                                "type": "function",
-                                                "index": index,
-                                                "function": {"name": "", "arguments": ""},
-                                            },
-                                        )
-
-                                    # Collect tool call ID (used for subsequent function calls)
-                                    if tool_call.id:
-                                        tool_info[index]["id"] += tool_call.id
-
-                                    # Collect function name (used for subsequent routing to specific functions)
-                                    if tool_call.function and tool_call.function.name:
-                                        tool_info[index]["function"]["name"] += tool_call.function.name
-
-                                    # Collect function parameters (in JSON string format, need subsequent parsing)
-                                    if tool_call.function and tool_call.function.arguments:
-                                        tool_info[index]["function"]["arguments"] += tool_call.function.arguments
-                    msg = {
-                        "role": "assistant",
-                        "content": answer_content,
-                        "reasoning_content": reasoning_content,
-                    }
-                    if tool_info:
-                        msg["tool_calls"] = tool_info
-                    return msg
-            except Exception as e:
-                logger.exception(f"encounter error with {e.args}")
-                time.sleep(1 + i * 10)
-
-        return "call llm error"
-
-    def env_step(self, run_id: int, index: int, messages: str) -> str:
-        """
-        Process one step in the conversation.
-        Both single turn and multi turn are supported.
-
-        Args:
-            messages: List of conversation messages, with the last one being assistant response
-            test_entry: Test entry containing initial_config, involved_classes, question etc.
-            **kwargs: Additional arguments for compatibility
-
-        Returns:
-            Dict containing next message and tools if applicable
-        """
-        try:
-            if not messages:
-                return handle_user_turn(self.original_test_entry[run_id][index], self.current_turn[run_id][index])
-
-            if messages[-1]["role"] != "assistant":
-                return create_error_response(
-                    "Last message must be from assistant",
-                )
-
-            if "tool_calls" in messages[-1] and len(messages[-1]["tool_calls"]) > 0:
-                try:
-                    tool_calls = messages[-1]["tool_calls"]
-                    decoded_calls = self._convert_tool_calls_to_execution_format(
-                        tool_calls,
-                    )
-                    # decoded_calls:[function(param=xxx)]
-                    print(f"decoded_calls: {decoded_calls}")
-                    if is_empty_execute_response(decoded_calls):
-                        warnings.warn(
-                            f"is_empty_execute_response: {is_empty_execute_response(decoded_calls)}",
-                        )
-                        return handle_user_turn(
-                            self.original_test_entry[run_id][index],
-                            self.current_turn[run_id][index],
-                        )
-                    return handle_tool_calls(
-                        tool_calls,
-                        decoded_calls,
-                        self.original_test_entry[run_id][index],
-                        self.current_turn[run_id][index],
-                    )
-                except Exception as e:
-                    warnings.warn(f"Errors during tool invocation: {str(e)}")
-                    return handle_user_turn(self.original_test_entry[run_id][index], self.current_turn[run_id][index])
-            else:
-                return handle_user_turn(self.original_test_entry[run_id][index], self.current_turn[run_id][index])
-
-        except Exception as e:
-            return create_error_response(f"Failed to process request: {str(e)}")
-
-    def _convert_tool_calls_to_execution_format(
-        self,
-        tool_calls: List[Dict[str, Any]],
-    ) -> List[str]:
-        """
-        Convert OpenAI format tool calls to execution format.
-
-        Args:
-            tool_calls: List of tool calls in OpenAI format
-
-        Returns:
-            List of function calls in string format
-        """
-        execution_list = []
-
-        for tool_call in tool_calls:
-            function = tool_call.get("function", {})
-            function_name = function.get("name", "")
-
-            try:
-                arguments = function.get("arguments", "{}")
-                if isinstance(arguments, str):
-                    args_dict = json.loads(arguments)
-                else:
-                    args_dict = arguments
-
-                args_str = ", ".join([f"{k}={repr(v)}" for k, v in args_dict.items()])
-                execution_list.append(f"{function_name}({args_str})")
-
-            except Exception:
-                execution_list.append(f"{function_name}()")
-
-        return execution_list
-
-    def get_reward(self, run_id, index) -> float:
-        """Get the reward."""
-        try:
-            if not self.history[run_id][index] or not self.original_test_entry[run_id][index]:
-                return 0.0
-
-            model_name = "env_handler"
-            handler = QwenAPIHandler(
-                model_name,
-                temperature=1.0,
-            )  # FIXME: magic number
-
-            model_result_data = self._convert_conversation_to_eval_format(run_id, index)
-
-            prompt_data = [self.original_test_entry[run_id][index]]
-
-            state = {"leaderboard_table": {}}
-            record_cost_latency(
-                state["leaderboard_table"],
-                model_name,
-                [model_result_data],
-            )
-
-            if is_relevance_or_irrelevance(self.categories[index]):
-                accuracy, _ = self._eval_relevance_test(
-                    handler,
-                    model_result_data,
-                    prompt_data,
-                    model_name,
-                    self.category,
-                )
-            else:
-                # Find the corresponding possible answer file
-
-                possible_answer_file = find_file_with_suffix(
-                    self.answer_path,
-                    self.categories[index],
-                )
-                possible_answer = load_file(possible_answer_file, sort_by_id=True)
-                possible_answer = [item for item in possible_answer if item["id"] == self.task_ids[index]]
-                if is_multi_turn(self.categories[index]):
-                    accuracy, _ = self._eval_multi_turn_test(
-                        handler,
-                        model_result_data,
-                        prompt_data,
-                        possible_answer,
-                        model_name,
-                        self.categories[index],
-                    )
-                else:
-                    accuracy, _ = self._eval_single_turn_test(
-                        handler,
-                        model_result_data,
-                        prompt_data,
-                        possible_answer,
-                        model_name,
-                        self.categories[index],
-                    )
-            print(f"model_result_data: {model_result_data}")
-            if possible_answer:
-                print(f"possible_answer: {possible_answer}")
-            else:
-                print("possible_answer: None")
-
-            return accuracy
-
-        except Exception:
-            import traceback
-
-            traceback.print_exc()
-            return 0
-
-    def _convert_conversation_to_eval_format(self, run_id, index) -> Dict[str, Any]:
-        """
-        Convert conversation history to evaluation format.
-
-        Args:
-            conversation_result: Result from run_conversation
-            original_test_entry: Original test entry data
-
-        Returns:
-            Data in format expected by multi_turn_runner or other runners
-        """
-        if is_multi_turn(self.categories[index]):
-            turns_data = extract_multi_turn_responses(self.history[run_id][index])
-        else:
-            turns_data = extract_single_turn_response(self.history[run_id][index])
-
-        model_result_data = {
-            "id": self.task_ids[index],
-            "result": turns_data,
-            "latency": 0,
-            "input_token_count": 0,
-            "output_token_count": 0,
-        }
-
-        return model_result_data
-
-    def _eval_multi_turn_test(
-        self,
-        handler,
-        model_result_data,
-        prompt_data,
-        possible_answer,
-        model_name,
-        test_category,
-    ):
-        """
-        Evaluate multi-turn test.
-
-        Args:
-            handler: Model handler instance
-            model_result_data: Model result data
-            prompt_data: Prompt data
-            possible_answer: Possible answer data
-            model_name: Name of the model
-            test_category: Category of the test
-
-        Returns:
-            Tuple of (accuracy, total_count)
-        """
-        with tempfile.TemporaryDirectory() as temp_dir:
-            score_dir = Path(temp_dir)
-            accuracy, total_count = multi_turn_runner(
-                handler=handler,
-                model_result=[model_result_data],
-                prompt=prompt_data,
-                possible_answer=possible_answer,
-                model_name=model_name,
-                test_category=test_category,
-                score_dir=score_dir,
-            )
-            capture_and_print_score_files(
-                score_dir,
-                model_name,
-                test_category,
-                "multi_turn",
-            )
-            return accuracy, total_count
-
-    def _eval_single_turn_test(
-        self,
-        handler,
-        model_result_data,
-        prompt_data,
-        possible_answer,
-        model_name,
-        test_category,
-    ):
-        """
-        Evaluate single-turn AST test.
-
-        Args:
-            handler: Model handler instance
-            model_result_data: Model result data
-            prompt_data: Prompt data
-            possible_answer: Possible answer data
-            model_name: Name of the model
-            test_category: Category of the test
-
-        Returns:
-            Tuple of (accuracy, total_count)
-        """
-        language = "Python"
-        if "java" in test_category.lower():
-            language = "Java"
-        elif "js" in test_category.lower() or "javascript" in test_category.lower():
-            language = "JavaScript"
-
-        with tempfile.TemporaryDirectory() as temp_dir:
-            score_dir = Path(temp_dir)
-            accuracy, total_count = ast_file_runner(
-                handler=handler,
-                model_result=[model_result_data],
-                prompt=prompt_data,
-                possible_answer=possible_answer,
-                language=language,
-                test_category=test_category,
-                model_name=model_name,
-                score_dir=score_dir,
-            )
-            capture_and_print_score_files(
-                score_dir,
-                model_name,
-                test_category,
-                "single_turn",
-            )
-            return accuracy, total_count
-
-    def execute(self):
-        """Execute the agent."""
-        result = []
-        counter = 0
-        for task_index, task_id in enumerate(tqdm(self.task_ids, desc=f"ray_index={self.index}")):
-            t_result = None
-            previous_memories = []
-            for run_id in range(self.num_trials):
-                try:
-                    start_time = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-                    for i in range(self.max_interactions):
-                        if self.use_memory and i == 0:
-                            self.update_task_history_with_memory(run_id, task_index, previous_memories)
-                        llm_output = self.call_llm(
-                            self.history[run_id][task_index],
-                            self.tool_schema[run_id][task_index],
-                        )
-                        self.history[run_id][task_index].append(llm_output)
-
-                        env_output = self.env_step(run_id, task_index, self.history[run_id][task_index])
-                        # Possible env_output returns after environment interaction:
-                        # 1. Triggers a query with available tools list:
-                        #    {"messages": [{"role": "user", "content": user_query}], "tools": tools}
-                        # 2. Returns tool invocation result: {"messages":
-                        #    [{"role": "tool", "content": {<exec_results>}, 'tool_call_id': 'chatcmpl-tool-xxx'}]}
-                        #    <exec_results>: when success, returns result dicts, e.g., {"travel_cost_list": [x]},
-                        #    when error, returns error message,
-                        #    e.g., {"error": "cd: temporary: No such directory. You cannot use path ..."}
-                        # 3. Conversation completion:
-                        #    {"messages": [{"role": "env", "content": "[CONVERSATION_COMPLETED]"}]}
-                        # 4. Program error: {"messages": [{"role": "env", "content": f"[ERROR] {error_message}"}]}
-
-                        # tool_list update
-                        if "tools" in env_output:
-                            self.tool_schema[run_id][task_index] = extract_tool_schema(env_output["tools"])
-
-                        new_tool_calls = []
-                        new_tool_call_ids = []
-                        next_user_msg = ""
-                        for idx, msg in enumerate(env_output.get("messages", [])):
-                            if msg["role"] == "tool" and len(msg["content"]) > 0:
-                                new_tool_calls.append(msg.get("content", ""))
-                                new_tool_call_ids.append(msg.get("tool_call_id", ""))
-                            elif msg["role"] == "user":
-                                next_user_msg = msg.get("content", "")
-                                self.current_turn[run_id][task_index] += 1
-                            else:  # for env role messages
-                                next_user_msg = msg.get("content", "")
-
-                        if new_tool_calls:
-                            for idx, call in enumerate(new_tool_calls):
-                                self.history[run_id][task_index].append(
-                                    {"role": "tool", "content": str(call), "tool_call_id": new_tool_call_ids[idx]},
-                                )
-                        else:
-                            self.history[run_id][task_index].append({"role": "user", "content": next_user_msg})
-
-                        logger.info(f"index={self.index} task_id={task_id} iteration={i}")
-
-                        if self.task_completed(run_id, task_index):
-                            break
-
-                    reward = self.get_reward(run_id, task_index)
-                    if self.use_memory:
-                        if self.use_memory_addition:
-                            new_traj_list = [
-                                self.get_traj_from_task_history(task_id, self.history[run_id][task_index], reward),
-                            ]
-                            previous_memories = self.summary_memory(new_traj_list)
-                            if reward == 1:
-                                self.add_memory(previous_memories)
-
-                        # update the freq & utility attributes of retrieved memories
-                        update_utility: bool = reward == 1
-                        self.update_memory_information(self.retrieved_memory_list[run_id][task_index], update_utility)
-
-                    counter += 1
-                    if self.use_memory_deletion and counter % self.delete_freq == 0:
-                        self.delete_memory()
-
-                    t_result = {
-                        "run_id": run_id,
-                        "task_id": self.task_ids[task_index],
-                        "experiment_name": self.experiment_name,
-                        "task_completed": self.task_completed(run_id, task_index),
-                        "reward": reward,
-                        "task_history": self.history[run_id][task_index],
-                        "task_start_time": start_time,
-                    }
-                    if reward == 1:
-                        break
-
-                except Exception as e:
-                    logger.exception(f"encounter error with {e.args}")
-            result.append(t_result)
-        return result
-
-    def task_completed(self, run_id, index):
-        """
-        Check if task is completed.
-
-        Returns:
-            True if task is completed, False otherwise
-        """
-        return self.history[run_id][index][-1]["content"] == "[CONVERSATION_COMPLETED]"
-
-
-def main():
-    """Main function to run the BFCLAgent."""
-    with open(os.getenv("BFCL_DATA_PATH"), "r", encoding="utf-8") as f:
-        task_ids = [json.loads(l)["id"] for l in f]
-    dataset_name = "dev"
-    agent = BFCLAgent(
-        index=0,
-        task_ids=[task_ids[0]],
-        experiment_name=f"qwen3_8b_{dataset_name}",
-    )
-    result = agent.execute()
-    logger.info(f"result={json.dumps(result)}")
-
-
-if __name__ == "__main__":
-    main()

benchmark/bfcl/bfcl_utils.py

@@ -1,399 +0,0 @@
-"""Utils for evaluation on BFCL tasks"""
-
-import json
-from pathlib import Path
-from typing import Dict, List, Any
-
-from bfcl_eval.constants.default_prompts import (
-    DEFAULT_USER_PROMPT_FOR_ADDITIONAL_FUNCTION_FC,
-)
-from bfcl_eval.constants.type_mappings import GORILLA_TO_OPENAPI
-from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_utils import (
-    execute_multi_turn_func_call,
-)
-from bfcl_eval.model_handler.model_style import ModelStyle
-from bfcl_eval.model_handler.utils import (
-    convert_to_tool,
-    default_decode_execute_prompting,
-    func_doc_language_specific_pre_processing,
-)
-
-
-def load_test_case(data_path: str, test_id: str | None) -> Dict[str, Any]:
-    """
-    load test cases by id
-    """
-    if not Path(data_path).exists():
-        raise FileNotFoundError(f"BFCL data file '{data_path}' not found")
-
-    if test_id is None:
-        raise ValueError("task_id is required")
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        if str(test_id).isdigit():  # pylint: disable=R1720
-            idx = int(test_id)
-            for line_no, line in enumerate(f):
-                if line_no == idx:
-                    return json.loads(line)
-            raise ValueError(f"Test case index {idx} not found in {data_path}")
-        else:
-            for line in f:
-                data = json.loads(line)
-                if data.get("id") == test_id:
-                    return data
-            raise ValueError(f"Test case id '{test_id}' not found in {data_path}")
-
-
-def handle_user_turn(
-    test_entry: Dict[str, Any],
-    current_turn: int,
-) -> Dict[str, Any]:
-    """
-    Handle user turn by returning appropriate content from test_entry["question"].
-    For non-first turns, processes user query and tools.
-
-    Args:
-        test_entry: Test entry containing conversation data
-        current_turn: Current turn number
-
-    Returns:
-        Response containing next user message and tools
-    """
-    try:
-        current_turn_message = []
-        tools = compile_tools(test_entry)
-        questions = test_entry.get("question", [])
-        holdout_function = test_entry.get("holdout_function", {})
-
-        if str(current_turn) in holdout_function:
-            test_entry["function"].extend(holdout_function[str(current_turn)])
-            tools = compile_tools(test_entry)
-            assert len(questions[current_turn]) == 0, "Holdout turn should not have user message."
-            current_turn_message = [
-                {
-                    "role": "user",
-                    "content": DEFAULT_USER_PROMPT_FOR_ADDITIONAL_FUNCTION_FC,
-                },
-            ]
-            return create_user_response(current_turn_message, tools)
-        if current_turn >= len(questions):
-            return create_completion_response()
-
-        current_turn_message = questions[current_turn]
-
-        return create_user_response(current_turn_message, tools)
-
-    except Exception as e:
-        return create_error_response(f"Failed to process user message: {str(e)}")
-
-
-def handle_tool_calls(  # pylint: disable=W0613
-    tool_calls: List[Dict[str, Any]],
-    decoded_calls: list[str],
-    test_entry: Dict[str, Any],
-    current_turn: int,
-) -> Dict[str, Any]:
-    """
-    Handle tool calls from assistant.
-
-    Args:
-        tool_calls: List of tool calls in OpenAI format
-        decoded_calls: List of decoded function calls
-        test_entry: Test entry containing environment data
-        current_turn: Current turn number
-
-    Returns:
-        Response containing tool execution results
-    """
-    execution_results, _ = execute_multi_turn_func_call(
-        func_call_list=decoded_calls,
-        initial_config=test_entry["initial_config"],
-        involved_classes=test_entry["involved_classes"],
-        model_name="env_handler",
-        test_entry_id=test_entry["id"],
-        long_context=("long_context" in test_entry["id"] or "composite" in test_entry["id"]),
-        is_evaL_run=False,
-    )
-    # print('execution_results in handler_tool_calls:', execution_results)
-
-    return create_tool_response(tool_calls, execution_results)
-
-
-def compile_tools(test_entry: dict) -> list:
-    """
-    Compile functions into tools format.
-
-    Args:
-        test_entry: Test entry containing functions
-
-    Returns:
-        List of tools in OpenAI format
-    """
-    functions: list = test_entry["function"]
-    test_category: str = test_entry["id"].rsplit("_", 1)[0]
-
-    functions = func_doc_language_specific_pre_processing(functions, test_category)
-    tools = convert_to_tool(functions, GORILLA_TO_OPENAPI, ModelStyle.OpenAI_Completions)
-
-    return tools
-
-
-def create_tool_response(
-    tool_calls: List[Dict[str, Any]],
-    execution_results: List[str],
-) -> Dict[str, Any]:
-    """
-    Create response for tool calls.
-
-    Args:
-        tool_calls: List of tool calls
-        execution_results: List of execution results
-
-    Returns:
-        Response containing tool execution results
-    """
-    tool_messages = []
-    for i, (tool_call, result) in enumerate(zip(tool_calls, execution_results)):
-        tool_messages.append(
-            {
-                "role": "tool",
-                "content": result,
-                "tool_call_id": tool_call.get("id", f"call_{i}"),
-            },
-        )
-
-    return {"messages": tool_messages}
-
-
-def create_user_response(
-    question_turn: List[Dict[str, Any]],
-    tools: List[Dict[str, Any]],
-) -> Dict[str, Any]:
-    """
-    Create response containing user message.
-
-    Args:
-        question_turn: List of messages for current turn
-        tools: List of available tools
-
-    Returns:
-        Response containing user message and tools
-    """
-    user_content = ""
-    for msg in question_turn:
-        if msg["role"] == "user":
-            user_content = msg["content"]
-            break
-
-    return {"messages": [{"role": "user", "content": user_content}], "tools": tools}
-
-
-def create_completion_response() -> Dict[str, Any]:
-    """
-    Create response indicating conversation completion.
-
-    Returns:
-        Response with completion message
-    """
-    return {"messages": [{"role": "env", "content": "[CONVERSATION_COMPLETED]"}]}
-
-
-def create_error_response(error_message: str) -> Dict[str, Any]:
-    """
-    Create response for error conditions.
-
-    Args:
-        error_message: Error message to include
-
-    Returns:
-        Response containing error message
-    """
-    return {"messages": [{"role": "env", "content": f"[ERROR] {error_message}"}]}
-
-
-def decode_execute(result):
-    """
-    Decode execute results for compatibility with evaluation framework.
-
-    Args:
-        result: Result to decode
-
-    Returns:
-        List of decoded function calls
-    """
-    return default_decode_execute_prompting(result)
-
-
-def extract_single_turn_response(messages: List[Dict[str, Any]]) -> str:
-    """
-    Extract single-turn response from conversation messages.
-
-    Args:
-        messages: List of conversation messages
-
-    Returns:
-        String representation of the response
-    """
-    for message in reversed(messages):
-        if message["role"] == "assistant":
-            if "tool_calls" in message and message["tool_calls"]:
-                formatted_calls = []
-                for tool_call in message["tool_calls"]:
-                    formatted_call = format_single_tool_call_for_eval(
-                        tool_call,
-                    )
-                    if formatted_call:
-                        formatted_calls.append(formatted_call)
-                return "\n".join(formatted_calls) if formatted_calls else ""
-            elif message.get("content"):
-                return message["content"]
-
-    return ""
-
-
-def extract_multi_turn_responses(
-    messages: List[Dict[str, Any]],
-) -> List[List[str]]:
-    """
-    Extract multi-turn responses from conversation messages.
-
-    Args:
-        messages: List of conversation messages
-
-    Returns:
-        List of turns, each turn is a list of function call strings
-    """
-    turns_data = []
-    current_turn_responses = []
-
-    i = 0
-    while i < len(messages):
-        message = messages[i]
-
-        if message["role"] == "user":
-            if current_turn_responses:
-                turns_data.append(current_turn_responses)
-                current_turn_responses = []
-
-            i += 1
-            while i < len(messages) and messages[i]["role"] == "assistant":
-                assistant_msg = messages[i]
-
-                if "tool_calls" in assistant_msg and assistant_msg["tool_calls"]:
-                    for tool_call in assistant_msg["tool_calls"]:
-                        formatted_call = format_single_tool_call_for_eval(
-                            tool_call,
-                        )
-                        if formatted_call:
-                            current_turn_responses.append(formatted_call)
-
-                i += 1
-
-                while i < len(messages) and messages[i]["role"] == "tool":
-                    i += 1
-        else:
-            i += 1
-
-    if current_turn_responses:
-        turns_data.append(current_turn_responses)
-
-    return turns_data
-
-
-def format_single_tool_call_for_eval(tool_call: Dict[str, Any]) -> str:
-    """
-    Format a single tool call into string representation for evaluation.
-
-    Args:
-        tool_call: Single tool call in OpenAI format
-
-    Returns:
-        Formatted string representation
-    """
-    function = tool_call.get("function", {})
-    function_name = function.get("name", "")
-
-    try:
-        arguments = function.get("arguments", "{}")
-        if isinstance(arguments, str):
-            args_dict = json.loads(arguments)
-        else:
-            args_dict = arguments
-
-        args_str = ", ".join([f"{k}={repr(v)}" for k, v in args_dict.items()])
-        return f"{function_name}({args_str})"
-
-    except Exception:
-        return f"{function_name}()"
-
-
-def capture_and_print_score_files(
-    score_dir: Path,
-    model_name: str,
-    test_category: str,
-    eval_type: str,
-):
-    """
-    Capture and print contents of score files written to score_dir.
-
-    Args:
-        score_dir: Directory containing score files
-        model_name: Name of the model
-        test_category: Category of the test
-        eval_type: Type of evaluation (relevance/multi_turn/single_turn)
-    """
-    try:
-        print(f"\n=== {eval_type.upper()} Evaluation Result Files ===")
-        print(f"Model: {model_name}")
-        print(f"Test Category: {test_category}")
-        print(f"Evaluation Type: {eval_type}")
-
-        for file_path in score_dir.rglob("*"):
-            if file_path.is_file():
-                relative_path = file_path.relative_to(score_dir)
-                print(f"\n--- File: {relative_path} ---")
-
-                try:
-                    with open(file_path, "r", encoding="utf-8") as f:
-                        content = f.read()
-
-                    if (
-                        file_path.suffix == ".json"
-                        or content.strip().startswith("{")
-                        or content.strip().startswith("[")
-                    ):
-                        try:
-                            lines = content.strip().split("\n")
-                            formatted_lines = []
-                            for line in lines:
-                                if line.strip():
-                                    parsed = json.loads(line)
-                                    formatted_lines.append(
-                                        json.dumps(
-                                            parsed,
-                                            ensure_ascii=False,
-                                            indent=2,
-                                        ),
-                                    )
-                            content = "\n".join(formatted_lines)
-                        except json.JSONDecodeError:
-                            pass
-
-                    print(content)
-
-                except UnicodeDecodeError:
-                    print(f"[Binary file, size: {file_path.stat().st_size} bytes]")
-                except Exception as e:
-                    print(f"[Error reading file: {str(e)}]")
-
-        print(f"=== {eval_type.upper()} Evaluation Result Files End ===\n")
-
-    except Exception as e:
-        print(f"Error capturing evaluation result files: {str(e)}")
-
-
-def extract_tool_schema(tools):
-    """Reformat tool schema"""
-    for i in range(len(tools)):  # pylint: disable=C0200
-        tools[i]["function"].pop("response")
-    return tools

benchmark/bfcl/init_task_memory_pool.py

@@ -1,246 +0,0 @@
-# pylint: disable=W0621,W1514
-"""Init task memory pool"""
-import argparse
-import json
-from collections import defaultdict
-from concurrent.futures import ThreadPoolExecutor, as_completed
-from pathlib import Path
-from typing import List, Dict, Any
-
-import requests
-
-
-def load_task_case(data_path: str, task_id: str | None) -> Dict[str, Any]:
-    """
-    load training cases by id
-    """
-    if not Path(data_path).exists():
-        raise FileNotFoundError(f"BFCL data file '{data_path}' not found")
-
-    if task_id is None:
-        raise ValueError("task_id is required")
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        if str(task_id).isdigit():  # pylint: disable=R1720
-            idx = int(task_id)
-            for line_no, line in enumerate(f):
-                if line_no == idx:
-                    return json.loads(line)
-            raise ValueError(f"Task case index {idx} not found in {data_path}")
-        else:
-            for line in f:
-                data = json.loads(line)
-                if data.get("id") == task_id:
-                    return data
-            raise ValueError(f"Task case id '{task_id}' not found in {data_path}")
-
-
-def get_tool_prompt(tools):
-    """Construct prompt with provided tools"""
-    tool_prompt = (
-        "\n\n# Tools\n\nYou may call one or more functions to assist with the user query."
-        "\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>"
-    )
-    for tool in tools:
-        tool_prompt += "\n" + json.dumps(tool)
-    tool_prompt += (
-        "\n</tools>\n\nFor each function call, return a json object with function name"
-        " and arguments within <tool_call></tool_call> XML tags:"
-        '\n<tool_call>\n{"name": <function-name>, "arguments": <args-json-object>}\n</tool_call>'
-    )
-    return tool_prompt
-
-
-def group_trajectories_by_task_id(jsonl_entries: List[Dict[str, Any]]) -> List[List[Any]]:
-    """
-    group trajectories by task_id
-
-    Args:
-        jsonl_entries: JSONL entry list
-
-    Returns:
-        List[List[Any]]: trajectory list grouped by task_id
-    """
-    grouped = defaultdict(list)
-
-    for entry in jsonl_entries:
-        task_id = entry.get("task_id", "")
-        taks_case = load_task_case("data/multiturn_data_base.jsonl", task_id)
-        tools = taks_case.get("tools", [{}])
-        from bfcl_utils import extract_tool_schema
-
-        tool_schema = extract_tool_schema(tools)
-        entry["task_history"][0]["content"] += get_tool_prompt(tool_schema)
-        grouped[task_id].append(entry)
-
-    # retain only the two with the highest and lowest rewards
-    filtered_groups = []
-    for _, trajectories in grouped.items():
-        if len(trajectories) == 1:
-            # when only one trajectory, retain it
-            filtered_groups.append(trajectories)
-        elif len(trajectories) == 2:
-            # when there are two trajectories, retain them
-            filtered_groups.append(trajectories)
-        else:
-            # when there are more than two trajectories, choose the two with the highest and lowest rewards
-            trajectories.sort(key=lambda t: t["reward"])
-            min_reward_traj = trajectories[0]  # highest reward
-            max_reward_traj = trajectories[-1]  # lowest reward
-            filtered_groups.append([min_reward_traj, max_reward_traj])
-
-    return filtered_groups
-
-
-def post_to_summarizer(trajectories: List[Any], service_url: str) -> Dict[str, Any]:
-    """
-    post trajectories to summarizer service
-
-    Args:
-        trajectories: trajectory list
-        service_url: summarizer service URL
-
-    Returns:
-        response json
-    """
-    trajectory_dicts = [
-        {
-            "task_id": traj["task_id"],
-            "messages": traj["task_history"],
-            "score": traj["reward"],
-        }
-        for traj in trajectories
-    ]
-
-    request_data = {
-        "trajectories": trajectory_dicts,
-    }
-
-    try:
-        response = requests.post(f"{service_url}/summary_task_memory", json=request_data)
-        response.raise_for_status()
-        return response.json()
-    except Exception as e:
-        return {"error": str(e), "trajectories_count": len(trajectories)}
-
-
-def process_trajectories_with_threads(
-    grouped_trajectories: List[List[Any]],
-    service_url: str,
-    n_threads: int = 4,
-) -> List[Dict[str, Any]]:
-    """
-    use threads to process trajectories
-
-    Args:
-        grouped_trajectories: group trajectory list by task_id
-        service_url: memory summarizer service URL
-        n_threads: number of threads
-
-    Returns:
-        all results
-    """
-    results = []
-
-    with ThreadPoolExecutor(max_workers=n_threads) as executor:
-        future_to_group = {
-            executor.submit(post_to_summarizer, group, service_url): i for i, group in enumerate(grouped_trajectories)
-        }
-
-        for future in as_completed(future_to_group):
-            group_index = future_to_group[future]
-            try:
-                result = future.result()
-                result["group_index"] = group_index
-                result["group_size"] = len(grouped_trajectories[group_index])
-                results.append(result)
-                if "memory_list" in result["metadata"]:
-                    print(f'✅ Group {group_index} processed: {result["metadata"].get("memory_list", 0)}')
-                else:
-                    print(f"❌ Group {group_index} processed: error")
-            except Exception as e:
-                error_result = {
-                    "group_index": group_index,
-                    "group_size": len(grouped_trajectories[group_index]),
-                    "error": str(e),
-                }
-                results.append(error_result)
-                print(f"❌ Group {group_index} failed: {e}")
-
-    return results
-
-
-def main():
-    """Main function to convert JSONL to memories using ReMe service."""
-    parser = argparse.ArgumentParser(description="Convert JSONL to memories using ReMe service")
-    parser.add_argument("--jsonl_file", type=str, required=True, help="Path to the JSONL file")
-    parser.add_argument("--service_url", type=str, default="http://localhost:8001", help="ReMe service URL")
-    parser.add_argument("--output_file", type=str, help="Output file to save results (optional)")
-    parser.add_argument("--n_threads", type=int, default=4, help="Number of threads for processing")
-
-    args = parser.parse_args()
-
-    print(f"Processing JSONL file: {args.jsonl_file}")
-    print(f"Service URL: {args.service_url}")
-    print(f"Threads: {args.n_threads}")
-
-    with open(args.jsonl_file, "r") as f:
-        data = [json.loads(line) for line in f]
-    print(f"Loaded {len(data)} entries from JSONL file")
-
-    grouped_trajectories = group_trajectories_by_task_id(data)
-    print(f"Total groups: {len(grouped_trajectories)}")
-
-    results = process_trajectories_with_threads(
-        grouped_trajectories,
-        args.service_url,
-        n_threads=args.n_threads,
-    )
-
-    print(f"Processed {len(results)} groups")
-
-    success_count = sum(1 for r in results if "error" not in r)
-    error_count = len(results) - success_count
-    total_memories = sum(len(r["metadata"].get("memory_list", [])) for r in results if "memory_list" in r["metadata"])
-
-    print(f"✅ Success: {success_count}")
-    print(f"❌ Errors: {error_count}")
-    print(f"📊 Total task memories created: {total_memories}")
-
-    if args.output_file:
-        try:
-            summary = {
-                "jsonl_file": args.jsonl_file,
-                "total_groups": len(grouped_trajectories),
-                "success_count": success_count,
-                "error_count": error_count,
-                "total_task_memories": total_memories,
-                "results": results,
-            }
-
-            with open(args.output_file, "w") as f:
-                json.dump(summary, f, indent=2)
-            print(f"Results saved to: {args.output_file}")
-        except Exception as e:
-            print(f"Error saving results: {e}")
-
-
-if __name__ == "__main__":
-    import sys
-
-    if len(sys.argv) > 1:
-        main()
-    else:
-        print("Running in compatibility mode...")
-        with open("exp_result/qwen3-8b/with_think/bfcl-multi-turn-base-train_wo-exp.jsonl", "r") as f:
-            data = [json.loads(line) for line in f]
-
-        grouped_trajectories = group_trajectories_by_task_id(data)
-        print(f"Total groups: {len(grouped_trajectories)}")
-
-        results = process_trajectories_with_threads(
-            grouped_trajectories,
-            "http://localhost:8001",
-            n_threads=4,
-        )
-        print(f"Processed {len(results)} groups")

benchmark/bfcl/local_file_to_library.py

@@ -1,30 +0,0 @@
-"""Load the library data and convert them to the new format"""
-
-import json
-
-with open("../../file_vector_store/bfcl_test.jsonl", "r", encoding="utf-8") as f:
-    bfcl = [json.loads(line) for line in f]
-
-new_bfcl = []
-for exp in bfcl:
-    new_exp = {}
-    new_exp["workspace_id"] = exp["workspace_id"]
-    new_exp["memory_id"] = exp["unique_id"]
-    new_exp["memory_type"] = exp["metadata"]["memory_type"]
-
-    new_exp["when_to_use"] = exp["content"]
-    new_exp["content"] = exp["metadata"]["content"]
-    new_exp["score"] = exp["metadata"]["score"]
-
-    new_exp["time_created"] = exp["metadata"]["time_created"]
-    new_exp["time_modified"] = exp["metadata"]["time_modified"]
-    new_exp["author"] = exp["metadata"]["author"]
-
-    new_exp["metadata"] = exp["metadata"]["metadata"]
-    new_exp["metadata"]["utility"] = 0
-    new_exp["metadata"]["freq"] = 0
-
-    new_bfcl.append(new_exp)
-
-with open("../../library/bfcl_test.jsonl", "w", encoding="utf-8") as f:
-    f.writelines(json.dumps(item, ensure_ascii=False) + "\n" for item in new_bfcl)

benchmark/bfcl/preprocess.py

@@ -1,73 +0,0 @@
-# pylint: disable=W0621
-"""Preprocess multi-turn test cases"""
-
-import json
-
-
-from pathlib import Path
-from bfcl_eval.model_handler.model_style import ModelStyle
-from bfcl_eval.eval_checker.eval_runner_helper import load_file
-from bfcl_eval.constants.type_mappings import GORILLA_TO_OPENAPI
-from bfcl_eval.constants.eval_config import MULTI_TURN_FUNC_DOC_PATH
-from bfcl_eval.constants.category_mapping import MULTI_TURN_FUNC_DOC_FILE_MAPPING
-from bfcl_eval.model_handler.utils import (
-    convert_to_tool,
-    func_doc_language_specific_pre_processing,
-)
-
-
-def process_multi_turn_test_case(file_path, output_path):
-    """
-    Multi-turn test cases don't have the function doc in the prompt. We need to add them here.
-    """
-    test_cases = []
-    with open(output_path, "w", encoding="utf-8") as outf:
-        with open(file_path, encoding="utf-8") as f:
-            file = f.readlines()
-            for line in file:
-                entry = json.loads(line)
-                if "multi_turn" not in entry["id"]:
-                    continue
-                test_category: str = entry["id"].rsplit("_", 1)[0]
-                involved_classes = entry["involved_classes"]
-                entry["function"] = []
-                for func_collection in involved_classes:
-                    # func_doc is a list of dict
-                    func_doc = load_file(
-                        MULTI_TURN_FUNC_DOC_PATH / MULTI_TURN_FUNC_DOC_FILE_MAPPING[func_collection],
-                    )
-                    entry["function"].extend(func_doc)
-
-                # Handle Miss Func category; we need to remove the holdout function doc
-                if "missed_function" in entry:
-                    for turn_index, missed_func_names in entry["missed_function"].items():
-                        entry["missed_function"][turn_index] = []
-                        for missed_func_name in missed_func_names:
-                            for i, func_doc in enumerate(entry["function"]):
-                                if func_doc["name"] == missed_func_name:
-                                    # Add the missed function doc to the missed_function list
-                                    entry["missed_function"][turn_index].append(func_doc)
-                                    # Remove it from the function list
-                                    entry["function"].pop(i)
-                                    break
-
-                functions = func_doc_language_specific_pre_processing(entry["function"], test_category)
-                tools = convert_to_tool(functions, GORILLA_TO_OPENAPI, ModelStyle.OpenAI_Completions)
-
-                test_cases.append(
-                    {
-                        "id": entry["id"],
-                        "messages": entry["question"][0],
-                        "tools": tools,
-                        "extra": entry,
-                    },
-                )
-                outf.write(json.dumps(test_cases[-1], ensure_ascii=False) + "\n")
-
-    return test_cases
-
-
-if __name__ == "__main__":
-    file_path = Path("./gorilla/berkeley-function-call-leaderboard/bfcl_eval/data/BFCL_v3_multi_turn_base.json")
-    output_path = "data/multiturn_data_base.jsonl"
-    preprocessed_test_cases = process_multi_turn_test_case(file_path, output_path)

benchmark/bfcl/requirements.txt

@@ -1,5 +0,0 @@
-jinja2
-loguru
-openai
-ray
-pandas

benchmark/bfcl/run_bfcl.py

@@ -1,151 +0,0 @@
-"""Run evaluation on BFCL-V3-Multi-Turn-Base dataset."""
-
-import time
-import json
-from pathlib import Path
-
-import ray
-import requests
-from loguru import logger
-from dotenv import load_dotenv
-from bfcl_agent import BFCLAgent
-
-load_dotenv("../../.env")
-
-
-def run_agent(
-    max_workers: int,
-    dataset_name: str,
-    experiment_suffix: str,
-    model_name: str = "qwen3-8b",
-    enable_thinking: bool = False,
-    data_path: str = "data/multiturn_data_base_val.jsonl",
-    answer_path: Path = Path("data/possible_answer"),
-    num_trials: int = 1,
-    use_memory: bool = False,
-    memory_base_url: str = "http://0.0.0.0:8002/",
-    use_memory_addition: bool = True,
-    use_memory_deletion: bool = False,
-    delete_freq: int = 10,
-    freq_threshold: int = 5,
-    utility_threshold: float = 0.5,
-):
-    """Run the agent"""
-    experiment_name = dataset_name + "_" + experiment_suffix
-    path: Path = Path(
-        f"./exp_result/{model_name}/with_think" if enable_thinking else f"./exp_result/{model_name}/no_think",
-    )
-    path.mkdir(parents=True, exist_ok=True)
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        task_ids = [json.loads(line)["id"] for line in f]
-
-    result: list = []
-
-    def dump_file():
-        with open(path / f"{experiment_name}.jsonl", "a", encoding="utf-8") as f:
-            for x in result:
-                f.write(json.dumps(x) + "\n")
-
-    future_list: list = []
-    for i in range(max_workers):
-        actor = BFCLAgent.remote(
-            index=i,
-            model_name=model_name,
-            task_ids=task_ids[i::max_workers],
-            experiment_name=experiment_name,
-            data_path=data_path,
-            answer_path=answer_path,
-            num_trials=num_trials,
-            use_memory=use_memory,
-            memory_base_url=memory_base_url,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=delete_freq,
-            freq_threshold=freq_threshold,
-            utility_threshold=utility_threshold,
-            enable_thinking=enable_thinking,
-        )
-        future = actor.execute.remote()
-        future_list.append(future)
-        time.sleep(1)
-    logger.info("submit complete")
-
-    for i, future in enumerate(future_list):
-        t_result = ray.get(future)
-        if t_result:
-            if isinstance(t_result, list):
-                result.extend(t_result)
-            else:
-                result.append(t_result)
-
-        logger.info(f"{i + 1}/{len(task_ids)} complete")
-    dump_file()
-
-
-def handle_api_response(response: requests.Response):
-    """Handle API response with proper error checking"""
-    if response.status_code != 200:
-        print(f"Error: {response.status_code}")
-        print(response.text)
-        return None
-
-    return response.json()
-
-
-def load_memory(path: str = "docs/library", api_url: str = "http://0.0.0.0:8002/"):
-    """Load memories from disk into the vector store"""
-    response = requests.post(
-        url=f"{api_url}load_memory",
-        json={
-            "load_file_path": path,
-            "clear_existing": True,
-        },
-    )
-
-    result = handle_api_response(response)
-    if result:
-        print(f"Memory loaded from {path}")
-
-
-def main():
-    """Main function"""
-    max_workers = 4
-    if max_workers > 1:
-        ray.init(num_cpus=max_workers)
-
-    num_runs = 4
-    num_trials = 1
-    model_name = "qwen3-8b"
-    enable_thinking = True
-    use_memory = True
-    use_memory_addition = False
-    use_memory_deletion = False
-    memory_base_url = "http://0.0.0.0:8003/"
-
-    if use_memory:
-        load_file_path = "docs/library/paper_data/task/bfcl_qwen3_8b.jsonl"
-        load_memory(load_file_path, memory_base_url)
-
-    for _ in range(num_runs):
-        run_agent(
-            max_workers=max_workers,
-            model_name=model_name,
-            dataset_name="bfcl-multi-turn-base",
-            experiment_suffix="w-fixed-memory",
-            data_path="data/multiturn_data_base_val.jsonl",
-            answer_path=Path("data/possible_answer"),
-            enable_thinking=enable_thinking,
-            num_trials=num_trials,
-            use_memory=use_memory,
-            memory_base_url=memory_base_url,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=5,
-            freq_threshold=5,
-            utility_threshold=0.5,
-        )
-
-
-if __name__ == "__main__":
-    main()

benchmark/bfcl/run_exp_statistic.py

@@ -1,163 +0,0 @@
-"""Run the experiment statistic."""
-
-import json
-from collections import defaultdict
-from pathlib import Path
-
-import pandas as pd
-from loguru import logger
-
-
-def calculate_best_at_k(scores: list, k: int) -> float:
-    """
-    Calculate best@k
-    Divide scores into groups of size k, take the maximum value in each group,
-    then average these maximum values
-
-    Args:
-        scores: List of after_score values for all runs of a task
-        k: Group size
-
-    Returns:
-        best@k value
-    """
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        group_maxs.append(max(group))
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def calculate_pass_at_k(scores: list, k: int) -> float:
-    """Calculate pass@k."""
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        is_pass = 1.0 if max(group) >= 1.0 else 0.0
-        group_maxs.append(is_pass)
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def get_possible_k_values(total_runs: int) -> list:
-    """
-    Get all possible k values (factors of total_runs)
-
-    Args:
-        total_runs: Total number of runs
-
-    Returns:
-        List of k values in descending order
-    """
-    k_values = []
-    for k in range(1, total_runs + 1):
-        if total_runs % k == 0:
-            k_values.append(k)
-    return sorted(k_values, reverse=True)  # Sort from large to small
-
-
-def run_exp_statistic():
-    """Run the experiment statistic."""
-    path: Path = Path("./exp_result/qwen3-8b/with_think")
-
-    # Store results for all experiments
-    all_results = {}
-    for file in path.glob("*.jsonl"):
-        # Group results by task_id
-        task_results = defaultdict(list)
-        print(file)
-        with open(file, "r", encoding="utf-8") as f:
-            for line in f:
-                if not line.strip():
-                    continue
-                data = json.loads(line)
-
-                if isinstance(data, list):
-                    for part_data in data:
-                        task_id = part_data["task_id"]
-                        after_score = part_data["reward"]
-                        task_results[task_id].append(after_score)
-                else:
-                    task_id = data["task_id"]
-                    after_score = data["reward"]
-                    task_results[task_id].append(after_score)
-
-        if not task_results:
-            logger.warning(f"No valid data found in file {file}")
-            continue
-
-        # Check if each task has consistent number of runs
-        run_counts = [len(scores) for scores in task_results.values()]
-        if len(set(run_counts)) > 1:
-            logger.warning(f"Inconsistent number of runs for different tasks in file {file}: {set(run_counts)}")
-            continue
-
-        num_runs = run_counts[0]
-        logger.info(f"File {file}: {len(task_results)} tasks, {num_runs} runs per task")
-
-        # Get all possible k values
-        k_values = get_possible_k_values(num_runs)
-        logger.info(f"Calculable best@k values: {k_values}")
-
-        # Calculate various best@k values
-        file_results = {"file": file.name}
-
-        for k in k_values:
-            best_at_k_scores = []
-            pass_at_k_scores = []
-            for task_id, scores in task_results.items():
-                try:
-                    best_k_score = calculate_best_at_k(scores, k)
-                    pass_at_k_score = calculate_pass_at_k(scores, k)
-                    pass_at_k_scores.append(pass_at_k_score)
-                    best_at_k_scores.append(best_k_score)
-                except ValueError as e:
-                    logger.error(f"Error calculating best@{k} for task {task_id}: {e}")
-                    continue
-
-            if best_at_k_scores:
-                avg_best_at_k = sum(best_at_k_scores) / len(best_at_k_scores)
-                file_results[f"best@{k}"] = avg_best_at_k
-                logger.info(f"file={file.name} best@{k}={avg_best_at_k:.4f}")
-
-            if pass_at_k_scores:
-                avg_pass_at_k = sum(pass_at_k_scores) / len(pass_at_k_scores)
-                file_results[f"pass@{k}"] = avg_pass_at_k
-                logger.info(f"file={file.name} pass@{k}={avg_pass_at_k:.4f}")
-
-        all_results[file.name] = file_results
-
-    # Create and display table
-    if all_results:
-        df = pd.DataFrame(list(all_results.values()))
-        df = df.set_index("file")
-
-        # Sort columns by the number in column name (best@8, best@4, best@2, best@1)
-        # best_columns = [col for col in df.columns if col.startswith('best@')]
-        best_columns = df.columns
-        best_columns.sort(key=lambda x: x, reverse=False)
-        df = df[best_columns]
-
-        print("\n" + "=" * 80)
-        print("Experiment Results Summary Table")
-        print("=" * 80)
-        print(df.round(4))
-        print("=" * 80)
-
-        # Save table to CSV
-        output_path = path / "experiment_summary.csv"
-        df.to_csv(output_path)
-        logger.info(f"Results table saved to: {output_path}")
-    else:
-        logger.warning("No valid experiment results found")
-
-
-if __name__ == "__main__":
-    run_exp_statistic()

benchmark/bfcl/split_into_trainval.py

@@ -1,34 +0,0 @@
-"""Split the JSONL file into train and validation sets."""
-
-import argparse
-import json
-import random
-
-
-def split_jsonl(input_file, train_file, val_file, ratio=0.8):
-    """Split the JSONL file into train and validation sets."""
-    with open(input_file, "r", encoding="utf-8") as f:
-        data = [json.loads(line) for line in f]
-    random.shuffle(data)
-
-    split_idx = int(len(data) * ratio)
-    train_data = data[:split_idx]
-    val_data = data[split_idx:]
-
-    with open(train_file, "w", encoding="utf-8") as f:
-        for item in train_data:
-            f.write(json.dumps(item, ensure_ascii=False) + "\n")
-    with open(val_file, "w", encoding="utf-8") as f:
-        for item in val_data:
-            f.write(json.dumps(item, ensure_ascii=False) + "\n")
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Split JSONL file into train and validation sets.")
-    parser.add_argument("--input", required=True, help="Path to input JSONL file")
-    parser.add_argument("--train", required=True, help="Path to output train file")
-    parser.add_argument("--val", required=True, help="Path to output validation file")
-    parser.add_argument("--ratio", type=float, default=0.5, help="Train ratio (default: 0.8)")
-
-    args = parser.parse_args()
-    split_jsonl(args.input, args.train, args.val, args.ratio)

benchmark/halumem/eval_reme.yaml

@@ -1,548 +0,0 @@
-TEMPLATE_MEMOS: |
-  Memories for user {user_id}:
-    {memories}
-
-PROMPT_MEMZERO_JSON: |
-  # CONTEXT:
-  {context}
-
-  # CONTEXT PRIORITY:
-  When the context contains information from multiple sources, follow this strict priority order:
-  1. **Historical Dialogue** (highest priority) - Direct conversation content
-  2. **Extracted Memories** (medium priority) - Summarized memory points
-  3. **User Profile** (lowest priority) - General user information
-
-  # Question:
-  {question}
-
-  # OUTPUT FORMAT:
-  Do not hallucinate; strictly answer the user's question based on the content of the CONTEXT.
-  Please provide your response in the following JSON format:
-
-  ```json
-  {{
-    "reasoning": "reasoning content",
-    "answer": "Provide a detailed answer"
-  }}
-  ```
-
-PROMPT_MEMZERO_JSON2: |
-  # CONTEXT:
-  {context}
-
-  # CONTEXT PRIORITY:
-  When the context contains information from multiple sources, follow this strict priority order:
-  1. **Historical Dialogue** (highest priority) - Direct conversation content
-  2. **Extracted Memories** (medium priority) - Summarized memory points
-  3. **User Profile** (lowest priority) - General user information
-
-  # Question:
-  {question}
-
-  # OUTPUT FORMAT:
-  Do not hallucinate; strictly answer the user's question based on the content of the CONTEXT.
-  Please provide your response in the following JSON format:
-
-  ```json
-  {{
-    "reasoning": "reasoning content",
-    "answer": "Provide a detailed answer"
-  }}
-  ```
-
-PROMPT_MEMZERO: |
-  You are an intelligent memory assistant tasked with retrieving accurate information from conversation memories.
-
-  # CONTEXT:
-  You have access to memories from two speakers in a conversation. These memories contain
-  timestamped information that may be relevant to answering the question.
-
-  # INSTRUCTIONS:
-  1. Carefully analyze all provided memories from both speakers
-  2. Pay special attention to the timestamps to determine the answer
-  3. If the question asks about a specific event or fact, look for direct evidence in the memories
-  4. If the memories contain contradictory information, prioritize the most recent memory
-  5. If there is a question about time references (like "last year", "two months ago", etc.),
-     calculate the actual date based on the memory timestamp. For example, if a memory from
-     4 May 2022 mentions "went to India last year," then the trip occurred in 2021.
-  6. Always convert relative time references to specific dates, months, or years. For example,
-     convert "last year" to "2022" or "two months ago" to "March 2023" based on the memory
-     timestamp. Ignore the reference while answering the question.
-  7. Focus only on the content of the memories from both speakers. Do not confuse character
-     names mentioned in memories with the actual users who created those memories.
-  8. The answer should be less than 5-6 words.
-
-  # APPROACH (Think step by step):
-  1. First, examine all memories that contain information related to the question
-  2. Examine the timestamps and content of these memories carefully
-  3. Look for explicit mentions of dates, times, locations, or events that answer the question
-  4. If the answer requires calculation (e.g., converting relative time references), show your work
-  5. Formulate a precise, concise answer based solely on the evidence in the memories
-  6. Double-check that your answer directly addresses the question asked
-  7. Ensure your final answer is specific and avoids vague time references
-
-  {context}
-
-  Question: {question}
-
-  Answer:
-
-PROMPT_ZEP: |
-  You are an intelligent memory assistant tasked with retrieving accurate information from conversation memories.
-
-  # CONTEXT:
-  You have access to memories from a conversation. These memories contain
-  timestamped information that may be relevant to answering the question.
-
-  # INSTRUCTIONS:
-  1. Carefully analyze all provided memories
-  2. Pay special attention to the timestamps to determine the answer
-  3. If the question asks about a specific event or fact, look for direct evidence in the memories
-  4. If the memories contain contradictory information, prioritize the most recent memory
-  5. If there is a question about time references (like "last year", "two months ago", etc.),
-     calculate the actual date based on the memory timestamp. For example, if a memory from
-     4 May 2022 mentions "went to India last year," then the trip occurred in 2021.
-  6. Always convert relative time references to specific dates, months, or years. For example,
-     convert "last year" to "2022" or "two months ago" to "March 2023" based on the memory
-     timestamp. Ignore the reference while answering the question.
-  7. Focus only on the content of the memories. Do not confuse character
-     names mentioned in memories with the actual users who created those memories.
-  8. The answer should be less than 5-6 words.
-
-  # APPROACH (Think step by step):
-  1. First, examine all memories that contain information related to the question
-  2. Examine the timestamps and content of these memories carefully
-  3. Look for explicit mentions of dates, times, locations, or events that answer the question
-  4. If the answer requires calculation (e.g., converting relative time references), show your work
-  5. Formulate a precise, concise answer based solely on the evidence in the memories
-  6. Double-check that your answer directly addresses the question asked
-  7. Ensure your final answer is specific and avoids vague time references
-
-  Context:
-
-  {context}
-
-  Question: {question}
-  Answer:
-
-PROMPT_MEMOS: |
-  You are a knowledgeable and helpful AI assistant.
-
-  # CONTEXT:
-  You have access to memories from two speakers in a conversation. These memories contain
-  timestamped information that may be relevant to answering the question.
-
-  # INSTRUCTIONS:
-  1. Carefully analyze all provided memories. Synthesize information across different entries if needed to form a complete answer.
-  2. Pay close attention to the timestamps to determine the answer. If memories contain contradictory information, the **most recent memory** is the source of truth.
-  3. If the question asks about a specific event or fact, look for direct evidence in the memories.
-  4. Your answer must be grounded in the memories. However, you may use general world knowledge to interpret or complete information found within a memory (e.g., identifying a landmark mentioned by description).
-  5. If the question involves time references (like "last year", "two months ago", etc.), you **must** calculate the actual date based on the memory's timestamp. For example, if a memory from 4 May 2022 mentions "went to India last year," then the trip occurred in 2021.
-  6. Always convert relative time references to specific dates, months, or years in your final answer.
-  7. Do not confuse character names mentioned in memories with the actual users who created them.
-  8. The answer must be brief (under 5-6 words) and direct, with no extra description.
-
-  # APPROACH (Think step by step):
-  1. First, examine all memories that contain information related to the question.
-  2. Synthesize findings from multiple memories if a single entry is insufficient.
-  3. Examine timestamps and content carefully, looking for explicit dates, times, locations, or events.
-  4. If the answer requires calculation (e.g., converting relative time references), perform the calculation.
-  5. Formulate a precise, concise answer based on the evidence from the memories (and allowed world knowledge).
-  6. Double-check that your answer directly addresses the question asked and adheres to all instructions.
-  7. Ensure your final answer is specific and avoids vague time references.
-
-  {context}
-
-  Question: {question}
-
-  Answer:
-
-PROMPT_MEMOBASE: |
-  You are an intelligent memory assistant tasked with retrieving accurate information from conversation memories.
-
-  # CONTEXT:
-  You have access to memories from two speakers in a conversation. These memories contain
-  timestamped information that may be relevant to answering the question.
-
-  # INSTRUCTIONS:
-  1. Carefully analyze all provided memories from both speakers
-  2. Pay special attention to the timestamps to determine the answer
-  3. If the question asks about a specific event or fact, look for direct evidence in the memories
-  4. If the memories contain contradictory information, prioritize the most recent memory
-  5. If there is a question about time references (like "last year", "two months ago", etc.), calculate the actual date based on the memory timestamp. For example, if a memory from 4 May 2022 mentions "went to India last year," then the trip occurred in 2021.
-  6. Always convert relative time references to specific dates, months, or years. For example, convert "last year" to "2022" or "two months ago" to "March 2023" based on the memory timestamp. Ignore the reference while answering the question.
-  7. Focus only on the content of the memories from both speakers. Do not confuse character names mentioned in memories with the actual users who created those memories.
-  8. The answer should be less than 5-6 words.
-
-  # APPROACH (Think step by step):
-  1. First, examine all memories that contain information related to the question
-  2. Examine the timestamps and content of these memories carefully
-  3. Look for explicit mentions of dates, times, locations, or events that answer the question
-  4. If the answer requires calculation (e.g., converting relative time references), show your work
-  5. Formulate a precise, concise answer based solely on the evidence in the memories
-  6. Double-check that your answer directly addresses the question asked
-  7. Ensure your final answer is specific and avoids vague time references
-
-  {context}
-
-  Question: {question}
-
-  Answer:
-
-
-EVALUATION_PROMPT_FOR_MEMORY_INTEGRITY: |
-  You are a strict **"Memory Integrity" evaluator**.
-  Your core task is to assess whether an AI memory system has **missed any key memory points** after processing a conversation. This evaluation measures the system’s **memory integrity**, i.e., its ability to resist **amnesia** or **omission**.
-
-  # Evaluation Context & Data:
-
-  1. **Extracted Memories:**
-     These are all the memory items actually extracted by the memory system.
-     {memories}
-
-  2. **Expected Memory Point:**
-     The key memory point that *should* have been extracted.
-     {expected_memory_point}
-
-  # Evaluation Instructions:
-
-  1. For each **Expected Memory Point**, search within the **Extracted Memories** list for corresponding or related information. Ignore unrelated items.
-  2. Based on the following scoring rubric, rate how well the memory system captured the **Expected Memory Point** and provide a detailed explanation.
-
-  # Scoring Rubric:
-
-  * **2:** Fully covered or implied.
-    One or more items in “Extracted Memories” fully cover or logically imply all information in the “Expected Memory Point.”
-
-  * **1:** Partially covered or mentioned.
-    Some information in “Extracted Memories” mentions part of the “Expected Memory Point,” but key information is missing, inaccurate, or slightly incorrect.
-
-  * **0:** Not mentioned or incorrect.
-    “Extracted Memories” contains no mention of the “Expected Memory Point,” or the corresponding information is entirely wrong.
-
-  # Scoring Notes:
-
-  * For **compound Expected Memory Points** (with multiple elements such as person/event/time/location/preference, etc.):
-
-    * All elements correct → **2 points**
-    * Some elements correct / uncertain → **1 point**
-    * Key elements missing or wrong → **0 points**
-
-  * Semantic matching is acceptable; exact wording is **not** required.
-
-  * If “Extracted Memories” contains **conflicting information**, assign the **best possible coverage score** and mention the conflict in your reasoning.
-
-  * Extra or stylistically different memories do **not** reduce the score; only the coverage of the **Expected Memory Point** matters.
-
-  * For uncertain wording (“might,” “probably,” “tends to,” etc.):
-
-    * If the Expected Memory Point is a definite statement, usually assign **1 point**.
-
-  * If critical fields (e.g., time, entity name, relationship) are partly wrong but others match → **1 point**.
-
-    * If all key fields are wrong or missing → **0 points**.
-
-  # Output Format:
-
-  Please output your result in the following JSON format:
-
-  ```json
-  {{
-    "reasoning": "Provide a concise justification for the score",
-    "score": "2|1|0"
-  }}
-  ```
-
-EVALUATION_PROMPT_FOR_MEMORY_ACCURACY: |
-  You are a **Dialogue Memory Accuracy Evaluator.** Your task is to evaluate the **accuracy** of a memory extracted by an AI memory system, based on three given inputs: the dialogue content, the *target (gold)* memory points (the correct annotated memories), and the *candidate* memory to be evaluated. The goal is to output a **structured evaluation result**.
-
-  # Input Content
-
-  * **Dialogue:**
-    {dialogue}
-
-  * **Golden Memories (Target Memory Points):**
-    The correct memory points pre-annotated for this dialogue in the evaluation dataset.
-    {golden_memories}
-
-  * **Candidate Memory:**
-    The memory extracted by the system to be evaluated.
-    {candidate_memory}
-
-  # Evaluation Principles and Definitions
-
-  ### 1) Support / Entailment
-
-  * An **information point** (atomic fact) in the candidate memory is considered *supported* if it can be directly stated or semantically entailed (via synonym, paraphrase, or equivalent expression) by the *Dialogue* or *Golden Memories*.
-  * Only the given dialogue and golden memories can be used for judgment — **no external knowledge** or assumptions are allowed.
-    Any information not appearing in or inferable from these two sources is considered *unsupported*.
-  * Pay careful attention to **negation**, **quantities**, **time**, and **subjects**.
-    If the candidate statement contradicts the dialogue or golden memories, it is considered a **conflict**.
-
-  ### 2) Memory Accuracy Score (integer: 0 / 1 / 2)
-
-  * **2 points:** Every information point in the candidate memory is supported by the dialogue or golden memories, with **no contradictions or hallucinations**.
-  * **1 point:** The candidate memory is *partially correct* (at least one supported information point) but also includes *unsupported* or *contradictory* content.
-  * **0 points:** The candidate memory is **entirely unsupported or contradictory** to the sources (i.e., a “hallucinated memory”).
-
-  > Note:
-  >
-  > * If a candidate memory contains multiple information points, **any unsupported or contradictory element** prevents a full score (2).
-  > * If both supported and unsupported/conflicting content appear, assign a score of **1**.
-
-  ### 3) Inclusion in Golden Memories (Boolean field-level judgment)
-
-  **Definition:**
-
-  * **Atomic information point:** the smallest factual unit in the candidate memory (e.g., *name = Li Si*, *age = 25*, *location = Beijing*, *preference = coffee*, *budget ≤ 2000*, *meeting_time = Wednesday 10:00*, *tool = Zoom*, etc.).
-  * **Field / Slot:** the semantic dimension of an information point (e.g., *name*, *age*, *residence*, *food preference*, *budget*, *meeting time*, *meeting tool*, etc.).
-
-  **Judgment Rules (independent of correctness):**
-
-  * **true:**
-    Every atomic information point in the candidate memory has a corresponding **field** in the golden memories (allowing for synonyms, paraphrases, or equivalent expressions; ignore value, polarity, or quantity differences).
-
-    * Note: A single field in the gold list may match multiple candidate points (e.g., multiple “drink preference” facts can be covered by one “drink preference” field in gold).
-  * **false:**
-    If **any** atomic information point’s field in the candidate memory cannot be found in the golden memories, mark as *false*.
-
-  **Important Notes:**
-
-  * Field matching is restricted to fields that are **explicitly present or semantically recognizable** in the golden memories — no external knowledge may be used to expand the field set.
-  * Differences in **values** (e.g., “Zhang San” vs. “Li Si”), **polarity** (like/dislike), or **exact number/time** do **not** affect this Boolean judgment.
-
-  # Evaluation Procedure
-
-  For each candidate memory:
-
-  1. **Decompose** it into atomic information points (e.g., name, number, location, preference).
-  2. For each information point, **search** the dialogue and golden memories for supporting or contradictory evidence.
-  3. Assign the **accuracy_score** (0 / 1 / 2) according to the rules above.
-  4. Determine **is_included_in_golden_memories (true/false)**:
-
-     * Identify each information point’s field;
-     * If *all* fields exist in the golden memories, mark as *true*; otherwise, *false*.
-  5. Provide a **concise Chinese explanation** in `"reason"`, citing key evidence (short excerpts allowed), and clearly state any unsupported or contradictory parts if applicable.
-
-  # Output Format (strictly required)
-
-  Output **only one JSON object**, with the following three fields:
-
-  * `"accuracy_score"`: `"0"` or `"1"` or `"2"`
-  * `"is_included_in_golden_memories"`: `"true"` or `"false"`
-  * `"reason"`: `"brief explanation in Chinese"`
-
-  Do **not** include any other text, explanation, or fields.
-  Do **not** include the candidate memory text inside the JSON.
-
-  Please output **only** the following JSON (in a code block):
-
-  ```json
-  {{
-    "accuracy_score": "2 | 1 | 0",
-    "is_included_in_golden_memories": "true | false",
-    "reason": "Brief explanation in Chinese"
-  }}
-  ```
-
-EVALUATION_PROMPT_FOR_UPDATE_MEMORY: |
-  Your task is to **evaluate the update accuracy** of an AI memory system.
-  Based on the information provided below, determine whether the system-generated **“Generated Memories”** correctly **includes** the **Target Memory for Update**.
-
-  # Background Information
-
-  The following information is provided for evaluation:
-
-  1. **Generated Memories:**
-     This is the list of memory points generated by the system after the current dialogue.
-     {memories}
-
-  2. **Target Memory for Update:**
-     This is the correct, updated version of the memory point that should have been produced — the one we focus on in this evaluation.
-     {updated_memory}
-
-  3. **Original Memory Content:**
-     This is the original version of the target memory before the update.
-     {original_memory}
-
-  # Evaluation Criteria
-
-  Please make your judgment **strictly based on the content update of the “Target Memory for Update.”**
-  Use the following categories:
-
-  ### Correct Update
-
-  * **Generated Memories** **contains all information points** from the “Target Memory for Update,” accurately and completely reflecting the intended update.
-  * **Key fields** (e.g., date, time, values, proper nouns, etc.) must match exactly.
-  * The **original memory** is effectively replaced or marked as outdated.
-  * Synonymous or slightly rephrased expressions are acceptable.
-
-  ### Hallucinated Update
-
-  * **Factual error:** The **Generated Memories** includes a new memory related to the “Target Memory for Update,” but its content contains factual mistakes or contradictions compared to the correct update.
-
-  ### Omitted Update
-
-  * **Completely omitted:** The **Generated Memories** contains no new memory related to the “Target Memory for Update.”
-  * **Partially omitted:** A related new memory was generated in **Generated Memories**, but it **misses key information** that should have been included.
-
-  ### Other
-
-  Used for update failures that do **not clearly fall** into the above categories of “Hallucination” or “Omission.”
-
-  # Output Requirements
-
-  Please return your evaluation strictly in the following JSON format and provide a concise explanation.
-
-  ```json
-  {{
-    "reason": "Briefly explain your reasoning here and why it fits this category.",
-    "evaluation_result": "Correct | Hallucination | Omission | Other"
-  }}
-  ```
-
-EVALUATION_PROMPT_FOR_QUESTION: |
-  You are an **evaluation expert for AI memory system question answering**.
-  Based **only** on the provided **“Question”**, **“Reference Answer”**, and **“Key Memory Points”** (the essential facts needed to derive the reference answer), strictly evaluate the **accuracy** of the **“Memory System Response.”** Classify it as one of **“Correct”**, **“Hallucination”**, or **“Omission.”** Do **not** use any external knowledge or subjective inference. Finally, output your judgment **strictly** in the specified JSON format.
-
-  # Evaluation Criteria
-
-  ## Answer Type Classification
-
-  ### 1. Correct
-
-  * The “Memory System Response” accurately answers the “Question,” and its content is **semantically equivalent** to the “Reference Answer.”
-  * It contains **no contradictions** with the “Key Memory Points” or “Reference Answer.”
-  * It introduces **no unsupported details** beyond the “Key Memory Points” that could alter the conclusion.
-  * Synonyms, paraphrasing, and reasonable summarization are acceptable.
-
-  ### 2. Hallucination
-
-  * The “Memory System Response” includes information or facts that **contradict or are inconsistent** with the “Reference Answer” or the “Key Memory Points.”
-  * When the “Reference Answer” is labeled as *unknown/uncertain*, yet the response provides a specific verifiable fact or conclusion.
-  * Extra irrelevant information that does **not change** the conclusion is **not** considered hallucination by itself; however, if it **changes or misleads** the conclusion, or **contradicts** the “Key Memory Points,” it should be judged as a **Hallucination**.
-
-  ### 3. Omission
-
-  * The response is **incomplete** compared to the “Reference Answer.”
-  * It explicitly states “don’t know,” “can’t remember,” or “no related memory,” even though relevant information exists in the “Key Memory Points.”
-  * For multi-element questions, **all elements must be correct and present**; omission of **any** element is considered an **Omission**.
-
-  ## Priority Rules (Conflict Handling)
-
-  * If the response contains **both missing necessary information** and **fabricated/contradictory information**, classify it as **Hallucination**.
-  * If there is **no fabrication/contradiction** but some necessary information is missing, classify it as **Omission**.
-  * Only when the meaning is **fully equivalent** to the reference answer should it be classified as **Correct**.
-
-  ## Detailed Guidelines and Tolerance
-
-  * Equivalent expressions of numbers, times, and units are acceptable, but the **numerical values themselves must not differ**.
-  * For multi-element questions, **all elements must be complete and accurate**; missing any element counts as **Omission**.
-  * If the reference answer is *“unknown / cannot be determined”* and the system provides a definite fact, that is a **Hallucination**.
-    If the system also answers *“unknown”* (without guessing), it may be **Correct**.
-  * The evaluation must rely **only** on the *Reference Answer*, *Key Memory Points*, and *System Response* — no external context, world knowledge, or speculative reasoning is allowed.
-
-  # Information for Evaluation
-
-  * **Question:**
-    {question}
-
-  * **Reference Answer:**
-    {reference_answer}
-
-  * **Key Memory Points:**
-    {key_memory_points}
-
-  * **Memory System Response:**
-    {response}
-
-  # Output Requirements
-
-  Please provide your evaluation result **strictly** in the JSON format below.
-  Do **not** add any extra explanation or comments outside the JSON block.
-
-  ```json
-  {{
-    "reasoning": "Provide a concise and traceable evaluation rationale: first compare the system’s response with the Key Memory Points (which were correctly used, which were missing, and whether there was any fabrication/contradiction), then assess its consistency with the Reference Answer, and finally state the classification basis.",
-    "evaluation_result": "Correct | Hallucination | Omission"
-  }}
-  ```
-
-
-EVALUATION_PROMPT_FOR_QUESTION2: |
-  You are an **evaluation expert for AI memory system question answering**.
-
-  Based **only** on the provided **"Question"**, **"Reference Answer"**, and **"Key Memory Points"** (the essential facts needed to derive the reference answer), strictly evaluate the **accuracy** of the **"Memory System Response."** Classify it as one of **"Correct"**, **"Hallucination"**, or **"Omission."** Do **not** use any external knowledge or subjective inference. Finally, output your judgment **strictly** in the specified JSON format.
-
-  # Evaluation Criteria
-
-  ## Answer Type Classification
-
-  ### 1. Correct
-
-  * The "Memory System Response" accurately answers the "Question," and its content is **semantically equivalent** to the "Reference Answer."
-  * It contains **no contradictions** with the "Key Memory Points" or "Reference Answer."
-  * **Extra details not present in the Key Memory Points are allowed and should not be penalized**, as long as they:
-    - Do not contradict the Key Memory Points or Reference Answer
-    - Do not change or mislead the core conclusion
-    - Are reasonable additional context that the memory system may have retained from the conversation
-  * The memory system may have stored additional information beyond the Key Memory Points. Such extra information should be treated as **supplementary context** rather than hallucination, provided it does not conflict with the core answer.
-  * Synonyms, paraphrasing, and reasonable summarization are acceptable.
-
-  ### 2. Hallucination
-
-  * The "Memory System Response" includes information or facts that **contradict or are inconsistent** with the "Reference Answer" or the "Key Memory Points."
-  * The response provides information that **directly contradicts** known facts from the Key Memory Points.
-  * When the "Reference Answer" is labeled as *unknown/uncertain*, yet the response provides a specific verifiable fact or conclusion.
-  * **Important:** Extra information that is NOT in Key Memory Points is **NOT automatically a hallucination**. Only classify as hallucination if the extra information:
-    - Directly contradicts the Key Memory Points or Reference Answer
-    - Changes or misleads the core conclusion in a way that makes the answer incorrect
-    - Provides a definitive answer when the Reference Answer indicates uncertainty
-
-  ### 3. Omission
-
-  * The response is **incomplete** compared to the "Reference Answer."
-  * It explicitly states "don't know," "can't remember," or "no related memory," even though relevant information exists in the "Key Memory Points."
-  * For multi-element questions, **all elements must be correct and present**; omission of **any** element is considered an **Omission**.
-
-  ## Priority Rules (Conflict Handling)
-
-  * If the response contains **both missing necessary information** and **fabricated/contradictory information**, classify it as **Hallucination**.
-  * If there is **no fabrication/contradiction** but some necessary information is missing, classify it as **Omission**.
-  * If the core answer is correct and complete, classify as **Correct** even if there are extra details not in Key Memory Points (as long as they don't contradict or mislead).
-
-  ## Detailed Guidelines and Tolerance
-
-  * Equivalent expressions of numbers, times, and units are acceptable, but the **numerical values themselves must not differ**.
-  * For multi-element questions, **all elements must be complete and accurate**; missing any element counts as **Omission**.
-  * If the reference answer is *"unknown / cannot be determined"* and the system provides a definite fact, that is a **Hallucination**.
-    If the system also answers *"unknown"* (without guessing), it may be **Correct**.
-  * **Focus on evaluating whether the core answer to the question is correct**, not whether the response is limited to only the Key Memory Points.
-  * Extra contextual information (e.g., additional preferences, related details) should be viewed as enrichment, not as errors, unless they contradict or mislead.
-
-  # Information for Evaluation
-
-  * **Question:**
-    {question}
-
-  * **Reference Answer:**
-    {reference_answer}
-
-  * **Key Memory Points:**
-    {key_memory_points}
-
-  * **Memory System Response:**
-    {response}
-
-  # Output Requirements
-
-  Please provide your evaluation result **strictly** in the JSON format below.
-  Do **not** add any extra explanation or comments outside the JSON block.
-
-  ```json
-  {{
-    "reasoning": "Provide a concise and traceable evaluation rationale: first verify that the system's response correctly includes all required elements from the Reference Answer, then check if any information contradicts the Key Memory Points or Reference Answer. Extra details not in Key Memory Points should be noted but not penalized unless they contradict or mislead. Finally state the classification basis.",
-    "evaluation_result": "Correct | Hallucination | Omission"
-  }}
-  ```
-  """

benchmark/halumem/scripts.sh

@@ -1 +0,0 @@
-cat bench_results/reme/Martin\ Mark/session* | grep '"result_type": "' | awk -F'"' '{total++; if($4=="Correct") count++} END {printf "Correct Rate: %.2f%% (%d/%d)\n", (count/total)*100, count, total}'

benchmark/longmemeval/README.md

@@ -0,0 +1,96 @@
+[中文版 / Chinese version](./README_ZH.md)
+
+# LongMemEval Benchmark
+
+LongMemEval is a benchmark for **long-term memory over multi-session chat
+histories**. Each item provides a chronologically ordered set of chat sessions
+between a user and an assistant, followed by a probing question whose answer is
+only recoverable by reasoning over the user-owned memory. ReMe ingests the
+sessions into an isolated per-item workspace, answers the question via an
+agentic (ReAct) mode, and scores the answer with an LLM-as-judge.
+
+Question types include single-session (user / assistant / preference),
+multi-session reasoning, knowledge update, and temporal reasoning.
+
+> For the shared setup (dependencies, credentials, log conventions) see the
+> [top-level benchmark README](../README.md).
+
+## 1. Get the Dataset
+
+ReMe uses only the **cleaned-S** split, hosted on HuggingFace:
+[agentscope-ai/ReMe_longmemeval_clean_s_v2](https://huggingface.co/datasets/agentscope-ai/ReMe_longmemeval_clean_s_v2).
+The download script fetches it via the hf-mirror.com mirror; to use a different
+mirror, modify `BASE_URL` in [`download.py`](./download.py).
+
+```bash
+cd benchmark/longmemeval
+python download.py            # saves dataset/longmemeval_s_reme_cleaned.json; skips if already present
+```
+
+Ground truth is embedded in the data file.
+
+## 2. Run
+
+From the repository root:
+
+```bash
+python benchmark/longmemeval/run.py
+python benchmark/longmemeval/run.py --config benchmark/longmemeval/config.yaml
+python benchmark/longmemeval/run.py -q                        # quiet: only eval-level logs
+python benchmark/longmemeval/run.py --log-level WARNING       # reduce eval runner logs
+python benchmark/longmemeval/run.py --reme-log-level WARNING  # reduce reme internal logs
+python benchmark/longmemeval/run.py --eval_only               # reuse existing workspaces, query + judge only
+```
+
+## 3. Pipeline
+
+1. Load the dataset (ground truth is embedded in the data file).
+2. For each item, create an isolated workspace and ingest sessions in chronological order.
+3. Trigger `auto_dream` when consecutive sessions cross the configured hour (default 23:00).
+4. Answer each question via agentic (ReAct) mode.
+5. Judge the answer (binary yes/no) with the `answer_judge` job and print per-type accuracy.
+
+## 4. Key config — `benchmark/longmemeval/config.yaml`
+
+| Key | Meaning |
+| --- | --- |
+| `dataset.path` | Dataset file to evaluate (e.g. `longmemeval_s_reme_cleaned.json`); ground truth is included. |
+| `dataset.start_index` / `num_items` | Slice of items to evaluate. |
+| `dataset.question_types` | Filter by question type; empty = all. |
+| `dataset.workspace_root` | Per-item workspace root (`benchmark/longmemeval/workspaces/longmemeval-s`). |
+| `evaluation.num_workers` | `0` = auto (cpu-2), `1` = sequential, `>1` = parallel. |
+| `evaluation.filter_future_sessions` | Only ingest sessions with timestamp ≤ `question_date`. |
+| `reme.config` | ReMe config used (`lme.yaml`). |
+| `reme.dream_trigger_hour` / `dream_scan_days` / `dream_max_units` | Dream triggering behavior. |
+| `output.dir` | Results directory (`benchmark/longmemeval/results`). |
+
+## 5. Outputs
+
+Results are JSON files written to `output.dir` as `results_<timestamp>.json`,
+with a per-type accuracy summary also printed to the console. Logging
+conventions are shared across benchmarks — see the
+[top-level README](../README.md#outputs--logs).
+
+## 6. Reference Results
+
+### cleaned-s
+
+**Basic settings**
+
+1. Modified auto-memory prompt, auto-dream disabled.
+2. All sessions in reme-memory are strictly earlier than the question time.
+
+**Results**
+
+agentscope==2.0.4.post1, conda reme env, 32 workers, eval-only (reusing prebuilt memory)
+(2026-08-06, 500 items, total 10.0 min)
+
+| Type | Agentic | input tok/q | output tok/q | total tok/q | tool calls/q |
+|---|---|---|---|---|---|
+| knowledge-update | 0.910 | 31,581 | 589 | 32,169 | 2.90 |
+| multi-session | 0.842 | 52,837 | 1,474 | 54,311 | 4.21 |
+| single-session-assistant | 1.000 | 15,596 | 279 | 15,875 | 1.89 |
+| single-session-preference | 0.633 | 36,802 | 818 | 37,620 | 3.60 |
+| single-session-user | 0.986 | 27,433 | 359 | 27,792 | 2.60 |
+| temporal-reasoning | 0.902 | 62,674 | 985 | 63,659 | 4.97 |
+| **OVERALL** | **0.894** | **43,448** | **876** | **44,324** | **3.69** |

benchmark/longmemeval/README_ZH.md

@@ -0,0 +1,90 @@
+# LongMemEval 评测
+
+[English version](./README.md)
+
+LongMemEval 是一个面向**多轮多会话历史的长期记忆能力**的评测基准。每个条目提供一组按时间
+顺序排列的用户与助手之间的会话，以及一个只能通过推理用户自有记忆才能回答的探测问题。ReMe
+将会话摄入按条目隔离的工作区，以 agentic（ReAct）模式回答问题，最后由 LLM-as-judge 打分。
+
+题型包括单会话（user / assistant / preference）、多会话推理、知识更新与时间推理等。
+
+> 公共设置（依赖、凭据、日志约定）见[总评测说明](../README_ZH.md)。
+
+## 1. 获取数据集
+
+ReMe 仅使用 **cleaned-S** 版本，数据托管在 HuggingFace：
+[agentscope-ai/ReMe_longmemeval_clean_s_v2](https://huggingface.co/datasets/agentscope-ai/ReMe_longmemeval_clean_s_v2)。
+下载脚本经 hf-mirror.com 镜像源获取，如需更换源请修改 [`download.py`](./download.py) 中的
+`BASE_URL`。
+
+```bash
+cd benchmark/longmemeval
+python download.py            # 保存为 dataset/longmemeval_s_reme_cleaned.json，已存在则自动跳过
+```
+
+ground truth 已内嵌在数据文件中。
+
+## 2. 运行
+
+在仓库根目录执行：
+
+```bash
+python benchmark/longmemeval/run.py
+python benchmark/longmemeval/run.py --config benchmark/longmemeval/config.yaml
+python benchmark/longmemeval/run.py -q                        # 安静模式：仅评测级日志
+python benchmark/longmemeval/run.py --log-level WARNING       # 降低评测 runner 日志
+python benchmark/longmemeval/run.py --reme-log-level WARNING  # 降低 reme 内部日志
+python benchmark/longmemeval/run.py --eval_only               # 复用已有工作区，仅执行查询 + 评判
+```
+
+## 3. 流程
+
+1. 加载数据集（ground truth 已内嵌在数据文件中）。
+2. 为每个条目创建独立工作区，按时间顺序摄入会话。
+3. 当相邻会话跨越配置的时刻（默认 23:00）时触发 `auto_dream`。
+4. 以 agentic（ReAct）模式回答每个问题。
+5. 通过 `answer_judge` 任务对答案做二元（yes/no）评判，并输出各类型准确率。
+
+## 4. 关键配置 —— `benchmark/longmemeval/config.yaml`
+
+| 配置项 | 含义 |
+| --- | --- |
+| `dataset.path` | 待评测的数据集文件（如 `longmemeval_s_reme_cleaned.json`），已包含 ground truth。 |
+| `dataset.start_index` / `num_items` | 评测条目的切片范围。 |
+| `dataset.question_types` | 按问题类型过滤，空表示全部。 |
+| `dataset.workspace_root` | 条目工作区根目录（`benchmark/longmemeval/workspaces/longmemeval-s`）。 |
+| `evaluation.num_workers` | `0` = 自动（cpu-2），`1` = 串行，`>1` = 并行。 |
+| `evaluation.filter_future_sessions` | 仅摄入时间戳 ≤ `question_date` 的会话。 |
+| `reme.config` | 使用的 ReMe 配置（`lme.yaml`）。 |
+| `reme.dream_trigger_hour` / `dream_scan_days` / `dream_max_units` | dream 触发行为。 |
+| `output.dir` | 结果目录（`benchmark/longmemeval/results`）。 |
+
+## 5. 输出
+
+结果以 JSON 文件写入 `output.dir`，文件名为 `results_<timestamp>.json`，
+同时控制台会打印含各类型准确率的汇总。日志约定在各基准间通用，见
+[总说明](../README_ZH.md#输出与日志)。
+
+## 6. 参考结果
+
+### cleaned-s
+
+**基础设置**
+
+1. 使用修改后的 auto-memory prompt，关闭 auto-dream 机制
+2. reme-memory 中的全部 session 的时间一定早于 question 的时间
+
+**结果**
+
+agentscope==2.0.4.post1, conda reme env, 32 workers, eval-only（复用预构建记忆）
+（2026-08-06，500 题，总计 10.0 min）
+
+| 类型 | Agentic | input tok/q | output tok/q | total tok/q | tool calls/q |
+|---|---|---|---|---|---|
+| knowledge-update | 0.910 | 31,581 | 589 | 32,169 | 2.90 |
+| multi-session | 0.842 | 52,837 | 1,474 | 54,311 | 4.21 |
+| single-session-assistant | 1.000 | 15,596 | 279 | 15,875 | 1.89 |
+| single-session-preference | 0.633 | 36,802 | 818 | 37,620 | 3.60 |
+| single-session-user | 0.986 | 27,433 | 359 | 27,792 | 2.60 |
+| temporal-reasoning | 0.902 | 62,674 | 985 | 63,659 | 4.97 |
+| **OVERALL** | **0.894** | **43,448** | **876** | **44,324** | **3.69** |

benchmark/longmemeval/config.yaml

@@ -0,0 +1,33 @@
+# LongMemEval evaluation configuration
+# This file controls what/how to evaluate.
+
+dataset:
+  path: "benchmark/longmemeval/dataset/longmemeval_s_reme_cleaned.json"
+  start_index: 0             # first item index
+  num_items: 500            # how many items to evaluate (starting from start_index)
+  max_sessions: 0            # 0 = all sessions; >0 = limit sessions per item for testing
+  question_types: []         # filter by question_type; empty list = no filtering (all types)
+  workspace_root: "benchmark/longmemeval/workspaces/longmemeval-s"  # workspace root for item workspaces
+
+evaluation:
+  # LLM-as-judge uses the 'judge' as_llm component defined in lme.yaml
+  # Model and credentials are configured there (reading from .env)
+  # Judgment is always binary (yes/no) — defined in lme/llm_judge.yaml
+  num_workers: 32            # 0 = auto (cpu_count - 2, min 1); 1 = sequential; >1 = parallel
+  filter_future_sessions: true  # true = only ingest sessions with timestamp <= question_date
+  compress_session: false   # true = compress session chunks in search_v2 (query-aware); false = no compression
+
+reme:
+  config: "lme.yaml" # reme config to use (in reme/config/)
+  # Dream trigger: when gap between consecutive sessions crosses this hour (23:00)
+  dream_trigger_hour: 23
+  # Dream scan_days for each trigger
+  dream_scan_days: 2
+  dream_max_units: 5
+
+output:
+  dir: "benchmark/longmemeval/results"
+  log_dir: "logs"                # log directory (relative to project root)
+  log_prefix: "longmemeval"      # benchmark name used in log filenames
+  log_to_console: true
+  log_to_file: true

benchmark/longmemeval/download.py

@@ -0,0 +1,67 @@
+"""Download the LongMemEval cleaned-S dataset used by ReMe.
+
+Source: https://huggingface.co/datasets/agentscope-ai/ReMe_longmemeval_clean_s_v2
+(downloaded via the hf-mirror.com mirror for reliability).
+
+The file ``longmemeval_s_reme_cleaned.json`` is saved under ``dataset/`` next to this
+script using the same name as on the remote (``benchmark/longmemeval/config.yaml``
+points to it).
+
+Usage:
+    python download.py           # download cleaned-S (skip if it already exists)
+"""
+
+import os
+import sys
+import urllib.request
+
+BASE_URL = "https://hf-mirror.com/datasets/agentscope-ai/ReMe_longmemeval_clean_s_v2/resolve/main"
+TARGET_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "dataset")
+
+# Files to download (saved with the same name as on the remote).
+FILES = [
+    "longmemeval_s_reme_cleaned.json",
+]
+
+
+def download_file(filename: str):
+    """Download a single file from the mirror to the target directory."""
+    url = f"{BASE_URL}/{filename}"
+    dest = os.path.join(TARGET_DIR, filename)
+
+    if os.path.exists(dest):
+        size = os.path.getsize(dest)
+        print(f"  [skip] {filename} already exists ({size / 1024 / 1024:.1f} MB)")
+        return
+
+    print(f"  [downloading] {filename} ...")
+    try:
+        urllib.request.urlretrieve(url, dest, reporthook=_progress)
+        size = os.path.getsize(dest)
+        print(f"\n  [done] {filename} ({size / 1024 / 1024:.1f} MB)")
+    except Exception as e:
+        print(f"\n  [error] {filename}: {e}")
+        if os.path.exists(dest):
+            os.remove(dest)
+        sys.exit(1)
+
+
+def _progress(block_num, block_size, total_size):
+    downloaded = block_num * block_size
+    if total_size > 0:
+        pct = min(100, downloaded * 100 / total_size)
+        mb = downloaded / 1024 / 1024
+        total_mb = total_size / 1024 / 1024
+        sys.stdout.write(f"\r    {mb:.1f}/{total_mb:.1f} MB ({pct:.1f}%)")
+    else:
+        mb = downloaded / 1024 / 1024
+        sys.stdout.write(f"\r    {mb:.1f} MB downloaded")
+    sys.stdout.flush()
+
+
+if __name__ == "__main__":
+    os.makedirs(TARGET_DIR, exist_ok=True)
+    print(f"Downloading LongMemEval cleaned-S dataset to: {TARGET_DIR}\n")
+    for fname in FILES:
+        download_file(fname)
+    print("\nAll files downloaded successfully!")

benchmark/longmemeval/kill.sh

@@ -0,0 +1,76 @@
+#!/bin/bash
+# 杀死指定进程及其所有子进程
+# Usage: bash kill.sh <PID>
+
+if [ -z "$1" ]; then
+    echo "Usage: bash kill.sh <PID>"
+    echo "  杀死指定进程及其所有子进程"
+    exit 1
+fi
+
+PID=$1
+
+# 检查进程是否存在
+if ! kill -0 "$PID" 2>/dev/null; then
+    echo "进程 $PID 不存在"
+    exit 1
+fi
+
+# 递归收集所有子进程（包括子进程的子进程）
+collect_children() {
+    local parent=$1
+    local children
+    children=$(ps -o pid= --ppid "$parent" 2>/dev/null | tr -d ' ')
+    for child in $children; do
+        collect_children "$child"
+    done
+    echo "$parent"
+}
+
+# 收集进程树（子进程在前，父进程在后，保证先杀子再杀父）
+PROCESS_TREE=$(collect_children "$PID")
+TOTAL=$(echo "$PROCESS_TREE" | wc -l | tr -d ' ')
+
+echo "进程树（共 $TOTAL 个进程）:"
+while read -r p; do
+    cmd=$(ps -o args= -p "$p" 2>/dev/null | head -c 80)
+    printf "  PID=%-8s %s\n" "$p" "$cmd"
+done <<< "$PROCESS_TREE"
+
+# 先 SIGTERM 优雅终止
+echo ""
+echo "发送 SIGTERM..."
+while read -r p; do
+    kill "$p" 2>/dev/null
+done <<< "$PROCESS_TREE"
+
+# 等待最多 5 秒
+for i in $(seq 1 5); do
+    alive=false
+    while read -r p; do
+        if kill -0 "$p" 2>/dev/null; then
+            alive=true
+        fi
+    done <<< "$PROCESS_TREE"
+    if [ "$alive" = false ]; then
+        break
+    fi
+    sleep 1
+done
+
+# 检查是否还有残留，强制 SIGKILL
+remaining=false
+while read -r p; do
+    if kill -0 "$p" 2>/dev/null; then
+        remaining=true
+    fi
+done <<< "$PROCESS_TREE"
+
+if [ "$remaining" = true ]; then
+    echo "部分进程未响应，发送 SIGKILL..."
+    while read -r p; do
+        kill -9 "$p" 2>/dev/null
+    done <<< "$PROCESS_TREE"
+fi
+
+echo "已终止进程树（根 PID=$PID，共 $TOTAL 个进程）"

benchmark/longmemeval/run.py

@@ -0,0 +1,816 @@
+"""LongMemEval evaluation runner for ReMe.
+
+Evaluates ReMe's long-term memory capability using the LongMemEval dataset.
+Each item gets an isolated workspace; sessions are ingested in chronological order;
+dream is triggered when sessions cross midnight (23:00); finally questions are
+answered via an agentic (ReAct) approach and judged by an LLM.
+
+Usage:
+    python benchmark/longmemeval/run.py
+    python benchmark/longmemeval/run.py --config benchmark/longmemeval/config.yaml
+    python benchmark/longmemeval/run.py -q                          # quiet: only eval-level logs
+    python benchmark/longmemeval/run.py --log-level WARNING         # reduce eval runner logs
+    python benchmark/longmemeval/run.py --reme-log-level WARNING    # reduce reme internal logs
+    python benchmark/longmemeval/run.py --eval_only                 # query+judge only, reuse existing workspace
+"""
+
+import json
+import logging
+import os
+import re
+import shutil
+import time
+import threading
+from datetime import datetime
+from pathlib import Path
+
+import yaml
+from dotenv import load_dotenv
+
+# Load .env from project root
+_PROJECT_ROOT = Path(__file__).parent.parent.parent
+load_dotenv(_PROJECT_ROOT / ".env")
+
+# Workspace root for evaluation items — read from config.yaml (dataset.workspace_root)
+_WORKSPACE_ROOT_DEFAULT = "benchmark/longmemeval/workspaces/longmemeval-s"
+
+# ---------------------------------------------------------------------------
+# Logging
+# ---------------------------------------------------------------------------
+_DEFAULT_LOG_FORMAT = "%(asctime)s | %(levelname)s | %(message)s"
+
+logging.basicConfig(level=logging.INFO, format=_DEFAULT_LOG_FORMAT)
+logger = logging.getLogger("longmemeval")
+
+# Noisy library loggers silenced by default
+_NOISY_LOGGERS = [
+    "httpx",
+    "httpcore",
+    "openai",
+    "uvicorn",
+    "multipart",
+    "asyncio",
+    "watchfiles",
+    "filelock",
+]
+
+
+def setup_logging(
+    log_level: str,
+    reme_log_level: str,
+    log_dir: str | None = None,
+):
+    """Configure logging for the eval runner and reme internals.
+
+    Args:
+        log_level: Level for the eval runner logger (DEBUG/INFO/WARNING/ERROR).
+        reme_log_level: Level for reme's internal loguru logger.
+        log_dir: Per-run log directory (absolute path). None = no file logging.
+    """
+    numeric = getattr(logging, log_level.upper(), logging.INFO)
+    # Eval runner logger
+    logging.getLogger().setLevel(numeric)
+    logger.setLevel(numeric)
+
+    # Suppress noisy library loggers when above DEBUG
+    if numeric > logging.DEBUG:
+        for name in _NOISY_LOGGERS:
+            lib_logger = logging.getLogger(name)
+            lib_logger.setLevel(max(numeric, logging.WARNING))
+
+    # Add file handler for eval runner if log_dir is specified
+    if log_dir:
+        os.makedirs(log_dir, exist_ok=True)
+        log_filepath = os.path.join(log_dir, "runner.log")
+        file_handler = logging.FileHandler(log_filepath, encoding="utf-8")
+        file_handler.setLevel(numeric)
+        file_handler.setFormatter(logging.Formatter(_DEFAULT_LOG_FORMAT))
+        logging.getLogger().addHandler(file_handler)
+        logger.info(f"Eval runner log file: {log_filepath}")
+
+    # Reme internal logger (loguru) — will be applied per-worker via _configure_worker
+    os.environ["REME_LOG_LEVEL"] = reme_log_level.upper()
+    if log_dir:
+        os.environ["REME_LOG_DIR"] = log_dir
+
+
+def _configure_worker(
+    log_level: str,
+    reme_log_level: str,
+    log_dir: str | None = None,
+):
+    """Set up logging inside a multiprocessing worker process.
+
+    Must be called at the top of each worker because child processes inherit
+    parent state but loguru sinks are NOT shared across fork/spawn.
+    """
+    numeric = getattr(logging, log_level.upper(), logging.INFO)
+    logging.basicConfig(level=numeric, format=_DEFAULT_LOG_FORMAT, force=True)
+    logging.getLogger("longmemeval").setLevel(numeric)
+    if numeric > logging.DEBUG:
+        for name in _NOISY_LOGGERS:
+            logging.getLogger(name).setLevel(max(numeric, logging.WARNING))
+
+    # Add file handler for eval runner in worker process
+    if log_dir:
+        os.makedirs(log_dir, exist_ok=True)
+        pid = os.getpid()
+        log_filepath = os.path.join(log_dir, f"worker-{pid}.log")
+        file_handler = logging.FileHandler(log_filepath, encoding="utf-8")
+        file_handler.setLevel(numeric)
+        file_handler.setFormatter(logging.Formatter(_DEFAULT_LOG_FORMAT))
+        logging.getLogger().addHandler(file_handler)
+
+    # Re-initialize loguru for reme internals at the desired level
+    from reme.utils import get_logger
+
+    reme_log_dir = log_dir or "logs"
+    get_logger(log_dir=reme_log_dir, level=reme_log_level.upper(), force_init=True)
+
+
+# ---------------------------------------------------------------------------
+# Config loading
+# ---------------------------------------------------------------------------
+def load_eval_config(config_path: str | None = None) -> dict:
+    """Load evaluation config yaml with env-var expansion."""
+    if config_path is None:
+        config_path = str(Path(__file__).parent / "config.yaml")
+    with open(config_path, encoding="utf-8") as f:
+        raw = f.read()
+
+    # Expand ${VAR} and ${VAR:-default}
+    def _expand(m):
+        expr = m.group(1)
+        if ":-" in expr:
+            key, default = expr.split(":-", 1)
+            return os.environ.get(key, default)
+        return os.environ.get(expr, "")
+
+    raw = re.sub(r"\$\{([^}]+)\}", _expand, raw)
+    return yaml.safe_load(raw)
+
+
+# ---------------------------------------------------------------------------
+# Date utilities
+# ---------------------------------------------------------------------------
+def parse_haystack_date(date_str: str) -> datetime:
+    """Parse LongMemEval date format: '2023/05/20 (Sat) 02:21' -> datetime."""
+    m = re.match(r"(\d{4}/\d{2}/\d{2})\s+\(\w+\)\s+(\d{2}:\d{2})", date_str)
+    if not m:
+        raise ValueError(f"Cannot parse haystack date: {date_str!r}")
+    return datetime.strptime(f"{m.group(1)} {m.group(2)}", "%Y/%m/%d %H:%M")
+
+
+def to_iso(dt: datetime) -> str:
+    """Convert datetime to ISO-8601 string precise to seconds."""
+    return dt.strftime("%Y-%m-%dT%H:%M:%S")
+
+
+def should_trigger_dream(prev_dt: datetime, curr_dt: datetime, _trigger_hour: int = 23) -> bool:
+    """Check if the time gap between two sessions crosses trigger_hour (e.g. 23:00)."""
+    if prev_dt.date() == curr_dt.date():
+        return False
+    # There's at least one midnight crossing; check if trigger_hour is between them
+    # Simple heuristic: if dates differ, dream should run for the previous day
+    return True
+
+
+def sessions_sorted_by_time(item: dict) -> list[tuple[int, datetime, str, list[dict]]]:
+    """Return (original_index, parsed_datetime, session_id, messages) sorted by time."""
+    entries = []
+    for i, (date_str, sid, msgs) in enumerate(
+        zip(item["haystack_dates"], item["haystack_session_ids"], item["haystack_sessions"]),
+    ):
+        dt = parse_haystack_date(date_str)
+        entries.append((i, dt, sid, msgs))
+    # Sort by time (ascending)
+    entries.sort(key=lambda x: x[1])
+    return entries
+
+
+# ---------------------------------------------------------------------------
+# Message formatting
+# ---------------------------------------------------------------------------
+def format_messages_for_reme(messages: list[dict], session_dt: datetime) -> list[dict]:
+    """Convert LongMemEval messages to ReMe auto_memory format.
+
+    Adds: name, created_at (ISO seconds). All messages in a session share the
+    same created_at (the session timestamp).
+    """
+    formatted = []
+    for msg in messages:
+        role = msg["role"]
+        formatted.append(
+            {
+                "name": role,
+                "role": role,
+                "content": msg["content"],
+                "created_at": to_iso(session_dt),
+            },
+        )
+    return formatted
+
+
+# ---------------------------------------------------------------------------
+# LLM-as-Judge (delegated to answer_judge_step via app.run_job)
+# ---------------------------------------------------------------------------
+
+
+async def judge_response_via_job(
+    app,
+    question: str,
+    ground_truth: str,
+    response: str,
+    question_type: str,
+) -> dict:
+    """Use the answer_judge_step to evaluate a response against the golden answer."""
+    judge_resp = await app.run_job(
+        "answer_judge",
+        query=question,
+        agent_answer=response,
+        golden_answer=ground_truth,
+        question_type=question_type,
+    )
+
+    verdict = (judge_resp.answer or "").strip().lower()
+    raw_answer = (judge_resp.metadata or {}).get("raw_answer_judgement", "")
+
+    return {
+        "verdict": verdict,
+        "reason": raw_answer if verdict not in ("yes", "no") else "",
+        "metric": "binary",
+        "question_type": question_type,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Main evaluation pipeline
+# ---------------------------------------------------------------------------
+async def evaluate_item(item: dict, eval_config: dict, item_index: int, eval_only: bool = False) -> dict:
+    """Evaluate a single LongMemEval item end-to-end.
+
+    Args:
+        item: The dataset item containing question, answer, sessions, etc.
+        eval_config: The evaluation configuration dict.
+        item_index: The index of this item in the dataset.
+        eval_only: If True, skip ingestion (phases 1-3) and only run query+judge
+            using the existing workspace. Useful for re-evaluating different query
+            configurations without re-ingesting sessions.
+    """
+    from reme import Application
+    from reme.config import resolve_app_config
+    from reme.utils.evaluation_interface import track_agent_token_usage, track_job_counts
+
+    reme_cfg = eval_config["reme"]
+    dream_trigger_hour = reme_cfg.get("dream_trigger_hour", 23)
+    dream_scan_days = reme_cfg.get("dream_scan_days", 2)
+    dream_max_units = reme_cfg.get("dream_max_units", 5)
+
+    # Sort sessions by time
+    sorted_sessions = sessions_sorted_by_time(item)
+
+    # Filter out sessions that occur after question_date (if enabled)
+    filter_future = eval_config["evaluation"].get("filter_future_sessions", True)
+    if filter_future and item.get("question_date"):
+        question_dt = parse_haystack_date(item["question_date"])
+        total_before_filter = len(sorted_sessions)
+        sorted_sessions = [(i, dt, sid, msgs) for i, dt, sid, msgs in sorted_sessions if dt <= question_dt]
+        if len(sorted_sessions) < total_before_filter:
+            logger.info(
+                f"[Item {item_index}] Filtered sessions: {total_before_filter} -> {len(sorted_sessions)} "
+                f"(removed {total_before_filter - len(sorted_sessions)} future sessions "
+                f"after question_date={item['question_date']})",
+            )
+
+    logger.info(
+        "[Item %s] question_id=%s type=%s sessions=%d%s",
+        item_index,
+        item["question_id"],
+        item["question_type"],
+        len(sorted_sessions),
+        " [eval_only]" if eval_only else "",
+    )
+
+    # Use fixed workspace directory (clean it for fresh evaluation)
+    workspace_root = _PROJECT_ROOT / eval_config["dataset"].get("workspace_root", _WORKSPACE_ROOT_DEFAULT)
+    item_dir = workspace_root / f"item_{item_index}"
+    workspace_dir = str(item_dir / ".reme")
+    if eval_only:
+        if not item_dir.exists() or not Path(workspace_dir).exists():
+            raise FileNotFoundError(
+                f"[Item {item_index}] eval_only: workspace not found at {item_dir}. "
+                f"Run without --eval_only first to build the workspace.",
+            )
+    else:
+        if item_dir.exists():
+            shutil.rmtree(item_dir)
+            logger.info(f"[Item {item_index}] Cleaned existing workspace: {item_dir}")
+        else:
+            logger.info(f"[Item {item_index}] Workspace not found, creating: {item_dir}")
+        item_dir.mkdir(parents=True, exist_ok=True)
+
+    # Pre-initialize ReMe's loguru logger with the correct log_dir
+    # (singleton — Application.__init__ will reuse this instance)
+    output_cfg = eval_config.get("output", {})
+    if output_cfg.get("log_to_file", False):
+        reme_log_dir = os.environ.get("REME_LOG_DIR")
+        if reme_log_dir:
+            from reme.utils import get_logger
+
+            get_logger(
+                log_dir=reme_log_dir,
+                level=os.environ.get("REME_LOG_LEVEL", "INFO"),
+                log_to_console=output_cfg.get("log_to_console", True),
+                log_to_file=True,
+                force_init=True,
+            )
+
+    cfg = resolve_app_config(
+        config=reme_cfg["config"],
+        workspace_dir=workspace_dir,
+        log_to_console=output_cfg.get("log_to_console", True),
+        log_to_file=output_cfg.get("log_to_file", False),
+        enable_logo=False,
+    )
+
+    app = Application(**cfg)
+    await app.start()
+
+    try:
+        dream_dates_triggered = set()
+        dream_available = True  # Set to False if auto_dream job is not found
+
+        if not eval_only:
+            # ── Phase 1: Ingest sessions ──────────────────────────────
+            prev_dt = None
+
+            for idx, (_, session_dt, session_id, messages) in enumerate(sorted_sessions):
+                # Check if dream should be triggered before this session
+                if (
+                    dream_available
+                    and prev_dt is not None
+                    and should_trigger_dream(prev_dt, session_dt, dream_trigger_hour)
+                ):
+                    dream_date = prev_dt.strftime("%Y-%m-%d")
+                    if dream_date not in dream_dates_triggered:
+                        logger.info(f"[Item {item_index}] Triggering dream for date={dream_date}")
+                        try:
+                            dream_resp = await app.run_job(
+                                "auto_dream",
+                                date=dream_date,
+                                scan_days=dream_scan_days,
+                                max_units=dream_max_units,
+                            )
+                            logger.info(
+                                f"[Item {item_index}] Dream done: success={dream_resp.success} "
+                                f"answer={dream_resp.answer[:100] if dream_resp.answer else ''}",
+                            )
+                        except Exception as e:
+                            if "not found" in str(e).lower():
+                                dream_available = False
+                                logger.warning(f"[Item {item_index}] auto_dream job not found, skipping all dreams")
+                            else:
+                                logger.warning(f"[Item {item_index}] Dream failed for {dream_date}: {e}")
+                        dream_dates_triggered.add(dream_date)
+                        # Index update after dream to pick up new digest nodes
+                        await app.run_job("index_update")
+
+                # Format and ingest the session
+                formatted_msgs = format_messages_for_reme(messages, session_dt)
+                date_str = session_dt.strftime("%Y-%m-%d")
+
+                logger.info(
+                    f"[Item {item_index}] Ingesting session {idx+1}/{len(sorted_sessions)} "
+                    f"id={session_id} date={date_str} msgs={len(formatted_msgs)}",
+                )
+                resp = await app.run_job(
+                    "auto_memory",
+                    messages=formatted_msgs,
+                    session_id=session_id,
+                    date=date_str,
+                )
+                if not resp.success:
+                    logger.warning(
+                        f"[Item {item_index}] auto_memory failed for session {session_id}: {resp.answer}",
+                    )
+
+                # Manual index update after each session
+                await app.run_job("index_update")
+
+                prev_dt = session_dt
+
+            # ── Phase 2: Final dream for the last day ─────────────────
+            if dream_available and prev_dt is not None:
+                last_dream_date = prev_dt.strftime("%Y-%m-%d")
+                if last_dream_date not in dream_dates_triggered:
+                    logger.info(f"[Item {item_index}] Final dream for date={last_dream_date}")
+                    try:
+                        await app.run_job(
+                            "auto_dream",
+                            date=last_dream_date,
+                            scan_days=dream_scan_days,
+                            max_units=dream_max_units,
+                        )
+                    except Exception as e:
+                        if "not found" in str(e).lower():
+                            dream_available = False
+                            logger.warning(f"[Item {item_index}] auto_dream job not found, skipping all dreams")
+                        else:
+                            logger.warning(f"[Item {item_index}] Final dream failed: {e}")
+                    dream_dates_triggered.add(last_dream_date)
+                    # Index update after final dream
+                    await app.run_job("index_update")
+
+            # ── Phase 3: Digest update ────────────────────────────────
+            await app.run_job("digest_update")
+
+        # ── Phase 4: Ask question via agentic_answer job (ReAct agent) ──
+        question = item["question"]
+        compress_session = bool(eval_config["evaluation"].get("compress_session", False))
+        question_date_raw = item.get("question_date", "")
+        question_dt = parse_haystack_date(question_date_raw) if question_date_raw else None
+        query_time = to_iso(question_dt) if question_dt else ""
+        logger.info(
+            f"[Item {item_index}] Asking (agentic): {question[:80]}... query_time={query_time}",
+        )
+
+        with (
+            track_job_counts(["search"], app.context) as tool_counts,
+            track_agent_token_usage(
+                ["bench"],
+                app.context,
+            ) as token_usages,
+        ):
+            query_resp = await app.run_job(
+                "agentic_answer",
+                query=question,
+                query_time=query_time,
+                compress_session=compress_session,
+            )
+        agentic_tool_counts = tool_counts
+        agentic_token_usage = token_usages["bench"]
+        agentic_response = (query_resp.answer or "").strip()
+        if not agentic_response:
+            agentic_response = "(no answer generated)"
+
+        logger.info(f"[Item {item_index}] Agentic response: {agentic_response[:200]}...")
+        logger.info(f"[Item {item_index}] Agentic tool calls: {agentic_tool_counts}")
+        logger.info(f"[Item {item_index}] Bench token usage: {agentic_token_usage}")
+
+        # ── Phase 5: Judge agentic response (via answer_judge_step) ──────────
+        logger.info(f"[Item {item_index}] Judging agentic (binary, type={item['question_type']})...")
+        agentic_judgment = await judge_response_via_job(
+            app=app,
+            question=question,
+            ground_truth=item["answer"],
+            response=agentic_response,
+            question_type=item["question_type"],
+        )
+        logger.info(f"[Item {item_index}] agentic binary result: {agentic_judgment}")
+
+    finally:
+        await app.close()
+
+    return {
+        "question_id": item["question_id"],
+        "question_type": item["question_type"],
+        "question": question,
+        "ground_truth": item["answer"],
+        "agentic_response": agentic_response,
+        "agentic_judgment": agentic_judgment,
+        "agentic_tool_counts": agentic_tool_counts,
+        "agentic_token_usage": agentic_token_usage,
+        "sessions_ingested": len(sorted_sessions),
+        "dreams_triggered": len(dream_dates_triggered),
+    }
+
+
+# ---------------------------------------------------------------------------
+# Worker: runs a single item in its own process with its own event loop
+# ---------------------------------------------------------------------------
+def _evaluate_item_worker(task_input: tuple) -> dict:
+    """Worker function for multiprocessing. Each process gets its own event loop."""
+    item, eval_config, item_index, log_level, reme_log_level, eval_only, log_dir = task_input
+    import asyncio  # pylint: disable=import-outside-toplevel
+
+    _configure_worker(log_level, reme_log_level, log_dir=log_dir)
+
+    # Permanently suppress "Task exception was never retrieved" /
+    # "Event loop is closed" noise from httpx AsyncClient GC cleanup.
+    # These fire AFTER asyncio.run() closes the loop, during Python's
+    # garbage collection of httpx connection-pool tasks — harmless.
+    logging.getLogger("asyncio").setLevel(logging.CRITICAL)
+
+    return asyncio.run(evaluate_item(item, eval_config, item_index, eval_only=eval_only))
+
+
+def _indexed_worker(indexed_input: tuple) -> tuple:
+    """Module-level wrapper for imap_unordered with index tracking."""
+    idx, task_input = indexed_input
+    return idx, _evaluate_item_worker(task_input)
+
+
+def _resolve_num_workers(configured: int) -> int:
+    """Resolve num_workers: 0=auto (cpu_count-2, min 1), 1=sequential, >1=parallel."""
+    if configured == 0:
+        return max(1, (os.cpu_count() or 4) - 2)
+    return max(1, configured)
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+def main(
+    config_path: str | None = None,
+    log_level: str = "INFO",
+    reme_log_level: str = "INFO",
+    eval_only: bool = False,
+):
+    """Run the LongMemEval evaluation pipeline.
+
+    Args:
+        config_path: Path to the YAML config file.
+        log_level: Log level for the eval runner.
+        reme_log_level: Log level for reme internal logs.
+        eval_only: If True, skip ingestion and only run query+judge using
+            existing workspaces.
+    """
+    from multiprocessing import Pool  # pylint: disable=import-outside-toplevel
+
+    # Load config BEFORE logging setup so log_dir is available
+    eval_config = load_eval_config(config_path)
+
+    # Resolve per-run log directory from config
+    output_cfg = eval_config.get("output", {})
+    log_dir_abs = None
+    if output_cfg.get("log_to_file", False):
+        log_dir_raw = output_cfg.get("log_dir", "logs")
+        log_prefix = output_cfg.get("log_prefix", "longmemeval")
+        run_ts = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+        log_dir_abs = str(_PROJECT_ROOT / log_dir_raw / f"{log_prefix}_{run_ts}")
+
+    setup_logging(log_level, reme_log_level, log_dir=log_dir_abs)
+    dataset_cfg = eval_config["dataset"]
+
+    # Load dataset
+    dataset_path = _PROJECT_ROOT / dataset_cfg["path"]
+    logger.info(f"Loading dataset from {dataset_path}")
+    with open(dataset_path, encoding="utf-8") as f:
+        data = json.load(f)
+
+    start = dataset_cfg.get("start_index", 0)
+    num_items = dataset_cfg.get("num_items", 0)
+    if num_items > 0:
+        raw_items = data[start : start + num_items]
+    else:
+        raw_items = data[start:]
+
+    # Build item list
+    items_with_idx = [(start + i, item) for i, item in enumerate(raw_items)]
+
+    # Filter by question_type if specified
+    question_types = dataset_cfg.get("question_types") or []
+    if question_types:
+        before_filter = len(items_with_idx)
+        items_with_idx = [(idx, item) for idx, item in items_with_idx if item.get("question_type") in question_types]
+        logger.info(
+            f"Filtered by question_types={question_types}: {before_filter} -> {len(items_with_idx)} items",
+        )
+
+    # Filter by question_id if specified
+    question_ids = dataset_cfg.get("question_ids") or []
+    if question_ids:
+        qid_set = set(question_ids)
+        before_filter = len(items_with_idx)
+        items_with_idx = [(idx, item) for idx, item in items_with_idx if item.get("question_id") in qid_set]
+        logger.info(
+            f"Filtered by question_ids ({len(qid_set)} ids): {before_filter} -> {len(items_with_idx)} items",
+        )
+
+    logger.info(
+        "Evaluating %d item(s) starting from index %d%s",
+        len(items_with_idx),
+        start,
+        " [eval_only: query+judge only]" if eval_only else "",
+    )
+
+    # Resolve parallelism
+    num_workers = _resolve_num_workers(eval_config["evaluation"].get("num_workers", 1))
+    logger.info(f"Using {num_workers} worker(s)")
+
+    # Create output directory
+    output_dir = _PROJECT_ROOT / output_cfg.get("dir", "benchmark/longmemeval/results")
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    # Create workspace root directory
+    workspace_root = _PROJECT_ROOT / dataset_cfg.get("workspace_root", _WORKSPACE_ROOT_DEFAULT)
+    workspace_root.mkdir(parents=True, exist_ok=True)
+
+    # Pre-check: verify all workspaces exist in eval_only mode
+    if eval_only:
+        missing_items = []
+        for orig_idx, _ in items_with_idx:
+            item_dir = workspace_root / f"item_{orig_idx}"
+            if not item_dir.exists() or not (item_dir / ".reme").exists():
+                missing_items.append(orig_idx)
+        if missing_items:
+            preview = missing_items[:10]
+            suffix = "..." if len(missing_items) > 10 else ""
+            raise FileNotFoundError(
+                f"eval_only: {len(missing_items)} workspace(s) not found under {workspace_root}. "
+                f"Missing item indices: {preview}{suffix}. "
+                f"Run without --eval_only first to build the workspaces.",
+            )
+
+    # Build task args — include log levels, eval_only flag, and log paths (use original index for workspace lookup)
+    task_args = [
+        (item, eval_config, orig_idx, log_level, reme_log_level, eval_only, log_dir_abs)
+        for orig_idx, item in items_with_idx
+    ]
+
+    # Progress tracking (force print regardless of log level, every 10 minutes)
+    total_items = len(task_args)
+    completed_count = [0]  # use list for mutability in closure
+    start_time = time.time()
+    progress_lock = threading.Lock()
+
+    def _print_progress(prefix: str = "PROGRESS"):
+        elapsed = time.time() - start_time
+        elapsed_min = elapsed / 60
+        done = completed_count[0]
+        pct = 100.0 * done / total_items if total_items else 0
+        eta_str = "N/A"
+        if done > 0:
+            eta_sec = elapsed / done * (total_items - done)
+            eta_str = f"{eta_sec/60:.1f}min"
+        print(
+            f"[{prefix}] {datetime.now().strftime('%Y-%m-%d %H:%M:%S')} | "
+            f"{done}/{total_items} ({pct:.1f}%) completed | "
+            f"elapsed={elapsed_min:.1f}min | ETA={eta_str}",
+            flush=True,
+        )
+
+    def _progress_timer():
+        """Background thread: print progress every 10 minutes."""
+        while not _timer_stop.is_set():
+            _timer_stop.wait(600)  # 10 minutes
+            if not _timer_stop.is_set():
+                with progress_lock:
+                    _print_progress()
+
+    _timer_stop = threading.Event()
+    timer_thread = threading.Thread(target=_progress_timer, daemon=True)
+    timer_thread.start()
+
+    # Run evaluation
+    if num_workers == 1:
+        # Sequential mode
+        results = []
+        for task_input in task_args:
+            result = _evaluate_item_worker(task_input)
+            results.append(result)
+            with progress_lock:
+                completed_count[0] += 1
+    else:
+        # Parallel mode — use imap_unordered for progress tracking
+        results = [None] * total_items
+        indexed_args = list(enumerate(task_args))
+
+        with Pool(processes=num_workers) as pool:
+            for idx, result in pool.imap_unordered(_indexed_worker, indexed_args):
+                results[idx] = result
+                with progress_lock:
+                    completed_count[0] += 1
+
+    # Stop progress timer
+    _timer_stop.set()
+    timer_thread.join(timeout=2)
+
+    # Save results
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    output_file = output_dir / f"results_{timestamp}.json"
+    with open(output_file, "w", encoding="utf-8") as f:
+        json.dump(results, f, ensure_ascii=False, indent=2)
+    logger.info(f"Results saved to {output_file}")
+
+    # Final progress
+    _print_progress("FINAL")
+
+    _print_summary(results, start_time)
+
+
+# ---------------------------------------------------------------------------
+# Summary printing
+# ---------------------------------------------------------------------------
+def _print_summary(results: list[dict], start_time: float) -> None:
+    """Print per-item verdicts and per-type accuracy."""
+    print("\n" + "=" * 60)
+    print("EVALUATION RESULTS")
+    print("=" * 60)
+
+    def _accumulate(judgment_key):
+        correct = 0
+        stats: dict = {}  # {question_type: {correct: int, total: int}}
+        for r in results:
+            qtype = r["question_type"]
+            verdict = r.get(judgment_key, {}).get("verdict", "N/A")
+            if qtype not in stats:
+                stats[qtype] = {"correct": 0, "total": 0}
+            stats[qtype]["total"] += 1
+            if verdict == "yes":
+                correct += 1
+                stats[qtype]["correct"] += 1
+        return correct, stats
+
+    agentic_correct, agentic_type_stats = _accumulate("agentic_judgment")
+
+    total = len(results)
+
+    # Per-item verdict rows
+    for r in results:
+        a_verdict = r.get("agentic_judgment", {}).get("verdict", "N/A")
+        print(f"  [{r['question_id']}] type={r['question_type']}  agentic={a_verdict}")
+
+    print("\n" + "-" * 60)
+    print(f"  Items: {total}")
+
+    # Agentic stats
+    print("\n  ── Agentic (ReAct) ──")
+    print(f"  Overall accuracy: {agentic_correct}/{total} ({100*agentic_correct/total:.1f}%)")
+    tool_call_totals = [sum(r.get("agentic_tool_counts", {}).values()) for r in results]
+    tool_call_mean, tool_call_std = _mean_and_std(tool_call_totals)
+    print(f"  Tool calls/query: mean={tool_call_mean:.2f} std={tool_call_std:.2f}")
+    token_usages = [r.get("agentic_token_usage", {}) for r in results]
+    print("  Bench reported tokens/query:")
+    for metric in _TOKEN_USAGE_METRICS:
+        values = [usage[metric] for usage in token_usages if usage.get(metric) is not None]
+        if values:
+            mean, std = _mean_and_std(values)
+            print(f"    {metric}: mean={mean:.2f} std={std:.2f}")
+        else:
+            print(f"    {metric}: unavailable")
+    print("  Per-type accuracy:")
+    for qtype, stats in sorted(agentic_type_stats.items()):
+        acc = 100 * stats["correct"] / stats["total"] if stats["total"] else 0
+        print(f"    {qtype}: {stats['correct']}/{stats['total']} ({acc:.1f}%)")
+
+    print("=" * 60)
+    total_elapsed = time.time() - start_time
+    print(f"\n  Total time: {total_elapsed/60:.1f} min")
+    print("\n" + "=" * 60)
+    print("  [DONE] EVALUATION COMPLETED SUCCESSFULLY")
+    print("=" * 60 + "\n")
+
+
+_TOKEN_USAGE_METRICS = (
+    "input_tokens",
+    "output_tokens",
+    "total_tokens",
+)
+
+
+def _mean_and_std(values: list[int]) -> tuple[float, float]:
+    """Return population mean and standard deviation for one per-query metric."""
+    if not values:
+        return 0.0, 0.0
+    mean = sum(values) / len(values)
+    return mean, (sum((value - mean) ** 2 for value in values) / len(values)) ** 0.5
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser(description="LongMemEval evaluation runner")
+    parser.add_argument("--config", type=str, default=None, help="Path to config.yaml")
+    parser.add_argument(
+        "--log-level",
+        type=str,
+        default="INFO",
+        choices=["DEBUG", "INFO", "WARNING", "ERROR"],
+        help="Log level for the eval runner (default: INFO)",
+    )
+    parser.add_argument(
+        "--reme-log-level",
+        type=str,
+        default="INFO",
+        choices=["DEBUG", "INFO", "WARNING", "ERROR"],
+        help="Log level for reme internal logs — loguru (default: INFO)",
+    )
+    parser.add_argument(
+        "-q",
+        "--quiet",
+        action="store_true",
+        help="Shortcut for --log-level WARNING --reme-log-level WARNING",
+    )
+    parser.add_argument(
+        "--eval_only",
+        action="store_true",
+        help="Skip ingestion (phases 1-3). Reuse existing workspaces and only run query+judge.",
+    )
+    args = parser.parse_args()
+
+    if args.quiet:
+        args.log_level = "WARNING"
+        args.reme_log_level = "WARNING"
+
+    main(args.config, args.log_level, args.reme_log_level, eval_only=args.eval_only)

benchmark/pibench/.gitignore

@@ -0,0 +1,14 @@
+# 含真实 API key，绝不入库
+env.sh
+
+# 运行时产物（含对话内容，勿入库）
+logs/
+outputs/
+reme_workspace/
+nanobot_workspace/
+
+# 数据符号链接（指向外部 π-Bench 仓库）
+data
+
+__pycache__/
+*.pyc

benchmark/pibench/README.md

@@ -0,0 +1,327 @@
+[中文版 / Chinese version](./README_ZH.md)
+
+# π-Bench Evaluation Suite
+
+A glue layer that connects the **ReMe agent (with persistent memory)** to
+**π-Bench** (Proactive Personal Assistant Benchmark). This directory contains
+only the minimal code and configuration needed for the integration: the
+π-Bench framework (`src/`), evaluation data (`data/`), the AppWorld tool
+environment, and ReMe itself are all **external third-party dependencies**,
+referenced in place via symlink and environment variables and never bundled
+with this suite.
+
+- π-Bench: https://github.com/Simplified-Reasoning/Pi-Bench (arXiv: 2605.14678)
+- ReMe: the root of the ReMe repository this suite lives in (recommended
+  location: `ReMe/benchmark/pibench/`)
+
+## 1. Architecture
+
+```
+π-Bench runner (src.main --mode run)
+  │  user_agent (simulated-user LLM) walks data/{persona}/episode.yaml
+  │  task by task, chatting with the agent over multiple turns and judging
+  │  hidden intents (PROC) during the run phase
+  ▼
+test server (π-Bench scripts/test_server.py, HTTP long-polling)
+  ▲ /send            │ /poll
+  │                  ▼
+bridge_reme.py ──────────────► ReMe Application (embedded as a library)
+  │                             ├─ agent_wrapper: agent under test (AgentScope)
+  │                             ├─ jobs: search / auto_memory / daily_write
+  │                             └─ workspace: reme_workspace/{persona}/
+  │                                (isolated persistent memory per persona)
+  └──── MCP ────► AppWorld MCP ────► AppWorld APIs (tool/app environment)
+
+π-Bench runner (src.main --mode eval)
+  judger (judge LLM) reads the traces and scores each checklist item (COMP)
+```
+
+Key points:
+- The bridge runs on **ReMe's own venv python** and uses ReMe as a library
+  (`resolve_app_config` + `Application`); **no ReMe source modification** is
+  required.
+- Every incoming user message automatically triggers a ReMe memory `search`
+  and injects the matched memories (tuning knobs in §8); on task end (reset)
+  the session is distilled into daily notes by `auto_memory`.
+- Tool calls executed by the agent (AppWorld MCP + ReMe job tools) are
+  captured per turn into the trace as `tool_steps`, so π-Bench
+  `tools_evaluation_path` scripts can score tool behavior (§7).
+- π-Bench's `data/`, `src/` and AppWorld are not part of this suite; install
+  π-Bench first (§3.1).
+
+## 2. Directory layout
+
+```
+pibench/
+├── README.md / README_ZH.md  # this document (English / Chinese)
+├── env.sh.example            # environment template (copy to env.sh, fill TODOs)
+├── bridge_reme.py            # ReMe ↔ test server bridge (memory inject/save,
+│                             #   profile injection, tool-trace capture)
+├── run_persona.sh            # full pipeline for ONE persona (5 services + run + eval)
+├── run_all.sh                # batch over 5 personas (fresh/resume, default parallel=2)
+├── resume.py                 # checkpoint resume: completion detection + surgical
+│                             #   cleanup of interrupted tasks' residual memory
+├── fix_trace_logs.py         # run outputs → ~/.nanobot/trace_logs conversion,
+│                             #   merging tool sidecars into turn files (pre-eval)
+├── .gitignore                # excludes env.sh and all runtime artifacts
+└── config/
+    ├── models/reme.yaml      # runner model config (model_id=reme)
+    └── bench/evaluation/trace_history.yaml   # trace render policy (shipped with
+                                              #   the suite; passed via --history-config-path)
+```
+
+Generated at runtime (all git-ignored): `data` (symlink), `logs/`, `outputs/`,
+`reme_workspace/`, `nanobot_workspace/`.
+
+## 3. Prerequisites (third-party, install first)
+
+### 3.1 π-Bench repository (with AppWorld)
+
+```bash
+git clone https://github.com/Simplified-Reasoning/Pi-Bench.git <pi-bench-dir>
+cd <pi-bench-dir>
+python3.11 -m venv .venv            # scripts expect exactly this venv name
+source .venv/bin/activate
+pip install -e .                    # pibench runner (src.main)
+bash scripts/setup_appworld.sh      # install AppWorld and download its data (large)
+```
+
+Post-install sanity checks:
+```bash
+ls data/                    # should contain researcher marketer pharmacist law_trainee Financier
+.venv/bin/python -c "import src" && echo OK
+.venv/bin/appworld --help >/dev/null && echo OK
+```
+
+### 3.2 ReMe repository
+
+```bash
+cd <reme-dir>               # ReMe repository root (contains the reme/ package)
+python3.11 -m venv .venv    # scripts expect exactly this venv name
+source .venv/bin/activate
+pip install -e .            # or ReMe's own install flow; `import reme` must work
+```
+
+Sanity check: `.venv/bin/python -c "import reme; print('ok')"`
+
+## 4. Install this suite (step by step)
+
+1. **Place the suite** (recommended inside the ReMe repo so `REME_DIR` is
+   inferred automatically):
+   ```bash
+   cp -r pibench <reme-dir>/benchmark/pibench
+   cd <reme-dir>/benchmark/pibench
+   ```
+   If placed elsewhere, set `REME_DIR` explicitly in env.sh later.
+
+2. **Create the environment file and fill in the custom parameters**:
+   ```bash
+   cp env.sh.example env.sh
+   ```
+   Open `env.sh`; required items (marked TODO):
+   | Variable | Description |
+   |---|---|
+   | `PI_BENCH_ROOT` | π-Bench repo root (contains `src/` `data/` `.venv` `third_party/appworld`) |
+   | `USER_API_KEY` | API key of the simulated-user LLM (run phase, hidden-intent judging) |
+   | `JUDGER_API_KEY` | API key of the judger LLM (eval phase, checklist scoring) |
+   | `BRAVE_SEARCH_API_KEY` | optional; for the agent's web_search tool, `dummy` when unused |
+
+   Optional tuning: `REME_MODEL_NAME` (base model of the agent under test),
+   `REME_DIR`, `REME_LLM_BASE_URL` (default: DashScope OpenAI-compatible
+   endpoint).
+
+3. **Link the evaluation data** (referenced in place, never copied):
+   ```bash
+   ln -s "$PI_BENCH_ROOT/data" data
+   ```
+
+4. **(Optional) adjust model config** `config/models/reme.yaml`:
+   - `user_agent.model` / `judger.model`: model names for the simulated user
+     and the judger (literal values; π-Bench only expands `${ENV}` in
+     base_url/api_key).
+   - `run.turn_timeout`, `max_tool_iterations`, etc. as needed.
+
+5. **Smoke check** (does not start the evaluation):
+   ```bash
+   bash -n run_all.sh && bash -n run_persona.sh
+   source env.sh && "$REME_DIR/.venv/bin/python" -c "import reme; print('reme ok')"
+   ```
+
+## 5. Run the evaluation
+
+> ⚠️ For long runs use `screen`, **not nohup** (nohup loses the permission
+> context in sandboxed/restricted environments and breaks child processes).
+
+```bash
+# Full official run: wipe ALL personas' memory/outputs/traces first (default
+# fresh mode, parallel=2)
+mkdir -p logs   # on a fresh deployment logs/ does not exist yet
+screen -dmS pibench_suite bash -c "cd $(pwd) && bash run_all.sh > logs/run_all_master.log 2>&1"
+
+# Checkpoint continuation (after an interruption; no wipe, completed tasks skipped)
+bash run_all.sh --resume
+
+# Other usages
+bash run_all.sh --parallel 1          # sequential
+bash run_all.sh --resume --skip-eval  # run phase only
+bash run_persona.sh researcher        # single persona (default --resume semantics)
+bash run_persona.sh researcher --fresh
+```
+
+Time reference: 5 personas × 20 tasks, parallel=2, fresh full run ≈ 12–14 hours.
+
+`run_all.sh` exits non-zero when any persona fails, so upstream automation
+cannot mistake a partially failed suite run for a success.
+
+## 6. Port allocation (parallel personas never collide)
+
+| persona     | AppWorld API | AppWorld MCP | Test Server | ReMe internal service |
+|-------------|------|-------|------|-------|
+| marketer    | 9001 | 10001 | 9998 | 18766 |
+| law_trainee | 9002 | 10002 | 9997 | 18767 |
+| pharmacist  | 9003 | 10003 | 9996 | 18768 |
+| researcher  | 9004 | 10004 | 9995 | 18765 |
+| Financier   | 9005 | 10005 | 9994 | 18769 |
+
+## 7. Outputs and scores
+
+- **Results**: `outputs/reme/{persona}/{task}/eval/results/*_result.json`
+  - `overall_average_score`: checklist completeness (COMP; the judger scores
+    each criterion YES/NO, weighted across dependency groups)
+  - `overall_proactiveness_average_score`: proactiveness (PROC; the
+    user_agent judges hidden-intent coverage during the run phase; each task
+    file also carries the global average)
+- **Traces**: `~/.nanobot/trace_logs/reme/{persona}/{task}/...` (the scoring
+  input of the eval phase)
+- **Logs**: `logs/` (`suite_<persona>.log` per persona; `bridge_*`,
+  `runner_run/eval_*`, `appworld_*`, `test_server_*` per service)
+- **Memory store**: `reme_workspace/{persona}/` (daily/digest notes, raw
+  session dialogs, BM25 index, etc.; persistent across runs, wiped only in
+  fresh mode)
+
+Score summary:
+```bash
+grep -h "overall_average_score\|overall_proactiveness" \
+  outputs/reme/*/*/eval/results/*_result.json | head
+```
+
+### Tool-trace capture (tools_evaluation support)
+
+Some tasks define `objectives.tools_evaluation_path`: Python scripts that
+score tool behavior (e.g. "the temporary Todoist board was created and
+removed"). They need the executed tool calls in the trace. The pipeline:
+
+1. During `reply()`, the bridge reads the persisted AgentScope session state
+   after each turn and extracts the new `tool_call` / `tool_result` blocks
+   (tool name, arguments, result).
+2. Records are appended to
+   `outputs/reme/{persona}/{task}/history/{ts}-tools.jsonl`, tagged with the
+   turn number; AgentScope MCP names (`mcp__AppWorld__<tool>`) are normalized
+   to the π-Bench convention (`mcp_appworld_<tool>`).
+3. `fix_trace_logs.py` pairs each `{ts}-messages.jsonl` run with the
+   temporally closest tools sidecar and merges the records into the generated
+   `turn_N.json` files under the `tool_steps` key — one of the two
+   tool-history formats understood by π-Bench's `collect_tool_history()`.
+4. The eval phase then feeds `tool_steps` to both the tools_evaluation
+   scripts and the rendered `<tool_trace_extracts>` seen by the judger.
+
+## 8. Memory mechanism (core design of this suite)
+
+- **Persona isolation**: each persona has its own workspace
+  (`reme_workspace/{persona}/`); the bridge takes an exclusive
+  `.bridge.lock` on it at startup, so two bridges can never share one memory
+  store, and one persona's memory search can never reach another's memories.
+- **Writes**: on task end (runner sends reset), the session is distilled by
+  the `auto_memory` job into daily notes and indexed by the background
+  watcher (BM25). Saves are non-blocking background tasks; the first message
+  of a new session waits for in-flight writes before searching.
+- **Reads**: on every incoming user message the bridge runs one `search` and
+  injects matched memories (`[Relevant memories from previous sessions]`
+  prefix); without matches the message passes through unchanged. Retrieval
+  tuning (bridge CLI flags, adjustable in run_persona.sh):
+  - `--search-limit 3`: at most 3 memory chunks injected per message;
+  - `--search-min-score 2.0`: weak BM25 hits are filtered out;
+  - `tool_context_id` rotates per task: chunks already injected within the
+    same task are not re-injected (ReMe's seen-chunk dedup, 24h TTL); normal
+    recall resumes after task boundaries.
+- **No self-leakage**: the in-progress session is not in the store yet
+  (saves happen on reset), so a task can never retrieve its own unfinished
+  content.
+- The agent also holds `search`/`daily_write` tools and can retrieve/record
+  proactively.
+- **System prompt**: `bridge_reme.py:build_system_prompt()` embeds the
+  HIDDEN-NEEDS protocol (proactiveness-oriented) and injects the persona
+  profile from `data/{persona}/profile.yaml` into every turn's system prompt.
+
+## 9. Checkpoint resume and memory-cleanup semantics
+
+- **Completion detection** (resume.py): scans
+  `outputs/reme/{persona}/**/history/*-log.jsonl` and
+  `outputs/reme/{persona}/run/*-log.jsonl` for
+  `Task finished task_id=X status=Y`. The status with the **newest event
+  timestamp** wins per task (record `timestamp`, falling back to
+  `timestamp_iso`, then to the timestamp embedded in the log file name) —
+  file category and read order alone can never override a newer record, so an
+  old run-level SUCCESS cannot mask a newer per-task ERROR. `SUCCESS /
+  MAX_TURNS / TIMEOUT` count as completed; `ERROR` and never-started tasks
+  are re-run (passed to the runner as repeated `--task-id` flags in episode
+  order).
+- **Answer-leak prevention**: an interrupted task may already have been
+  distilled into daily notes during graceful shutdown; re-running it with
+  that memory injected would inflate scores. Before resuming,
+  `resume.py cleanup` therefore removes residual memory **only for tasks
+  about to be re-run** (daily/digest notes, session/dialog, mem_session;
+  matched via `session_id = pibench_{task}_*`). Completed tasks' memories are
+  never touched. Daily index files are refreshed **only for the dates that
+  lost notes**, by full workspace-relative wikilink path — and when the ReMe
+  package is importable, the refresh reuses ReMe's own daily-index rebuild
+  logic (`refresh_day_index`), so same-named notes on other dates are never
+  modified.
+- **fresh vs resume are mutually exclusive**: a full memory wipe belongs to
+  fresh mode only (`run_all.sh` default, executed before any service starts);
+  resume never wipes.
+
+## 10. Customization entry points
+
+| Goal | Location |
+|---|---|
+| Base model of the agent under test | `REME_MODEL_NAME` in `env.sh` |
+| user_agent / judger models | `config/models/reme.yaml` |
+| Agent system prompt | `bridge_reme.py` `build_system_prompt()` |
+| Memory retrieval limit/threshold | `--search-limit/--search-min-score` on the bridge command in `run_persona.sh` |
+| ReMe internal parameters | **Do not modify ReMe source**; write a dedicated config modeled on `reme/config/beam.yaml` and override via `resolve_app_config(config=...)` (see bridge `_init_reme_app`) |
+| Turn timeout / tool iteration cap | `config/models/reme.yaml` `run.turn_timeout`, `model.max_tool_iterations` |
+
+## 11. Troubleshooting
+
+- **Port already in use**: the scripts auto-kill residual processes on the
+  four port groups above; if another suite (e.g. a different π-Bench
+  experiment) holds them, stop it first or change the port table in
+  run_persona.sh.
+- **Bridge exits immediately with workspace locked**: another bridge already
+  holds the same workspace; make sure each persona uses its own
+  `--workspace-dir` (the scripts allocate one per persona).
+- **Runner reports `${USER_API_KEY} ... empty`**: env.sh is unfilled or not
+  sourced; run_persona.sh sources env.sh automatically — when running the
+  runner manually, `source env.sh` first.
+- **`Cannot import 'reme'`**: the bridge must run with
+  `${REME_DIR}/.venv/bin/python` (run_persona.sh already does); otherwise
+  check that `REME_DIR` points at the ReMe repository root.
+- **AppWorld fails to start**: run `bash scripts/setup_appworld.sh` in the
+  π-Bench repo first (downloads data); inspect
+  `logs/appworld_*_<persona>.log`.
+- **trace_history.yaml not found**: the runner needs
+  `config/bench/evaluation/trace_history.yaml`; this suite ships the file and
+  passes it explicitly via `--history-config-path`, and run_persona.sh fails
+  fast with a clear error if it is missing. Always launch run_persona.sh /
+  run_all.sh from the suite directory.
+
+## 12. Privacy and security
+
+- The suite code and config templates contain **no real API keys, user names
+  or absolute paths**; real keys live only in your local `env.sh`
+  (git-ignored).
+- `logs/`, `outputs/`, `reme_workspace/` and `nanobot_workspace/` contain
+  full conversations and model outputs; never commit or share them.
+- The `data` symlink points at the official π-Bench evaluation data; respect
+  its data license terms.

benchmark/pibench/README_ZH.md

@@ -0,0 +1,284 @@
+# π-Bench 评测说明
+
+[English version](./README.md)
+
+将 **ReMe agent（带持久记忆）** 接入 **π-Bench**（Proactive Personal Assistant
+Benchmark）的胶水层评测套件。只含对接所需的最小代码与配置；π-Bench 框架
+（`src/`）、评测数据（`data/`）、AppWorld 工具环境、ReMe 本体均为**外部第三方
+依赖**，通过符号链接与环境变量原位引用，不随本套件分发。
+
+- π-Bench: https://github.com/Simplified-Reasoning/Pi-Bench （arXiv: 2605.14678）
+- ReMe: 你所在 ReMe 仓库的根目录（本套件推荐放在 `ReMe/benchmark/pibench/`）
+
+## 1. 架构总览
+
+```
+π-Bench runner (src.main --mode run)
+  │  user_agent（模拟用户 LLM）按 data/{persona}/episode.yaml 顺序
+  │  逐任务、多轮地与 agent 对话，并在 run 阶段判定隐藏意图(PROC)
+  ▼
+test server (π-Bench scripts/test_server.py, HTTP 长轮询)
+  ▲ /send            │ /poll
+  │                  ▼
+bridge_reme.py ──────────────► ReMe Application（以库方式内嵌启动）
+  │                             ├─ agent_wrapper: 被测 agent（AgentScope）
+  │                             ├─ jobs: search / auto_memory / daily_write
+  │                             └─ workspace: reme_workspace/{persona}/
+  │                                （每 persona 独立持久记忆库，互不可见）
+  └──── MCP ────► AppWorld MCP ────► AppWorld API（工具/应用环境）
+
+π-Bench runner (src.main --mode eval)
+  judger（裁判 LLM）读取 trace，按 checklist 逐条 YES/NO 打分(COMP)
+```
+
+要点：
+- bridge 用 **ReMe 自己的 venv python** 运行，把 ReMe 当库用（`resolve_app_config`
+  + `Application`），**ReMe 源码零改动**。
+- 每条用户消息都会自动触发一次 ReMe memory `search` 并把命中记忆注入当前消息
+  （参数见 §8）；任务结束（reset）时会话被 `auto_memory` 提炼为 daily 笔记落盘。
+- agent 执行的每一轮工具调用（AppWorld MCP + ReMe job 工具）都会被采集并以
+  `tool_steps` 形式写入 trace，供 π-Bench 的 `tools_evaluation_path` 脚本
+  对工具行为评分（§7）。
+- π-Bench 的 `data/`、`src/`、AppWorld 均不属于本套件，需先装好 π-Bench（§3.1）。
+
+## 2. 目录结构
+
+```
+pibench/
+├── README.md / README_ZH.md  # 本文档（英文 / 中文）
+├── env.sh.example            # 环境配置模板（复制为 env.sh 后填写 TODO 项）
+├── bridge_reme.py            # ReMe ↔ test server 桥接（记忆注入/保存、
+│                             #   profile 注入、工具调用轨迹采集）
+├── run_persona.sh            # 单 persona 全流程（5 个服务 + run + eval）
+├── run_all.sh                # 5 个 persona 批跑（fresh/resume，默认 2 并行）
+├── resume.py                 # 断点续跑：完成判定 + 中断任务残留记忆的外科清理
+├── fix_trace_logs.py         # run 输出 → ~/.nanobot/trace_logs 转换，
+│                             #   并把工具轨迹合并进 turn 文件（eval 前置）
+├── .gitignore                # 排除 env.sh 与全部运行产物
+└── config/
+    ├── models/reme.yaml      # runner 模型配置（model_id=reme）
+    └── bench/evaluation/trace_history.yaml   # trace 渲染策略（随套件提供，
+                                              #   经 --history-config-path 显式传入）
+```
+
+运行时自动生成（均被 .gitignore 排除）：`data`（符号链接）、`logs/`、
+`outputs/`、`reme_workspace/`、`nanobot_workspace/`。
+
+## 3. 前置依赖（第三方，先装好）
+
+### 3.1 π-Bench 仓库（含 AppWorld）
+
+```bash
+git clone https://github.com/Simplified-Reasoning/Pi-Bench.git <pi-bench-dir>
+cd <pi-bench-dir>
+python3.11 -m venv .venv            # 脚本约定使用 .venv 这个目录名
+source .venv/bin/activate
+pip install -e .                    # pibench runner（src.main）
+bash scripts/setup_appworld.sh      # 安装 AppWorld 并下载其数据（体积较大，需网络）
+```
+
+装完自检：
+```bash
+ls data/                    # 应含 researcher marketer pharmacist law_trainee Financier
+.venv/bin/python -c "import src" && echo OK
+.venv/bin/appworld --help >/dev/null && echo OK
+```
+
+### 3.2 ReMe 仓库
+
+```bash
+cd <reme-dir>               # ReMe 仓库根目录（含 reme/ 包）
+python3.11 -m venv .venv    # 脚本约定使用 .venv 这个目录名
+source .venv/bin/activate
+pip install -e .            # 或按 ReMe 自身安装方式，保证 `import reme` 可用
+```
+
+自检：`.venv/bin/python -c "import reme; print('ok')"`
+
+## 4. 安装本套件（逐步）
+
+1. **放置套件**（推荐放进 ReMe 仓库，`REME_DIR` 可自动推断）：
+   ```bash
+   cp -r pibench <reme-dir>/benchmark/pibench
+   cd <reme-dir>/benchmark/pibench
+   ```
+   若放在其他位置，稍后在 env.sh 中显式设置 `REME_DIR`。
+
+2. **创建环境文件并填写自定义参数**：
+   ```bash
+   cp env.sh.example env.sh
+   ```
+   打开 `env.sh`，必填项（标 TODO 的）：
+   | 变量 | 说明 |
+   |---|---|
+   | `PI_BENCH_ROOT` | π-Bench 仓库根目录（含 `src/` `data/` `.venv` `third_party/appworld`） |
+   | `USER_API_KEY` | 模拟用户 LLM 的 API key（run 阶段判定隐藏意图） |
+   | `JUDGER_API_KEY` | 裁判 LLM 的 API key（eval 阶段 checklist 打分） |
+   | `BRAVE_SEARCH_API_KEY` | 可选；agent 的 web_search 工具用，不用填 `dummy` |
+
+   可选调整：`REME_MODEL_NAME`（被测 agent 基模）、`REME_DIR`、
+   `REME_LLM_BASE_URL`（默认 DashScope OpenAI 兼容端点）。
+
+3. **链接评测数据**（π-Bench 数据原位引用，不复制）：
+   ```bash
+   ln -s "$PI_BENCH_ROOT/data" data
+   ```
+
+4. **（可选）调整模型配置** `config/models/reme.yaml`：
+   - `user_agent.model` / `judger.model`：模拟用户与裁判的模型名（字面量，
+     π-Bench 仅对 base_url/api_key 做 `${ENV}` 展开）。
+   - `run.turn_timeout`、`max_tool_iterations` 等按需。
+
+5. **冒烟自检**（不启动评测）：
+   ```bash
+   bash -n run_all.sh && bash -n run_persona.sh
+   source env.sh && "$REME_DIR/.venv/bin/python" -c "import reme; print('reme ok')"
+   ```
+
+## 5. 运行评测
+
+> ⚠️ 长时间运行请放进 `screen`，**不要用 nohup**（nohup 在沙箱/受限环境下
+> 会丢失权限上下文导致子进程异常）。
+
+```bash
+# 完整正式评测：先清空全部 persona 的记忆/输出/trace，再从头跑（默认 fresh，2 并行）
+mkdir -p logs   # 全新部署时 logs/ 尚不存在，先建再重定向
+screen -dmS pibench_suite bash -c "cd $(pwd) && bash run_all.sh > logs/run_all_master.log 2>&1"
+
+# 断点续跑（中断后继续；不清记忆，跳过已完成任务）
+bash run_all.sh --resume
+
+# 其他用法
+bash run_all.sh --parallel 1        # 串行
+bash run_all.sh --resume --skip-eval  # 只跑 run 阶段
+bash run_persona.sh researcher      # 单 persona（默认 --resume 语义）
+bash run_persona.sh researcher --fresh
+```
+
+耗时参考：5 persona × 20 任务、2 并行，fresh 全量约 12–14 小时。
+
+任一 persona 失败时 `run_all.sh` 以非零状态退出，上层自动化不会把部分失败
+的评测误判为成功。
+
+## 6. 端口分配（多 persona 并行互不冲突）
+
+| persona     | AppWorld API | AppWorld MCP | Test Server | ReMe 内部服务 |
+|-------------|------|-------|------|-------|
+| marketer    | 9001 | 10001 | 9998 | 18766 |
+| law_trainee | 9002 | 10002 | 9997 | 18767 |
+| pharmacist  | 9003 | 10003 | 9996 | 18768 |
+| researcher  | 9004 | 10004 | 9995 | 18765 |
+| Financier   | 9005 | 10005 | 9994 | 18769 |
+
+## 7. 输出与分数
+
+- **结果**：`outputs/reme/{persona}/{task}/eval/results/*_result.json`
+  - `overall_average_score`：checklist 完整度（COMP，judger 逐条 YES/NO 按依赖组加权）
+  - `overall_proactiveness_average_score`：主动性（PROC，run 阶段 user_agent
+    判定隐藏意图覆盖率；每个任务文件同时携带全局均值）
+- **trace**：`~/.nanobot/trace_logs/reme/{persona}/{task}/...`（eval 的判分输入）
+- **日志**：`logs/`（`suite_<persona>.log` 为每 persona 总日志，`bridge_*`、
+  `runner_run/eval_*`、`appworld_*`、`test_server_*` 分服务）
+- **记忆库**：`reme_workspace/{persona}/`（daily/digest 笔记、session 原始对话、
+  BM25 索引等；跨运行持久，fresh 才清空）
+
+查看汇总：
+```bash
+grep -h "overall_average_score\|overall_proactiveness" \
+  outputs/reme/*/*/eval/results/*_result.json | head
+```
+
+### 工具轨迹采集（tools_evaluation 支持）
+
+部分任务定义了 `objectives.tools_evaluation_path`：用 Python 脚本对工具行为
+打分（例如"临时 Todoist 看板已创建并被删除"）。这些脚本需要 trace 里有真实
+的工具调用记录。采集链路：
+
+1. 每轮 `reply()` 之后，bridge 读取 AgentScope 落盘的会话状态，提取本轮新增
+   的 `tool_call` / `tool_result` 块（工具名、参数、结果）。
+2. 记录按 turn 编号追加写入
+   `outputs/reme/{persona}/{task}/history/{ts}-tools.jsonl`；AgentScope 的
+   MCP 工具名（`mcp__AppWorld__<tool>`）会规范化为 π-Bench 约定
+   （`mcp_appworld_<tool>`）。
+3. `fix_trace_logs.py` 将每个 `{ts}-messages.jsonl` 运行与时间上最接近的
+   tools 旁路文件配对，把记录合并进生成的 `turn_N.json` 的 `tool_steps`
+   字段——这是 π-Bench `collect_tool_history()` 支持的两种工具轨迹格式之一。
+4. eval 阶段 `tool_steps` 既提供给 tools_evaluation 脚本，也会被渲染为
+   judger 可见的 `<tool_trace_extracts>`。
+
+## 8. 记忆机制（本套件的核心设计）
+
+- **persona 隔离**：每个 persona 独立 workspace（`reme_workspace/{persona}/`），
+  bridge 启动时对 workspace 加 `.bridge.lock` 排他锁，两个 bridge 不可能共用
+  同一记忆库；一个 persona 的 memory search 永远接触不到其他 persona 的记忆。
+- **写入**：任务结束（runner 发送 reset）时，会话经 `auto_memory` job 提炼为
+  daily 笔记落盘，后台 watcher 建 BM25 索引。保存为非阻塞后台任务，
+  新会话首条消息会先等待在途写入完成再检索。
+- **读取**：bridge 每收到一条用户消息自动 `search` 一次并注入命中记忆
+  （`[Relevant memories from previous sessions]` 前缀），无命中则原样透传。
+  检索参数（bridge 命令行，可在 run_persona.sh 中调整）：
+  - `--search-limit 3`：每条消息最多注入 3 个记忆块；
+  - `--search-min-score 2.0`：过滤弱 BM25 命中；
+  - `tool_context_id` 按任务轮换：同一任务内已注入的记忆块不重复注入
+    （ReMe 自带 seen-chunk 去重，24h TTL），任务边界后恢复正常召回。
+- **无自泄漏**：进行中的会话尚未入库（save 发生在 reset），任务不会检索到
+  自己未完成的内容。
+- agent 同时持有 `search`/`daily_write` 工具，可主动检索/记录。
+- **system prompt**：`bridge_reme.py:build_system_prompt()` 内置
+  HIDDEN-NEEDS 协议（面向 proactiveness），并把 `data/{persona}/profile.yaml`
+  的 persona profile 注入每轮 system prompt。
+
+## 9. 断点续跑与记忆清理语义
+
+- **完成判定**（resume.py）：扫描 `outputs/reme/{persona}/**/history/*-log.jsonl`
+  与 `outputs/reme/{persona}/run/*-log.jsonl` 中的
+  `Task finished task_id=X status=Y`。每个任务以**事件时间最新**的记录为准
+  （优先取记录的 `timestamp`，回退 `timestamp_iso`，再回退日志文件名中的
+  时间戳）——文件类别与读取顺序本身不能覆盖更新的记录，因此旧的 run 级
+  SUCCESS 不会掩盖更新的 per-task ERROR。`SUCCESS/MAX_TURNS/TIMEOUT` 记为
+  完成，`ERROR`/未开始的任务重跑（按 episode 顺序以 `--task-id` 传给 runner）。
+- **防答案泄漏**：被中断的任务可能已在优雅退出时提炼成 daily 笔记，直接重跑会
+  把答案注入、抬高分数。因此 resume 启动前 `resume.py cleanup` **只删除待重跑
+  任务**的残留记忆（daily/digest 笔记、session/dialog、mem_session，按
+  `session_id = pibench_{task}_*` 匹配），已完成任务的记忆一律不动。daily
+  索引**只刷新实际发生删除的日期**，按完整的 workspace 相对 wikilink 路径
+  匹配；当 ReMe 包可导入时，刷新直接复用 ReMe 自带的 daily 索引重建逻辑
+  （`refresh_day_index`），不会误改其他日期下的同名笔记条目。
+- **fresh vs resume 互斥**：全量清记忆只属于 fresh 模式（`run_all.sh` 默认，
+  在任何服务启动前执行）；resume 永不清全量。
+
+## 10. 自定义与调优入口
+
+| 目标 | 位置 |
+|---|---|
+| 被测 agent 基模 | `env.sh` 的 `REME_MODEL_NAME` |
+| user_agent / judger 模型 | `config/models/reme.yaml` |
+| agent system prompt | `bridge_reme.py` `build_system_prompt()` |
+| 记忆检索条数/阈值 | `run_persona.sh` bridge 启动命令的 `--search-limit/--search-min-score` |
+| ReMe 内部参数 | **不要改 ReMe 源码**；仿照 `reme/config/beam.yaml` 写专有配置，经 `resolve_app_config(config=...)` 覆盖（见 bridge `_init_reme_app`） |
+| 轮超时/工具迭代上限 | `config/models/reme.yaml` `run.turn_timeout`、`model.max_tool_iterations` |
+
+## 11. 故障排查
+
+- **端口被占用**：脚本会自动 kill 上述 4 组端口上的残留进程；若与其他套件
+  （如别的 π-Bench 实验）冲突，请先停掉对方或改 run_persona.sh 的端口表。
+- **bridge 启动即退出，提示 workspace locked**：另一个 bridge 正占用同一
+  workspace；确认每个 persona 用各自的 `--workspace-dir`（脚本已按 persona 分配）。
+- **runner 报 `${USER_API_KEY} ... empty`**：env.sh 未填写或未生效；
+  run_persona.sh 会自动 source env.sh，手动运行 runner 时请先 `source env.sh`。
+- **`Cannot import 'reme'`**：bridge 必须用 `${REME_DIR}/.venv/bin/python` 运行
+  （run_persona.sh 已如此），或检查 `REME_DIR` 是否指向 ReMe 仓库根目录。
+- **AppWorld 启动失败**：先在 π-Bench 仓库执行 `bash scripts/setup_appworld.sh`
+  下载数据；查看 `logs/appworld_*_<persona>.log`。
+- **trace_history.yaml 找不到**：runner 需要
+  `config/bench/evaluation/trace_history.yaml`；本套件已随附该文件并通过
+  `--history-config-path` 显式传入，run_persona.sh 启动前会做存在性检查，
+  缺失时立即报出清晰错误。请始终从套件目录启动 run_persona.sh / run_all.sh。
+
+## 12. 隐私与安全
+
+- 套件代码与配置模板中**不含任何真实 API key、用户名或绝对路径**；
+  真实 key 只存在于你本地的 `env.sh`（已被 .gitignore 排除）。
+- `logs/`、`outputs/`、`reme_workspace/`、`nanobot_workspace/` 含完整对话内容
+  与模型输出，请勿提交仓库或外传。
+- `data` 符号链接指向 π-Bench 官方评测数据，请遵守其数据许可条款。

benchmark/pibench/config/bench/evaluation/trace_history.yaml

@@ -0,0 +1,53 @@
+version: 1
+
+format:
+  root_tag: trace
+  turn_tag: turn
+  message_tag: message
+  file_tag: file
+  tool_call_tag_prefix: tool_call
+  tool_result_tag_prefix: tool_result
+
+text_policy:
+  default:
+    truncate_chars: 1200
+    mask_newlines: false
+  field_overrides:
+    files_read:
+      truncate_chars: 40000
+    assistant_content:
+      truncate_chars: 40000
+    tool_result_content:
+      truncate_chars: 40000
+
+fields:
+  turn:
+    include_session_key: false
+
+  files:
+    enabled: true
+
+  messages:
+    enabled: true
+    include_message_role_attr: true
+    include_message_index_attr: false
+    include_system: false
+    include_user: true
+    include_assistant_thinking_content: false
+    include_assistant_thinking_reasoning: false
+    include_assistant_content: true
+    include_assistant_reasoning: false
+    include_assistant_tool_calls: false
+    require_matching_tool_call: true
+
+    tool_calls:
+      include_tool_call_id: false
+      tools:
+        web_fetch:
+          enabled: true
+          include_tool_call_keys: [url]
+          include_tool_result: false
+        web_search:
+          enabled: true
+          include_tool_call_keys: [query]
+          include_tool_result: false

benchmark/pibench/config/models/reme.yaml

@@ -0,0 +1,40 @@
+# ReMe model configuration for Pi-Bench
+# Uses ReMe's AgentScope agent with Dashscope as the LLM backend
+
+model:
+  model: reme
+  base_url: "http://localhost:8088"
+  api_key: "dummy"
+  provider: custom
+  max_tokens: 16384
+  max_tool_iterations: 120
+  memory_window: 100
+
+user_agent:
+  model: qwen3.8-max
+  base_url: "${USER_BASE_URL}"
+  api_key: "${USER_API_KEY}"
+  temperature: 0.0
+  request_timeout: 360.0
+
+judger:
+  model: qwen3.8-max
+  base_url: "${JUDGER_BASE_URL}"
+  api_key: "${JUDGER_API_KEY}"
+  temperature: 0.0
+  request_timeout: 360.0
+
+tools:
+  brave_search_api_key: "${BRAVE_SEARCH_API_KEY}"
+  web_search_max_results: 10
+
+nanobot:
+  trace_logs_dir: "~/.nanobot/trace_logs"
+  workspace_dir: "~/.nanobot/workspace"
+  copy_task_assets_to_workspace: true
+
+run:
+  output_dir: outputs
+  log_level: INFO
+  user_mode: llm
+  turn_timeout: 2400.0

benchmark/pibench/env.sh.example

@@ -0,0 +1,57 @@
+#!/bin/bash
+# ═══════════════════════════════════════════════════════════════════════
+# pibench evaluation suite - environment configuration template
+# Usage: cp env.sh.example env.sh, then fill in the TODO items below.
+# ⚠️  env.sh contains real API keys; never commit or share it
+#    (already excluded via .gitignore).
+# ═══════════════════════════════════════════════════════════════════════
+
+SUITE_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# ─── TODO: π-Bench repository root ────────────────────────────────────
+# Must contain src/, data/, scripts/test_server.py, third_party/appworld
+# and .venv (see README setup).
+export PI_BENCH_ROOT=""
+
+# ─── ReMe repository ──────────────────────────────────────────────────
+# Defaults to two levels above this directory (the layout this suite uses
+# when placed at ReMe/benchmark/pibench); point it at the actual ReMe
+# repository root if the suite lives elsewhere.
+export REME_DIR="${REME_DIR:-$(cd "${SUITE_DIR}/../.." && pwd)}"
+
+# ─── Base model of the agent under test (LLM used by the ReMe agent) ──
+export REME_MODEL_NAME="${REME_MODEL_NAME:-qwen3.6-plus}"
+
+# ─── LLM service endpoint (default: DashScope OpenAI-compatible; any
+#     OpenAI-compatible endpoint works) ────────────────────────────────
+DASHSCOPE_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
+export REME_LLM_BASE_URL="${REME_LLM_BASE_URL:-${DASHSCOPE_BASE_URL}}"
+
+# ─── TODO: API keys ───────────────────────────────────────────────────
+# USER_API_KEY : drives the simulated user LLM (run phase; judges whether
+#                hidden intents are satisfied and asks follow-ups)
+# JUDGER_API_KEY: drives the judger LLM (eval phase; scores the checklist)
+# The two may be identical; one strong model is recommended for both.
+export USER_BASE_URL="${DASHSCOPE_BASE_URL}"
+export USER_API_KEY="TODO-fill-in-user-agent-api-key"
+
+export JUDGER_BASE_URL="${DASHSCOPE_BASE_URL}"
+export JUDGER_API_KEY="TODO-fill-in-judger-api-key"
+
+# The ReMe agent's key reuses USER_API_KEY by default (no need to repeat
+# it when both use the same service and key).
+export REME_LLM_API_KEY="${REME_LLM_API_KEY:-${USER_API_KEY}}"
+
+# Brave Search (optional; used by the agent's web_search tool - use
+# "dummy" when not needed).
+export BRAVE_SEARCH_API_KEY="TODO-optional-brave-search-key-or-dummy"
+
+# ─── Persistent memory workspaces (one subdirectory per persona,
+#     created automatically) ───────────────────────────────────────────
+export REME_WORKSPACE_ROOT="${REME_WORKSPACE_ROOT:-${SUITE_DIR}/reme_workspace}"
+
+# ─── Variables consumed by ReMe's default.yaml model config expansion;
+#     do not remove ────────────────────────────────────────────────────
+export LLM_MODEL_NAME="${REME_MODEL_NAME}"
+export LLM_BASE_URL="${REME_LLM_BASE_URL}"
+export LLM_API_KEY="${REME_LLM_API_KEY}"

benchmark/pibench/fix_trace_logs.py

@@ -0,0 +1,198 @@
+#!/usr/bin/env python3
+"""Convert reme_eval run outputs into eval-compatible trace logs.
+
+outputs/{model_id}/{user_id}/{task_id}/history/{ts}-messages.jsonl
+  ->  ~/.nanobot/trace_logs/{model_id}/{user_id}/{task_id}/{ts}/turn_N.json
+
+The bridge additionally writes {ts}-tools.jsonl sidecar files next to the
+message histories: one JSON object per executed tool call with fields
+{turn, name, arguments, result}. Each messages run is paired with the
+temporally closest sidecar, and the records are merged into the generated
+turn files under the "tool_steps" key, which is one of the tool-history
+formats π-Bench's collect_tool_history() understands. Without this step,
+tools_evaluation scripts would see no tool evidence at all.
+
+Usage: python fix_trace_logs.py [user_id ...]   (no args = all users)
+"""
+
+import json
+import re
+import sys
+from datetime import datetime
+from pathlib import Path
+
+SUITE_DIR = Path(__file__).resolve().parent
+OUTPUTS_DIR = SUITE_DIR / "outputs"
+TRACE_LOGS_DIR = Path.home() / ".nanobot" / "trace_logs"
+
+MESSAGES_FILE_RE = re.compile(r"^(\d{8}_\d{6})-messages\.jsonl$")
+TOOLS_FILE_RE = re.compile(r"^(\d{8}_\d{6})-tools\.jsonl$")
+TIME_FORMAT = "%Y%m%d_%H%M%S"
+# A tool sidecar belongs to the messages run that started at most this many
+# seconds earlier (the bridge stamps the sidecar when the task's first user
+# message arrives, shortly after the runner opened the messages file).
+MAX_PAIR_DELTA_SECONDS = 6 * 3600
+
+
+def _to_epoch(timestamp: str) -> float:
+    """Parse a YYYYMMDD_HHMMSS timestamp into epoch seconds."""
+    try:
+        return datetime.strptime(timestamp, TIME_FORMAT).timestamp()
+    except ValueError:
+        return 0.0
+
+
+def load_tool_records(tools_file: Path) -> dict:
+    """Group sidecar tool records by turn number."""
+    by_turn: dict = {}
+    try:
+        with open(tools_file, "r", encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if not line:
+                    continue
+                try:
+                    record = json.loads(line)
+                except json.JSONDecodeError:
+                    continue
+                if not isinstance(record, dict) or not record.get("name"):
+                    continue
+                turn = int(record.get("turn") or 0)
+                by_turn.setdefault(turn, []).append(
+                    {
+                        "name": record["name"],
+                        "arguments": record.get("arguments", {}),
+                        "result": record.get("result", ""),
+                    },
+                )
+    except OSError as exc:
+        print(f"  WARNING: cannot read tool sidecar {tools_file}: {exc}")
+    return by_turn
+
+
+def pair_tool_sidecars(message_runs: list, tool_runs: list) -> dict:
+    """Pair each messages run with the temporally closest unused tool sidecar.
+
+    Fresh runs produce exactly one messages file and one sidecar per task;
+    re-runs append matching pairs, so sorted greedy nearest-timestamp
+    matching is stable. Sidecars farther away than MAX_PAIR_DELTA_SECONDS
+    (e.g. leftovers of a crashed bridge) stay unpaired.
+    """
+    pairing: dict = {}
+    unused = list(tool_runs)
+    for msg_ts, _ in message_runs:
+        best_delta = None
+        best_item = None
+        for tool_ts, tool_path in unused:
+            delta = abs(_to_epoch(tool_ts) - _to_epoch(msg_ts))
+            if best_delta is None or delta < best_delta:
+                best_delta = delta
+                best_item = (tool_ts, tool_path)
+        if best_delta is not None and best_item is not None and best_delta <= MAX_PAIR_DELTA_SECONDS:
+            pairing[msg_ts] = best_item[1]
+            unused.remove(best_item)
+    return pairing
+
+
+def build_turns(messages: list) -> list:
+    """Split the flat message list into per-turn [user, assistant] groups."""
+    turns = []
+    i = 0
+    while i < len(messages):
+        turn_msgs = []
+        if messages[i]["role"] == "user":
+            turn_msgs.append({"role": "user", "content": messages[i]["message"]})
+            i += 1
+        if i < len(messages) and messages[i]["role"] == "assistant":
+            turn_msgs.append({"role": "assistant", "content": messages[i]["message"]})
+            i += 1
+        if not turn_msgs:
+            i += 1  # defensive: never spin on unexpected roles
+            continue
+        turns.append(turn_msgs)
+    return turns
+
+
+def convert_task(model_id: str, user_id: str, task_dir: Path) -> None:
+    """Convert one task's history dir into trace turn files with tool_steps."""
+    history_dir = task_dir / "history"
+    if not history_dir.is_dir():
+        return
+
+    message_runs = []
+    tool_runs = []
+    for msg_file in history_dir.glob("*-messages.jsonl"):
+        match = MESSAGES_FILE_RE.match(msg_file.name)
+        if match:
+            message_runs.append((match.group(1), msg_file))
+    for tools_file in history_dir.glob("*-tools.jsonl"):
+        match = TOOLS_FILE_RE.match(tools_file.name)
+        if match:
+            tool_runs.append((match.group(1), tools_file))
+    if not message_runs:
+        return
+
+    message_runs.sort(key=lambda item: item[0])
+    tool_runs.sort(key=lambda item: item[0])
+    pairing = pair_tool_sidecars(message_runs, tool_runs)
+
+    print(f"\n{model_id}/{user_id}/{task_dir.name}")
+    for timestamp, msg_file in message_runs:
+        trace_dir = TRACE_LOGS_DIR / model_id / user_id / task_dir.name / timestamp
+        trace_dir.mkdir(parents=True, exist_ok=True)
+
+        messages = []
+        with open(msg_file, "r", encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if not line:
+                    continue
+                msg = json.loads(line)
+                if msg.get("role") == "user" and msg.get("message") == "/new":
+                    continue
+                messages.append(msg)
+
+        tools_file = pairing.get(timestamp)
+        tools_by_turn = load_tool_records(tools_file) if tools_file else {}
+        if tools_file is not None:
+            print(f"  {timestamp}: paired tool sidecar {tools_file.name}")
+
+        turns = build_turns(messages)
+        for turn_idx, turn_msgs in enumerate(turns, start=1):
+            turn_data = {"messages": turn_msgs}
+            tool_steps = tools_by_turn.get(turn_idx)
+            if tool_steps:
+                turn_data["tool_steps"] = tool_steps
+            turn_file = trace_dir / f"turn_{turn_idx}.json"
+            with open(turn_file, "w", encoding="utf-8") as f:
+                json.dump(turn_data, f, indent=2, ensure_ascii=False)
+        tool_total = sum(len(steps) for steps in tools_by_turn.values())
+        print(f"  {timestamp}: {len(turns)} turns, {tool_total} tool step(s) -> {trace_dir}")
+
+
+def convert_outputs(user_filter=None):
+    """Convert message history JSONL files into per-turn trace JSON files."""
+    if not OUTPUTS_DIR.exists():
+        print(f"outputs dir not found: {OUTPUTS_DIR}")
+        return
+
+    for model_dir in sorted(OUTPUTS_DIR.iterdir()):
+        if not model_dir.is_dir():
+            continue
+        model_id = model_dir.name
+
+        for user_dir in sorted(model_dir.iterdir()):
+            if not user_dir.is_dir():
+                continue
+            user_id = user_dir.name
+            if user_filter and user_id not in user_filter:
+                continue
+
+            for task_dir in sorted(user_dir.iterdir()):
+                if task_dir.is_dir():
+                    convert_task(model_id, user_id, task_dir)
+
+
+if __name__ == "__main__":
+    convert_outputs(set(sys.argv[1:]) or None)
+    print("\ndone")

benchmark/pibench/resume.py

@@ -0,0 +1,332 @@
+#!/usr/bin/env python3
+"""Checkpoint-resume support for the reme_eval suite.
+
+Completion source of truth:
+    - outputs/reme/<persona>/<task_id>/history/*-log.jsonl  (per-task logs,
+      flushed incrementally, survive mid-run kills)
+    - outputs/reme/<persona>/run/*-log.jsonl                (run-level logs,
+      may be truncated if the process was killed before flush)
+    lines: "Task finished task_id=<id> status=<STATUS>"
+    A task counts as COMPLETED when its latest terminal status is one of
+    SUCCESS / MAX_TURNS / TIMEOUT. ERROR or never-started tasks stay pending.
+
+    "Latest" is decided by EVENT TIME, not by file category or read order:
+    each record's "timestamp" (epoch seconds, or "timestamp_iso" as fallback)
+    is compared across per-task and run-level logs alike, with the timestamp
+    embedded in the log file name as a last-resort fallback. This keeps an
+    old run-level SUCCESS from overriding a newer per-task ERROR when the
+    re-run died before the new run-level log captured the task.
+
+Commands:
+    remaining <persona> [--json]
+        Print task_ids still to run, in data/<persona>/episode.yaml order
+        (one per line; --json prints {"completed": [...], "remaining": [...]}).
+
+    cleanup <persona> [--dry-run]
+        Surgically remove residual memory artifacts of tasks that are about
+        to be RE-RUN (i.e. pending tasks that left partial state because a
+        previous run was interrupted). This prevents answer leakage: an
+        interrupted task's conversation may already have been distilled into
+        daily notes during graceful shutdown, and re-running the task with
+        that memory injected would inflate scores.
+
+        Removed artifacts (only for pending tasks with residual state):
+          - daily/<date>/<note>.md whose frontmatter session_id matches
+            pibench_<task_id>_*, plus a refresh of ONLY the daily index of
+            the affected date(s) (daily/<date>.md), matched by the full
+            workspace-relative note path, never by bare file name
+          - digest notes with matching session_id
+          - session/dialog/pibench_<task_id>_*.jsonl
+          - mem_session/**.jsonl files containing pibench_<task_id>_
+        When the ReMe package is importable, the daily index refresh reuses
+        ReMe's own rebuild logic (reme.steps.file_io._daily_index.
+        refresh_day_index); otherwise index lines are dropped by exact
+        wikilink path match. Either way, indexes of other dates are never
+        touched. The ReMe watcher (init_changes_step) detects the deleted
+        daily notes on next bridge startup and removes them from the BM25
+        index itself.
+
+        Completed tasks' memories are NEVER touched by this command.
+
+Design note (resume vs memory-wipe conflict):
+    A full memory wipe is a suite-level action of fresh mode (run_all.sh
+    without --resume) and happens before any service starts. Resume mode
+    never wipes; it only performs the surgical cleanup above. The two modes
+    are mutually exclusive, so a resumed run can never lose the cross-session
+    memory accumulated by completed tasks.
+"""
+
+import asyncio
+import json
+import os
+import re
+import sys
+from datetime import datetime
+from pathlib import Path
+
+import yaml
+
+try:  # Reuse ReMe's daily-index rebuild when running inside the ReMe venv.
+    from reme.steps.file_io._daily_index import refresh_day_index
+except ImportError:  # pragma: no cover - depends on runtime venv
+    refresh_day_index = None
+
+SUITE_DIR = Path(__file__).resolve().parent
+DATA_DIR = Path(os.environ.get("REME_EVAL_DATA_DIR", SUITE_DIR / "data")).resolve()
+OUTPUTS_DIR = Path(os.environ.get("REME_EVAL_OUTPUTS_DIR", SUITE_DIR / "outputs")) / "reme"
+WORKSPACE_ROOT = Path(
+    os.environ.get("REME_WORKSPACE_ROOT", SUITE_DIR / "reme_workspace"),
+).resolve()
+
+COMPLETED_STATUSES = {"SUCCESS", "MAX_TURNS", "TIMEOUT"}
+TASK_FINISHED_RE = re.compile(r"Task finished task_id=(\S+) status=(\S+)")
+SESSION_ID_RE = re.compile(r"^session_id:\s*(\S+)", re.MULTILINE)
+NOTE_COUNT_RE = re.compile(r"(description:\s*)\d+(\s*note\(s\) today)")
+LOG_FILE_TS_RE = re.compile(r"^(\d{8}_\d{6})-log\.jsonl$")
+TIME_FORMAT = "%Y%m%d_%H%M%S"
+
+
+def log(msg: str) -> None:
+    """Print a status message to stderr."""
+    print(msg, file=sys.stderr)
+
+
+def episode_task_order(persona: str) -> list[str]:
+    """Return the ordered task ids from the persona's episode.yaml."""
+    episode_path = DATA_DIR / persona / "episode.yaml"
+    with open(episode_path, "r", encoding="utf-8") as f:
+        episode = yaml.safe_load(f)
+    return [task["task_id"] for task in episode.get("tasks", [])]
+
+
+def _event_time(record: dict, file_ts: str) -> float:
+    """Best-effort event time (epoch seconds) of one log record.
+
+    Prefers the record's own timestamp fields; falls back to the timestamp
+    embedded in the log file name so that even stripped records keep a
+    meaningful order. Returns 0.0 when nothing is parseable.
+    """
+    timestamp = record.get("timestamp")
+    if isinstance(timestamp, (int, float)) and not isinstance(timestamp, bool):
+        return float(timestamp)
+    iso = record.get("timestamp_iso")
+    if isinstance(iso, str):
+        try:
+            return datetime.fromisoformat(iso).timestamp()
+        except ValueError:
+            pass
+    if file_ts:
+        try:
+            return datetime.strptime(file_ts, TIME_FORMAT).timestamp()
+        except ValueError:
+            pass
+    return 0.0
+
+
+def latest_task_statuses(persona: str) -> dict[str, str]:
+    """Scan per-task and run-level logs; the newest EVENT TIME wins per task.
+
+    Every "Task finished" record across both log categories is keyed by
+    (event_time, file timestamp, file order, line number); the record with
+    the highest key decides the task's status. File category and read order
+    alone can never override a newer record from the other category.
+    """
+    persona_dir = OUTPUTS_DIR / persona
+    if not persona_dir.is_dir():
+        return {}
+
+    log_files = sorted(persona_dir.glob("*/history/*-log.jsonl"))
+    log_files += sorted(persona_dir.glob("run/*-log.jsonl"))
+
+    best: dict[str, tuple[tuple, str]] = {}
+    for file_order, log_file in enumerate(log_files):
+        ts_match = LOG_FILE_TS_RE.match(log_file.name)
+        file_ts = ts_match.group(1) if ts_match else ""
+        try:
+            with open(log_file, "r", encoding="utf-8") as f:
+                for line_no, line in enumerate(f):
+                    if "Task finished" not in line:
+                        continue
+                    try:
+                        record = json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    match = TASK_FINISHED_RE.search(str(record.get("message", "")))
+                    if not match:
+                        continue
+                    task_id, status = match.group(1), match.group(2)
+                    sort_key = (_event_time(record, file_ts), file_ts, file_order, line_no)
+                    current = best.get(task_id)
+                    if current is None or sort_key > current[0]:
+                        best[task_id] = (sort_key, status)
+        except OSError:
+            continue
+    return {task_id: status for task_id, (_, status) in best.items()}
+
+
+def split_tasks(persona: str) -> tuple[list[str], list[str]]:
+    """Split the episode task order into completed and remaining tasks."""
+    order = episode_task_order(persona)
+    statuses = latest_task_statuses(persona)
+    completed = [t for t in order if statuses.get(t) in COMPLETED_STATUSES]
+    remaining = [t for t in order if t not in set(completed)]
+    return completed, remaining
+
+
+def _daily_note_session_id(note_path: Path) -> str:
+    try:
+        text = note_path.read_text(encoding="utf-8")
+    except OSError:
+        return ""
+    match = SESSION_ID_RE.search(text)
+    return match.group(1) if match else ""
+
+
+class _WorkspaceFileStoreShim:
+    """Structural stand-in for ReMe's file store; only workspace_path is read."""
+
+    def __init__(self, workspace_path: Path):
+        self.workspace_path = workspace_path
+
+
+def _refresh_daily_indexes(
+    workspace: Path,
+    removed_by_date: dict[str, set[str]],
+    removed: list[str],
+) -> None:
+    """Rebuild the daily index of each affected date via ReMe's own logic."""
+    for date in sorted(removed_by_date):
+        result = asyncio.run(
+            refresh_day_index(_WorkspaceFileStoreShim(workspace), date, "daily"),
+        )
+        if result.get("error"):
+            log(f"[resume] WARNING: daily index refresh failed for {date}: {result['error']}")
+            continue
+        removed.append(f"daily/{date}.md (refreshed, {len(removed_by_date[date])} note(s) removed)")
+
+
+def _strip_index_lines(
+    workspace: Path,
+    removed_by_date: dict[str, set[str]],
+    removed: list[str],
+    dry_run: bool,
+) -> None:
+    """Fallback index edit: drop lines that reference removed notes by full
+    workspace-relative wikilink path, and fix the note count. Only the index
+    files of affected dates are touched."""
+    for date in sorted(removed_by_date):
+        index_path = workspace / "daily" / f"{date}.md"
+        if not index_path.is_file():
+            continue
+        wikilinks = [f"[[{rel_path}]]" for rel_path in sorted(removed_by_date[date])]
+        lines = index_path.read_text(encoding="utf-8").splitlines()
+        kept = [line for line in lines if not any(link in line for link in wikilinks)]
+        if len(kept) == len(lines):
+            continue
+        note_count = sum(1 for line in kept if line.startswith("- [[daily/"))
+        kept = [NOTE_COUNT_RE.sub(rf"\g<1>{note_count}\2", line) for line in kept]
+        removed.append(f"{index_path.relative_to(workspace)} (rewritten)")
+        if not dry_run:
+            index_path.write_text("\n".join(kept) + "\n", encoding="utf-8")
+
+
+def cleanup_partial_memory(persona: str, remaining: list[str], dry_run: bool = False) -> list[str]:
+    """Remove partial memory artifacts of remaining tasks so they can be re-run cleanly."""
+    workspace = WORKSPACE_ROOT / persona
+    removed: list[str] = []
+    if not workspace.is_dir() or not remaining:
+        return removed
+
+    prefixes = tuple(f"pibench_{task_id}_" for task_id in remaining)
+
+    def act(path: Path, label: str) -> None:
+        removed.append(label)
+        if not dry_run:
+            path.unlink()
+
+    # 1) daily / digest notes distilled from interrupted sessions. For daily
+    # notes, remember the full workspace-relative path grouped by date so only
+    # the affected daily indexes are refreshed below.
+    removed_by_date: dict[str, set[str]] = {}
+    for section in ("daily", "digest"):
+        section_root = workspace / section
+        if not section_root.is_dir():
+            continue
+        for note_path in section_root.rglob("*.md"):
+            if note_path.parent == section_root:
+                continue  # index files handled below
+            session_id = _daily_note_session_id(note_path)
+            if session_id.startswith(prefixes):
+                rel_path = note_path.relative_to(workspace).as_posix()
+                act(note_path, rel_path)
+                if section == "daily":
+                    removed_by_date.setdefault(note_path.parent.name, set()).add(rel_path)
+
+    # 2) daily index files: refresh only the dates that lost notes, matching
+    # notes by their full wikilink path instead of their bare file name.
+    if removed_by_date:
+        if dry_run:
+            for date in sorted(removed_by_date):
+                removed.append(f"daily/{date}.md (would refresh index)")
+        elif refresh_day_index is not None:
+            _refresh_daily_indexes(workspace, removed_by_date, removed)
+        else:
+            _strip_index_lines(workspace, removed_by_date, removed, dry_run)
+
+    # 3) raw dialog logs of interrupted sessions
+    dialog_dir = workspace / "session" / "dialog"
+    if dialog_dir.is_dir():
+        for task_id in remaining:
+            for dialog_path in dialog_dir.glob(f"pibench_{task_id}_*.jsonl"):
+                act(dialog_path, str(dialog_path.relative_to(workspace)))
+
+    # 4) agent-scope session states that contain interrupted-task sessions
+    mem_session_dir = workspace / "mem_session"
+    if mem_session_dir.is_dir():
+        for session_path in mem_session_dir.rglob("*.jsonl"):
+            try:
+                content = session_path.read_text(encoding="utf-8", errors="ignore")
+            except OSError:
+                continue
+            if any(prefix in content for prefix in prefixes):
+                act(session_path, str(session_path.relative_to(workspace)))
+
+    return removed
+
+
+def main() -> int:
+    """CLI entrypoint: run 'remaining' or 'cleanup' action for a persona."""
+    args = sys.argv[1:]
+    if len(args) < 2 or args[0] not in {"remaining", "cleanup"}:
+        print(__doc__, file=sys.stderr)
+        return 2
+
+    command, persona = args[0], args[1]
+    completed, remaining = split_tasks(persona)
+
+    if command == "remaining":
+        if "--json" in args:
+            print(json.dumps({"completed": completed, "remaining": remaining}))
+        else:
+            for task_id in remaining:
+                print(task_id)
+        log(
+            f"[resume] {persona}: completed={len(completed)} "
+            f"({', '.join(completed) if completed else '-'}) remaining={len(remaining)}",
+        )
+        return 0
+
+    dry_run = "--dry-run" in args
+    removed = cleanup_partial_memory(persona, remaining, dry_run=dry_run)
+    if removed:
+        verb = "would remove" if dry_run else "removed"
+        log(f"[resume] {persona}: {verb} {len(removed)} partial-memory artifact(s):")
+        for item in removed:
+            log(f"  - {item}")
+    else:
+        log(f"[resume] {persona}: no partial-memory artifacts to clean")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

benchmark/pibench/run_all.sh

@@ -0,0 +1,119 @@
+#!/bin/bash
+# Run all 5 personas with the ReMe agent, PARALLEL at a time (default 2).
+# Each persona's tasks follow data/{persona}/episode.yaml order.
+#
+# Usage:
+#   bash run_all.sh                  # FRESH official run: wipes ALL personas'
+#                                    # ReMe memory/outputs/trace logs first,
+#                                    # then runs everything from scratch.
+#   bash run_all.sh --resume         # Checkpoint continuation: no wipe; every
+#                                    # persona skips already-completed tasks.
+#   bash run_all.sh --parallel 1     # sequential (original behavior)
+#   bash run_all.sh --skip-eval      # run phase only
+#
+# Memory-wipe vs resume conflict resolution:
+#   The full ReMe memory wipe happens ONLY here, ONLY in fresh mode (the
+#   default), and ONLY before any service/bridge starts. --resume never
+#   wipes; run_persona.sh then additionally performs a surgical cleanup of
+#   residual memory belonging to interrupted (to-be-re-run) tasks, so a
+#   resumed run keeps all completed-task memory but never inherits a partial
+#   task's own answer. The two modes are mutually exclusive.
+set -uo pipefail
+
+SUITE_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PERSONAS=(researcher marketer law_trainee pharmacist Financier)
+TRACE_ROOT="${HOME}/.nanobot/trace_logs"
+
+PARALLEL=2
+MODE="fresh"
+PASS_ARGS=()
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        --parallel)
+            PARALLEL="${2:-}"; shift 2 || true
+            case "$PARALLEL" in (""|*[!0-9]*) echo "--parallel needs a positive integer"; exit 2 ;; esac
+            [ "$PARALLEL" -lt 1 ] && PARALLEL=1
+            [ "$PARALLEL" -gt ${#PERSONAS[@]} ] && PARALLEL=${#PERSONAS[@]}
+            ;;
+        --resume)
+            if [ "$MODE" = "fresh_set" ]; then echo "--fresh and --resume are mutually exclusive"; exit 2; fi
+            MODE="resume"; shift ;;
+        --fresh)
+            if [ "$MODE" = "resume" ]; then echo "--fresh and --resume are mutually exclusive"; exit 2; fi
+            MODE="fresh_set"; shift ;;
+        --skip-eval) PASS_ARGS+=(--skip-eval); shift ;;
+        *) echo "Unknown option: $1"; exit 1 ;;
+    esac
+done
+[ "$MODE" = "fresh_set" ] && MODE="fresh"
+
+START_TS=$(date +%Y%m%d_%H%M%S)
+SUMMARY_LOG="${SUITE_DIR}/logs/run_all_${START_TS}.summary"
+mkdir -p "${SUITE_DIR}/logs"
+
+echo "############################################################"
+echo "# reme_eval suite | mode=${MODE} parallel=${PARALLEL} | ${START_TS}"
+echo "############################################################"
+
+# ─── Fresh mode: suite-level wipe BEFORE anything starts ──────────────
+if [ "$MODE" = "fresh" ]; then
+    echo "[fresh] wiping ALL personas' memory workspaces, outputs and trace logs..."
+    for persona in "${PERSONAS[@]}"; do
+        rm -rf "${SUITE_DIR}/reme_workspace/${persona}"
+        rm -rf "${SUITE_DIR}/outputs/reme/${persona}"
+        rm -rf "${TRACE_ROOT}/reme/${persona}"
+        rm -rf "${SUITE_DIR}/nanobot_workspace/${persona}"
+    done
+    echo "[fresh] wipe done."
+else
+    echo "[resume] no memory wipe; personas resume after their last completed task."
+fi
+
+# ─── Run personas in batches of PARALLEL ──────────────────────────────
+STATUS_LIST=()
+ANY_FAILED=0
+OVERALL_START=$(date +%s)
+TOTAL=${#PERSONAS[@]}
+
+for ((i = 0; i < TOTAL; i += PARALLEL)); do
+    BATCH=("${PERSONAS[@]:i:PARALLEL}")
+    BATCH_PIDS=()
+    BATCH_NAMES=()
+    echo ""
+    echo "============================================================"
+    echo "# BATCH $(( i / PARALLEL + 1 )): ${BATCH[*]}   started $(date '+%F %T')"
+    echo "============================================================"
+    for persona in "${BATCH[@]}"; do
+        bash "${SUITE_DIR}/run_persona.sh" "${persona}" --resume ${PASS_ARGS[@]+"${PASS_ARGS[@]}"} \
+            > "${SUITE_DIR}/logs/suite_${persona}.log" 2>&1 &
+        BATCH_PIDS+=($!)
+        BATCH_NAMES+=("$persona")
+    done
+    for j in $(seq 0 $(( ${#BATCH[@]} - 1 ))); do
+        pid=${BATCH_PIDS[$j]}
+        persona=${BATCH_NAMES[$j]}
+        if wait "$pid"; then
+            STATUS_LIST+=("${persona}: OK")
+        else
+            rc=$?
+            ANY_FAILED=1
+            STATUS_LIST+=("${persona}: FAILED rc=${rc}")
+            echo "[run_all] ${persona} FAILED (rc=${rc}); see logs/suite_${persona}.log"
+        fi
+    done
+done
+
+total=$(( $(date +%s) - OVERALL_START ))
+echo ""
+echo "================ FINAL SUMMARY (${total}s total) ================" | tee -a "${SUMMARY_LOG}"
+for line in "${STATUS_LIST[@]}"; do
+    echo "  ${line}" | tee -a "${SUMMARY_LOG}"
+done
+echo "Summary: ${SUMMARY_LOG}"
+
+if [ "${ANY_FAILED}" -ne 0 ]; then
+    FAILED_COUNT=$(printf '%s\n' "${STATUS_LIST[@]}" | grep -c "FAILED")
+    echo "[run_all] ${FAILED_COUNT} persona(s) FAILED; suite run is marked as failed." | tee -a "${SUMMARY_LOG}"
+    exit 1
+fi
+exit 0

benchmark/pibench/run_persona.sh

@@ -0,0 +1,301 @@
+#!/bin/bash
+# Run the full pi-bench evaluation for ONE persona with the ReMe agent.
+# Tasks follow data/{persona}/episode.yaml order (runner-native).
+#
+# Usage: bash run_persona.sh <persona> [--fresh|--resume] [--skip-eval]
+#
+# Modes (default: --resume):
+#   --resume  Checkpoint continuation. Never wipes memory. Tasks already
+#             finished (SUCCESS/MAX_TURNS/TIMEOUT in the task history logs)
+#             are skipped via repeated --task-id flags. Before starting, any
+#             residual memory of tasks that are about to be RE-RUN (partial
+#             sessions from an interrupted run) is surgically removed by
+#             resume.py cleanup, so re-runs don't inherit leaked answers.
+#   --fresh   Wipes THIS persona's ReMe memory, outputs and trace logs first,
+#             then runs all tasks from scratch.
+# The two flags are mutually exclusive. A full multi-persona memory wipe is a
+# suite-level action of `run_all.sh` (fresh mode), never done here implicitly.
+set -uo pipefail
+
+SUITE_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+TRACE_ROOT="${HOME}/.nanobot/trace_logs"
+
+# ─── External dependencies (pi-bench / ReMe are NOT bundled; see README) ──
+if [ ! -f "${SUITE_DIR}/env.sh" ]; then
+    echo "env.sh not found. Run: cp env.sh.example env.sh  (then fill in the TODO items)"
+    exit 1
+fi
+source "${SUITE_DIR}/env.sh"
+
+PIBENCH_DIR="${PI_BENCH_ROOT:-}"
+if [ -z "${PIBENCH_DIR}" ] || [ ! -f "${PIBENCH_DIR}/src/main.py" ]; then
+    echo "PI_BENCH_ROOT is unset or invalid (src/main.py not found). Set it in env.sh."
+    exit 1
+fi
+if [ ! -x "${PIBENCH_DIR}/.venv/bin/python" ] || [ ! -x "${PIBENCH_DIR}/.venv/bin/appworld" ]; then
+    echo "pi-bench venv incomplete: ${PIBENCH_DIR}/.venv must provide python + appworld (see README setup)."
+    exit 1
+fi
+if [ ! -x "${REME_DIR}/.venv/bin/python" ]; then
+    echo "ReMe venv not found: ${REME_DIR}/.venv/bin/python (check REME_DIR in env.sh)"
+    exit 1
+fi
+if [ ! -e "${SUITE_DIR}/data" ]; then
+    echo 'Benchmark data not linked. Run: ln -s "$PI_BENCH_ROOT/data" data'
+    exit 1
+fi
+
+# ─── Pre-flight: files the runner needs before any service starts ─────
+MODEL_CONFIG="${SUITE_DIR}/config/models/reme.yaml"
+HISTORY_CONFIG="${SUITE_DIR}/config/bench/evaluation/trace_history.yaml"
+if [ ! -f "${MODEL_CONFIG}" ]; then
+    echo "Model config not found: ${MODEL_CONFIG} (see README directory layout)."
+    exit 1
+fi
+if [ ! -f "${HISTORY_CONFIG}" ]; then
+    echo "Trace history config not found: ${HISTORY_CONFIG}"
+    echo "pi-bench requires config/bench/evaluation/trace_history.yaml; see README."
+    exit 1
+fi
+
+APPWORLD_DIR="${PIBENCH_DIR}/third_party/appworld"
+PI_PYTHON="${PIBENCH_DIR}/.venv/bin/python"
+APPWORLD_BIN="${PIBENCH_DIR}/.venv/bin/appworld"
+# resume.py runs on the ReMe venv so it can reuse ReMe's daily-index rebuild.
+REME_PYTHON="${REME_DIR}/.venv/bin/python"
+
+PERSONA="${1:-}"
+if [ -z "$PERSONA" ]; then
+    echo "Usage: $0 <persona> [--fresh|--resume] [--skip-eval]"
+    exit 1
+fi
+shift
+
+MODE="resume"
+SKIP_EVAL=false
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        --fresh)
+            if [ "$MODE" = "resume_set" ]; then echo "--fresh and --resume are mutually exclusive"; exit 2; fi
+            MODE="fresh"; shift ;;
+        --resume)
+            if [ "$MODE" = "fresh" ]; then echo "--fresh and --resume are mutually exclusive"; exit 2; fi
+            MODE="resume_set"; shift ;;
+        --skip-eval) SKIP_EVAL=true; shift ;;
+        *) echo "Unknown option: $1"; exit 1 ;;
+    esac
+done
+[ "$MODE" = "resume_set" ] && MODE="resume"
+
+# ─── Per-persona ports (pi-bench AGENTS.md convention) ────────────────
+# REME_PORT: ReMe's internal HTTP service; must be unique per concurrent bridge.
+case "$PERSONA" in
+    marketer)     API_PORT=9001; MCP_PORT=10001; TEST_PORT=9998; REME_PORT=18766 ;;
+    law_trainee)  API_PORT=9002; MCP_PORT=10002; TEST_PORT=9997; REME_PORT=18767 ;;
+    pharmacist)   API_PORT=9003; MCP_PORT=10003; TEST_PORT=9996; REME_PORT=18768 ;;
+    researcher)   API_PORT=9004; MCP_PORT=10004; TEST_PORT=9995; REME_PORT=18765 ;;
+    Financier)    API_PORT=9005; MCP_PORT=10005; TEST_PORT=9994; REME_PORT=18769 ;;
+    *) echo "Unknown persona: $PERSONA"; exit 1 ;;
+esac
+
+API_URL="http://127.0.0.1:${API_PORT}"
+MCP_URL="http://127.0.0.1:${MCP_PORT}/mcp"
+TEST_URL="http://127.0.0.1:${TEST_PORT}"
+LOG_DIR="${SUITE_DIR}/logs"
+mkdir -p "${LOG_DIR}"
+
+# ─── Environment (env.sh already sourced at the top) ──────────────────
+WORKSPACE_DIR="${REME_WORKSPACE_ROOT}/${PERSONA}"
+NANOBOT_WORKSPACE_DIR="${SUITE_DIR}/nanobot_workspace/${PERSONA}"
+mkdir -p "${WORKSPACE_DIR}" "${NANOBOT_WORKSPACE_DIR}"
+
+echo "========================================="
+echo "ReMe x Pi-Bench | persona=${PERSONA} | mode=${MODE}"
+echo "  api=${API_PORT} mcp=${MCP_PORT} test=${TEST_PORT} reme=${REME_PORT}"
+echo "  model=${REME_MODEL_NAME}"
+echo "  memory workspace=${WORKSPACE_DIR}  (persistent)"
+echo "========================================="
+
+# ─── Fresh mode: wipe this persona's state ────────────────────────────
+if [ "$MODE" = "fresh" ]; then
+    echo "[fresh] wiping persona state: memory workspace, outputs, trace logs"
+    rm -rf "${WORKSPACE_DIR}"
+    rm -rf "${SUITE_DIR}/outputs/reme/${PERSONA}"
+    rm -rf "${TRACE_ROOT}/reme/${PERSONA}"
+    rm -rf "${NANOBOT_WORKSPACE_DIR}"
+    mkdir -p "${WORKSPACE_DIR}" "${NANOBOT_WORKSPACE_DIR}"
+fi
+
+# ─── Resume: determine remaining tasks + clean partial memories ───────
+TASK_ARGS=()
+RUN_PHASE_NEEDED=true
+if [ "$MODE" = "resume" ]; then
+    REMAINING_JSON="$("${REME_PYTHON}" "${SUITE_DIR}/resume.py" remaining "${PERSONA}" --json)"
+    if [ -z "$REMAINING_JSON" ]; then
+        echo "Failed to compute remaining tasks"; exit 1
+    fi
+    echo "[resume] ${REMAINING_JSON}"
+    REMAINING_TASKS=()
+    while IFS= read -r tid_line; do
+        [ -n "$tid_line" ] && REMAINING_TASKS+=("$tid_line")
+    done < <("${REME_PYTHON}" "${SUITE_DIR}/resume.py" remaining "${PERSONA}" 2>/dev/null)
+    if [ ${#REMAINING_TASKS[@]} -eq 0 ]; then
+        RUN_PHASE_NEEDED=false
+        echo "[resume] all tasks already completed; skipping run phase"
+    else
+        # Remove residual memory of interrupted (to-be-re-run) tasks so
+        # re-runs don't get their own partial answers injected.
+        "${REME_PYTHON}" "${SUITE_DIR}/resume.py" cleanup "${PERSONA}"
+        for tid in "${REMAINING_TASKS[@]}"; do
+            TASK_ARGS+=(--task-id "$tid")
+        done
+        echo "[resume] running ${#REMAINING_TASKS[@]} remaining task(s): ${REMAINING_TASKS[*]}"
+    fi
+fi
+
+# ─── Port cleanup from previous runs ──────────────────────────────────
+for port in ${API_PORT} ${MCP_PORT} ${TEST_PORT} ${REME_PORT}; do
+    pids=$(lsof -ti :${port} 2>/dev/null || true)
+    if [ -n "$pids" ]; then
+        echo "Killing stale processes on port ${port}: ${pids}"
+        kill -9 $pids 2>/dev/null || true
+    fi
+done
+sleep 2
+
+PIDS=()
+cleanup() {
+    echo "[${PERSONA}] cleaning up services..."
+    for pid in "${PIDS[@]:-}"; do
+        kill "$pid" 2>/dev/null || true
+    done
+    wait 2>/dev/null || true
+}
+trap cleanup EXIT INT TERM
+
+wait_for_service() {
+    local url="$1" name="$2" port="$3" timeout="${4:-180}"
+    echo -n "  waiting for ${name}..."
+    local start=$(date +%s)
+    while true; do
+        if curl -sf --max-time 5 "${url}" > /dev/null 2>&1; then
+            echo " ready"; return 0
+        fi
+        if [ -n "$port" ] && lsof -ti :${port} > /dev/null 2>&1; then
+            local elapsed=$(( $(date +%s) - start ))
+            if [ "$elapsed" -ge 10 ]; then echo " ready (port)"; return 0; fi
+        fi
+        if [ $(( $(date +%s) - start )) -ge "$timeout" ]; then
+            echo " TIMEOUT"; return 1
+        fi
+        sleep 2
+    done
+}
+
+# ─── [1/5] AppWorld API ────────────────────────────────────────────────
+echo "[1/5] AppWorld API (:${API_PORT})"
+(cd "${APPWORLD_DIR}" && exec "${APPWORLD_BIN}" serve apis --root . \
+    --port ${API_PORT}) > "${LOG_DIR}/appworld_api_${PERSONA}.log" 2>&1 &
+PIDS+=($!)
+if ! wait_for_service "${API_URL}/docs" "AppWorld API" "${API_PORT}" 180; then
+    tail -20 "${LOG_DIR}/appworld_api_${PERSONA}.log"; exit 1
+fi
+
+# ─── [2/5] AppWorld MCP ────────────────────────────────────────────────
+echo "[2/5] AppWorld MCP (:${MCP_PORT})"
+TOOLS_CONFIG="${SUITE_DIR}/data/${PERSONA}/tools.yaml"
+(cd "${APPWORLD_DIR}" && exec "${APPWORLD_BIN}" serve mcp http --root . \
+    --remote-apis-url "${API_URL}" --port ${MCP_PORT} \
+    --tools-config-file "${TOOLS_CONFIG}") > "${LOG_DIR}/appworld_mcp_${PERSONA}.log" 2>&1 &
+PIDS+=($!)
+if ! wait_for_service "${MCP_URL}" "AppWorld MCP" "${MCP_PORT}" 180; then
+    tail -20 "${LOG_DIR}/appworld_mcp_${PERSONA}.log"; exit 1
+fi
+
+# ─── [3/5] Test Server ─────────────────────────────────────────────────
+echo "[3/5] Test Server (:${TEST_PORT})"
+PORT=${TEST_PORT} "${PI_PYTHON}" "${PIBENCH_DIR}/scripts/test_server.py" \
+    > "${LOG_DIR}/test_server_${PERSONA}.log" 2>&1 &
+PIDS+=($!)
+if ! wait_for_service "${TEST_URL}/sent?after=-1" "Test Server" "${TEST_PORT}" 30; then
+    tail -20 "${LOG_DIR}/test_server_${PERSONA}.log"; exit 1
+fi
+
+# ─── [4/5] ReMe Bridge (ReMe venv) ─────────────────────────────────────
+echo "[4/5] ReMe Bridge (reme service port ${REME_PORT})"
+"${REME_DIR}/.venv/bin/python" "${SUITE_DIR}/bridge_reme.py" \
+    --test-server-url "${TEST_URL}" \
+    --appworld-mcp-url "${MCP_URL}" \
+    --reme-dir "${REME_DIR}" \
+    --data-root "${SUITE_DIR}/data" \
+    --user-id "${PERSONA}" \
+    --workspace-dir "${WORKSPACE_DIR}" \
+    --reme-port "${REME_PORT}" \
+    --model-name "${REME_MODEL_NAME}" \
+    --model-base-url "${REME_LLM_BASE_URL}" \
+    --model-api-key "${REME_LLM_API_KEY}" \
+    > "${LOG_DIR}/bridge_${PERSONA}.log" 2>&1 &
+BRIDGE_PID=$!
+PIDS+=(${BRIDGE_PID})
+sleep 5
+if ! kill -0 "${BRIDGE_PID}" 2>/dev/null; then
+    echo "Bridge failed to start:"; tail -30 "${LOG_DIR}/bridge_${PERSONA}.log"; exit 1
+fi
+for i in $(seq 1 12); do
+    if grep -q "Bridge started:" "${LOG_DIR}/bridge_${PERSONA}.log" 2>/dev/null; then
+        echo "  bridge initialized"; break
+    fi
+    sleep 5
+done
+grep -q "Bridge started:" "${LOG_DIR}/bridge_${PERSONA}.log" 2>/dev/null || {
+    echo "WARNING: bridge may not be ready:"; tail -20 "${LOG_DIR}/bridge_${PERSONA}.log"; }
+
+# ─── [5/5] Runner (run phase) ──────────────────────────────────────────
+if [ "$RUN_PHASE_NEEDED" = true ]; then
+    echo "[5/5] Runner: run phase (episode order from data/${PERSONA}/episode.yaml)"
+    cd "${SUITE_DIR}"
+    BENCH_TEST_SERVER_URL="${TEST_URL}" PYTHONPATH="${PIBENCH_DIR}" \
+        "${PI_PYTHON}" -m src.main \
+        --model-config "${MODEL_CONFIG}" \
+        --history-config-path "${HISTORY_CONFIG}" \
+        --mode run --user-id "${PERSONA}" \
+        --workspace-dir "${NANOBOT_WORKSPACE_DIR}" \
+        ${TASK_ARGS[@]+"${TASK_ARGS[@]}"} \
+        2>&1 | tee "${LOG_DIR}/runner_run_${PERSONA}.log"
+    RUN_EXIT=${PIPESTATUS[0]}
+    if [ ${RUN_EXIT} -ne 0 ]; then
+        echo "Run phase failed (exit ${RUN_EXIT}). Logs: ${LOG_DIR}/"
+        exit ${RUN_EXIT}
+    fi
+else
+    echo "[5/5] Runner: run phase skipped (all tasks completed)"
+fi
+
+if [ "$SKIP_EVAL" = true ]; then
+    echo "Skipping eval (--skip-eval)"
+    exit 0
+fi
+
+# ─── Trace conversion + eval phase (always over all available traces) ──
+echo "Converting trace logs..."
+"${PI_PYTHON}" "${SUITE_DIR}/fix_trace_logs.py" "${PERSONA}"
+
+echo "Runner: eval phase"
+cd "${SUITE_DIR}"
+BENCH_TEST_SERVER_URL="${TEST_URL}" PYTHONPATH="${PIBENCH_DIR}" \
+    "${PI_PYTHON}" -m src.main \
+    --model-config "${MODEL_CONFIG}" \
+    --history-config-path "${HISTORY_CONFIG}" \
+    --mode eval --user-id "${PERSONA}" \
+    --workspace-dir "${NANOBOT_WORKSPACE_DIR}" \
+    2>&1 | tee "${LOG_DIR}/runner_eval_${PERSONA}.log"
+EVAL_EXIT=${PIPESTATUS[0]}
+
+echo ""
+echo "========================================="
+echo "persona=${PERSONA} finished (eval exit=${EVAL_EXIT})"
+echo "  results : ${SUITE_DIR}/outputs/reme/${PERSONA}/"
+echo "  memory  : ${WORKSPACE_DIR}/"
+echo "  logs    : ${LOG_DIR}/"
+echo "========================================="
+exit ${EVAL_EXIT}

benchmark/toolmemory/README.md

@@ -0,0 +1,98 @@
+## Towards Robust Tool Use in Agents via Experience-Driven Adaptive Guidance
+
+**Language**: English (default) / [中文](./README_ZH.md)
+
+> Paper: [arXiv:2608.03403](https://arxiv.org/abs/2608.03403)
+> Code: [https://github.com/WangCan1178/ExpG](https://github.com/WangCan1178/ExpG)
+
+<p align="center">
+  <img src="gitcha.png" alt="ExpG challenges and overview" width="85%">
+</p>
+
+### Overview
+
+This folder archives **ExpG**, a tool-use enhancement built on [Agentscope ReMe](https://github.com/agentscope-ai/ReMe). ExpG mines, distills, and reuses experience from historical tool calls to provide **capability boundaries** and **best-practice guidance**, which helps agents:
+
+- Select and invoke tools more robustly under dynamic or noisy environments;
+- Let smaller models with guidance outperform larger, memoryless baselines;
+- Improve consistently across tool selection, tool calling, and response generation.
+
+**How ReMe is used:** Start the Tool Memory service; historical tool calls are written and evaluated via `add_tool_call_result`, distilled into tool-level guidance via `summary_tool_memory`, then retrieved and injected into later reasoning via `retrieve_tool_memory`. ReMe provides the vector store and service APIs; the acquisition / distillation / reuse strategy is implemented by ExpG. Full implementation and experiments are in [WangCan1178/ExpG](https://github.com/WangCan1178/ExpG).
+
+---
+
+### ExpG Mechanism
+
+ExpG treats tool invocations as learnable experience and runs a three-stage pipeline:
+
+1. **Experience Acquisition**
+   - Analyze invocation quality from historical trajectories (success/failure, cost, latency, etc.);
+   - Build structured experience units per tool, recording context, parameter patterns, and outcomes.
+
+2. **Experience Distillation**
+   - Filter noisy or unhelpful experiences and keep representative patterns;
+   - Aggregate by equivalence classes to cover common and rare failure modes;
+   - Summarize with an LLM into generalizable textual guidance.
+
+3. **Experience Reuse**
+   - Retrieve relevant experience / guidance for future tasks;
+   - Inject guidance into tool selection, argument generation, and response synthesis;
+   - Improve stability under dynamic environments and imperfect feedback.
+
+---
+
+### Main Results
+
+Performance comparison (%) across MetaTool, API-Bank, and BFCL-V3. **Bold** indicates the best results within each model.
+
+| Model | Method | MetaTool Pass@1 | MetaTool Avg@3 | MetaTool Pass@3 | API-Bank Pass@1 | API-Bank Avg@3 | API-Bank Pass@3 | BFCL-V3 Pass@1 | BFCL-V3 Avg@3 | BFCL-V3 Pass@3 | Total Pass@1 | Total Avg@3 | Total Pass@3 |
+| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| GPT-5 nano | No Method | 72.62 | 72.76 | 78.49 | 82.96 | 83.46 | 86.97 | 53.80 | 53.00 | 60.95 | 70.82 | 70.62 | 76.63 |
+| GPT-5 nano | Few-shot | 74.12 | 75.11 | 82.32 | 83.71 | 83.96 | **87.22** | 56.18 | 55.24 | 61.39 | 72.36 | 72.65 | 79.28 |
+| GPT-5 nano | DRAFT | 73.94 | 73.04 | 78.97 | 84.21 | 83.46 | **87.22** | 57.27 | 57.27 | 62.26 | 72.52 | 71.58 | 77.23 |
+| GPT-5 nano | Mem0 | 74.96 | 76.13 | 82.92 | 84.96 | 85.21 | **87.22** | 60.95 | 61.61 | 65.08 | 73.98 | 74.67 | 80.35 |
+| GPT-5 nano | **ExpG** | **81.67** | **82.07** | **84.60** | **86.72** | **86.55** | **87.22** | **64.43** | **63.99** | **66.38** | **79.32** | **79.22** | **81.69** |
+| DeepSeek-V3 | No Method | 83.10 | 82.94 | 84.66 | 84.71 | 84.38 | 85.46 | 58.79 | 59.65 | 65.94 | 78.92 | 78.66 | 81.37 |
+| DeepSeek-V3 | Few-shot | 82.74 | 83.90 | 86.28 | 85.21 | 84.63 | 86.22 | 60.52 | 60.30 | 67.90 | 79.08 | 79.45 | 82.92 |
+| DeepSeek-V3 | DRAFT | 80.23 | 80.79 | 82.44 | 84.96 | 85.63 | 86.47 | 62.26 | 61.61 | 68.55 | 77.70 | 77.80 | 80.54 |
+| DeepSeek-V3 | Mem0 | 83.88 | 84.56 | 86.40 | 85.46 | 85.55 | 86.47 | 65.08 | 65.15 | 68.33 | 80.70 | 80.91 | 83.12 |
+| DeepSeek-V3 | **ExpG** | **85.26** | **85.38** | **86.52** | **87.72** | **87.39** | **87.97** | **69.41** | **69.92** | **72.02** | **82.76** | **82.61** | **84.11** |
+| Qwen3-8B | No Method | 76.51 | 76.97 | 77.71 | 83.96 | 83.88 | 84.21 | 58.79 | 58.28 | 60.30 | 74.46 | 74.41 | 75.56 |
+| Qwen3-8B | Few-shot | 79.93 | 79.83 | 82.92 | 83.71 | 82.62 | 84.96 | 60.09 | 59.29 | 61.39 | 76.91 | 76.27 | 79.32 |
+| Qwen3-8B | DRAFT | 78.19 | 77.33 | 77.89 | 85.71 | 84.96 | 85.46 | 60.74 | 60.30 | 62.91 | 76.20 | 75.18 | 76.35 |
+| Qwen3-8B | Mem0 | 75.07 | 75.47 | 82.38 | 86.22 | 86.05 | 86.47 | 63.34 | 64.93 | 66.16 | 74.69 | 74.98 | 80.07 |
+| Qwen3-8B | **ExpG** | **83.52** | **84.88** | **85.08** | **86.47** | **87.89** | **87.97** | **67.46** | **66.96** | **68.33** | **81.06** | **81.82** | **82.48** |
+| Qwen3-32B | No Method | 80.05 | 79.43 | 80.17 | 84.71 | 84.88 | 85.21 | 65.15 | 65.08 | 66.16 | 78.05 | 77.55 | 78.41 |
+| Qwen3-32B | **ExpG** | **84.68** | **85.02** | **86.28** | **86.97** | **87.30** | **87.72** | **70.72** | **71.01** | **73.32** | **82.48** | **82.56** | **84.14** |
+| Qwen3-235B | No Method | 78.25 | 79.23 | 80.29 | 85.46 | 85.46 | 85.71 | 71.37 | 71.15 | 73.54 | 78.13 | 78.49 | 79.91 |
+| Qwen3-235B | **ExpG** | **86.34** | **86.70** | **86.94** | **87.47** | **86.97** | **88.22** | **79.61** | **78.52** | **80.04** | **85.29** | **84.98** | **85.69** |
+
+---
+
+### Reference Code
+
+| Path | Role |
+| --- | --- |
+| [`tool_memory.py`](./tool_memory.py) | HTTP client for official ReMe Tool Memory APIs (`add_tool_call_result` / `summary_tool_memory` / `retrieve_tool_memory`) |
+| [`parse_tool_call_result_prompt.yaml`](./parse_tool_call_result_prompt.yaml) | Prompt for multi-aspect evaluation of each tool call |
+| [`summary_tool_memory_prompt.yaml`](./summary_tool_memory_prompt.yaml) | Prompt for summarizing tool call history into guidance |
+| [`tool_memory_flows.yaml`](./tool_memory_flows.yaml) | Tool Memory flow / op config excerpt |
+
+These are reference snippets. For the full runnable codebase, see [WangCan1178/ExpG](https://github.com/WangCan1178/ExpG).
+
+---
+
+### Citation
+
+```bibtex
+@misc{wang2026expg,
+  title         = {Towards Robust Tool Use in Agents via Experience-Driven Adaptive Guidance},
+  author        = {Can Wang and Haoran Chen and Li Yu and Ding Hao and Bohai Zhao and Zhaoyang Liu and Zhiying Tu},
+  year          = {2026},
+  eprint        = {2608.03403},
+  archivePrefix = {arXiv},
+  primaryClass  = {cs.AI},
+  url           = {https://arxiv.org/abs/2608.03403},
+  howpublished  = {\url{https://github.com/WangCan1178/ExpG}}
+}
+```

benchmark/toolmemory/README_ZH.md

@@ -0,0 +1,98 @@
+## Towards Robust Tool Use in Agents via Experience-Driven Adaptive Guidance
+
+**语言**：中文 / [English](./README.md)
+
+> 论文：[arXiv:2608.03403](https://arxiv.org/abs/2608.03403)
+> 代码：[https://github.com/WangCan1178/ExpG](https://github.com/WangCan1178/ExpG)
+
+<p align="center">
+  <img src="gitcha.png" alt="ExpG 挑战与概览" width="85%">
+</p>
+
+### 简介
+
+本目录归档基于 [Agentscope ReMe](https://github.com/agentscope-ai/ReMe) 的工具使用增强工作 **ExpG**：在 ReMe 记忆框架之上，从历史工具调用中挖掘、提炼并复用经验，为智能体提供工具的 **能力边界** 与 **最佳实践指导**，从而：
+
+- 在动态或有噪环境下更鲁棒地选择和调用工具；
+- 让较小模型在带有经验指导时超越更大、但无记忆的基线；
+- 在工具选择、工具调用和响应生成等多个阶段带来一致收益。
+
+**如何使用 ReMe：** 启动 Tool Memory 服务后，历史工具调用经 `add_tool_call_result` 写入并评估，经 `summary_tool_memory` 蒸馏成工具级指导，再经 `retrieve_tool_memory` 取回并注入后续推理。向量存储与服务接口由 ReMe 提供，经验获取 / 蒸馏 / 复用策略由 ExpG 实现。完整实现与实验见 [WangCan1178/ExpG](https://github.com/WangCan1178/ExpG)。
+
+---
+
+### ExpG 机制概览
+
+ExpG 将工具调用视为可学习经验，并通过三阶段流水线完成经验的获取、提炼与复用：
+
+1. **经验获取（Experience Acquisition）**
+   - 从历史工具调用轨迹中分析调用质量（成功/失败、代价、时间等）；
+   - 针对不同工具构建结构化的经验单元，记录调用上下文、参数模式和结果。
+
+2. **经验蒸馏（Experience Distillation）**
+   - 过滤无效 / 噪声经验，保留具有代表性的调用模式；
+   - 基于“等价类”视角对经验进行聚合，覆盖常见模式与稀有失败模式；
+   - 使用 LLM 对经验进行总结，形成可泛化的文本化指导（guidance）。
+
+3. **经验复用（Experience Reuse）**
+   - 在未来任务中，根据当前工具调用上下文检索相关经验 / 指导；
+   - 将经验引导融入到工具选择、参数生成和响应整理等环节；
+   - 使得代理在面对动态环境和不完美反馈时仍能保持稳定表现。
+
+---
+
+### 主实验结果
+
+MetaTool、API-Bank、BFCL-V3 上的性能对比（%）。**加粗**为各模型组内最优。
+
+| Model | Method | MetaTool Pass@1 | MetaTool Avg@3 | MetaTool Pass@3 | API-Bank Pass@1 | API-Bank Avg@3 | API-Bank Pass@3 | BFCL-V3 Pass@1 | BFCL-V3 Avg@3 | BFCL-V3 Pass@3 | Total Pass@1 | Total Avg@3 | Total Pass@3 |
+| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| GPT-5 nano | No Method | 72.62 | 72.76 | 78.49 | 82.96 | 83.46 | 86.97 | 53.80 | 53.00 | 60.95 | 70.82 | 70.62 | 76.63 |
+| GPT-5 nano | Few-shot | 74.12 | 75.11 | 82.32 | 83.71 | 83.96 | **87.22** | 56.18 | 55.24 | 61.39 | 72.36 | 72.65 | 79.28 |
+| GPT-5 nano | DRAFT | 73.94 | 73.04 | 78.97 | 84.21 | 83.46 | **87.22** | 57.27 | 57.27 | 62.26 | 72.52 | 71.58 | 77.23 |
+| GPT-5 nano | Mem0 | 74.96 | 76.13 | 82.92 | 84.96 | 85.21 | **87.22** | 60.95 | 61.61 | 65.08 | 73.98 | 74.67 | 80.35 |
+| GPT-5 nano | **ExpG** | **81.67** | **82.07** | **84.60** | **86.72** | **86.55** | **87.22** | **64.43** | **63.99** | **66.38** | **79.32** | **79.22** | **81.69** |
+| DeepSeek-V3 | No Method | 83.10 | 82.94 | 84.66 | 84.71 | 84.38 | 85.46 | 58.79 | 59.65 | 65.94 | 78.92 | 78.66 | 81.37 |
+| DeepSeek-V3 | Few-shot | 82.74 | 83.90 | 86.28 | 85.21 | 84.63 | 86.22 | 60.52 | 60.30 | 67.90 | 79.08 | 79.45 | 82.92 |
+| DeepSeek-V3 | DRAFT | 80.23 | 80.79 | 82.44 | 84.96 | 85.63 | 86.47 | 62.26 | 61.61 | 68.55 | 77.70 | 77.80 | 80.54 |
+| DeepSeek-V3 | Mem0 | 83.88 | 84.56 | 86.40 | 85.46 | 85.55 | 86.47 | 65.08 | 65.15 | 68.33 | 80.70 | 80.91 | 83.12 |
+| DeepSeek-V3 | **ExpG** | **85.26** | **85.38** | **86.52** | **87.72** | **87.39** | **87.97** | **69.41** | **69.92** | **72.02** | **82.76** | **82.61** | **84.11** |
+| Qwen3-8B | No Method | 76.51 | 76.97 | 77.71 | 83.96 | 83.88 | 84.21 | 58.79 | 58.28 | 60.30 | 74.46 | 74.41 | 75.56 |
+| Qwen3-8B | Few-shot | 79.93 | 79.83 | 82.92 | 83.71 | 82.62 | 84.96 | 60.09 | 59.29 | 61.39 | 76.91 | 76.27 | 79.32 |
+| Qwen3-8B | DRAFT | 78.19 | 77.33 | 77.89 | 85.71 | 84.96 | 85.46 | 60.74 | 60.30 | 62.91 | 76.20 | 75.18 | 76.35 |
+| Qwen3-8B | Mem0 | 75.07 | 75.47 | 82.38 | 86.22 | 86.05 | 86.47 | 63.34 | 64.93 | 66.16 | 74.69 | 74.98 | 80.07 |
+| Qwen3-8B | **ExpG** | **83.52** | **84.88** | **85.08** | **86.47** | **87.89** | **87.97** | **67.46** | **66.96** | **68.33** | **81.06** | **81.82** | **82.48** |
+| Qwen3-32B | No Method | 80.05 | 79.43 | 80.17 | 84.71 | 84.88 | 85.21 | 65.15 | 65.08 | 66.16 | 78.05 | 77.55 | 78.41 |
+| Qwen3-32B | **ExpG** | **84.68** | **85.02** | **86.28** | **86.97** | **87.30** | **87.72** | **70.72** | **71.01** | **73.32** | **82.48** | **82.56** | **84.14** |
+| Qwen3-235B | No Method | 78.25 | 79.23 | 80.29 | 85.46 | 85.46 | 85.71 | 71.37 | 71.15 | 73.54 | 78.13 | 78.49 | 79.91 |
+| Qwen3-235B | **ExpG** | **86.34** | **86.70** | **86.94** | **87.47** | **86.97** | **88.22** | **79.61** | **78.52** | **80.04** | **85.29** | **84.98** | **85.69** |
+
+---
+
+### 参考代码
+
+| 路径 | 作用 |
+| --- | --- |
+| [`tool_memory.py`](./tool_memory.py) | 官方风格 ReMe Tool Memory HTTP 客户端（`add_tool_call_result` / `summary_tool_memory` / `retrieve_tool_memory`） |
+| [`parse_tool_call_result_prompt.yaml`](./parse_tool_call_result_prompt.yaml) | 单次工具调用多维评估用的 prompt |
+| [`summary_tool_memory_prompt.yaml`](./summary_tool_memory_prompt.yaml) | 将工具调用历史总结为 guidance 的 prompt |
+| [`tool_memory_flows.yaml`](./tool_memory_flows.yaml) | Tool Memory 相关的 flow / op 配置摘录 |
+
+以上为参考片段。完整可运行代码见 [WangCan1178/ExpG](https://github.com/WangCan1178/ExpG)。
+
+---
+
+### 引用
+
+```bibtex
+@misc{wang2026expg,
+  title         = {Towards Robust Tool Use in Agents via Experience-Driven Adaptive Guidance},
+  author        = {Can Wang and Haoran Chen and Li Yu and Ding Hao and Bohai Zhao and Zhaoyang Liu and Zhiying Tu},
+  year          = {2026},
+  eprint        = {2608.03403},
+  archivePrefix = {arXiv},
+  primaryClass  = {cs.AI},
+  url           = {https://arxiv.org/abs/2608.03403},
+  howpublished  = {\url{https://github.com/WangCan1178/ExpG}}
+}
+```

benchmark/toolmemory/parse_tool_call_result_prompt.yaml

@@ -0,0 +1,49 @@
+prompt: |
+  You are an expert in evaluating tool invocation process. The tool is invoked by an AI agent.
+
+  Tool invocation Information:
+  - Tool Name: {tool_name}
+  - Success Flag: {success_flag}
+  - Time Cost: {time_cost}s
+  - Token Cost: {token_cost} tokens
+  - Agent Context: {context}
+  - Input Parameters: {input_params}
+  - Tool Response: {response}
+  - Tool Schema: {schema}
+
+  Evaluation Method:
+  Start from a default score list of scores = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0].
+  For each item below that is satisfied, assign 1 point to the corresponding index.
+  The final scores should be a list of 10 integers, each being either 0 or 1.
+
+  1. Use Quality (total 2 points. If context is provided, use it as an aid when evaluating):
+  - Index 1: Should the tool be invoked now? Consider whether all necessary information for the tool's invocation is ready, and whether the tool execution environment is correct. If it is a multi-round conversation, also consider the dependency relationships of the tool chain.
+  - Index 2: If should, is the chosen tool appropriate?
+
+  2. Input Quality (total 4 points. When evaluating, consider both the context and the tool schema):
+  - Index 3: Are all required parameters provided?
+  - Index 4: Are the input parameters valid and supported by the tool?
+  - Index 5: Are the input parameters in the correct format for their respective fields?
+  - Index 6: Does the value (content) of input parameter correctly reflect and match the given context?
+
+  3. Response Quality (total 4 points):
+  - Index 7: Does the response provide meaningful and useful information? Or are there any error messages or information that can be used as guidance for agent invoking tool better?
+  - Index 8: Does the response match the tool's intended purpose/function?
+  - Index 9: Does the response value correct (content appropriate) given the input parameters?
+  - Index 10: Does the response help accomplish the task within the given context?
+
+  Important:
+  1. Sometimes there is not enough information in the context or schema to make a complete evaluation. In such cases, make your best judgment based on the available information.
+  2. Some tools (commonly system tools such as mkdir, touch, echo, etc.) modify the external environment. Since these results cannot be obtained, they return "None" as the response. At this point, all the scores in the quality of the response should be obtained and should not be seen as a problem for the tool.
+  3. Evaluation independently from the success flag. The success_flag indicates whether the tool executed without technical errors. The evaluation should evaluate the quality of the tool invocation. A tool can execute successfully (Success Flag=1) but still produce low-quality or irrelevant responses, leading to a low evaluation score.
+  4. Sometimes an agent will execute multiple steps and invoke multiple tools to complete a task, but you only need to evaluate the use of one tool for one of the steps, not whether the final task is completed or not.
+
+  Answer Format:
+  Please provide your answer in the following JSON format:
+
+  ```json
+  {
+    "scores": [0,0,0,0,0,0,0,0,0,0],
+    "explanation": "A brief evaluation (2-3 sentences) explaining the quality of the tool invocation, based on your evaluation. Low-quality aspects need to be reified, especially the causes of tool invocation errors."
+  }
+  ```

benchmark/toolmemory/summary_tool_memory_prompt.yaml

@@ -0,0 +1,32 @@
+prompt: |
+  You are an expert in analyzing tool usage patterns and generating practical usage guidance for agents.
+
+  Tool Information:
+  - Tool Name: {tool_name}
+  - Tool Schema: {tool_schema}
+
+  Recent Tool Invocation Experiences:
+  {experiences}
+
+  Important:
+  1. Assume the tool (tool schema) can't be changed, your task is to guide agent to use it better.
+  2. Your answer must be based on the information given, don't make it up. If not enough data, state "Not enough data to determine Core Function/Success Patterns/Common Issues/Best Practices."
+  3. Your answer will be used to guide the use of the tool in the future, so do not include content related to recent tool invocation experience such as "case #3" or "Call #2", but some values can be used as examples.
+  4. Pay attention to information not mentioned in the tool schema, such as the response upon successful tool invocation. It's also welcome to uncover insights, such as how tools can be used more effectively, and possible dependencies between tools. But if they aren't, don't make them up.
+  5. Finally, to avoid deriving incorrect guidance from individual invocation, check whether, if the agent follows the proposed guidance, it can perform better on all recent invocation histories. If not, revise the guidance until it can. Specifically:
+     - Don't write guidance in an absolute tone without a very deterministic message (meaning that all invocation histories are satisfied, otherwise it will result in failure).
+     - Sometimes there may be inconsistencies. Consider whether this is due to the context in which the tool is being used.
+
+  Your Task:
+  Based on the tool invocation history, generate a concise and logical tool usage guidance following this structure:
+  1. Core Function: What this tool does and when to use it.
+  2. Success Patterns: Parameter patterns and usage scenarios that work well.
+  3. Common Issues: Main pitfalls to avoid and why they fail.
+  4. Best Practices: 2-3 actionable recommendations.
+
+  Answer Format:
+  Provide a structured, concise guidance (max 200 words). Focus on actionable insights derived from actual usage data. Avoid generic advice and think step by step.
+
+  ```txt
+  Your concise, data-driven tool usage guidance
+  ```

benchmark/toolmemory/tool_memory.py

@@ -0,0 +1,234 @@
+"""Official-style ReMe Tool Memory HTTP helpers.
+
+Aligned with ReMe Tool Memory HTTP APIs (see ReMe cookbook
+``use_tool_memory_demo.py`` and docs under ``docs/tool_memory/``):
+
+- ``add_tool_call_result``
+- ``summary_tool_memory``
+- ``retrieve_tool_memory``
+
+Response memories are read from ``metadata.memory_list[].content``.
+This module does not use ExpG-only fields such as ``no_persist``,
+``source_task``, or ``add_to``.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, Dict, List, Optional
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_BASE_URL = "http://localhost:8002"
+
+
+class ToolMemoryFetcher:
+    """HTTP client for ReMe Tool Memory endpoints."""
+
+    def __init__(
+        self,
+        workspace_id: str,
+        base_url: str = DEFAULT_BASE_URL,
+        timeout: float = 60.0,
+    ) -> None:
+        self.workspace_id = workspace_id
+        self.base_url = base_url.rstrip("/")
+        self.timeout = timeout
+
+    def _url(self, endpoint: str) -> str:
+        return f"{self.base_url}/{endpoint.lstrip('/')}"
+
+    @staticmethod
+    def _join_tool_names(tool_names: List[str] | str) -> str:
+        if isinstance(tool_names, str):
+            return tool_names
+        return ",".join(tool_names)
+
+    @staticmethod
+    def _memory_list(payload: Dict[str, Any]) -> List[Dict[str, Any]]:
+        metadata = payload.get("metadata") or {}
+        if not isinstance(metadata, dict):
+            return []
+        memory_list = metadata.get("memory_list") or []
+        return memory_list if isinstance(memory_list, list) else []
+
+    @classmethod
+    def _content_by_tool(cls, payload: Dict[str, Any]) -> Dict[str, str]:
+        result: Dict[str, str] = {}
+        for memory in cls._memory_list(payload):
+            if not isinstance(memory, dict):
+                continue
+            tool_name = str(memory.get("when_to_use") or "").strip()
+            content = memory.get("content") or ""
+            if tool_name:
+                result[tool_name] = str(content)
+        return result
+
+    async def add_tool_call_result_async(
+        self,
+        tool_call_results: List[Dict[str, Any]],
+    ) -> Dict[str, Any]:
+        """Call ``add_tool_call_result``."""
+        async with httpx.AsyncClient() as client:
+            response = await client.post(
+                self._url("add_tool_call_result"),
+                json={
+                    "workspace_id": self.workspace_id,
+                    "tool_call_results": tool_call_results,
+                },
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+
+    async def summary_tool_memory_async(
+        self,
+        tool_names: List[str] | str,
+    ) -> Dict[str, Any]:
+        """Call ``summary_tool_memory``."""
+        async with httpx.AsyncClient() as client:
+            response = await client.post(
+                self._url("summary_tool_memory"),
+                json={
+                    "workspace_id": self.workspace_id,
+                    "tool_names": self._join_tool_names(tool_names),
+                },
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+
+    async def retrieve_tool_memory_async(
+        self,
+        tool_names: List[str] | str,
+    ) -> Dict[str, Any]:
+        """Call ``retrieve_tool_memory``."""
+        async with httpx.AsyncClient() as client:
+            response = await client.post(
+                self._url("retrieve_tool_memory"),
+                json={
+                    "workspace_id": self.workspace_id,
+                    "tool_names": self._join_tool_names(tool_names),
+                },
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+
+    async def collect_memory_async(
+        self,
+        tool_names: List[str],
+    ) -> Dict[str, str]:
+        """Summarize then retrieve guidance for tools.
+
+        Returns:
+            Mapping from tool name to memory ``content`` string.
+        """
+        if not tool_names:
+            return {}
+
+        names = self._join_tool_names(tool_names)
+        try:
+            summary = await self.summary_tool_memory_async(names)
+            if not summary.get("success"):
+                logger.warning("summary_tool_memory failed for %s", names)
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("summary_tool_memory error for %s: %s", names, exc)
+
+        try:
+            retrieved = await self.retrieve_tool_memory_async(names)
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("retrieve_tool_memory error for %s: %s", names, exc)
+            return {}
+
+        if not retrieved.get("success"):
+            logger.warning("retrieve_tool_memory failed for %s", names)
+            return {}
+
+        return self._content_by_tool(retrieved)
+
+    def add_tool_call_result(
+        self,
+        tool_call_results: List[Dict[str, Any]],
+    ) -> Dict[str, Any]:
+        """Sync wrapper for ``add_tool_call_result``."""
+        with httpx.Client() as client:
+            response = client.post(
+                self._url("add_tool_call_result"),
+                json={
+                    "workspace_id": self.workspace_id,
+                    "tool_call_results": tool_call_results,
+                },
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+
+    def summary_tool_memory(self, tool_names: List[str] | str) -> Dict[str, Any]:
+        """Sync wrapper for ``summary_tool_memory``."""
+        with httpx.Client() as client:
+            response = client.post(
+                self._url("summary_tool_memory"),
+                json={
+                    "workspace_id": self.workspace_id,
+                    "tool_names": self._join_tool_names(tool_names),
+                },
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+
+    def retrieve_tool_memory(self, tool_names: List[str] | str) -> Dict[str, Any]:
+        """Sync wrapper for ``retrieve_tool_memory``."""
+        with httpx.Client() as client:
+            response = client.post(
+                self._url("retrieve_tool_memory"),
+                json={
+                    "workspace_id": self.workspace_id,
+                    "tool_names": self._join_tool_names(tool_names),
+                },
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+
+    def collect_memory(self, tool_names: List[str]) -> Dict[str, str]:
+        """Sync wrapper for summarize + retrieve.
+
+        Prefer ``collect_memory_async`` inside an existing event loop.
+        """
+        if not tool_names:
+            return {}
+
+        names = self._join_tool_names(tool_names)
+        try:
+            summary = self.summary_tool_memory(names)
+            if not summary.get("success"):
+                logger.warning("summary_tool_memory failed for %s", names)
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("summary_tool_memory error for %s: %s", names, exc)
+
+        try:
+            retrieved = self.retrieve_tool_memory(names)
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("retrieve_tool_memory error for %s: %s", names, exc)
+            return {}
+
+        if not retrieved.get("success"):
+            logger.warning("retrieve_tool_memory failed for %s", names)
+            return {}
+
+        return self._content_by_tool(retrieved)
+
+    def get_memory_content(
+        self,
+        tool_names: List[str] | str,
+    ) -> Optional[str]:
+        """Retrieve and join memory contents for the given tools."""
+        payload = self.retrieve_tool_memory(tool_names)
+        if not payload.get("success"):
+            return None
+        contents = [content for content in self._content_by_tool(payload).values() if content]
+        return "\n\n".join(contents) if contents else None

benchmark/toolmemory/tool_memory_flows.yaml

@@ -0,0 +1,45 @@
+# Tool Memory flow / op config excerpt used by ExpG.
+# Full runnable code: https://github.com/WangCan1178/ExpG
+
+flow:
+  retrieve_tool_memory:
+    flow_content: retrieve_tool_memory_op
+    description: "Retrieves tool memories from the vector database based on tool names to provide tool usage patterns and best practices"
+    input_schema:
+      tool_names:
+        type: string
+        description: "Comma-separated tool names (e.g., 'tool_name1,tool_name2')"
+        required: true
+
+  add_tool_call_result:
+    flow_content: parse_tool_call_result_op >> update_vector_store_op
+    description: "Evaluates and adds tool call results to the tool memory database, creating new memory or updating existing memory for the specified tool"
+    input_schema:
+      tool_call_results:
+        type: array
+        description: "List of tool call result objects, each containing: tool_name, input, output, success, time_cost, token_cost, create_time"
+        required: true
+
+  summary_tool_memory:
+    flow_content: summary_tool_memory_op >> update_vector_store_op
+    description: "Analyzes tool call history and generates comprehensive usage patterns, best practices, and recommendations for the specified tools"
+    input_schema:
+      tool_names:
+        type: string
+        description: "Comma-separated tool names to summarize (e.g., 'tool_name1,tool_name2')"
+        required: true
+
+op:
+  parse_tool_call_result_op:
+    backend: parse_tool_call_result_op
+    llm: default
+    params:
+      max_history_tool_call_cnt: 100
+      evaluation_sleep_interval: 1.0
+
+  summary_tool_memory_op:
+    backend: summary_tool_memory_op
+    llm: default
+    params:
+      data_from: '2025-09-10 10:56:58'
+      summary_sleep_interval: 1.0

cookbook/appworld/appworld_react_agent.py

@@ -1,330 +0,0 @@
-# flake8: noqa: E402, E501
-import os
-from typing import List, Any
-
-from tqdm import tqdm
-
-os.environ["APPWORLD_ROOT"] = "."
-from dotenv import load_dotenv
-
-load_dotenv("../../.env")
-
-import re
-import time
-import json
-import ray
-import requests
-import datetime
-
-from appworld import AppWorld, load_task_ids
-from jinja2 import Template
-from loguru import logger
-from openai import OpenAI
-
-from prompt import NEW_PROMPT_TEMPLATE
-
-
-@ray.remote
-class AppworldReactAgent:
-    """A minimal ReAct Agent for AppWorld tasks."""
-
-    def __init__(
-        self,
-        index: int,
-        task_ids: List[str],
-        experiment_name: str,
-        model_name: str = "qwen3-8b",
-        temperature: float = 0.9,
-        max_interactions: int = 30,
-        max_response_size: int = 129024,
-        num_trials: int = 1,
-        use_memory: bool = False,
-        use_memory_addition: bool = False,
-        use_memory_deletion: bool = False,
-        delete_freq: int = 10,
-        freq_threshold: int = 5,
-        utility_threshold: float = 0.5,
-        memory_base_url: str = "http://0.0.0.0:8002/",
-        memory_workspace_id: str = "appworld_v1",
-    ):
-
-        self.index: int = index
-        self.task_ids: List[str] = task_ids
-        self.experiment_name: str = experiment_name
-        self.model_name: str = model_name
-        self.temperature: float = temperature
-        self.max_interactions: int = max_interactions
-        self.max_response_size: int = max_response_size
-        self.num_trials: int = num_trials
-        self.use_memory: bool = use_memory
-        self.use_memory_addition: bool = use_memory_addition if use_memory else False
-        self.use_memory_deletion: bool = use_memory_deletion if use_memory else False
-        self.delete_freq: int = delete_freq
-        self.freq_threshold: int = freq_threshold
-        self.utility_threshold: float = utility_threshold
-        self.memory_base_url: str = memory_base_url
-        self.memory_workspace_id: str = memory_workspace_id
-
-        self.llm_client = OpenAI()
-
-        self.history: List[List[List[dict]]] = [[] for _ in range(num_trials)]
-        self.retrieved_memory_list: List[List[List[Any]]] = [[] for _ in range(num_trials)]
-
-        for run_id in range(num_trials):
-            for _ in range(len(task_ids)):
-                self.retrieved_memory_list[run_id].append([])
-                self.history[run_id].append([])
-
-    def call_llm(self, messages: list) -> str:
-        for i in range(100):
-            try:
-                response = self.llm_client.chat.completions.create(
-                    model=self.model_name,
-                    messages=messages,
-                    temperature=self.temperature,
-                    extra_body={"enable_thinking": False},
-                    seed=0,
-                )
-
-                return response.choices[0].message.content
-
-            except Exception as e:
-                logger.exception(f"encounter error with {e.args}")
-                time.sleep(1 + i * 10)
-
-        return "call llm error"
-
-    def prompt_messages(self, run_id, task_index, previous_memories: None, world: AppWorld):
-        app_descriptions = json.dumps(
-            [
-                {"name": k, "description": v}
-                for (k, v) in world.task.app_descriptions.items()
-            ],
-            indent=1,
-        )
-        dictionary = {"supervisor": world.task.supervisor, "app_descriptions": app_descriptions}
-        sys_prompt = Template(NEW_PROMPT_TEMPLATE.lstrip()).render(dictionary)
-        query = world.task.instruction
-        if self.use_memory:
-            if len(previous_memories) == 0:
-                response = self.get_memory(world.task.instruction)
-                if response and "memory_list" in response["metadata"]:
-                    self.retrieved_memory_list[run_id][task_index] = response["metadata"]["memory_list"]
-                    task_memory = response["answer"]
-                    logger.info(f"loaded task_memory: {task_memory}")
-                    query = "Task:\n" + query + "\n\nSome Related Experience to help you to complete the task:\n" + re.sub(r'(?i)\bMemory\s*(\d+)\s*[:]', r'Experience \1:', task_memory)
-            else:
-                formatted_memories = []
-                for i, memory in enumerate(previous_memories, 1):
-                    condition = memory["when_to_use"]
-                    memory_content = memory["content"]
-                    memory_text = f"Experience {i}:\n When to use: {condition}\n Content: {memory_content}\n"
-                    formatted_memories.append(memory_text)
-                query = "Task:\n" + query + "\n\nSome Related Experience to help you to complete the task:\n" + "\n".join(formatted_memories)
-        messages = [
-            {"role": "system", "content": sys_prompt},
-            {"role": "user", "content": query}
-        ]
-        self.history[run_id][task_index] = messages
-
-
-    @staticmethod
-    def get_reward(world) -> float:
-        tracker = world.evaluate()
-        num_passes = len(tracker.passes)
-        num_failures = len(tracker.failures)
-        return num_passes / (num_passes + num_failures)
-
-    def extract_code_and_fix_content(
-        self, text: str, ignore_multiple_calls=True
-    ) -> tuple[str, str]:
-        full_code_regex = r"```python\n(.*?)```"
-        partial_code_regex = r".*```python\n(.*)"
-
-        original_text = text
-        output_code = ""
-        match_end = 0
-        # Handle multiple calls
-        for re_match in re.finditer(full_code_regex, original_text, flags=re.DOTALL):
-            code = re_match.group(1).strip()
-            if ignore_multiple_calls:
-                text = original_text[: re_match.end()]
-                return code, text
-            output_code += code + "\n"
-            match_end = re_match.end()
-        # check for partial code match at end (no terminating ```)  following the last match
-        partial_match = re.match(
-            partial_code_regex, original_text[match_end:], flags=re.DOTALL
-        )
-        if partial_match:
-            output_code += partial_match.group(1).strip()
-            # terminated due to stop condition. Add stop condition to output.
-            if not text.endswith("\n"):
-                text = text + "\n"
-            text = text + "```"
-        if len(output_code) == 0:
-            return text, text
-        else:
-            return output_code, text
-
-    def execute(self):
-        result = []
-        counter = 0
-        for task_index, task_id in enumerate(tqdm(self.task_ids, desc=f"ray_index={self.index}")):
-            t_result = None
-            previous_memories = []
-            # Run each task num_trials times
-            for run_id in range(self.num_trials):
-                start_time = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-                with AppWorld(task_id=task_id, experiment_name=f"{self.experiment_name}_run_{run_id}") as world:
-                    before_score = self.get_reward(world)
-                    for i in range(self.max_interactions):
-                        if i == 0:
-                            self.prompt_messages(run_id=run_id, task_index=task_index, previous_memories=previous_memories, world=world)
-                        code_msg = self.call_llm(self.history[run_id][task_index])
-                        code, text = self.extract_code_and_fix_content(code_msg)
-                        self.history[run_id][task_index].append({"role": "assistant", "content": code})
-
-                        output = world.execute(code)
-                        # if len(output) > self.max_response_size:
-                        #     # logger.warning(f"output exceed max size={len(output)}")
-                        #     output = output[: self.max_response_size]
-                        self.history[run_id][task_index].append({"role": "user", "content": "Output:\n```\n" + output + "```\n\n"})
-
-                        if world.task_completed():
-                            break
-
-                    after_score = self.get_reward(world)
-                    uplift_score = after_score - before_score
-
-                    if self.use_memory:
-                        if self.use_memory_addition:
-                            new_traj_list = [self.get_traj_from_task_history(task_id, self.history[run_id][task_index], after_score)]
-                            previous_memories = self.add_memory(new_traj_list)
-                            if after_score != 1:
-                                self.delete_memory_by_ids([mem["memory_id"] for mem in previous_memories])
-
-                        # update the freq & utility attributes of retrieved memories
-                        update_utility: bool = after_score == 1
-                        self.update_memory_information(self.retrieved_memory_list[run_id][task_index], update_utility)
-
-                    counter += 1
-                    if self.use_memory_deletion: # and counter % self.delete_freq == 0:
-                        self.delete_memory()
-
-                    t_result = {
-                        "task_id": world.task_id,
-                        "run_id": run_id,
-                        "experiment_name": self.experiment_name,
-                        "task_completed": world.task_completed(),
-                        "before_score": before_score,
-                        "after_score": after_score,
-                        "uplift_score": uplift_score,
-                        "task_history": self.history[run_id][task_index],
-                        "task_start_time": start_time,
-                    }
-                    if after_score == 1:
-                        break
-            result.append(t_result)
-
-        return result
-
-    def handle_api_response(self, response: requests.Response):
-        """Handle API response with proper error checking"""
-        if response.status_code != 200:
-            print(f"Error: {response.status_code}")
-            print(response.text)
-            return None
-
-        return response.json()
-
-    def get_memory(self, query: str):
-        """Retrieve relevant task memories based on a query"""
-        response = requests.post(
-            url=f"{self.memory_base_url}retrieve_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "query": query,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return None
-
-        logger.info(f"query: {query}, response: {result}")
-        return result
-
-    def get_traj_from_task_history(self, task_id: str, task_history: list, reward: float):
-        pattern = r"\n\nSome Related Experience to help you to complete the task:.*"
-        task_history[1]["content"] = re.sub(pattern, "", task_history[1]["content"], flags=re.DOTALL)
-        return {
-            "task_id": task_id,
-            "messages": task_history,
-            "score": reward
-        }
-
-    def add_memory(self, trajectories):
-        """Generate a summary of conversation messages and create task memories"""
-
-        response = requests.post(
-            url=f"{self.memory_base_url}summary_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "trajectories": trajectories,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return []
-
-        # Extract memory list from response
-        memory_list = result.get("metadata", {}).get("memory_list", [])
-        print(f"Task memory list created: {len(memory_list)} memories")
-        return memory_list
-
-    def delete_memory_by_ids(self, memory_ids):
-        response = requests.post(
-            url=f"{self.memory_base_url}vector_store",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "action": "delete_ids",
-                "memory_ids": memory_ids
-            }
-        )
-        response.raise_for_status()
-
-    def update_memory_information(self, memory_list, update_utility: bool = False):
-        response = requests.post(
-            url=f"{self.memory_base_url}record_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "memory_dicts": memory_list,
-                "update_utility": update_utility,
-            },
-        )
-        response.raise_for_status()
-        logger.info(response.json())
-
-    def delete_memory(self):
-        response = requests.post(
-            url=f"{self.memory_base_url}delete_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "freq_threshold": self.freq_threshold,
-                "utility_threshold": self.utility_threshold,
-            },
-        )
-        response.raise_for_status()
-
-def main():
-    dataset_name = "train"
-    task_ids = load_task_ids(dataset_name)
-    agent = AppworldReactAgent(index=0, task_ids=task_ids[0:1], experiment_name=dataset_name, num_trials=1)
-    result = agent.execute()
-    logger.info(f"result={json.dumps(result)}")
-
-
-if __name__ == "__main__":
-    main()

cookbook/appworld/prompt.py

@@ -1,659 +0,0 @@
-# flake8: noqa: E402, E501
-# This is a basic prompt template containing all the necessary onboarding information to solve AppWorld tasks. It explains the role of the agent and the supervisor, how to explore the API documentation, how to operate the interactive coding environment and call APIs via a simple task, and provides key instructions and disclaimers.
-
-# You can adapt it as needed by your agent. You can also choose to bypass API docs app and build your own API retrieval, e.g., for FullCodeRefl, IPFunCall, etc, we asked an LLM to predict relevant APIs separately and put its documentation directly in the prompt.
-PROMPT_TEMPLATE = """
-USER:
-I am your supervisor and you are a super intelligent AI Assistant whose job is to achieve my day-to-day tasks completely autonomously.
-
-To do this, you will need to interact with app/s (e.g., spotify, venmo, etc) using their associated APIs on my behalf. For this you will undertake a *multi-step conversation* using a python REPL environment. That is, you will write the python code and the environment will execute it and show you the result, based on which, you will write python code for the next step and so on, until you've achieved the goal. This environment will let you interact with app/s using their associated APIs on my behalf.
-
-Here are three key APIs that you need to know to get more information
-
-# To get a list of apps that are available to you.
-print(apis.api_docs.show_app_descriptions())
-
-# To get the list of apis under any app listed above, e.g. supervisor
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-# To get the specification of a particular api, e.g. supervisor app's show_account_passwords
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-Each code execution will produce an output that you can use in subsequent calls. Using these APIs, you can now generate code, that the environment will execute, to solve the task.
-
-For example, consider the task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-What is the password for my Spotify account?
-
-ASSISTANT:
-# Okay. Lets first find which apps are available to get the password by looking at the app descriptions.
-print(apis.api_docs.show_app_descriptions())
-
-USER:
-[
-  {
-    "name": "api_docs",
-    "description": "An app to search and explore API documentation."
-  },
-  {
-    "name": "supervisor",
-    "description": "An app to access supervisor's personal information, account credentials, addresses, payment cards, and manage the assigned task."
-  },
-  ...
-  {
-    "name": "spotify",
-    "description": "A music streaming app to stream songs and manage song, album and playlist libraries."
-  },
-  {
-    "name": "venmo",
-    "description": "A social payment app to send, receive and request money to and from others."
-  },
-  ...
-]
-
-
-ASSISTANT:
-# Looks like the supervisor app could help me with that. Lets see what apis are available under this app.
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-
-USER:
-[
-  ...
-  "show_account_passwords : Show your supervisor's account passwords."
-  ...
-]
-
-
-ASSISTANT:
-# I can use `show_account_passwords` to get the passwords. Let me see its detailed specification to understand its arguments and output structure.
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-USER:
-{
-  'app_name': 'supervisor',
-  'api_name': 'show_account_passwords',
-  'path': '/account_passwords',
-  'method': 'GET',
-  'description': "Show your supervisor's app account passwords.",
-  'parameters': [],
-  'response_schemas': {
-    'success': [{'account_name': 'string', 'password': 'string'}],
-    'failure': {'message': 'string'}
-  }
-}
-
-ASSISTANT:
-# Okay, it requires no arguments. So I can just call it directly.
-print(apis.supervisor.show_account_passwords())
-
-USER:
-[
-  {
-    "account_name": "spotify",
-    "password": "dummy_spotify_pass"
-  },
-  {
-    "account_name": "file_system",
-    "password": "dummy_fs_pass"
-  },
-  ...
-]
-
-
-ASSISTANT:
-# So the Spotify password is an entry in the `passwords` list with the account_name=spotify.
-spotify_password = [account_password["account_name"] == "spotify" for account_password in passwords][0]["password"]
-print(spotify_password)
-
-
-USER:
-dummy_spotify_pass
-
-ASSISTANT:
-# When the task is completed, I need to call apis.supervisor.complete_task(). If there is an answer, I need to pass it as an argument `answer`. I will pass the spotify_password as an answer.
-apis.supervisor.complete_task(answer=spotify_password)
-
-
-USER:
-Marked the active task complete.
-
-
-----------------------------------------------
-
-USER:
-**Key instructions and disclaimers**:
-
-1. The email addresses, access tokens and variables (e.g. spotify_password) in the example above were only for demonstration. Obtain the correct information by calling relevant APIs yourself.
-2. Only generate valid code blocks, i.e., do not put them in ```...``` or add any extra formatting. Any thoughts should be put as code comments.
-3. You can use the variables from the previous code blocks in the subsequent code blocks.
-4. Write small chunks of code and only one chunk of code in every step. Make sure everything is working correctly before making any irreversible change.
-5. The provided Python environment has access to its standard library. But modules and functions that have a risk of affecting the underlying OS, file system or process are disabled. You will get an error if do call them.
-6. Any reference to a file system in the task instructions means the file system *app*, operable via given APIs, and not the actual file system the code is running on. So do not write code making calls to os-level modules and functions.
-7. To interact with apps, only use the provided APIs, and not the corresponding Python packages. E.g., do NOT use `spotipy` for Spotify. Remember, the environment only has the standard library.
-8. The provided API documentation has both the input arguments and the output JSON schemas. All calls to APIs and parsing its outputs must be as per this documentation.
-9. For APIs that return results in "pages", make sure to consider all pages.
-10. To obtain current date or time, use Python functions like `datetime.now()` or obtain it from the phone app. Do not rely on your existing knowledge of what the current date or time is.
-11. For all temporal requests, use proper time boundaries, e.g., if I ask for something that happened yesterday, make sure to consider the time between 00:00:00 and 23:59:59. All requests are concerning a single, default (no) time zone.
-12. Any reference to my friends, family or any other person or relation refers to the people in my phone's contacts list.
-13. All my personal information, and information about my app account credentials, physical addresses and owned payment cards are stored in the "supervisor" app. You can access them via the APIs provided by the supervisor app.
-14. Once you have completed the task, call `apis.supervisor.complete_task()`. If the task asks for some information, return it as the answer argument, i.e. call `apis.supervisor.complete_task(answer=<answer>)`. For tasks that do not require an answer, just skip the answer argument or pass it as None.
-15. The answers, when given, should be just entity or number, not full sentences, e.g., `answer=10` for "How many songs are in the Spotify queue?". When an answer is a number, it should be in numbers, not in words, e.g., "10" and not "ten".
-16. You can also pass `status="fail"` in the complete_task API if you are sure you cannot solve it and want to exit.
-17. You must make all decisions completely autonomously and not ask for any clarifications or confirmations from me or anyone else.
-
-USER:
-Using these APIs, now generate code to solve the actual task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-{{ instruction }}
-"""
-
-PROMPT_TEMPLATE_WITH_EXPERIENCE = """
-USER:
-I am your supervisor and you are a super intelligent AI Assistant whose job is to achieve my day-to-day tasks completely autonomously.
-
-To do this, you will need to interact with app/s (e.g., spotify, venmo, etc) using their associated APIs on my behalf. For this you will undertake a *multi-step conversation* using a python REPL environment. That is, you will write the python code and the environment will execute it and show you the result, based on which, you will write python code for the next step and so on, until you've achieved the goal. This environment will let you interact with app/s using their associated APIs on my behalf.
-
-Here are three key APIs that you need to know to get more information
-
-# To get a list of apps that are available to you.
-print(apis.api_docs.show_app_descriptions())
-
-# To get the list of apis under any app listed above, e.g. supervisor
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-# To get the specification of a particular api, e.g. supervisor app's show_account_passwords
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-Each code execution will produce an output that you can use in subsequent calls. Using these APIs, you can now generate code, that the environment will execute, to solve the task.
-
-For example, consider the task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-What is the password for my Spotify account?
-
-ASSISTANT:
-# Okay. Lets first find which apps are available to get the password by looking at the app descriptions.
-print(apis.api_docs.show_app_descriptions())
-
-USER:
-[
-  {
-    "name": "api_docs",
-    "description": "An app to search and explore API documentation."
-  },
-  {
-    "name": "supervisor",
-    "description": "An app to access supervisor's personal information, account credentials, addresses, payment cards, and manage the assigned task."
-  },
-  ...
-  {
-    "name": "spotify",
-    "description": "A music streaming app to stream songs and manage song, album and playlist libraries."
-  },
-  {
-    "name": "venmo",
-    "description": "A social payment app to send, receive and request money to and from others."
-  },
-  ...
-]
-
-
-ASSISTANT:
-# Looks like the supervisor app could help me with that. Lets see what apis are available under this app.
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-
-
-USER:
-[
-  ...
-  "show_account_passwords : Show your supervisor's account passwords."
-  ...
-]
-
-
-ASSISTANT:
-# I can use `show_account_passwords` to get the passwords. Let me see its detailed specification to understand its arguments and output structure.
-print(apis.api_docs.show_api_doc(app_name='supervisor', api_name='show_account_passwords'))
-
-USER:
-{
-  'app_name': 'supervisor',
-  'api_name': 'show_account_passwords',
-  'path': '/account_passwords',
-  'method': 'GET',
-  'description': "Show your supervisor's app account passwords.",
-  'parameters': [],
-  'response_schemas': {
-    'success': [{'account_name': 'string', 'password': 'string'}],
-    'failure': {'message': 'string'}
-  }
-}
-
-ASSISTANT:
-# Okay, it requires no arguments. So I can just call it directly.
-print(apis.supervisor.show_account_passwords())
-
-USER:
-[
-  {
-    "account_name": "spotify",
-    "password": "dummy_spotify_pass"
-  },
-  {
-    "account_name": "file_system",
-    "password": "dummy_fs_pass"
-  },
-  ...
-]
-
-
-ASSISTANT:
-# So the Spotify password is an entry in the `passwords` list with the account_name=spotify.
-spotify_password = [account_password["account_name"] == "spotify" for account_password in passwords][0]["password"]
-print(spotify_password)
-
-
-USER:
-dummy_spotify_pass
-
-ASSISTANT:
-# When the task is completed, I need to call apis.supervisor.complete_task(). If there is an answer, I need to pass it as an argument `answer`. I will pass the spotify_password as an answer.
-apis.supervisor.complete_task(answer=spotify_password)
-
-
-USER:
-Marked the active task complete.
-
-
-----------------------------------------------
-
-USER:
-**Key instructions and disclaimers**:
-
-1. The email addresses, access tokens and variables (e.g. spotify_password) in the example above were only for demonstration. Obtain the correct information by calling relevant APIs yourself.
-2. Only generate valid code blocks, i.e., do not put them in ```...``` or add any extra formatting. Any thoughts should be put as code comments.
-3. You can use the variables from the previous code blocks in the subsequent code blocks.
-4. Write small chunks of code and only one chunk of code in every step. Make sure everything is working correctly before making any irreversible change.
-5. The provided Python environment has access to its standard library. But modules and functions that have a risk of affecting the underlying OS, file system or process are disabled. You will get an error if do call them.
-6. Any reference to a file system in the task instructions means the file system *app*, operable via given APIs, and not the actual file system the code is running on. So do not write code making calls to os-level modules and functions.
-7. To interact with apps, only use the provided APIs, and not the corresponding Python packages. E.g., do NOT use `spotipy` for Spotify. Remember, the environment only has the standard library.
-8. The provided API documentation has both the input arguments and the output JSON schemas. All calls to APIs and parsing its outputs must be as per this documentation.
-9. For APIs that return results in "pages", make sure to consider all pages.
-10. To obtain current date or time, use Python functions like `datetime.now()` or obtain it from the phone app. Do not rely on your existing knowledge of what the current date or time is.
-11. For all temporal requests, use proper time boundaries, e.g., if I ask for something that happened yesterday, make sure to consider the time between 00:00:00 and 23:59:59. All requests are concerning a single, default (no) time zone.
-12. Any reference to my friends, family or any other person or relation refers to the people in my phone's contacts list.
-13. All my personal information, and information about my app account credentials, physical addresses and owned payment cards are stored in the "supervisor" app. You can access them via the APIs provided by the supervisor app.
-14. Once you have completed the task, call `apis.supervisor.complete_task()`. If the task asks for some information, return it as the answer argument, i.e. call `apis.supervisor.complete_task(answer=<answer>)`. For tasks that do not require an answer, just skip the answer argument or pass it as None.
-15. The answers, when given, should be just entity or number, not full sentences, e.g., `answer=10` for "How many songs are in the Spotify queue?". When an answer is a number, it should be in numbers, not in words, e.g., "10" and not "ten".
-16. You can also pass `status="fail"` in the complete_task API if you are sure you cannot solve it and want to exit.
-17. You must make all decisions completely autonomously and not ask for any clarifications or confirmations from me or anyone else.
-18. Some Related Experience to help you to complete the task:
-{{experience}}
-
-USER:
-Using these APIs, now generate code to solve the actual task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-Task:
-
-{{ instruction }}
-"""
-
-NEW_PROMPT_TEMPLATE = """
-USER:
-I am your supervisor and you are a super intelligent AI Assistant whose job is to achieve my day-to-day tasks completely autonomously.
-
-To do this, you will need to interact with app/s (e.g., spotify, venmo etc) using their associated APIs on my behalf. For this you will undertake a *multi-step conversation* using a python REPL environment. That is, you will write the python code and the environment will execute it and show you the result, based on which, you will write python code for the next step and so on, until you've achieved the goal. This environment will let you interact with app/s using their associated APIs on my behalf.
-
-Here are three key APIs that you need to know to get more information
-
-# To get a list of apps that are available to you.
-
-```python
-print(apis.api_docs.show_app_descriptions())
-```
-
-# To get the list of apis under any app listed above, e.g. spotify
-
-```python
-print(apis.api_docs.show_api_descriptions(app_name='spotify'))
-```
-
-# To get the specification of a particular api, e.g. spotify app's login api
-
-```python
-print(apis.api_docs.show_api_doc(app_name='spotify', api_name='login'))
-```
-
-Each code execution will produce an output that you can use in subsequent calls. Using these APIs, you can now generate code, that I will execute, to solve the task. Let's start with the task
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-Task: How many playlists do I have in Spotify?
-
-ASSISTANT:
-Okay. Lets first find which APIs are available to use in Spotify.
-Code:
-```python
-print(apis.api_docs.show_api_descriptions(app_name='spotify'))
-```
-
-USER:
-Output:
-```
-[
- ...
- "login : Login to your account.",
- "logout : Logout from your account.",
- ...
-]
-```
-
-
-ASSISTANT:
-Okay. Looks like I can use the `login` api. Lets find its specifications.
-
-Code:
-```python
-print(apis.api_docs.show_api_doc(app_name='spotify', api_name='login'))
-```
-
-
-USER:
-Output:
-```
-{
- "app_name": "spotify",
- "api_name": "login",
- "path": "/auth/token",
- "method": "POST",
- "description": "Login to your account.",
- "parameters": [
-  {
-   "name": "username",
-   "type": "string",
-   "required": true,
-   "description": "Your account email.",
-   "default": null,
-   "constraints": []
-  },
-  {
-   "name": "password",
-   "type": "string",
-   "required": true,
-   "description": "Your account password.",
-   "default": null,
-   "constraints": []
-  }
- ],
- "response_schemas": {
-  "success": {
-   "token_type": "string",
-   "access_token": "string"
-  },
-  "failure": {
-   "message": "string"
-  }
- }
-}
-```
-
-
-ASSISTANT:
-I need the supervisor's username and password. Lets see if any app can help me get that.
-
-Code:
-```python
-print(apis.api_docs.show_app_descriptions())
-```
-
-
-USER:
-Output:
-```
-{{ app_descriptions }}
-```
-
-
-ASSISTANT:
-Looks like the supervisor app could help me with that.
-
-Code:
-```python
-print(apis.api_docs.show_api_descriptions(app_name='supervisor'))
-```
-
-
-USER:
-Output:
-```
-[
-  ...
-  "show_account_passwords : Show your supervisor's account passwords."
-  ...
-]
-```
-
-
-ASSISTANT:
-I can use `show_account_passwords` to get the passwords. So I will call that and save it in a variable for future reference.
-Code:
-```python
-passwords=apis.supervisor.show_account_passwords()
-print(passwords)
-```
-
-
-USER:
-Output:
-```
-[
-  {
-    "account_name": "spotify",
-    "password": "dummy_spotify_pass"
-  },
-  {
-    "account_name": "file_system",
-    "password": "dummy_fs_pass"
-  }
-]
-```
-
-
-ASSISTANT:
-So the Spotify password is an entry in the `passwords` list with the account_name=spotify. I can use it to login now with the provided email address: {{ supervisor.email }}.
-
-Code:
- ```python
-spotify_password = [account_password["account_name"] == "spotify" for account_password in passwords][0]["password"]
-login_result = apis.spotify.login(username='{{ supervisor.email }}', password=spotify_password)
-print(login_result)
-```
-
-
-USER:
-Output:
-```
-{
- "token_type": "Bearer",
- "access_token": "fake_access_token"
-}
-```
-
-
-ASSISTANT:
-First, let's check the available APIs for the Spotify app.
-Code:
-```python
-print(apis.api_docs.show_api_descriptions(app_name='spotify'))
-```
-
-
-USER:
-Output:
-```
-[
-...
-{
-  "name": "show_playlist_library",
-  "description": "Get a list of playlists in the user's playlist library."
-},
-{
-  "name": "show_playlist",
-  "description": "Get detailed information about a specific playlist. You can view your own playlists or others' playlists if they are public."
- },
-...
-]
-```
-
-
-ASSISTANT:
-The `show_playlist_library` API seems to be the one we need. Let's check its specifications.
-Code:
-```python
-print(apis.api_docs.show_api_doc(app_name='spotify', api_name='show_playlist_library'))
-```
-
-
-USER:
-Output:
-```
-{
- "app_name": "spotify",
- "api_name": "show_playlist_library",
- "path": "/private_playlists",
- "method": "GET",
- "description": "Get a list of playlists in the user's playlist library.",
- "parameters": [
-  {
-   "name": "access_token",
-   "type": "string",
-   "required": true,
-   "description": "Access token obtained from spotify app login.",
-   "default": null,
-   "constraints": []
-  },
-  {
-   "name": "page_index",
-   "type": "integer",
-   "required": false,
-   "description": "The index of the page to retrieve.",
-   "default": 0,
-   "constraints": [
-    "value >= 0.0"
-   ]
-  },
-  {
-   "name": "page_limit",
-   "type": "integer",
-   "required": false,
-   "description": "The maximum number of results to return per page.",
-   "default": 5,
-   "constraints": [
-    "value >= 1.0, <= 20.0"
-   ]
-  },
-  {
-   "name": "is_public",
-   "type": "boolean",
-   "required": false,
-   "description": "Whether to show public playlists or private playlists.",
-   "default": null,
-   "constraints": []
-  }
- ],
- "response_schema": [
-  {
-   "title": "string",
-   "created_at": "2019-01-01T00:00:00",
-   "is_public": true,
-   "rating": 0.0,
-   "like_count": 1,
-   "owner_email": "user@example.com",
-   "playlist_id": 1,
-   "song_ids": [
-    1
-   ]
-  }
- ]
-}
-```
-
-
-ASSISTANT:
-I need to page through all the playlists to get the list of playlists and save it in `playlists`.
-Code:
-```python
-page_index = 0
-playlists = []
-while page_index < 10:
-  playlist_page = apis.spotify.show_playlist_library(access_token=spotify_access_token, page_index=page_index)
-  if playlist_page:
-    playlists.extend(playlist_page)
-    page_index += 1
-  else:
-    break
-num_playlists = len(playlists)
-print(num_playlists)
-
-```
-
-USER:
-Output:
-```
-23
-```
-
-
-ASSISTANT:
-Now that the task is completed, I can call apis.supervisor.complete_task(). Since this task has an answer to be returned, I will pass that as an argument.
-
-Code:
-```python
-apis.supervisor.complete_task(answer=num_playlists)
-```
-
-
-USER:
-Output:
-Marked the active task complete.
-
-
-----------------------------------------------
-
-USER:
-**Key instructions**:
-(1) Make sure to end code blocks with ``` followed by a newline(\n).
-
-(2) Remember you can use the variables in your code in subsequent code blocks.
-
-(3) Remember that the email addresses, access tokens and variables (e.g. spotify_password) in the example above are not valid anymore.
-
-(4) You can use the "supervisor" app to get information about my accounts and use the "phone" app to get information about friends and family.
-
-(5) Always look at API specifications (using apis.api_docs.show_api_doc) before calling an API.
-
-(6) Write small chunks of code and only one chunk of code in every step. Make sure everything is working correctly before making any irreversible change.
-
-(7) Many APIs return items in "pages". Make sure to run through all the pages by looping over `page_index`.
-
-(8) Once you have completed the task, make sure to call apis.supervisor.complete_task(). If the task asked for some information, return it as the answer argument, i.e. call apis.supervisor.complete_task(answer=<answer>). Many tasks do not require an answer, so in those cases, just call apis.supervisor.complete_task() i.e. do not pass any argument.
-
-USER:
-Using these APIs, now generate code to solve the actual task:
-
-My name is: {{ supervisor.first_name }} {{ supervisor.last_name }}. My personal email is {{ supervisor.email }} and phone number is {{ supervisor.phone_number }}.
-
-"""

cookbook/appworld/requirements.txt

@@ -1,5 +0,0 @@
-jinja2
-loguru
-openai
-ray
-pandas

cookbook/appworld/run_appworld.py

@@ -1,239 +0,0 @@
-# flake8: noqa: E402
-import os
-import time
-
-import ray
-import requests
-from ray import logger
-
-os.environ["APPWORLD_ROOT"] = "."
-from dotenv import load_dotenv
-
-load_dotenv("../../.env")
-
-import json
-from pathlib import Path
-
-from appworld import load_task_ids
-
-from appworld_react_agent import AppworldReactAgent
-
-
-def handle_api_response(response: requests.Response):
-    """Handle API response with proper error checking"""
-    if response.status_code != 200:
-        print(f"Error: {response.status_code}")
-        print(response.text)
-        return None
-
-    return response.json()
-
-
-def delete_workspace(workspace_id: str, api_url: str = "http://0.0.0.0:8002/"):
-    """Delete the current workspace from the vector store"""
-    response = requests.post(
-        url=f"{api_url}vector_store",
-        json={
-            "workspace_id": workspace_id,
-            "action": "delete",
-        },
-    )
-
-    result = handle_api_response(response)
-    if result:
-        print(f"Workspace '{workspace_id}' deleted successfully")
-
-
-def dump_memory(workspace_id: str, path: str = "./", api_url: str = "http://0.0.0.0:8002/"):
-    """Dump the vector store memories to disk"""
-    response = requests.post(
-        url=f"{api_url}vector_store",
-        json={
-            "workspace_id": workspace_id,
-            "action": "dump",
-            "path": path,
-        },
-    )
-
-    result = handle_api_response(response)
-    if result:
-        print(f"Memory dumped to {path}")
-
-
-def load_memory(workspace_id: str, path: str = "docs/library", api_url: str = "http://0.0.0.0:8002/"):
-    """Load memories from disk into the vector store"""
-    response = requests.post(
-        url=f"{api_url}vector_store",
-        json={
-            "workspace_id": workspace_id,
-            "action": "load",
-            "path": path,
-        },
-    )
-
-    result = handle_api_response(response)
-    if result:
-        print(f"Memory loaded from {path}")
-
-
-def run_agent(
-    model_name: str,
-    dataset_name: str,
-    experiment_suffix: str,
-    max_workers: int,
-    num_trials: int = 1,
-    use_memory: bool = False,
-    use_memory_addition: bool = False,
-    use_memory_deletion: bool = False,
-    delete_freq: int = 10,
-    freq_threshold: int = 5,
-    utility_threshold: float = 0.5,
-    workspace_id: str = "appworld_v1",
-    api_url: str = "http://0.0.0.0:8002/",
-    batch_size: int = 4
-):
-    experiment_name = dataset_name + "_" + experiment_suffix
-    path: Path = Path(f"./exp_result/{model_name}")
-    path.mkdir(parents=True, exist_ok=True)
-
-    task_ids = load_task_ids(dataset_name)
-    result: list = []
-
-    def dump_file():
-        with open(path / f"{experiment_name}.jsonl", "a") as f:
-            for x in result:
-                f.write(json.dumps(x) + "\n")
-
-    if max_workers > 1:
-        # Process tasks in batches
-        total_tasks = len(task_ids)
-        num_batches = (total_tasks + batch_size - 1) // batch_size  # Ceiling division
-
-        logger.info(f"Total tasks: {total_tasks}, Batch size: {batch_size}, Number of batches: {num_batches}")
-
-        for batch_idx in range(num_batches):
-            # Initialize Ray for this batch
-            start_idx = batch_idx * batch_size
-            end_idx = min(start_idx + batch_size, total_tasks)
-            batch_task_ids = task_ids[start_idx:end_idx]
-
-            logger.info(f"Starting batch {batch_idx + 1}/{num_batches} with {len(batch_task_ids)} tasks")
-
-            # Initialize Ray with the number of CPUs needed for this batch
-            ray.init(num_cpus=len(batch_task_ids))
-
-            future_list: list = []
-            for i, task_id in enumerate(batch_task_ids):
-                actor = AppworldReactAgent.remote(
-                    index=start_idx+i,
-                    model_name=model_name,
-                    task_ids=[task_id],
-                    experiment_name=experiment_name,
-                    num_trials=num_trials,
-                    use_memory=use_memory,
-                    use_memory_addition=use_memory_addition,
-                    use_memory_deletion=use_memory_deletion,
-                    delete_freq=delete_freq,
-                    freq_threshold=freq_threshold,
-                    utility_threshold=utility_threshold,
-                    memory_workspace_id=workspace_id,
-                    memory_base_url=api_url,
-                )
-                future = actor.execute.remote()
-                future_list.append(future)
-                time.sleep(1)
-
-            logger.info(f"Batch {batch_idx + 1} submit complete, waiting for results...")
-
-            # Collect results from this batch
-            for i, (task_id, future) in enumerate(zip(batch_task_ids, future_list)):
-                try:
-                    t_result = ray.get(future)
-                    if t_result:
-                        if isinstance(t_result, list):
-                            result.extend(t_result)
-                        else:
-                            result.append(t_result)
-                except Exception as e:
-                    logger.exception(f"run ray error with task_id={task_id}")
-
-                logger.info(f"Batch {batch_idx + 1}: task {i + 1}/{len(batch_task_ids)} complete")
-
-            # Shutdown Ray to free resources before next batch
-            ray.shutdown()
-            logger.info(f"Batch {batch_idx + 1}/{num_batches} complete, Ray resources released")
-
-            # Optional: small delay between batches
-            if batch_idx < num_batches - 1:
-                time.sleep(2)
-
-        dump_file()
-
-    else:
-        for index, task_id in enumerate(task_ids):
-            agent = AppworldReactAgent(
-                index=index,
-                model_name=model_name,
-                task_ids=[task_id],
-                experiment_name=experiment_name,
-                num_trials=num_trials,
-                use_memory=use_memory,
-                use_memory_addition=use_memory_addition,
-                use_memory_deletion=use_memory_deletion,
-                delete_freq=delete_freq,
-                freq_threshold=freq_threshold,
-                utility_threshold=utility_threshold,
-                workspace_id=workspace_id,
-                api_url=api_url,
-            )
-            task_results = agent.execute()
-            if isinstance(task_results, list):
-                result.extend(task_results)
-            else:
-                result.append(task_results)
-        dump_file()
-
-def main():
-    max_workers = 8
-    num_runs = 1 # Number of runs
-    batch_size = 8  # Number of concurrent tasks per batch
-
-    num_trials = 2
-    model_name = "qwen3-8b"
-    use_memory = True
-    use_memory_addition = True
-    use_memory_deletion = True
-    workspace_id = "appworld"
-    api_url = "http://0.0.0.0:8002/"
-
-
-    # Clean up workspace before starting
-    logger.info("Deleting workspace...")
-    delete_workspace(workspace_id=workspace_id, api_url=api_url)
-    time.sleep(5)
-
-    # First run to build task memories
-    logger.info("Start load experiments to build task memories")
-    load_memory(workspace_id=workspace_id, api_url=api_url)
-
-
-    for i in range(num_runs):
-        run_agent(
-            model_name=model_name,
-            dataset_name="test_normal",
-            experiment_suffix=f"with-memory",
-            max_workers=max_workers,
-            num_trials=num_trials,
-            use_memory=use_memory,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=5,
-            freq_threshold=5,
-            utility_threshold=0.5,
-            workspace_id=workspace_id,
-            api_url=api_url,
-            batch_size=batch_size
-        )
-
-if __name__ == "__main__":
-    main()

cookbook/appworld/run_exp_statistic.py

@@ -1,160 +0,0 @@
-import json
-from collections import defaultdict
-from pathlib import Path
-
-import pandas as pd
-from loguru import logger
-
-
-def calculate_best_at_k(scores: list, k: int) -> float:
-    """
-    Calculate best@k
-    Divide scores into groups of size k, take the maximum value in each group,
-    then average these maximum values
-
-    Args:
-        scores: List of after_score values for all runs of a task
-        k: Group size
-
-    Returns:
-        best@k value
-    """
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        group_maxs.append(max(group))
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def calculate_pass_at_k(scores: list, k: int) -> float:
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        is_pass = 1.0 if max(group) >= 1.0 else 0.0
-        group_maxs.append(is_pass)
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def get_possible_k_values(total_runs: int) -> list:
-    """
-    Get all possible k values (factors of total_runs)
-
-    Args:
-        total_runs: Total number of runs
-
-    Returns:
-        List of k values in descending order
-    """
-    k_values = []
-    for k in range(1, total_runs + 1):
-        if total_runs % k == 0:
-            k_values.append(k)
-    return sorted(k_values, reverse=True)  # Sort from large to small
-
-
-def run_exp_statistic():
-    path: Path = Path("./exp_result")
-
-    # Store results for all experiments
-    all_results = {}
-
-    for file in [f for f in path.glob("*.jsonl") if not f.stem[-1].isdigit()]:
-        # Group results by task_id
-        task_results = defaultdict(list)
-
-        with open(file, "r") as f:
-            for line in f:
-                if not line.strip():
-                    continue
-                data = json.loads(line)
-
-                if isinstance(data, list):
-                    for part_data in data:
-                        task_id = part_data["task_id"]
-                        after_score = part_data["after_score"]
-                        task_results[task_id].append(after_score)
-                else:
-                    task_id = data["task_id"]
-                    after_score = data["after_score"]
-                    task_results[task_id].append(after_score)
-
-        if not task_results:
-            logger.warning(f"No valid data found in file {file}")
-            continue
-
-        # Check if each task has consistent number of runs
-        run_counts = [len(scores) for scores in task_results.values()]
-        if len(set(run_counts)) > 1:
-            logger.warning(f"Inconsistent number of runs for different tasks in file {file}: {set(run_counts)}")
-            continue
-
-        num_runs = run_counts[0]
-        logger.info(f"File {file}: {len(task_results)} tasks, {num_runs} runs per task")
-
-        # Get all possible k values
-        k_values = get_possible_k_values(num_runs)
-        logger.info(f"Calculable best@k values: {k_values}")
-
-        # Calculate various best@k values
-        file_results = {"file": file.name}
-
-        for k in k_values:
-            best_at_k_scores = []
-            pass_at_k_scores = []
-            for task_id, scores in task_results.items():
-                try:
-                    best_k_score = calculate_best_at_k(scores, k)
-                    pass_at_k_score = calculate_pass_at_k(scores, k)
-                    pass_at_k_scores.append(pass_at_k_score)
-                    best_at_k_scores.append(best_k_score)
-                except ValueError as e:
-                    logger.error(f"Error calculating best@{k} for task {task_id}: {e}")
-                    continue
-
-            if best_at_k_scores:
-                avg_best_at_k = sum(best_at_k_scores) / len(best_at_k_scores)
-                file_results[f"best@{k}"] = avg_best_at_k
-                logger.info(f"file={file.name} best@{k}={avg_best_at_k:.4f}")
-
-            if pass_at_k_scores:
-                avg_pass_at_k = sum(pass_at_k_scores) / len(pass_at_k_scores)
-                file_results[f"pass@{k}"] = avg_pass_at_k
-                logger.info(f"file={file.name} pass@{k}={avg_pass_at_k:.4f}")
-
-        all_results[file.name] = file_results
-
-    # Create and display table
-    if all_results:
-        df = pd.DataFrame(list(all_results.values()))
-        df = df.set_index("file")
-
-        # Sort columns by the number in column name (best@8, best@4, best@2, best@1)
-        pass_columns = [col for col in df.columns if col.startswith("pass@")]
-        # best_columns = [col for col in df.columns]
-        pass_columns.sort(key=lambda x: x, reverse=False)
-        df = df[pass_columns]
-
-        print("\n" + "=" * 80)
-        print("Experiment Results Summary Table")
-        print("=" * 80)
-        print(df.round(4))
-        print("=" * 80)
-
-        # Save table to CSV
-        output_path = path / "experiment_summary.csv"
-        df.to_csv(output_path)
-        logger.info(f"Results table saved to: {output_path}")
-    else:
-        logger.warning("No valid experiment results found")
-
-
-if __name__ == "__main__":
-    run_exp_statistic()

cookbook/bfcl/bfcl_agent.py

@@ -1,705 +0,0 @@
-# flake8: noqa: E402
-import os
-
-os.environ["BFCL_DATA_PATH"] = "data/multiturn_data_base_val.jsonl"
-os.environ["BFCL_ANSWER_PATH"] = "data/possible_answer"
-from dotenv import load_dotenv
-
-load_dotenv("../../.env")
-
-import re
-import time
-import json
-import ray
-import warnings
-import tempfile
-import requests
-import datetime
-
-from tqdm import tqdm
-from pathlib import Path
-from loguru import logger
-from openai import OpenAI
-from typing import Dict, List, Any
-
-from bfcl_utils import (
-    load_test_case,
-    handle_user_turn,
-    handle_tool_calls,
-    extract_tool_schema,
-    extract_single_turn_response,
-    extract_multi_turn_responses,
-    capture_and_print_score_files,
-    create_error_response,
-)
-from bfcl_eval.model_handler.api_inference.qwen import QwenAPIHandler
-from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_utils import (
-    is_empty_execute_response,
-)
-from bfcl_eval.eval_checker.eval_runner import (
-    multi_turn_runner,
-    ast_file_runner,
-)
-from bfcl_eval.eval_checker.eval_runner_helper import record_cost_latency
-from bfcl_eval.utils import (
-    is_multi_turn,
-    is_relevance_or_irrelevance,
-    find_file_with_suffix,
-    load_file,
-)
-
-
-@ray.remote
-class BFCLAgent:
-    """A minimal ReAct Agent for BFCL-v3(multi-turn) tasks."""
-
-    def __init__(
-        self,
-        index: int,
-        task_ids: List[str],
-        experiment_name: str,
-        data_path: str = os.getenv("BFCL_DATA_PATH"),
-        answer_path: Path = Path(os.getenv("BFCL_ANSWER_PATH")),
-        model_name: str = "qwen3-8b",
-        temperature: float = 0.9,
-        max_interactions: int = 30,
-        max_response_size: int = 2000,
-        num_trials: int = 1,
-        enable_thinking: bool = False,
-        use_memory: bool = False,
-        use_memory_addition: bool = False,
-        use_memory_deletion: bool = False,
-        delete_freq: int = 10,
-        freq_threshold: int = 5,
-        utility_threshold: float = 0.5,
-        memory_base_url: str = "http://0.0.0.0:8002/",
-        memory_workspace_id: str = "bfcl_v3",
-    ):
-
-        self.index: int = index
-        self.task_ids: List[str] = task_ids
-        self.categories: List[str] = [task_id.rsplit("_", 1)[0] if "_" in task_id else task_id for task_id in task_ids]
-        self.experiment_name: str = experiment_name
-        self.data_path: str = data_path
-        self.answer_path: Path = answer_path
-        self.model_name: str = model_name
-        self.temperature: float = temperature
-        self.max_interactions: int = max_interactions
-        self.max_response_size: int = max_response_size
-        self.num_trials: int = num_trials
-        self.enable_thinking: bool = enable_thinking
-        self.use_memory: bool = use_memory
-        self.use_memory_addition: bool = use_memory_addition if use_memory else False
-        self.use_memory_deletion: bool = use_memory_deletion if use_memory else False
-        self.delete_freq: int = delete_freq
-        self.freq_threshold: int = freq_threshold
-        self.utility_threshold: float = utility_threshold
-        self.memory_base_url: str = memory_base_url
-        self.memory_workspace_id: str = memory_workspace_id
-
-        self.history: List[List[List[dict]]] = [[] for _ in range(num_trials)]
-        self.retrieved_memory_list: List[List[List[Any]]] = [[] for _ in range(num_trials)]
-        self.test_entry: List[List[Dict[str, Any]]] = [[] for _ in range(num_trials)]
-        self.original_test_entry: List[List[Dict[str, Any]]] = [[] for _ in range(num_trials)]
-        self.tool_schema: List[List[List[dict]]] = [[] for _ in range(num_trials)]
-        self.current_turn = [[0 for _ in range(len(task_ids))] for _ in range(num_trials)]
-
-        for run_id in range(num_trials):
-            for task_index in range(len(task_ids)):
-                self.init_state(run_id, task_index)
-
-    def init_state(self, run_id, i) -> Dict[str, Any]:
-        self.test_entry[run_id].append(load_test_case(self.data_path, self.task_ids[i]))
-        self.original_test_entry[run_id].append(self.test_entry[run_id][i].get("extra", {}))
-        self.tool_schema[run_id].append(extract_tool_schema(self.test_entry[run_id][i].get("tools", [{}])))
-
-        msg = self.test_entry[run_id][i].get("messages", [])
-        self.history[run_id].append(msg)
-        self.retrieved_memory_list[run_id].append([])
-        self.current_turn[run_id][i] = 1
-
-    def update_task_history_with_memory(self, run_id, task_index, previous_memories: None):
-        query = self.history[run_id][task_index][0]["content"]
-        if len(previous_memories) == 0:
-            response = self.get_memory(query)
-            if response and "memory_list" in response["metadata"]:
-                self.retrieved_memory_list[run_id][task_index] = response["metadata"]["memory_list"]
-                task_memory = response["answer"]
-                logger.info(f"loaded task_memory: {task_memory}")
-                self.history[run_id][task_index][0] = self.get_query_with_memory(query, task_memory)
-        else:
-            formatted_memories = []
-            for i, memory in enumerate(previous_memories, 1):
-                condition = memory["when_to_use"]
-                memory_content = memory["content"]
-                memory_text = f"Experience {i}:\n When to use: {condition}\n Content: {memory_content}\n"
-                formatted_memories.append(memory_text)
-            self.history[run_id][task_index][0] = self.get_query_with_memory(query, "\n".join(formatted_memories))
-
-    def get_query_with_memory(self, query: str, memory: str):
-        return {
-            "role": "user",
-            "content": "Task:\n" + query + "\n\nSome Related Experience to help you to complete the task:\n" + memory,
-        }
-
-    def get_query_without_experience(self, query: str):
-        if "\n\nSome Related Experience" in query:
-            query = query.split("\n\nSome Related Experience")[0].split("Task:\n")[-1]
-        return query
-
-    def get_traj_from_task_history(self, task_id: str, task_history: list, reward: float):
-        return {
-            "task_id": task_id,
-            "messages": task_history,
-            "score": reward,
-        }
-
-    def handle_api_response(self, response: requests.Response):
-        """Handle API response with proper error checking"""
-        if response.status_code != 200:
-            print(f"Error: {response.status_code}")
-            print(response.text)
-            return None
-
-        return response.json()
-
-    def get_memory(self, query: str):
-        response = requests.post(
-            url=self.memory_base_url + "retrieve_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "query": query,
-                "top_k": 5,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return None
-
-        logger.info(f"query: {query}, response: {result}")
-        return result
-
-    def add_memory(self, trajectories):
-        response = requests.post(
-            url=self.memory_base_url + "summary_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "trajectories": trajectories,
-            },
-        )
-
-        result = self.handle_api_response(response)
-        if not result:
-            return []
-
-        # Extract memory list from response
-        memory_list = result.get("metadata", {}).get("memory_list", [])
-        logger.info(f'add new memories: {memory_list}')
-        return memory_list
-
-    def delete_memory_by_ids(self, memory_ids):
-        response = requests.post(
-            url=self.memory_base_url + "vector_store",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "action": "delete_ids",
-                "memory_ids": memory_ids
-            }
-        )
-        response.raise_for_status()
-
-    def update_memory_information(self, memory_list, update_utility: bool = False):
-        response = requests.post(
-            url=self.memory_base_url + "record_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "memory_dicts": memory_list,
-                "update_utility": update_utility,
-            },
-        )
-        response.raise_for_status()
-        logger.info(response.json())
-
-    def delete_memory(self):
-        response = requests.post(
-            url=self.memory_base_url + "delete_task_memory",
-            json={
-                "workspace_id": self.memory_workspace_id,
-                "freq_threshold": self.freq_threshold,
-                "utility_threshold": self.utility_threshold,
-            },
-        )
-        response.raise_for_status()
-
-    def call_llm(self, messages: list, tool_schemas: list[dict]) -> str:
-        for i in range(100):
-            try:
-                client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
-                # Change this function to modify the base llm
-                response = client.chat.completions.create(
-                    model=self.model_name,
-                    messages=messages,
-                    tools=tool_schemas,
-                    temperature=self.temperature,
-                    seed=0,
-                    extra_body={"enable_thinking": self.enable_thinking},
-                    stream=self.enable_thinking,
-                    parallel_tool_calls=True,
-                )
-                if not self.enable_thinking:
-                    out_msg = response.choices[0].message
-                    return out_msg.model_dump(exclude_unset=True, exclude_none=True)
-                else:
-                    reasoning_content = ""  # Complete reasoning process
-                    answer_content = ""  # Define complete response
-                    tool_info = []  # Store tool invocation information
-                    is_answering = (
-                        False  # Determine whether the reasoning process has finished and response has started
-                    )
-
-                    for chunk in response:
-                        if not chunk.choices:
-                            # Handle usage information
-                            continue
-                        else:
-                            delta = chunk.choices[0].delta
-                            # Handle AI's thought process (chain reasoning)
-                            if hasattr(delta, "reasoning_content") and delta.reasoning_content is not None:
-                                reasoning_content += delta.reasoning_content
-
-                            # Handle final response content
-                            else:
-                                if not is_answering:  # Print title when entering the response phase for the first time
-                                    is_answering = True
-                                if delta.content is not None:
-                                    answer_content += delta.content
-
-                                # Handle tool invocation information (support parallel tool calls)
-                                if delta.tool_calls is not None:
-                                    for tool_call in delta.tool_calls:
-                                        index = tool_call.index  # Tool call index, used for parallel calls
-
-                                        # Dynamically expand tool information storage list
-                                        while len(tool_info) <= index:
-                                            tool_info.append(
-                                                {
-                                                    "id": "",
-                                                    "type": "function",
-                                                    "index": index,
-                                                    "function": {"name": "", "arguments": ""},
-                                                },
-                                            )
-
-                                        # Collect tool call ID (used for subsequent function calls)
-                                        if tool_call.id:
-                                            tool_info[index]["id"] += tool_call.id
-
-                                        # Collect function name (used for subsequent routing to specific functions)
-                                        if tool_call.function and tool_call.function.name:
-                                            tool_info[index]["function"]["name"] += tool_call.function.name
-
-                                        # Collect function parameters (in JSON string format, need subsequent parsing)
-                                        if tool_call.function and tool_call.function.arguments:
-                                            tool_info[index]["function"]["arguments"] += tool_call.function.arguments
-                    msg = {
-                        "role": "assistant",
-                        "content": answer_content,
-                        "reasoning_content": reasoning_content,
-                    }
-                    if tool_info:
-                        msg["tool_calls"] = tool_info
-                    return msg
-            except Exception as e:
-                logger.exception(f"encounter error with {e.args}")
-                time.sleep(1 + i * 10)
-
-        return "call llm error"
-
-    def env_step(self, run_id: int, index: int, messages: str) -> str:
-        """
-        Process one step in the conversation.
-        Both single turn and multi turn are supported.
-
-        Args:
-            messages: List of conversation messages, with the last one being assistant response
-            test_entry: Test entry containing initial_config, involved_classes, question etc.
-            **kwargs: Additional arguments for compatibility
-
-        Returns:
-            Dict containing next message and tools if applicable
-        """
-        try:
-            if not messages:
-                return handle_user_turn(self.original_test_entry[run_id][index], self.current_turn[run_id][index])
-
-            if messages[-1]["role"] != "assistant":
-                return create_error_response(
-                    "Last message must be from assistant",
-                )
-
-            if "tool_calls" in messages[-1] and len(messages[-1]["tool_calls"]) > 0:
-                try:
-                    tool_calls = messages[-1]["tool_calls"]
-                    decoded_calls = self._convert_tool_calls_to_execution_format(
-                        tool_calls,
-                    )
-                    # decoded_calls:[function(param=xxx)]
-                    print(f"decoded_calls: {decoded_calls}")
-                    if is_empty_execute_response(decoded_calls):
-                        warnings.warn(
-                            f"is_empty_execute_response: {is_empty_execute_response(decoded_calls)}",
-                        )
-                        return handle_user_turn(
-                            self.original_test_entry[run_id][index],
-                            self.current_turn[run_id][index],
-                        )
-                    return handle_tool_calls(
-                        tool_calls,
-                        decoded_calls,
-                        self.original_test_entry[run_id][index],
-                        self.current_turn[run_id][index],
-                    )
-                except Exception as e:
-                    warnings.warn(f"Errors during tool invocation: {str(e)}")
-                    return handle_user_turn(self.original_test_entry[run_id][index], self.current_turn[run_id][index])
-            else:
-                return handle_user_turn(self.original_test_entry[run_id][index], self.current_turn[run_id][index])
-
-        except Exception as e:
-            return create_error_response(f"Failed to process request: {str(e)}")
-
-    def _convert_tool_calls_to_execution_format(
-        self,
-        tool_calls: List[Dict[str, Any]],
-    ) -> List[str]:
-        """
-        Convert OpenAI format tool calls to execution format.
-
-        Args:
-            tool_calls: List of tool calls in OpenAI format
-
-        Returns:
-            List of function calls in string format
-        """
-        execution_list = []
-
-        for tool_call in tool_calls:
-            function = tool_call.get("function", {})
-            function_name = function.get("name", "")
-
-            try:
-                arguments = function.get("arguments", "{}")
-                if isinstance(arguments, str):
-                    args_dict = json.loads(arguments)
-                else:
-                    args_dict = arguments
-
-                args_str = ", ".join([f"{k}={repr(v)}" for k, v in args_dict.items()])
-                execution_list.append(f"{function_name}({args_str})")
-
-            except Exception as e:
-                execution_list.append(f"{function_name}()")
-
-        return execution_list
-
-    def get_reward(self, run_id, index) -> float:
-        try:
-            if not self.history[run_id][index] or not self.original_test_entry[run_id][index]:
-                return 0.0
-
-            model_name = "env_handler"
-            handler = QwenAPIHandler(
-                model_name,
-                temperature=1.0,
-            )  # FIXME: magic number
-
-            model_result_data = self._convert_conversation_to_eval_format(run_id, index)
-
-            prompt_data = [self.original_test_entry[run_id][index]]
-
-            state = {"leaderboard_table": {}}
-            record_cost_latency(
-                state["leaderboard_table"],
-                model_name,
-                [model_result_data],
-            )
-
-            if is_relevance_or_irrelevance(self.categories[index]):
-                accuracy, _ = self._eval_relevance_test(
-                    handler,
-                    model_result_data,
-                    prompt_data,
-                    model_name,
-                    self.category,
-                )
-            else:
-                # Find the corresponding possible answer file
-
-                possible_answer_file = find_file_with_suffix(
-                    self.answer_path,
-                    self.categories[index],
-                )
-                possible_answer = load_file(possible_answer_file, sort_by_id=True)
-                possible_answer = [item for item in possible_answer if item["id"] == self.task_ids[index]]
-                if is_multi_turn(self.categories[index]):
-                    accuracy, _ = self._eval_multi_turn_test(
-                        handler,
-                        model_result_data,
-                        prompt_data,
-                        possible_answer,
-                        model_name,
-                        self.categories[index],
-                    )
-                else:
-                    accuracy, _ = self._eval_single_turn_test(
-                        handler,
-                        model_result_data,
-                        prompt_data,
-                        possible_answer,
-                        model_name,
-                        self.categories[index],
-                    )
-            print(f"model_result_data: {model_result_data}")
-            print(f"possible_answer: {possible_answer}") if possible_answer else None
-
-            return accuracy
-
-        except Exception as e:
-            import traceback
-
-            traceback.print_exc()
-            return 0
-
-    def _convert_conversation_to_eval_format(self, run_id, index) -> Dict[str, Any]:
-        """
-        Convert conversation history to evaluation format.
-
-        Args:
-            conversation_result: Result from run_conversation
-            original_test_entry: Original test entry data
-
-        Returns:
-            Data in format expected by multi_turn_runner or other runners
-        """
-        if is_multi_turn(self.categories[index]):
-            turns_data = extract_multi_turn_responses(self.history[run_id][index])
-        else:
-            turns_data = extract_single_turn_response(self.history[run_id][index])
-
-        model_result_data = {
-            "id": self.task_ids[index],
-            "result": turns_data,
-            "latency": 0,
-            "input_token_count": 0,
-            "output_token_count": 0,
-        }
-
-        return model_result_data
-
-    def _eval_multi_turn_test(
-        self,
-        handler,
-        model_result_data,
-        prompt_data,
-        possible_answer,
-        model_name,
-        test_category,
-    ):
-        """
-        Evaluate multi-turn test.
-
-        Args:
-            handler: Model handler instance
-            model_result_data: Model result data
-            prompt_data: Prompt data
-            possible_answer: Possible answer data
-            model_name: Name of the model
-            test_category: Category of the test
-
-        Returns:
-            Tuple of (accuracy, total_count)
-        """
-        with tempfile.TemporaryDirectory() as temp_dir:
-            score_dir = Path(temp_dir)
-            accuracy, total_count = multi_turn_runner(
-                handler=handler,
-                model_result=[model_result_data],
-                prompt=prompt_data,
-                possible_answer=possible_answer,
-                model_name=model_name,
-                test_category=test_category,
-                score_dir=score_dir,
-            )
-            capture_and_print_score_files(
-                score_dir,
-                model_name,
-                test_category,
-                "multi_turn",
-            )
-            return accuracy, total_count
-
-    def _eval_single_turn_test(
-        self,
-        handler,
-        model_result_data,
-        prompt_data,
-        possible_answer,
-        model_name,
-        test_category,
-    ):
-        """
-        Evaluate single-turn AST test.
-
-        Args:
-            handler: Model handler instance
-            model_result_data: Model result data
-            prompt_data: Prompt data
-            possible_answer: Possible answer data
-            model_name: Name of the model
-            test_category: Category of the test
-
-        Returns:
-            Tuple of (accuracy, total_count)
-        """
-        language = "Python"
-        if "java" in test_category.lower():
-            language = "Java"
-        elif "js" in test_category.lower() or "javascript" in test_category.lower():
-            language = "JavaScript"
-
-        with tempfile.TemporaryDirectory() as temp_dir:
-            score_dir = Path(temp_dir)
-            accuracy, total_count = ast_file_runner(
-                handler=handler,
-                model_result=[model_result_data],
-                prompt=prompt_data,
-                possible_answer=possible_answer,
-                language=language,
-                test_category=test_category,
-                model_name=model_name,
-                score_dir=score_dir,
-            )
-            capture_and_print_score_files(
-                score_dir,
-                model_name,
-                test_category,
-                "single_turn",
-            )
-            return accuracy, total_count
-
-    def execute(self):
-        result = []
-        counter = 0
-        for task_index, task_id in enumerate(tqdm(self.task_ids, desc=f"ray_index={self.index}")):
-            t_result = None
-            previous_memories = []
-            for run_id in range(self.num_trials):
-                try:
-                    start_time = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-                    for i in range(self.max_interactions):
-                        if self.use_memory and i == 0:
-                            self.update_task_history_with_memory(run_id, task_index, previous_memories)
-                        llm_output = self.call_llm(
-                            self.history[run_id][task_index],
-                            self.tool_schema[run_id][task_index],
-                        )
-                        self.history[run_id][task_index].append(llm_output)
-
-                        env_output = self.env_step(run_id, task_index, self.history[run_id][task_index])
-                        # Possible env_output returns after environment interaction:
-                        # 1. Triggers a query with available tools list: {"messages": [{"role": "user", "content": user_query}], "tools": tools}
-                        # 2. Returns tool invocation result: {"messages": [{"role": "tool", "content": {<execution_results>}, 'tool_call_id': 'chatcmpl-tool-xxx'}]}
-                        #    <execution_results>: when success, returns result dicts, e.g., {"travel_cost_list": [1140.0]}, when error, returns error message, e.g., {"error": "cd: temporary: No such directory. You cannot use path to change directory."}
-                        # 3. Conversation completion: {"messages": [{"role": "env", "content": "[CONVERSATION_COMPLETED]"}]}
-                        # 4. Program error: {"messages": [{"role": "env", "content": f"[ERROR] {error_message}"}]}
-
-                        # tool_list update
-                        if "tools" in env_output:
-                            self.tool_schema[run_id][task_index] = extract_tool_schema(env_output["tools"])
-
-                        new_tool_calls = []
-                        new_tool_call_ids = []
-                        next_user_msg = ""
-                        for idx, msg in enumerate(env_output.get("messages", [])):
-                            if msg["role"] == "tool" and len(msg["content"]) > 0:
-                                new_tool_calls.append(msg.get("content", ""))
-                                new_tool_call_ids.append(msg.get("tool_call_id", ""))
-                            elif msg["role"] == "user":
-                                next_user_msg = msg.get("content", "")
-                                self.current_turn[run_id][task_index] += 1
-                            else:  # for env role messages
-                                next_user_msg = msg.get("content", "")
-
-                        if new_tool_calls:
-                            for idx, call in enumerate(new_tool_calls):
-                                self.history[run_id][task_index].append(
-                                    {"role": "tool", "content": str(call), "tool_call_id": new_tool_call_ids[idx]},
-                                )
-                        else:
-                            self.history[run_id][task_index].append({"role": "user", "content": next_user_msg})
-
-                        logger.info(f"index={self.index} task_id={task_id} iteration={i}")
-
-                        if self.task_completed(run_id, task_index):
-                            break
-
-                    reward = self.get_reward(run_id, task_index)
-                    if self.use_memory:
-                        if self.use_memory_addition:  # selectively add memories when succeed
-                            new_traj_list = [self.get_traj_from_task_history(task_id, self.history[run_id][task_index], reward)]
-                            previous_memories = self.add_memory(new_traj_list)
-                            if reward != 1:
-                                self.delete_memory_by_ids([mem["memory_id"] for mem in previous_memories])
-
-                        # update the freq & utility attributes of retrieved memories
-                        update_utility: bool = reward == 1
-                        self.update_memory_information(self.retrieved_memory_list[run_id][task_index], update_utility)
-
-                    counter += 1
-                    if self.use_memory_deletion and counter % self.delete_freq == 0:
-                        self.delete_memory()
-
-                    t_result = {
-                        "run_id": run_id,
-                        "task_id": self.task_ids[task_index],
-                        "experiment_name": self.experiment_name,
-                        "task_completed": self.task_completed(run_id, task_index),
-                        "reward": reward,
-                        "task_history": self.history[run_id][task_index],
-                        "task_start_time": start_time,
-                    }
-                    if reward == 1:
-                        break
-
-                except Exception as e:
-                    logger.exception(f"encounter error with {e.args}")
-                    result.append({})
-            result.append(t_result)
-        return result
-
-    def task_completed(self, run_id, index):
-        """
-        Check if task is completed.
-
-        Returns:
-            True if task is completed, False otherwise
-        """
-        return self.history[run_id][index][-1]["content"] == "[CONVERSATION_COMPLETED]"
-
-
-def main():
-    with open(os.getenv("BFCL_DATA_PATH"), "r", encoding="utf-8") as f:
-        task_ids = [json.loads(l)["id"] for l in f]
-    dataset_name = "dev"
-    agent = BFCLAgent(
-        index=0,
-        task_id=task_ids[0],
-        experiment_name=f"qwen3_8b_{dataset_name}",
-    )
-    result = agent.execute()
-    logger.info(f"result={json.dumps(result)}")
-
-
-if __name__ == "__main__":
-    main()

cookbook/bfcl/bfcl_utils.py

@@ -1,395 +0,0 @@
-import json
-from pathlib import Path
-from typing import Dict, List, Any
-
-from bfcl_eval.constants.default_prompts import (
-    DEFAULT_USER_PROMPT_FOR_ADDITIONAL_FUNCTION_FC,
-)
-from bfcl_eval.constants.type_mappings import GORILLA_TO_OPENAPI
-from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_utils import (
-    execute_multi_turn_func_call,
-)
-from bfcl_eval.model_handler.model_style import ModelStyle
-from bfcl_eval.model_handler.utils import (
-    convert_to_tool,
-    default_decode_execute_prompting,
-    func_doc_language_specific_pre_processing,
-)
-
-
-def load_test_case(data_path: str, test_id: str | None) -> Dict[str, Any]:
-    if not Path(data_path).exists():
-        raise FileNotFoundError(f"BFCL data file '{data_path}' not found")
-
-    if test_id is None:
-        raise ValueError("task_id is required")
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        if str(test_id).isdigit():
-            idx = int(test_id)
-            for line_no, line in enumerate(f):
-                if line_no == idx:
-                    return json.loads(line)
-            raise ValueError(f"Test case index {idx} not found in {data_path}")
-        else:
-            for line in f:
-                data = json.loads(line)
-                if data.get("id") == test_id:
-                    return data
-            raise ValueError(f"Test case id '{test_id}' not found in {data_path}")
-
-
-def handle_user_turn(
-    test_entry: Dict[str, Any],
-    current_turn: int,
-) -> Dict[str, Any]:
-    """
-    Handle user turn by returning appropriate content from test_entry["question"].
-    For non-first turns, processes user query and tools.
-
-    Args:
-        test_entry: Test entry containing conversation data
-        current_turn: Current turn number
-
-    Returns:
-        Response containing next user message and tools
-    """
-    try:
-        current_turn_message = []
-        tools = compile_tools(test_entry)
-        questions = test_entry.get("question", [])
-        holdout_function = test_entry.get("holdout_function", {})
-
-        if str(current_turn) in holdout_function:
-            test_entry["function"].extend(holdout_function[str(current_turn)])
-            tools = compile_tools(test_entry)
-            assert len(questions[current_turn]) == 0, "Holdout turn should not have user message."
-            current_turn_message = [
-                {
-                    "role": "user",
-                    "content": DEFAULT_USER_PROMPT_FOR_ADDITIONAL_FUNCTION_FC,
-                },
-            ]
-            return create_user_response(current_turn_message, tools)
-        if current_turn >= len(questions):
-            return create_completion_response()
-
-        current_turn_message = questions[current_turn]
-
-        return create_user_response(current_turn_message, tools)
-
-    except Exception as e:
-        return create_error_response(f"Failed to process user message: {str(e)}")
-
-
-def handle_tool_calls(
-    tool_calls: List[Dict[str, Any]],
-    decoded_calls: list[str],
-    test_entry: Dict[str, Any],
-    current_turn: int,
-) -> Dict[str, Any]:
-    """
-    Handle tool calls from assistant.
-
-    Args:
-        tool_calls: List of tool calls in OpenAI format
-        decoded_calls: List of decoded function calls
-        test_entry: Test entry containing environment data
-        current_turn: Current turn number
-
-    Returns:
-        Response containing tool execution results
-    """
-    execution_results, _ = execute_multi_turn_func_call(
-        func_call_list=decoded_calls,
-        initial_config=test_entry["initial_config"],
-        involved_classes=test_entry["involved_classes"],
-        model_name="env_handler",
-        test_entry_id=test_entry["id"],
-        long_context=("long_context" in test_entry["id"] or "composite" in test_entry["id"]),
-        is_evaL_run=False,
-    )
-    # print('execution_results in handler_tool_calls:', execution_results)
-
-    return create_tool_response(tool_calls, execution_results)
-
-
-def compile_tools(test_entry: dict) -> list:
-    """
-    Compile functions into tools format.
-
-    Args:
-        test_entry: Test entry containing functions
-
-    Returns:
-        List of tools in OpenAI format
-    """
-    functions: list = test_entry["function"]
-    test_category: str = test_entry["id"].rsplit("_", 1)[0]
-
-    functions = func_doc_language_specific_pre_processing(functions, test_category)
-    tools = convert_to_tool(functions, GORILLA_TO_OPENAPI, ModelStyle.OpenAI_Completions)
-
-    return tools
-
-
-def create_tool_response(
-    tool_calls: List[Dict[str, Any]],
-    execution_results: List[str],
-) -> Dict[str, Any]:
-    """
-    Create response for tool calls.
-
-    Args:
-        tool_calls: List of tool calls
-        execution_results: List of execution results
-
-    Returns:
-        Response containing tool execution results
-    """
-    tool_messages = []
-    for i, (tool_call, result) in enumerate(zip(tool_calls, execution_results)):
-        tool_messages.append(
-            {
-                "role": "tool",
-                "content": result,
-                "tool_call_id": tool_call.get("id", f"call_{i}"),
-            },
-        )
-
-    return {"messages": tool_messages}
-
-
-def create_user_response(
-    question_turn: List[Dict[str, Any]],
-    tools: List[Dict[str, Any]],
-) -> Dict[str, Any]:
-    """
-    Create response containing user message.
-
-    Args:
-        question_turn: List of messages for current turn
-        tools: List of available tools
-
-    Returns:
-        Response containing user message and tools
-    """
-    user_content = ""
-    for msg in question_turn:
-        if msg["role"] == "user":
-            user_content = msg["content"]
-            break
-
-    return {"messages": [{"role": "user", "content": user_content}], "tools": tools}
-
-
-def create_completion_response() -> Dict[str, Any]:
-    """
-    Create response indicating conversation completion.
-
-    Returns:
-        Response with completion message
-    """
-    return {"messages": [{"role": "env", "content": "[CONVERSATION_COMPLETED]"}]}
-
-
-def create_error_response(error_message: str) -> Dict[str, Any]:
-    """
-    Create response for error conditions.
-
-    Args:
-        error_message: Error message to include
-
-    Returns:
-        Response containing error message
-    """
-    return {"messages": [{"role": "env", "content": f"[ERROR] {error_message}"}]}
-
-
-def decode_execute(result):
-    """
-    Decode execute results for compatibility with evaluation framework.
-
-    Args:
-        result: Result to decode
-
-    Returns:
-        List of decoded function calls
-    """
-    return default_decode_execute_prompting(result)
-
-
-def extract_single_turn_response(messages: List[Dict[str, Any]]) -> str:
-    """
-    Extract single-turn response from conversation messages.
-
-    Args:
-        messages: List of conversation messages
-
-    Returns:
-        String representation of the response
-    """
-    for message in reversed(messages):
-        if message["role"] == "assistant":
-            if "tool_calls" in message and message["tool_calls"]:
-                formatted_calls = []
-                for tool_call in message["tool_calls"]:
-                    formatted_call = format_single_tool_call_for_eval(
-                        tool_call,
-                    )
-                    if formatted_call:
-                        formatted_calls.append(formatted_call)
-                return "\n".join(formatted_calls) if formatted_calls else ""
-            elif message.get("content"):
-                return message["content"]
-
-    return ""
-
-
-def extract_multi_turn_responses(
-    messages: List[Dict[str, Any]],
-) -> List[List[str]]:
-    """
-    Extract multi-turn responses from conversation messages.
-
-    Args:
-        messages: List of conversation messages
-
-    Returns:
-        List of turns, each turn is a list of function call strings
-    """
-    turns_data = []
-    current_turn_responses = []
-
-    i = 0
-    while i < len(messages):
-        message = messages[i]
-
-        if message["role"] == "user":
-            if current_turn_responses:
-                turns_data.append(current_turn_responses)
-                current_turn_responses = []
-
-            i += 1
-            while i < len(messages) and messages[i]["role"] == "assistant":
-                assistant_msg = messages[i]
-
-                if "tool_calls" in assistant_msg and assistant_msg["tool_calls"]:
-                    for tool_call in assistant_msg["tool_calls"]:
-                        formatted_call = format_single_tool_call_for_eval(
-                            tool_call,
-                        )
-                        if formatted_call:
-                            current_turn_responses.append(formatted_call)
-
-                i += 1
-
-                while i < len(messages) and messages[i]["role"] == "tool":
-                    i += 1
-        else:
-            i += 1
-
-    if current_turn_responses:
-        turns_data.append(current_turn_responses)
-
-    return turns_data
-
-
-def format_single_tool_call_for_eval(tool_call: Dict[str, Any]) -> str:
-    """
-    Format a single tool call into string representation for evaluation.
-
-    Args:
-        tool_call: Single tool call in OpenAI format
-
-    Returns:
-        Formatted string representation
-    """
-    function = tool_call.get("function", {})
-    function_name = function.get("name", "")
-
-    try:
-        arguments = function.get("arguments", "{}")
-        if isinstance(arguments, str):
-            args_dict = json.loads(arguments)
-        else:
-            args_dict = arguments
-
-        args_str = ", ".join([f"{k}={repr(v)}" for k, v in args_dict.items()])
-        return f"{function_name}({args_str})"
-
-    except Exception:
-        return f"{function_name}()"
-
-
-def capture_and_print_score_files(
-    score_dir: Path,
-    model_name: str,
-    test_category: str,
-    eval_type: str,
-):
-    """
-    Capture and print contents of score files written to score_dir.
-
-    Args:
-        score_dir: Directory containing score files
-        model_name: Name of the model
-        test_category: Category of the test
-        eval_type: Type of evaluation (relevance/multi_turn/single_turn)
-    """
-    try:
-        print(f"\n=== {eval_type.upper()} Evaluation Result Files ===")
-        print(f"Model: {model_name}")
-        print(f"Test Category: {test_category}")
-        print(f"Evaluation Type: {eval_type}")
-
-        for file_path in score_dir.rglob("*"):
-            if file_path.is_file():
-                relative_path = file_path.relative_to(score_dir)
-                print(f"\n--- File: {relative_path} ---")
-
-                try:
-                    with open(file_path, "r", encoding="utf-8") as f:
-                        content = f.read()
-
-                    if (
-                        file_path.suffix == ".json"
-                        or content.strip().startswith("{")
-                        or content.strip().startswith("[")
-                    ):
-                        try:
-                            import json
-
-                            lines = content.strip().split("\n")
-                            formatted_lines = []
-                            for line in lines:
-                                if line.strip():
-                                    parsed = json.loads(line)
-                                    formatted_lines.append(
-                                        json.dumps(
-                                            parsed,
-                                            ensure_ascii=False,
-                                            indent=2,
-                                        ),
-                                    )
-                            content = "\n".join(formatted_lines)
-                        except json.JSONDecodeError:
-                            pass
-
-                    print(content)
-
-                except UnicodeDecodeError:
-                    print(f"[Binary file, size: {file_path.stat().st_size} bytes]")
-                except Exception as e:
-                    print(f"[Error reading file: {str(e)}]")
-
-        print(f"=== {eval_type.upper()} Evaluation Result Files End ===\n")
-
-    except Exception as e:
-        print(f"Error capturing evaluation result files: {str(e)}")
-
-
-def extract_tool_schema(tools):
-    for i in range(len(tools)):
-        tools[i]["function"].pop("response")
-    return tools

cookbook/bfcl/init_exp_pool.py

@@ -1,233 +0,0 @@
-import argparse
-import json
-from collections import defaultdict
-from concurrent.futures import ThreadPoolExecutor, as_completed
-from pathlib import Path
-from typing import List, Dict, Any
-
-import requests
-
-
-def load_task_case(data_path: str, task_id: str | None) -> Dict[str, Any]:
-    """
-    load training cases by id
-    """
-    if not Path(data_path).exists():
-        raise FileNotFoundError(f"BFCL data file '{data_path}' not found")
-
-    if task_id is None:
-        raise ValueError("task_id is required")
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        if str(task_id).isdigit():
-            idx = int(task_id)
-            for line_no, line in enumerate(f):
-                if line_no == idx:
-                    return json.loads(line)
-            raise ValueError(f"Task case index {idx} not found in {data_path}")
-        else:
-            for line in f:
-                data = json.loads(line)
-                if data.get("id") == task_id:
-                    return data
-            raise ValueError(f"Task case id '{task_id}' not found in {data_path}")
-
-
-def get_tool_prompt(tools):
-    tool_prompt = "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>"
-    for tool in tools:
-        tool_prompt += "\n" + json.dumps(tool)
-    tool_prompt += '\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{"name": <function-name>, "arguments": <args-json-object>}\n</tool_call>'
-    return tool_prompt
-
-
-def group_trajectories_by_task_id(jsonl_entries: List[Dict[str, Any]]) -> List[List[Any]]:
-    """
-    group trajectories by task_id
-
-    Args:
-        jsonl_entries: JSONL entry list
-
-    Returns:
-        List[List[Any]]: trajectory list grouped by task_id
-    """
-    grouped = defaultdict(list)
-
-    for entry in jsonl_entries:
-        task_id = entry.get("task_id", "")
-        taks_case = load_task_case("data/multiturn_data_base.jsonl", task_id)
-        tools = taks_case.get("tools", [{}])
-        from bfcl_utils import extract_tool_schema
-
-        tool_schema = extract_tool_schema(tools)
-        entry["task_history"][0]["content"] += get_tool_prompt(tool_schema)
-        grouped[task_id].append(entry)
-
-    # retain only the two with the highest and lowest rewards
-    filtered_groups = []
-    for key, trajectories in grouped.items():
-        if len(trajectories) == 1:
-            # when only one trajectory, retain it
-            filtered_groups.append(trajectories)
-        elif len(trajectories) == 2:
-            # when there are two trajectories, retain them
-            filtered_groups.append(trajectories)
-        else:
-            # when there are more than two trajectories, choose the two with the highest and lowest rewards
-            trajectories.sort(key=lambda t: t["reward"])
-            min_reward_traj = trajectories[0]  # highest reward
-            max_reward_traj = trajectories[-1]  # lowest reward
-            filtered_groups.append([min_reward_traj, max_reward_traj])
-
-    return filtered_groups
-
-
-def post_to_summarizer(trajectories: List[Any], service_url: str, workspace_id: str) -> Dict[str, Any]:
-    trajectory_dicts = [
-        {
-            "task_id": traj["task_id"],
-            "messages": traj["task_history"],
-            "score": traj["reward"],
-        }
-        for traj in trajectories
-    ]
-
-    request_data = {
-        "trajectories": trajectory_dicts,
-        "workspace_id": workspace_id,
-    }
-
-    try:
-        response = requests.post(f"{service_url}/summary_task_memory", json=request_data)
-        response.raise_for_status()
-        return response.json()
-    except Exception as e:
-        return {"error": str(e), "trajectories_count": len(trajectories)}
-
-
-def process_trajectories_with_threads(
-    grouped_trajectories: List[List[Any]],
-    service_url: str,
-    workspace_id: str,
-    n_threads: int = 4,
-) -> List[Dict[str, Any]]:
-    """
-    use threads to process trajectories
-
-    Args:
-        grouped_trajectories: group trajectory list by task_id
-        service_url: memory summarizer service URL
-        workspace_id: workspace ID
-        n_threads: number of threads
-
-    Returns:
-        all results
-    """
-    results = []
-
-    with ThreadPoolExecutor(max_workers=n_threads) as executor:
-        future_to_group = {
-            executor.submit(post_to_summarizer, group, service_url, workspace_id): i
-            for i, group in enumerate(grouped_trajectories)
-        }
-
-        for future in as_completed(future_to_group):
-            group_index = future_to_group[future]
-            try:
-                result = future.result()
-                result["group_index"] = group_index
-                result["group_size"] = len(grouped_trajectories[group_index])
-                results.append(result)
-                print(
-                    f'✅ Group {group_index} processed: {result["metadata"].get("memory_list", 0) if "memory_list" in result["metadata"] else "error"}',
-                )
-            except Exception as e:
-                error_result = {
-                    "group_index": group_index,
-                    "group_size": len(grouped_trajectories[group_index]),
-                    "error": str(e),
-                }
-                results.append(error_result)
-                print(f"❌ Group {group_index} failed: {e}")
-
-    return results
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Convert JSONL to memories using ReMe service")
-    parser.add_argument("--jsonl_file", type=str, required=True, help="Path to the JSONL file")
-    parser.add_argument("--service_url", type=str, default="http://localhost:8001", help="ReMe service URL")
-    parser.add_argument("--workspace_id", type=str, required=True, help="Workspace ID for the task memory pool")
-    parser.add_argument("--output_file", type=str, help="Output file to save results (optional)")
-    parser.add_argument("--n_threads", type=int, default=4, help="Number of threads for processing")
-
-    args = parser.parse_args()
-
-    print(f"Processing JSONL file: {args.jsonl_file}")
-    print(f"Service URL: {args.service_url}")
-    print(f"Workspace ID: {args.workspace_id}")
-    print(f"Threads: {args.n_threads}")
-
-    with open(args.jsonl_file, "r") as f:
-        data = [json.loads(line) for line in f]
-    print(f"Loaded {len(data)} entries from JSONL file")
-
-    grouped_trajectories = group_trajectories_by_task_id(data)
-    print(f"Total groups: {len(grouped_trajectories)}")
-
-    results = process_trajectories_with_threads(
-        grouped_trajectories,
-        args.service_url,
-        args.workspace_id,
-        n_threads=args.n_threads,
-    )
-
-    print(f"Processed {len(results)} groups")
-
-    success_count = sum(1 for r in results if "error" not in r)
-    error_count = len(results) - success_count
-    total_memories = sum(len(r["metadata"].get("memory_list", [])) for r in results if "memory_list" in r["metadata"])
-
-    print(f"✅ Success: {success_count}")
-    print(f"❌ Errors: {error_count}")
-    print(f"📊 Total task memories created: {total_memories}")
-
-    if args.output_file:
-        try:
-            summary = {
-                "workspace_id": args.workspace_id,
-                "jsonl_file": args.jsonl_file,
-                "total_groups": len(grouped_trajectories),
-                "success_count": success_count,
-                "error_count": error_count,
-                "total_task_memories": total_memories,
-                "results": results,
-            }
-
-            with open(args.output_file, "w") as f:
-                json.dump(summary, f, indent=2)
-            print(f"Results saved to: {args.output_file}")
-        except Exception as e:
-            print(f"Error saving results: {e}")
-
-
-if __name__ == "__main__":
-    import sys
-
-    if len(sys.argv) > 1:
-        main()
-    else:
-        print("Running in compatibility mode...")
-        with open("exp_result/qwen3-8b/with_think/bfcl-multi-turn-base-train50_wo-exp.jsonl", "r") as f:
-            data = [json.loads(line) for line in f]
-
-        grouped_trajectories = group_trajectories_by_task_id(data)
-        print(f"Total groups: {len(grouped_trajectories)}")
-
-        results = process_trajectories_with_threads(
-            grouped_trajectories,
-            "http://localhost:8001",
-            "bfcl_train50_qwen3_8b_extract_compare_validate",
-            n_threads=4,
-        )
-        print(f"Processed {len(results)} groups")

cookbook/bfcl/init_task_memory_pool.py

@@ -1,233 +0,0 @@
-import argparse
-import json
-from collections import defaultdict
-from concurrent.futures import ThreadPoolExecutor, as_completed
-from pathlib import Path
-from typing import List, Dict, Any
-
-import requests
-
-
-def load_task_case(data_path: str, task_id: str | None) -> Dict[str, Any]:
-    """
-    load training cases by id
-    """
-    if not Path(data_path).exists():
-        raise FileNotFoundError(f"BFCL data file '{data_path}' not found")
-
-    if task_id is None:
-        raise ValueError("task_id is required")
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        if str(task_id).isdigit():
-            idx = int(task_id)
-            for line_no, line in enumerate(f):
-                if line_no == idx:
-                    return json.loads(line)
-            raise ValueError(f"Task case index {idx} not found in {data_path}")
-        else:
-            for line in f:
-                data = json.loads(line)
-                if data.get("id") == task_id:
-                    return data
-            raise ValueError(f"Task case id '{task_id}' not found in {data_path}")
-
-
-def get_tool_prompt(tools):
-    tool_prompt = "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>"
-    for tool in tools:
-        tool_prompt += "\n" + json.dumps(tool)
-    tool_prompt += '\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{"name": <function-name>, "arguments": <args-json-object>}\n</tool_call>'
-    return tool_prompt
-
-
-def group_trajectories_by_task_id(jsonl_entries: List[Dict[str, Any]]) -> List[List[Any]]:
-    """
-    group trajectories by task_id
-
-    Args:
-        jsonl_entries: JSONL entry list
-
-    Returns:
-        List[List[Any]]: trajectory list grouped by task_id
-    """
-    grouped = defaultdict(list)
-
-    for entry in jsonl_entries:
-        task_id = entry.get("task_id", "")
-        taks_case = load_task_case("data/multiturn_data_base.jsonl", task_id)
-        tools = taks_case.get("tools", [{}])
-        from bfcl_utils import extract_tool_schema
-
-        tool_schema = extract_tool_schema(tools)
-        entry["task_history"][0]["content"] += get_tool_prompt(tool_schema)
-        grouped[task_id].append(entry)
-
-    # retain only the two with the highest and lowest rewards
-    filtered_groups = []
-    for key, trajectories in grouped.items():
-        if len(trajectories) == 1:
-            # when only one trajectory, retain it
-            filtered_groups.append(trajectories)
-        elif len(trajectories) == 2:
-            # when there are two trajectories, retain them
-            filtered_groups.append(trajectories)
-        else:
-            # when there are more than two trajectories, choose the two with the highest and lowest rewards
-            trajectories.sort(key=lambda t: t["reward"])
-            min_reward_traj = trajectories[0]  # highest reward
-            max_reward_traj = trajectories[-1]  # lowest reward
-            filtered_groups.append([min_reward_traj, max_reward_traj])
-
-    return filtered_groups
-
-
-def post_to_summarizer(trajectories: List[Any], service_url: str, workspace_id: str) -> Dict[str, Any]:
-    trajectory_dicts = [
-        {
-            "task_id": traj["task_id"],
-            "messages": traj["task_history"],
-            "score": traj["reward"],
-        }
-        for traj in trajectories
-    ]
-
-    request_data = {
-        "trajectories": trajectory_dicts,
-        "workspace_id": workspace_id,
-    }
-
-    try:
-        response = requests.post(f"{service_url}/summary_task_memory", json=request_data)
-        response.raise_for_status()
-        return response.json()
-    except Exception as e:
-        return {"error": str(e), "trajectories_count": len(trajectories)}
-
-
-def process_trajectories_with_threads(
-    grouped_trajectories: List[List[Any]],
-    service_url: str,
-    workspace_id: str,
-    n_threads: int = 4,
-) -> List[Dict[str, Any]]:
-    """
-    use threads to process trajectories
-
-    Args:
-        grouped_trajectories: group trajectory list by task_id
-        service_url: memory summarizer service URL
-        workspace_id: workspace ID
-        n_threads: number of threads
-
-    Returns:
-        all results
-    """
-    results = []
-
-    with ThreadPoolExecutor(max_workers=n_threads) as executor:
-        future_to_group = {
-            executor.submit(post_to_summarizer, group, service_url, workspace_id): i
-            for i, group in enumerate(grouped_trajectories)
-        }
-
-        for future in as_completed(future_to_group):
-            group_index = future_to_group[future]
-            try:
-                result = future.result()
-                result["group_index"] = group_index
-                result["group_size"] = len(grouped_trajectories[group_index])
-                results.append(result)
-                print(
-                    f'✅ Group {group_index} processed: {result["metadata"].get("memory_list", 0) if "memory_list" in result["metadata"] else "error"}',
-                )
-            except Exception as e:
-                error_result = {
-                    "group_index": group_index,
-                    "group_size": len(grouped_trajectories[group_index]),
-                    "error": str(e),
-                }
-                results.append(error_result)
-                print(f"❌ Group {group_index} failed: {e}")
-
-    return results
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Convert JSONL to memories using ReMe service")
-    parser.add_argument("--jsonl_file", type=str, required=True, help="Path to the JSONL file")
-    parser.add_argument("--service_url", type=str, default="http://localhost:8001", help="ReMe service URL")
-    parser.add_argument("--workspace_id", type=str, required=True, help="Workspace ID for the task memory pool")
-    parser.add_argument("--output_file", type=str, help="Output file to save results (optional)")
-    parser.add_argument("--n_threads", type=int, default=4, help="Number of threads for processing")
-
-    args = parser.parse_args()
-
-    print(f"Processing JSONL file: {args.jsonl_file}")
-    print(f"Service URL: {args.service_url}")
-    print(f"Workspace ID: {args.workspace_id}")
-    print(f"Threads: {args.n_threads}")
-
-    with open(args.jsonl_file, "r") as f:
-        data = [json.loads(line) for line in f]
-    print(f"Loaded {len(data)} entries from JSONL file")
-
-    grouped_trajectories = group_trajectories_by_task_id(data)
-    print(f"Total groups: {len(grouped_trajectories)}")
-
-    results = process_trajectories_with_threads(
-        grouped_trajectories,
-        args.service_url,
-        args.workspace_id,
-        n_threads=args.n_threads,
-    )
-
-    print(f"Processed {len(results)} groups")
-
-    success_count = sum(1 for r in results if "error" not in r)
-    error_count = len(results) - success_count
-    total_memories = sum(len(r["metadata"].get("memory_list", [])) for r in results if "memory_list" in r["metadata"])
-
-    print(f"✅ Success: {success_count}")
-    print(f"❌ Errors: {error_count}")
-    print(f"📊 Total task memories created: {total_memories}")
-
-    if args.output_file:
-        try:
-            summary = {
-                "workspace_id": args.workspace_id,
-                "jsonl_file": args.jsonl_file,
-                "total_groups": len(grouped_trajectories),
-                "success_count": success_count,
-                "error_count": error_count,
-                "total_task_memories": total_memories,
-                "results": results,
-            }
-
-            with open(args.output_file, "w") as f:
-                json.dump(summary, f, indent=2)
-            print(f"Results saved to: {args.output_file}")
-        except Exception as e:
-            print(f"Error saving results: {e}")
-
-
-if __name__ == "__main__":
-    import sys
-
-    if len(sys.argv) > 1:
-        main()
-    else:
-        print("Running in compatibility mode...")
-        with open("exp_result/qwen3-8b/no_think/bfcl-multi-turn-base_wo-exp.jsonl", "r") as f:
-            data = [json.loads(line) for line in f]
-
-        grouped_trajectories = group_trajectories_by_task_id(data)
-        print(f"Total groups: {len(grouped_trajectories)}")
-
-        results = process_trajectories_with_threads(
-            grouped_trajectories,
-            "http://localhost:8001",
-            "bfcl_v3",
-            n_threads=4,
-        )
-        print(f"Processed {len(results)} groups")

cookbook/bfcl/local_file_to_library.py

@@ -1,28 +0,0 @@
-import json
-
-with open("../../file_vector_store/bfcl_test.jsonl", "r") as f:
-    bfcl = [json.loads(line) for line in f]
-
-new_bfcl = []
-for exp in bfcl:
-    new_exp = {}
-    new_exp["workspace_id"] = exp["workspace_id"]
-    new_exp["memory_id"] = exp["unique_id"]
-    new_exp["memory_type"] = exp["metadata"]["memory_type"]
-
-    new_exp["when_to_use"] = exp["content"]
-    new_exp["content"] = exp["metadata"]["content"]
-    new_exp["score"] = exp["metadata"]["score"]
-
-    new_exp["time_created"] = exp["metadata"]["time_created"]
-    new_exp["time_modified"] = exp["metadata"]["time_modified"]
-    new_exp["author"] = exp["metadata"]["author"]
-
-    new_exp["metadata"] = exp["metadata"]["metadata"]
-    new_exp["metadata"]["utility"] = 0
-    new_exp["metadata"]["freq"] = 0
-
-    new_bfcl.append(new_exp)
-
-with open("../../library/bfcl_test.jsonl", "w", encoding="utf-8") as f:
-    f.writelines(json.dumps(item, ensure_ascii=False) + "\n" for item in new_bfcl)

cookbook/bfcl/requirements.txt

@@ -1,5 +0,0 @@
-jinja2
-loguru
-openai
-ray
-pandas

cookbook/bfcl/run_bfcl.py

@@ -1,126 +0,0 @@
-import time
-
-import ray
-from dotenv import load_dotenv
-
-# from ray import logger
-from loguru import logger
-
-load_dotenv("../../.env")
-
-import json
-from pathlib import Path
-
-
-from bfcl_agent import BFCLAgent
-
-
-def run_agent(
-    dataset_name: str,
-    experiment_suffix: str,
-    max_workers: int,
-    num_trials: int = 1,
-    model_name: str = "qwen3-8b",
-    data_path: str = "data/multiturn_data_base_val.jsonl",
-    answer_path: Path = Path("data/possible_answer"),
-    use_memory: bool = False,
-    use_memory_addition: bool = True,
-    use_memory_deletion: bool = False,
-    delete_freq: int = 10,
-    freq_threshold: int = 5,
-    utility_threshold: float = 0.5,
-    enable_thinking: bool = False,
-    memory_base_url: str = "http://0.0.0.0:8002/",
-    memory_workspace_id: str = "bfcl_v3",
-):
-    experiment_name = dataset_name + "_" + experiment_suffix
-    path: Path = Path(
-        f"./exp_result/{model_name}/with_think" if enable_thinking else f"./exp_result/{model_name}/no_think",
-    )
-    path.mkdir(parents=True, exist_ok=True)
-
-    with open(data_path, "r", encoding="utf-8") as f:
-        task_ids = [json.loads(l)["id"] for l in f]
-
-    result: list = []
-
-    def dump_file():
-        with open(path / f"{experiment_name}.jsonl", "a") as f:
-            for x in result:
-                f.write(json.dumps(x) + "\n")
-
-    future_list: list = []
-    for i in range(max_workers):
-        actor = BFCLAgent.remote(
-            index=i,
-            task_ids=task_ids[i::max_workers],
-            experiment_name=experiment_name,
-            data_path=data_path,
-            answer_path=answer_path,
-            model_name=model_name,
-            num_trials=num_trials,
-            use_memory=use_memory,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=delete_freq,
-            freq_threshold=freq_threshold,
-            utility_threshold=utility_threshold,
-            enable_thinking=enable_thinking,
-            memory_base_url=memory_base_url,
-            memory_workspace_id=memory_workspace_id,
-        )
-        future = actor.execute.remote()
-        future_list.append(future)
-        time.sleep(1)
-    logger.info("submit complete")
-
-    for i, future in enumerate(future_list):
-        t_result = ray.get(future)
-        if t_result:
-            if isinstance(t_result, list):
-                result.extend(t_result)
-            else:
-                result.append(t_result)
-
-        logger.info(f"{i + 1}/{len(task_ids)} complete")
-    dump_file()
-
-
-def main():
-    max_workers = 4
-    num_runs = 1
-
-    num_trials = 2
-    model_name="qwen3-8b"
-    use_memory = False
-    use_memory_addition = False
-    use_memory_deletion = False
-    memory_base_url = "http://0.0.0.0:8002/"
-    memory_workspace_id = "bfcl_v3"
-
-    if max_workers > 1:
-        ray.init(num_cpus=max_workers)
-
-    for run_id in range(num_runs):
-        run_agent(
-            dataset_name="bfcl-multi-turn-base",
-            experiment_suffix=f"wo-exp",
-            model_name=model_name,
-            max_workers=max_workers,
-            num_trials=num_trials,
-            data_path="data/multiturn_data_base_val.jsonl",
-            answer_path=Path("data/possible_answer"),
-            enable_thinking=False,
-            use_memory=use_memory,
-            use_memory_addition=use_memory_addition,
-            use_memory_deletion=use_memory_deletion,
-            delete_freq=5,
-            freq_threshold=5,
-            utility_threshold=0.5,
-            memory_base_url=memory_base_url,
-            memory_workspace_id=memory_workspace_id,
-        )
-
-
-if __name__ == "__main__":
-    main()

cookbook/bfcl/run_exp_statistic.py

@@ -1,159 +0,0 @@
-import json
-from collections import defaultdict
-from pathlib import Path
-
-import pandas as pd
-from loguru import logger
-
-
-def calculate_best_at_k(scores: list, k: int) -> float:
-    """
-    Calculate best@k
-    Divide scores into groups of size k, take the maximum value in each group,
-    then average these maximum values
-
-    Args:
-        scores: List of after_score values for all runs of a task
-        k: Group size
-
-    Returns:
-        best@k value
-    """
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        group_maxs.append(max(group))
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def calculate_pass_at_k(scores: list, k: int) -> float:
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        is_pass = 1.0 if max(group) >= 1.0 else 0.0
-        group_maxs.append(is_pass)
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def get_possible_k_values(total_runs: int) -> list:
-    """
-    Get all possible k values (factors of total_runs)
-
-    Args:
-        total_runs: Total number of runs
-
-    Returns:
-        List of k values in descending order
-    """
-    k_values = []
-    for k in range(1, total_runs + 1):
-        if total_runs % k == 0:
-            k_values.append(k)
-    return sorted(k_values, reverse=True)  # Sort from large to small
-
-
-def run_exp_statistic():
-    path: Path = Path(f"./exp_result/qwen3-8b/no_think")
-
-    # Store results for all experiments
-    all_results = {}
-    for file in [f for f in path.glob("*.jsonl")]:
-        # Group results by task_id
-        task_results = defaultdict(list)
-        print(file)
-        with open(file, "r") as f:
-            for line in f:
-                if not line.strip():
-                    continue
-                data = json.loads(line)
-
-                if isinstance(data, list):
-                    for part_data in data:
-                        task_id = part_data["task_id"]
-                        after_score = part_data["reward"]
-                        task_results[task_id].append(after_score)
-                else:
-                    task_id = data["task_id"]
-                    after_score = data["reward"]
-                    task_results[task_id].append(after_score)
-
-        if not task_results:
-            logger.warning(f"No valid data found in file {file}")
-            continue
-
-        # Check if each task has consistent number of runs
-        run_counts = [len(scores) for scores in task_results.values()]
-        if len(set(run_counts)) > 1:
-            logger.warning(f"Inconsistent number of runs for different tasks in file {file}: {set(run_counts)}")
-            continue
-
-        num_runs = run_counts[0]
-        logger.info(f"File {file}: {len(task_results)} tasks, {num_runs} runs per task")
-
-        # Get all possible k values
-        k_values = get_possible_k_values(num_runs)
-        logger.info(f"Calculable best@k values: {k_values}")
-
-        # Calculate various best@k values
-        file_results = {"file": file.name}
-
-        for k in k_values:
-            best_at_k_scores = []
-            pass_at_k_scores = []
-            for task_id, scores in task_results.items():
-                try:
-                    best_k_score = calculate_best_at_k(scores, k)
-                    pass_at_k_score = calculate_pass_at_k(scores, k)
-                    pass_at_k_scores.append(pass_at_k_score)
-                    best_at_k_scores.append(best_k_score)
-                except ValueError as e:
-                    logger.error(f"Error calculating best@{k} for task {task_id}: {e}")
-                    continue
-
-            if best_at_k_scores:
-                avg_best_at_k = sum(best_at_k_scores) / len(best_at_k_scores)
-                file_results[f"best@{k}"] = avg_best_at_k
-                logger.info(f"file={file.name} best@{k}={avg_best_at_k:.4f}")
-
-            if pass_at_k_scores:
-                avg_pass_at_k = sum(pass_at_k_scores) / len(pass_at_k_scores)
-                file_results[f"pass@{k}"] = avg_pass_at_k
-                logger.info(f"file={file.name} pass@{k}={avg_pass_at_k:.4f}")
-
-        all_results[file.name] = file_results
-
-    # Create and display table
-    if all_results:
-        df = pd.DataFrame(list(all_results.values()))
-        df = df.set_index("file")
-
-        # Sort columns by the number in column name (best@8, best@4, best@2, best@1)
-        # best_columns = [col for col in df.columns if col.startswith('best@')]
-        best_columns = [col for col in df.columns]
-        best_columns.sort(key=lambda x: x, reverse=False)
-        df = df[best_columns]
-
-        print("\n" + "=" * 80)
-        print("Experiment Results Summary Table")
-        print("=" * 80)
-        print(df.round(4))
-        print("=" * 80)
-
-        # Save table to CSV
-        output_path = path / "experiment_summary.csv"
-        df.to_csv(output_path)
-        logger.info(f"Results table saved to: {output_path}")
-    else:
-        logger.warning("No valid experiment results found")
-
-
-if __name__ == "__main__":
-    run_exp_statistic()

cookbook/bfcl/split_into_trainval.py

@@ -1,31 +0,0 @@
-import argparse
-import json
-import random
-
-
-def split_jsonl(input_file, train_file, val_file, ratio=0.8):
-    with open(input_file, "r", encoding="utf-8") as f:
-        data = [json.loads(line) for line in f]
-    random.shuffle(data)
-
-    split_idx = int(len(data) * ratio)
-    train_data = data[:split_idx]
-    val_data = data[split_idx:]
-
-    with open(train_file, "w", encoding="utf-8") as f:
-        for item in train_data:
-            f.write(json.dumps(item, ensure_ascii=False) + "\n")
-    with open(val_file, "w", encoding="utf-8") as f:
-        for item in val_data:
-            f.write(json.dumps(item, ensure_ascii=False) + "\n")
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Split JSONL file into train and validation sets.")
-    parser.add_argument("--input", required=True, help="Path to input JSONL file")
-    parser.add_argument("--train", required=True, help="Path to output train file")
-    parser.add_argument("--val", required=True, help="Path to output validation file")
-    parser.add_argument("--ratio", type=float, default=0.5, help="Train ratio (default: 0.8)")
-
-    args = parser.parse_args()
-    split_jsonl(args.input, args.train, args.val, args.ratio)

cookbook/frozenlake/frozenlake_prompts.yaml

@@ -1,47 +0,0 @@
-frozenlake_sys_prompt_no_slippery: |
-  You are an AI agent playing FrozenLake game. Your goal is to navigate from Start (S) to Goal (G) while avoiding Holes (H).
-
-  Game Rules:
-  - S: Starting position (safe)
-  - F: Frozen surface (safe to walk on)
-  - H: Hole (you fall in and lose)
-  - G: Goal (you win!)
-  - []: Your current position
-
-  Actions:
-  - 0: Move LEFT
-  - 1: Move DOWN
-  - 2: Move RIGHT
-  - 3: Move UP
-
-  Your task: Analyze the current state and choose the best action (0-3) to reach the Goal while avoiding Holes.
-  While ensuring a safe arrival at the goal, you should aim to complete the task in as few steps as possible.
-  Think step by step, and respond with your thoughts and then clearly state your action as a number (0-3) in format {"action":"(0-3)"}.
-
-frozenlake_sys_prompt_slippery: |
-  You are an AI agent playing FrozenLake game. Your goal is to navigate from Start (S) to Goal (G) while avoiding Holes (H).
-
-  Game Rules:
-  - S: Starting position (safe)
-  - F: Frozen surface (safe to walk on)
-  - H: Hole (you fall in and lose)
-  - G: Goal (you win!)
-  - []: Your current position
-
-  Actions:
-  - 0: Move LEFT
-  - 1: Move DOWN
-  - 2: Move RIGHT
-  - 3: Move UP
-
-  The ice is slippery, so you might not always move in the intended direction!
-  you will move in intended direction with probability of 1/3 else will move in either perpendicular direction with equal probability of 1/3 in both directions.
-
-  For example, if action is left, then:
-  - P(move left)=1/3
-  - P(move up)=1/3
-  - P(move down)=1/3
-
-  Your task: Analyze the current state and choose the best action (0-3) to reach the Goal while avoiding Holes.
-  While ensuring a safe arrival at the goal, you should aim to complete the task in as few steps as possible.
-  Think step by step, and respond with your thoughts and then clearly state your action as a number (0-3) in format {{"action":"(0-3)"}}.

cookbook/frozenlake/frozenlake_react_agent.py

@@ -1,378 +0,0 @@
-import random
-import re
-import time
-from dataclasses import dataclass
-from typing import List, Dict, Any
-
-import gymnasium as gym
-import ray
-import requests
-import yaml
-from dotenv import load_dotenv
-from gymnasium.envs.toy_text.frozen_lake import generate_random_map
-from loguru import logger
-from openai import OpenAI
-from tqdm import tqdm
-
-load_dotenv("../../.env")
-
-
-@dataclass
-class GameResult:
-    task_id: str
-    run_id: int
-    experiment_name: str
-    success: bool
-    steps: int
-    reward: float
-    trajectory: List[Dict]
-    map_config: Dict[str, Any]
-
-
-@ray.remote
-class FrozenLakeReactAgent:
-    """A ReAct Agent for FrozenLake game with task memory learning."""
-
-    def __init__(
-        self,
-        index: int,
-        task_configs: List[Dict],
-        experiment_name: str,
-        model_name: str = "qwen3-8b",
-        temperature: float = 0.7,
-        max_steps: int = 50,
-        num_runs: int = 1,
-        use_task_memory: bool = False,
-        make_task_memory: bool = False,
-    ):
-
-        self.index = index
-        self.task_configs = task_configs
-        self.experiment_name = experiment_name
-        self.model_name = model_name
-        self.temperature = temperature
-        self.max_steps = max_steps
-        self.num_runs = num_runs
-        self.use_task_memory = use_task_memory
-        self.make_task_memory = make_task_memory
-
-        self.llm_client = OpenAI()
-        self.action_map = {0: "LEFT", 1: "DOWN", 2: "RIGHT", 3: "UP"}
-
-        # Load prompts
-        self.prompts = self._load_prompts()
-
-    def _load_prompts(self) -> Dict[str, str]:
-        """Load prompts from yaml file"""
-        try:
-            with open("frozenlake_prompts.yaml", "r", encoding="utf-8") as f:
-                return yaml.safe_load(f)
-        except FileNotFoundError:
-            logger.warning("Prompt file not found, using default prompts")
-            raise FileNotFoundError(
-                "Prompt file not found. Please check your current path (should be ./cook/frozenlake) and try again.",
-            )
-
-    def call_llm(self, messages: List[Dict]) -> str:
-        """Call LLM with retry logic"""
-        for i in range(5):
-            try:
-                response = self.llm_client.chat.completions.create(
-                    model=self.model_name,
-                    messages=messages,
-                    temperature=self.temperature,
-                    extra_body={"enable_thinking": False},
-                    seed=0,
-                )
-                return response.choices[0].message.content
-            except Exception as e:
-                logger.warning(f"LLM call failed (attempt {i + 1}): {e}")
-                time.sleep(1 + i * 2)
-        return "LLM call failed"
-
-    def observe_state(self, env, observation: int) -> str:
-        """Convert environment observation to text description"""
-        desc = env.unwrapped.desc
-        nrow, ncol = desc.shape
-
-        # Convert to string grid
-        grid = [[cell.decode("utf-8") for cell in row] for row in desc]
-
-        # Get current position
-        row, col = observation // ncol, observation % ncol
-
-        # Create visual representation
-        state_text = "Current State:\n"
-        for i in range(nrow):
-            for j in range(ncol):
-                if i == row and j == col:
-                    state_text += f"[{grid[i][j]}]"
-                else:
-                    state_text += f" {grid[i][j]} "
-            state_text += "\n"
-
-        state_text += "\nLegend: S=Start, F=Frozen, H=Hole, G=Goal, []=Your Position"
-        return state_text
-
-    def build_system_prompt(self, is_slippery: bool) -> str:
-        """Build system prompt based on game configuration"""
-        if is_slippery:
-            return self.prompts["frozenlake_sys_prompt_slippery"]
-        else:
-            return self.prompts["frozenlake_sys_prompt_no_slippery"]
-
-    def get_task_memory(self, map_desc: str, is_slippery: bool) -> str:
-        """Retrieve relevant task memory from task memory service"""
-        if not self.use_task_memory:
-            return ""
-
-        try:
-            query = f"FrozenLake game map: {map_desc}, slippery: {is_slippery}"
-            base_url = "http://0.0.0.0:8002/"
-            workspace_id = self.experiment_name
-
-            response = requests.post(
-                url=base_url + "retrieve_task_memory",
-                json={
-                    "workspace_id": workspace_id,
-                    "query": query,
-                },
-                timeout=60,
-            )
-
-            if response.status_code == 200:
-                data = response.json()
-                return data.get("answer", "")
-            else:
-                logger.warning(f"Task memory retrieval failed: {response.status_code}")
-                return ""
-
-        except Exception as e:
-            logger.warning(f"Failed to get task memory: {e}")
-            return ""
-
-    def action_parser(self, response: str) -> int:
-        """Parse action from LLM response"""
-        # Look for {"action":"X"} pattern
-        patterns = [
-            r'["\']action["\']\s*:\s*["\']([0-3])["\']',
-            r'"action"\s*:\s*"([0-3])"',
-            r"'action'\s*:\s*'([0-3])'",
-            r'\baction["\']?\s*[:=]\s*["\']?([0-3])',
-        ]
-
-        for pattern in patterns:
-            match = re.search(pattern, response)
-            if match:
-                action = int(match.group(1))
-                if 0 <= action <= 3:
-                    return action
-
-        # Random fallback
-        action = random.randint(0, 3)
-        logger.warning(f"Could not parse action from response, using random: {action}")
-        return action
-
-    def run_single_episode(self, task_config: Dict, run_id: int) -> GameResult:
-        """Run a single episode of the game"""
-        map_size = task_config.get("map_size", 4)
-        is_slippery = task_config.get("is_slippery", True)
-        map_desc = task_config.get("map_desc", None)
-
-        # Create environment
-        env_kwargs = {
-            "render_mode": None,
-            "is_slippery": is_slippery,
-        }
-
-        if map_desc is not None:
-            env_kwargs["desc"] = map_desc
-        else:
-            env_kwargs["desc"] = generate_random_map(size=map_size)
-
-        env = gym.make("FrozenLake-v1", **env_kwargs)
-
-        # Get map description for task memory
-        map_str = "\n".join(["".join([cell.decode("utf-8") for cell in row]) for row in env.unwrapped.desc])
-
-        # Build messages
-        system_prompt = self.build_system_prompt(is_slippery)
-        task_memory = self.get_task_memory(map_str, is_slippery)
-
-        messages = [{"role": "system", "content": system_prompt}]
-
-        if task_memory:
-            memory_content = f"Here are some relevant tips from previous successful games:\n\n{task_memory}\n\nUse these tips to help you succeed."
-            messages.append({"role": "user", "content": memory_content})
-            messages.append(
-                {"role": "assistant", "content": "I'll use these tips to navigate the frozen lake successfully."},
-            )
-
-        # Initialize game
-        observation, info = env.reset()
-        trajectory = []
-
-        # Add initial state
-        initial_state = self.observe_state(env, observation)
-        messages.append({"role": "user", "content": initial_state})
-
-        success = False
-        total_reward = 0
-
-        for step in range(self.max_steps):
-            # Get action from LLM
-            response = self.call_llm(messages)
-            logger.info(response)
-            action = self.action_parser(response)
-
-            messages.append({"role": "assistant", "content": response})
-
-            # Take action
-            next_observation, reward, terminated, truncated, info = env.step(action)
-            total_reward += reward
-            done = terminated or truncated
-
-            # Record trajectory step
-            trajectory.append(
-                {
-                    "step": step,
-                    "state": observation,
-                    "action": action,
-                    "action_name": self.action_map[action],
-                    "reward": reward,
-                    "next_state": next_observation,
-                    "done": done,
-                    "llm_response": response,
-                },
-            )
-
-            if done:
-                if terminated and reward > 0:
-                    success = True
-                    result_msg = f"Success! You reached the goal in {step + 1} steps!"
-                else:
-                    result_msg = f"Game over! You fell into a hole or ran out of time."
-
-                messages.append({"role": "user", "content": result_msg})
-                break
-            else:
-                # Continue game
-                next_state = self.observe_state(env, next_observation)
-                step_msg = f"Step {step + 1}: You moved {self.action_map[action]}. Reward: {reward}\n{next_state}"
-                messages.append({"role": "user", "content": step_msg})
-                observation = next_observation
-
-        env.close()
-
-        # Create result
-        map_id = task_config.get("map_id", f"unknown_{self.index}_{run_id}")
-        task_id = f"{task_config.get('task_type', 'test')}_map{map_id}_{run_id}"
-        result = GameResult(
-            task_id=task_id,
-            run_id=run_id,
-            experiment_name=self.experiment_name,
-            success=success,
-            steps=len(trajectory),
-            reward=total_reward,
-            trajectory=trajectory,
-            map_config={
-                "map_desc": map_str,
-                "map_id": map_id,
-                "is_slippery": is_slippery,
-                "map_size": map_size,
-                "use_task_memory": self.use_task_memory,
-            },
-        )
-
-        return result, messages
-
-    def save_task_memory(self, results: List[GameResult], messages_list: List[List[Dict]]):
-        """Save successful trajectories as task memory"""
-        if not self.make_task_memory:
-            return
-
-        trajectories = []
-        for result, messages in zip(results, messages_list):
-            if result.success:
-                # Create trajectory for task memory service
-                traj = {
-                    "messages": messages,
-                    "score": 1.0,  # Success
-                }
-                trajectories.append(traj)
-            else:
-                traj = {
-                    "messages": messages,
-                    "score": 0.0,  # Failure
-                }
-                trajectories.append(traj)
-
-        if trajectories:
-            try:
-                base_url = "http://0.0.0.0:8002/"
-                workspace_id = self.experiment_name
-
-                response = requests.post(
-                    url=base_url + "summary_task_memory",
-                    json={
-                        "workspace_id": workspace_id,
-                        "trajectories": trajectories,
-                    },
-                    timeout=300,
-                )
-
-                if response.status_code == 200:
-                    logger.info(f"Saved {len(trajectories)} trajectories as task memory")
-                else:
-                    logger.warning(f"Failed to save task memory: {response.status_code}")
-
-            except Exception as e:
-                logger.error(f"Error saving task memory: {e}")
-
-    def execute(self) -> List[Dict]:
-        """Execute all tasks"""
-        all_results = []
-        all_messages = []
-
-        for task_index, task_config in tqdm(enumerate(self.task_configs), desc="Processing tasks:"):
-            for run_id in range(self.num_runs):
-                logger.info(f"Ray {self.index}, Task {task_index}, Run {run_id}")
-
-                result, messages = self.run_single_episode(task_config, run_id)
-                all_results.append(result)
-                all_messages.append(messages)
-
-                # Convert result to dict for JSON serialization
-                result_dict = {
-                    "task_id": result.task_id,
-                    "run_id": result.run_id,
-                    "experiment_name": result.experiment_name,
-                    "task_completed": result.success,
-                    "success": result.success,
-                    "steps": result.steps,
-                    "reward": result.reward,
-                    "map_config": result.map_config,
-                    "trajectory": result.trajectory,
-                }
-                all_results[-1] = result_dict
-
-        # Save task memory if needed
-        if self.make_task_memory:
-            # Convert back to GameResult objects for task memory saving
-            game_results = []
-            for i, result_dict in enumerate(all_results):
-                game_result = GameResult(
-                    task_id=result_dict["task_id"],
-                    run_id=result_dict["run_id"],
-                    experiment_name=result_dict["experiment_name"],
-                    success=result_dict["success"],
-                    steps=result_dict["steps"],
-                    reward=result_dict["reward"],
-                    trajectory=result_dict["trajectory"],
-                    map_config=result_dict["map_config"],
-                )
-                game_results.append(game_result)
-
-            self.save_task_memory(game_results, all_messages)
-
-        return all_results

cookbook/frozenlake/map_manager.py

@@ -1,128 +0,0 @@
-#!/usr/bin/env python3
-"""
-Map Management Tool - Pre-generate and manage test maps
-"""
-
-import json
-from pathlib import Path
-from typing import List, Optional, Dict, Any
-
-import numpy as np
-from gymnasium.envs.toy_text.frozen_lake import generate_random_map
-from loguru import logger
-
-
-class MapManager:
-    """Map Manager - pre-generating, storing and loading test maps"""
-
-    def __init__(self, data_dir: str = "./map/"):
-        self.data_dir = Path(data_dir)
-        self.data_dir.mkdir(parents=True, exist_ok=True)
-
-    def generate_test_maps(
-        self,
-        num_maps: int,
-        map_size: int = 4,
-        base_seed: int = 10000,
-    ) -> str:
-        """
-        Generate test map collection and save
-
-        Args:
-            num_maps: Number of maps to generate
-            map_size: Map size
-            base_seed: Base random seed
-
-        Returns:
-            Path of saved file
-        """
-        logger.info(f"🗺️ Generating {num_maps} test maps (size={map_size})")
-
-        maps_data = []
-        for i in range(num_maps):
-            seed = base_seed + i
-            np.random.seed(seed)
-            map_desc = generate_random_map(size=map_size)
-
-            maps_data.append(
-                {
-                    "map_id": i,
-                    "seed": seed,
-                    "map_size": map_size,
-                    "map_desc": map_desc,  # Convert to list for JSON serialization
-                },
-            )
-
-        # Save to file
-        filename = f"test_maps_{num_maps}_{map_size}x{map_size}.jsonl"
-        filepath = self.data_dir / filename
-
-        with open(filepath, "w", encoding="utf-8") as f:
-            for map_data in maps_data:
-                f.write(json.dumps(map_data, ensure_ascii=False) + "\n")
-
-        logger.info(f"✅ Test maps saved to {filepath}")
-        return str(filepath)
-
-    def load_test_maps(self, filepath: str) -> List[Dict[str, Any]]:
-        """
-        Load test maps
-
-        Args:
-            filepath: Map file path
-
-        Returns:
-            Map data list
-        """
-        if not Path(filepath).exists():
-            raise FileNotFoundError(f"Map file not found: {filepath}")
-
-        maps_data = []
-        with open(filepath, "r", encoding="utf-8") as f:
-            for line in f:
-                if line.strip():
-                    map_data = json.loads(line)
-                    # Convert list back to numpy array
-                    maps_data.append(map_data)
-
-        logger.info(f"📖 Loaded {len(maps_data)} test maps from {filepath}")
-        return maps_data
-
-    def get_map_by_index(self, maps_data: List[Dict], index: int) -> Optional[list]:
-        """Get map by index"""
-        if 0 <= index < len(maps_data):
-            return maps_data[index]["map_desc"]
-        return None
-
-    def get_or_create_test_maps(self, num_maps: int, map_size: int = 4) -> List[Dict[str, Any]]:
-        """
-        Get or create test maps
-        If file exists and has sufficient quantity, load directly; otherwise regenerate
-        """
-        filename = f"test_maps_{num_maps}_{map_size}x{map_size}.jsonl"
-        filepath = self.data_dir / filename
-
-        if filepath.exists():
-            try:
-                maps_data = self.load_test_maps(str(filepath))
-                if len(maps_data) >= num_maps:
-                    logger.info(f"✅ Using existing test maps: {filepath}")
-                    return maps_data[:num_maps]  # Return required number of maps
-            except Exception as e:
-                logger.warning(f"⚠️ Failed to load existing maps: {e}, regenerating...")
-
-        # File doesn't exist or insufficient quantity, regenerate
-        self.generate_test_maps(num_maps, map_size)
-        return self.load_test_maps(str(filepath))
-
-
-if __name__ == "__main__":
-    # Usage example
-    manager = MapManager()
-
-    # Generate 100 4x4 test maps
-    manager.generate_test_maps(num_maps=100, map_size=4)
-
-    # Load and view the first map
-    maps = manager.load_test_maps("./map/test_maps_100_4x4.jsonl")
-    print(f"First map:\n{maps[0]['map_desc']}")

cookbook/frozenlake/run_exp_statistic.py

@@ -1,373 +0,0 @@
-import json
-from collections import defaultdict
-from pathlib import Path
-from typing import Dict, List, Tuple
-
-import pandas as pd
-from loguru import logger
-
-
-def calculate_best_at_k(scores: List[float], k: int) -> float:
-    """
-    Calculate best@k metric.
-    Divide scores into groups of size k, take the maximum value in each group,
-    then average these maximum values.
-
-    Args:
-        scores: List of success scores (0 or 1) for all runs of a task
-        k: Group size
-
-    Returns:
-        best@k value
-    """
-    if len(scores) % k != 0:
-        raise ValueError(f"Length of scores ({len(scores)}) must be divisible by k ({k})")
-
-    group_maxs = []
-    for i in range(0, len(scores), k):
-        group = scores[i : i + k]
-        group_maxs.append(max(group))
-
-    return sum(group_maxs) / len(group_maxs)
-
-
-def get_possible_k_values(total_runs: int) -> List[int]:
-    """Get all possible k values (divisors of total_runs)"""
-    k_values = []
-    for k in range(1, total_runs + 1):
-        if total_runs % k == 0:
-            k_values.append(k)
-    return sorted(k_values, reverse=True)
-
-
-def parse_task_config(task_id: str, map_config: Dict) -> Tuple[str, bool, bool]:
-    """
-    Parse task configuration from task_id and map_config.
-
-    Returns:
-        (condition, is_slippery, use_experience)
-    """
-    is_slippery = map_config.get("is_slippery", True)
-    use_experience = map_config.get("use_experience", False)
-
-    # Create condition string
-    slip_str = "slippery" if is_slippery else "no_slip"
-    exp_str = "with_exp" if use_experience else "no_exp"
-    condition = f"{slip_str}_{exp_str}"
-
-    return condition, is_slippery, use_experience
-
-
-def analyze_frozenlake_results():
-    """Analyze FrozenLake experiment results"""
-    path = Path("./exp_result")
-
-    if not path.exists():
-        logger.error("Experiment results directory not found!")
-        return
-
-    all_results = {}
-
-    # Process all result files
-    for file in path.glob("*test*.jsonl"):
-        logger.info(f"Processing {file.name}")
-
-        # Group results by condition and map
-        condition_results = defaultdict(lambda: defaultdict(list))
-
-        with open(file, "r") as f:
-            for line in f:
-                if not line.strip():
-                    continue
-
-                try:
-                    data = json.loads(line)
-
-                    if isinstance(data, list):
-                        for item in data:
-                            process_single_result(item, condition_results)
-                    else:
-                        process_single_result(data, condition_results)
-
-                except json.JSONDecodeError as e:
-                    logger.warning(f"Invalid JSON in {file.name}: {e}")
-                    continue
-
-        if not condition_results:
-            logger.warning(f"No valid data found in {file.name}")
-            continue
-
-        # Calculate metrics for this file
-        file_metrics = calculate_file_metrics(condition_results, file.name)
-        all_results[file.name] = file_metrics
-
-    # Generate comprehensive report
-    if all_results:
-        generate_analysis_report(all_results)
-    else:
-        logger.warning("No valid results found!")
-
-
-def process_single_result(data: Dict, condition_results: Dict):
-    """Process a single result entry"""
-    map_config = data.get("map_config", {})
-    task_id = data.get("task_id", "unknown")
-    success = data.get("success", False)
-
-    # Parse condition
-    condition, is_slippery, use_experience = parse_task_config(task_id, map_config)
-
-    # Extract map identifier - prefer map_id from map_config
-    map_id = map_config.get("map_id", "unknown")
-    if map_id == "unknown" and "test_map" in task_id:
-        # Fallback to parsing from task_id
-        parts = task_id.split("_")
-        for part in parts:
-            if part.startswith("map"):
-                try:
-                    # Extract number from "mapXX"
-                    map_num = "".join(filter(str.isdigit, part))
-                    if map_num:
-                        map_id = int(map_num)
-                        break
-                except:
-                    pass
-
-    # Store result
-    success_score = 1.0 if success else 0.0
-    condition_results[condition][f"map_{map_id}"].append(success_score)
-
-
-def calculate_file_metrics(condition_results: Dict, filename: str) -> Dict:
-    """Calculate metrics for a single file"""
-    file_metrics = {"file": filename}
-
-    for condition, map_results in condition_results.items():
-        condition_scores = []
-
-        # Collect all scores for this condition
-        for map_id, scores in map_results.items():
-            condition_scores.extend(scores)
-
-        if not condition_scores:
-            continue
-
-        # Check if all maps have the same number of runs
-        run_counts = [len(scores) for scores in map_results.values()]
-        if len(set(run_counts)) > 1:
-            logger.warning(f"Inconsistent runs for {condition}: {set(run_counts)}")
-            continue
-
-        num_runs = run_counts[0] if run_counts else 0
-        if num_runs == 0:
-            continue
-
-        # Calculate overall success rate
-        overall_success = sum(condition_scores) / len(condition_scores)
-        file_metrics[f"{condition}_success_rate"] = overall_success
-
-        # Calculate best@k metrics
-        k_values = get_possible_k_values(num_runs)
-        for k in k_values:
-            try:
-                # Calculate best@k for each map, then average
-                map_best_k_scores = []
-                for map_id, scores in map_results.items():
-                    map_best_k = calculate_best_at_k(scores, k)
-                    map_best_k_scores.append(map_best_k)
-
-                avg_best_k = sum(map_best_k_scores) / len(map_best_k_scores)
-                file_metrics[f"{condition}_best@{k}"] = avg_best_k
-
-            except ValueError as e:
-                logger.warning(f"Error calculating best@{k} for {condition}: {e}")
-
-        # Map-level analysis
-        map_success_rates = {}
-        for map_id, scores in map_results.items():
-            map_success_rate = sum(scores) / len(scores)
-            map_success_rates[map_id] = map_success_rate
-
-        file_metrics[f"{condition}_map_details"] = map_success_rates
-
-        logger.info(
-            f"{filename} - {condition}: {overall_success:.3f} success rate, "
-            f"{len(map_results)} maps, {num_runs} runs each",
-        )
-
-    return file_metrics
-
-
-def generate_analysis_report(all_results: Dict):
-    """Generate comprehensive analysis report"""
-    logger.info("Generating comprehensive analysis report...")
-
-    # 1. Create summary table
-    summary_data = []
-    for file_name, metrics in all_results.items():
-        row = {"file": file_name}
-
-        # Extract success rates and best@k metrics
-        for key, value in metrics.items():
-            if key != "file" and not key.endswith("_map_details"):
-                row[key] = value
-
-        summary_data.append(row)
-
-    if summary_data:
-        df_summary = pd.DataFrame(summary_data)
-        df_summary = df_summary.set_index("file")
-
-        print("\n" + "=" * 100)
-        print("FROZENLAKE EXPERIMENT RESULTS SUMMARY")
-        print("=" * 100)
-        print(df_summary.round(4))
-        print("=" * 100)
-
-        # Save summary table
-        output_path = Path("./exp_result") / "frozenlake_summary.csv"
-        df_summary.to_csv(output_path)
-        logger.info(f"Summary table saved to: {output_path}")
-
-    # 2. Condition comparison
-    print("\n" + "=" * 80)
-    print("CONDITION COMPARISON")
-    print("=" * 80)
-
-    condition_comparison = defaultdict(list)
-
-    for file_name, metrics in all_results.items():
-        for key, value in metrics.items():
-            if "_success_rate" in key:
-                condition = key.replace("_success_rate", "")
-                condition_comparison[condition].append(value)
-
-    # Calculate average performance per condition
-    condition_avg = {}
-    for condition, scores in condition_comparison.items():
-        if scores:
-            avg_score = sum(scores) / len(scores)
-            condition_avg[condition] = avg_score
-            print(f"{condition:20s}: {avg_score:.4f} (±{pd.Series(scores).std():.4f})")
-
-    # 3. Experience effect analysis
-    print("\n" + "=" * 80)
-    print("EXPERIENCE EFFECT ANALYSIS")
-    print("=" * 80)
-
-    experience_analysis = analyze_experience_effect(condition_avg)
-    for analysis_line in experience_analysis:
-        print(analysis_line)
-
-    # 4. Map difficulty analysis
-    print("\n" + "=" * 80)
-    print("MAP DIFFICULTY ANALYSIS")
-    print("=" * 80)
-
-    map_analysis = analyze_map_difficulty(all_results)
-    for map_id, difficulty in map_analysis.items():
-        print(f"{map_id:10s}: {difficulty:.4f} average success rate")
-
-    # 5. Detailed statistics
-    print("\n" + "=" * 80)
-    print("DETAILED STATISTICS")
-    print("=" * 80)
-
-    generate_detailed_stats(all_results)
-
-
-def analyze_experience_effect(condition_avg: Dict[str, float]) -> List[str]:
-    """Analyze the effect of experience on performance"""
-    analysis = []
-
-    # Compare with/without experience for each slippery condition
-    slippery_no_exp = condition_avg.get("slippery_no_exp", 0)
-    slippery_with_exp = condition_avg.get("slippery_with_exp", 0)
-    no_slip_no_exp = condition_avg.get("no_slip_no_exp", 0)
-    no_slip_with_exp = condition_avg.get("no_slip_with_exp", 0)
-
-    if slippery_no_exp > 0 and slippery_with_exp > 0:
-        improvement_slippery = (slippery_with_exp - slippery_no_exp) / slippery_no_exp * 100
-        analysis.append(f"Slippery condition - Experience effect: {improvement_slippery:+.1f}%")
-        analysis.append(f"  Without exp: {slippery_no_exp:.4f}")
-        analysis.append(f"  With exp:    {slippery_with_exp:.4f}")
-
-    if no_slip_no_exp > 0 and no_slip_with_exp > 0:
-        improvement_no_slip = (no_slip_with_exp - no_slip_no_exp) / no_slip_no_exp * 100
-        analysis.append(f"No-slip condition - Experience effect: {improvement_no_slip:+.1f}%")
-        analysis.append(f"  Without exp: {no_slip_no_exp:.4f}")
-        analysis.append(f"  With exp:    {no_slip_with_exp:.4f}")
-
-    # Overall experience effect
-    exp_conditions = [v for k, v in condition_avg.items() if "with_exp" in k]
-    no_exp_conditions = [v for k, v in condition_avg.items() if "no_exp" in k]
-
-    if exp_conditions and no_exp_conditions:
-        avg_with_exp = sum(exp_conditions) / len(exp_conditions)
-        avg_without_exp = sum(no_exp_conditions) / len(no_exp_conditions)
-        overall_improvement = (avg_with_exp - avg_without_exp) / avg_without_exp * 100
-        analysis.append(f"Overall experience effect: {overall_improvement:+.1f}%")
-
-    return analysis
-
-
-def analyze_map_difficulty(all_results: Dict) -> Dict[str, float]:
-    """Analyze difficulty of different maps"""
-    map_scores = defaultdict(list)
-
-    for file_name, metrics in all_results.items():
-        for key, value in metrics.items():
-            if key.endswith("_map_details") and isinstance(value, dict):
-                for map_id, success_rate in value.items():
-                    map_scores[map_id].append(success_rate)
-
-    # Calculate average difficulty per map
-    map_difficulty = {}
-    for map_id, scores in map_scores.items():
-        if scores:
-            avg_success = sum(scores) / len(scores)
-            map_difficulty[map_id] = avg_success
-
-    # Sort by difficulty (hardest first)
-    return dict(sorted(map_difficulty.items(), key=lambda x: x[1]))
-
-
-def generate_detailed_stats(all_results: Dict):
-    """Generate detailed statistics"""
-    total_experiments = len(all_results)
-    total_conditions = set()
-
-    for metrics in all_results.values():
-        for key in metrics.keys():
-            if "_success_rate" in key:
-                condition = key.replace("_success_rate", "")
-                total_conditions.add(condition)
-
-    print(f"Total experiment files: {total_experiments}")
-    print(f"Total conditions tested: {len(total_conditions)}")
-    print(f"Conditions: {', '.join(sorted(total_conditions))}")
-
-    # Best performing conditions
-    all_success_rates = []
-    for metrics in all_results.values():
-        for key, value in metrics.items():
-            if "_success_rate" in key and isinstance(value, (int, float)):
-                all_success_rates.append((key.replace("_success_rate", ""), value))
-
-    if all_success_rates:
-        best_condition = max(all_success_rates, key=lambda x: x[1])
-        worst_condition = min(all_success_rates, key=lambda x: x[1])
-
-        print(f"Best performance: {best_condition[0]} ({best_condition[1]:.4f})")
-        print(f"Worst performance: {worst_condition[0]} ({worst_condition[1]:.4f})")
-
-
-def main():
-    """Main function for statistics analysis"""
-    logger.info("🔍 Starting FrozenLake Results Analysis")
-    analyze_frozenlake_results()
-    logger.info("📊 Analysis completed!")
-
-
-if __name__ == "__main__":
-    main()

cookbook/frozenlake/run_frozenlake.py

@@ -1,310 +0,0 @@
-import json
-import time
-from pathlib import Path
-from typing import List, Dict
-
-import numpy as np
-import ray
-from gymnasium.envs.toy_text.frozen_lake import generate_random_map
-from loguru import logger
-
-from frozenlake_react_agent import FrozenLakeReactAgent
-from map_manager import MapManager
-
-
-def generate_training_configs(num_maps: int = 20, map_size: int = 4, is_slippery: bool = False) -> List[Dict]:
-    """Generate random maps for training/task memory generation"""
-    configs = []
-
-    for i in range(num_maps):
-        # Generate both slippery and non-slippery versions
-        random_map = generate_random_map(size=map_size)
-
-        config = {
-            "task_type": "training",
-            "map_desc": random_map,
-            "map_size": map_size,
-            "is_slippery": is_slippery,
-            "task_id": f"train_{i}_{is_slippery}",
-        }
-        configs.append(config)
-
-    return configs
-
-
-def generate_test_configs(num_test_maps: int = 100, is_slippery: bool = False) -> List[Dict]:
-    """Generate test configurations using MapManager"""
-    logger.info(f"📋 Generating test configurations for {num_test_maps} maps")
-
-    # Initialize MapManager and get test maps
-    map_manager = MapManager()
-    maps_data = map_manager.get_or_create_test_maps(num_maps=num_test_maps, map_size=4)
-
-    configs = []
-
-    for map_data in maps_data:
-        map_desc = np.array([list(row) for row in map_data["map_desc"]], dtype="c")
-        map_id = map_data["map_id"]
-
-        for use_memory in [True, False]:
-            config = {
-                "task_type": "test",
-                "map_desc": map_desc,
-                "map_size": 4,
-                "is_slippery": is_slippery,
-                "use_task_memory": use_memory,
-                "map_id": map_id,
-                "task_id": f"test_map{map_id}_slip{is_slippery}_mem{use_memory}",
-            }
-            configs.append(config)
-
-    logger.info(f"✅ Generated {len(configs)} test configurations")
-    return configs
-
-
-def train(
-    experiment_name: str,
-    max_workers: int = 2,
-    num_runs: int = 3,
-    num_training_maps=15,
-    is_slippery: bool = False,
-) -> None:
-    """Phase 1: Generate task memory from random maps"""
-    logger.info("🎯 Starting Training Phase - Generating Task Memory")
-    logger.info("=" * 60)
-
-    training_configs = generate_training_configs(num_maps=num_training_maps, map_size=4, is_slippery=is_slippery)
-    path = Path("./exp_result")
-    path.mkdir(parents=True, exist_ok=True)
-
-    results = []
-
-    def dump_results():
-        output_file = path / f"{experiment_name}_training.jsonl"
-        with open(output_file, "w") as f:
-            for result in results:
-                f.write(json.dumps(result) + "\n")
-        logger.info(f"Training results saved to {output_file}")
-
-    if max_workers > 1:
-        # Distributed training
-        future_list = []
-        for i in range(max_workers):
-            worker_configs = training_configs[i::max_workers]
-            if worker_configs:  # Only create worker if it has tasks
-                agent = FrozenLakeReactAgent.remote(
-                    index=i,
-                    task_configs=worker_configs,
-                    experiment_name=experiment_name,
-                    num_runs=num_runs,
-                    use_task_memory=False,  # No task memory in training phase
-                    make_task_memory=True,  # Generate task memory
-                )
-                future = agent.execute.remote()
-                future_list.append(future)
-                time.sleep(1)
-
-        logger.info(f"Started {len(future_list)} training workers")
-
-        for i, future in enumerate(future_list):
-            worker_results = ray.get(future)
-            if worker_results:
-                results.extend(worker_results)
-                logger.info(f"results: {results[0]}")
-            logger.info(f"Training worker {i + 1}/{len(future_list)} completed")
-            dump_results()
-
-    else:
-        # Single process training
-        agent = FrozenLakeReactAgent(
-            index=0,
-            task_configs=training_configs,
-            experiment_name=experiment_name,
-            num_runs=num_runs,
-            use_task_memory=False,
-            make_task_memory=True,
-        )
-        results = agent.execute()
-        dump_results()
-
-    # Calculate training statistics
-    successful_runs = [r for r in results if r["success"]]
-    total_runs = len(results)
-    success_rate = len(successful_runs) / total_runs if total_runs > 0 else 0
-
-    logger.info(f"Training completed: {len(successful_runs)}/{total_runs} successful ({success_rate:.2%})")
-    return results
-
-
-def test(
-    experiment_name: str,
-    max_workers: int = 2,
-    num_runs: int = 5,
-    num_test_maps: int = 100,
-    is_slippery: bool = False,
-) -> None:
-    """Phase 2: Test on fixed maps with/without task memory"""
-    logger.info("🧪 Starting Test Phase - Evaluating Performance")
-    logger.info(f"📊 Testing on {num_test_maps} maps with {num_runs} runs each")
-    logger.info("=" * 60)
-
-    test_configs = generate_test_configs(num_test_maps=num_test_maps, is_slippery=is_slippery)
-    path = Path("./exp_result")
-    path.mkdir(parents=True, exist_ok=True)
-
-    # Group configs by task memory usage for separate experiments
-    memory_configs = [c for c in test_configs if c.get("use_task_memory", False)]
-    no_memory_configs = [c for c in test_configs if not c.get("use_task_memory", False)]
-
-    logger.info(f"📝 Configs without task memory: {len(no_memory_configs)}")
-    logger.info(f"📝 Configs with task memory: {len(memory_configs)}")
-
-    def dump_results(suffix: str):
-        output_file = path / f"{experiment_name}_test_{suffix}.jsonl"
-        with open(output_file, "w") as f:
-            for result in all_results:
-                f.write(json.dumps(result) + "\n")
-        logger.info(f"💾 Test results saved to {output_file}")
-
-    # Test without task memory first
-    logger.info("🚫 Testing WITHOUT task memory...")
-    all_results = []
-    results_no_memory = run_test_configs(
-        configs=no_memory_configs,
-        experiment_name=experiment_name,
-        max_workers=max_workers,
-        num_runs=num_runs,
-        use_task_memory=False,
-    )
-    all_results.extend(results_no_memory)
-    dump_results("no_memory")
-
-    # Test with task memory
-    logger.info("✅ Testing WITH task memory...")
-    all_results = []
-    results_with_memory = run_test_configs(
-        configs=memory_configs,
-        experiment_name=experiment_name,
-        max_workers=max_workers,
-        num_runs=num_runs,
-        use_task_memory=True,
-    )
-    all_results.extend(results_with_memory)
-    dump_results("with_memory")
-
-    return all_results
-
-
-def run_test_configs(
-    configs: List[Dict],
-    experiment_name: str,
-    max_workers: int,
-    num_runs: int,
-    use_task_memory: bool,
-) -> List[Dict]:
-    """Run a set of test configurations"""
-    results = []
-
-    if max_workers > 1:
-        future_list = []
-        for i in range(max_workers):
-            worker_configs = configs[i::max_workers]
-            if worker_configs:
-                agent = FrozenLakeReactAgent.remote(
-                    index=i,
-                    task_configs=worker_configs,
-                    experiment_name=experiment_name,
-                    num_runs=num_runs,
-                    use_task_memory=use_task_memory,
-                    make_task_memory=False,
-                )
-                future = agent.execute.remote()
-                future_list.append(future)
-                time.sleep(1)
-
-        for i, future in enumerate(future_list):
-            worker_results = ray.get(future)
-            if worker_results:
-                results.extend(worker_results)
-            logger.info(f"Test worker {i + 1}/{len(future_list)} completed")
-
-    else:
-        agent = FrozenLakeReactAgent(
-            index=0,
-            task_configs=configs,
-            experiment_name=experiment_name,
-            num_runs=num_runs,
-            use_task_memory=use_task_memory,
-            make_task_memory=False,
-        )
-        results = agent.execute()
-
-    return results
-
-
-def main():
-    """Main execution function"""
-    experiment_name = "frozenlake_no_slippery"
-    max_workers = 4
-    training_runs = 4  # Runs per training map
-    num_training_maps = 50
-    test_runs = 1  # Runs per test configuration
-    num_test_maps = 100  # Number of test maps to use
-    is_slippery = False
-    # model_name = "qwen-max-latest"
-
-    # Initialize Ray if using multiple workers
-    if max_workers > 1:
-        ray.init(num_cpus=max_workers)
-
-    try:
-        # Phase 1: Training (Experience Generation)
-        logger.info("🚀 Starting FrozenLake Experiment")
-        logger.info(f"🎯 Experiment: {experiment_name}")
-        logger.info(f"🏃 Workers: {max_workers}")
-        logger.info(f"📊 Test maps: {num_test_maps}")
-        logger.info(f"🔄 Test runs per map: {test_runs}")
-
-        training_results = train(
-            experiment_name=experiment_name,
-            max_workers=max_workers,
-            num_runs=training_runs,
-            num_training_maps=num_training_maps,
-            is_slippery=is_slippery,
-        )
-
-        # Wait a bit for task memory service to process
-        logger.info("⏰ Waiting for task memory service to process data...")
-        time.sleep(10)
-
-        # Phase 2: Testing (Performance Evaluation)
-        test_results = test(
-            experiment_name=experiment_name,
-            max_workers=max_workers,
-            num_runs=test_runs,
-            num_test_maps=num_test_maps,
-            is_slippery=is_slippery,
-        )
-
-        # Summary
-        logger.info("🎉 Experiment completed!")
-        logger.info(f"📈 Training results: {len(training_results)} episodes")
-        logger.info(f"📈 Test results: {len(test_results)} episodes")
-
-        # Quick statistics
-        successful_training = sum(1 for r in training_results if r.get("success", False))
-        training_success_rate = successful_training / len(training_results) if training_results else 0
-
-        successful_test = sum(1 for r in test_results if r.get("success", False))
-        test_success_rate = successful_test / len(test_results) if test_results else 0
-
-        logger.info(f"📊 Training success rate: {training_success_rate:.2%}")
-        logger.info(f"📊 Test success rate: {test_success_rate:.2%}")
-
-    finally:
-        if max_workers > 1:
-            ray.shutdown()
-
-
-if __name__ == "__main__":
-    main()