fix(ci): bugbot — persist compat-results artifacts from PR gate

The conftest's pytest_sessionfinish writes per-cell tagged-union JSON
(compat-results.json) and the per-provider rate-limit summary to paths
controlled by COMPAT_RESULTS_PATH / COMPAT_RATE_LIMIT_SUMMARY_PATH.
The PR gate never set either env var, so the artifacts were written to
the working directory — but store_test_results only collects
test-results/junit.xml, leaving the per-cell JSON unreachable from the
CircleCI artifact browser. Reviewers triaging a red PR gate couldn't
pull the cell-level breakdown without re-running.

Point both env vars at a dedicated compat-artifacts/ directory and add
a store_artifacts step so the JSON blobs become downloadable. Also
extend the existing PR-gate wiring test to pin all three pieces:
COMPAT_RESULTS_PATH override, COMPAT_RATE_LIMIT_SUMMARY_PATH override,
and at least one store_artifacts step whose path matches the directory
the exports point at. Without the cross-check, a future refactor could
break the chain (e.g. only export the env vars, or only add the
store_artifacts step) and the wiring would silently drop the artifacts
again.

Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>

.circleci/config.yml

@@ -2370,7 +2370,15 @@ jobs:
           command: |
             export LITELLM_PROXY_BASE_URL="http://localhost:4000"
             export LITELLM_PROXY_API_KEY="sk-1234"
-            mkdir -p test-results
+            mkdir -p test-results compat-artifacts
+            # Point the conftest's tagged-union artifacts at a directory
+            # we explicitly persist below via store_artifacts. Without
+            # this override, the conftest defaults to compat-results.json
+            # / compat-rate-limit-summary.json in the working directory,
+            # which no CI step collects -- so reviewers can't pull the
+            # per-cell pass/fail breakdown when a red gate needs triage.
+            export COMPAT_RESULTS_PATH="$(pwd)/compat-artifacts/compat-results.json"
+            export COMPAT_RATE_LIMIT_SUMMARY_PATH="$(pwd)/compat-artifacts/compat-rate-limit-summary.json"
             uv run --no-sync python -m pytest -vv tests/claude_code/ \
               --ignore=tests/claude_code/_driver_unit_tests \
               --ignore=tests/claude_code/_builder_unit_tests \
@@ -2380,9 +2388,17 @@ jobs:
               --durations=10
           no_output_timeout: 30m
 
-      # Store test results
+      # Store test results (JUnit) for CircleCI's test-summary view.
       - store_test_results:
           path: test-results
+      # Persist the compat-matrix artifacts (per-cell tagged-union JSON
+      # + per-provider rate-limit summary) so a red gate can be triaged
+      # from the CircleCI UI without re-running. `store_test_results`
+      # only handles JUnit-shaped XML, so these JSON blobs need their
+      # own store_artifacts step.
+      - store_artifacts:
+          path: compat-artifacts
+          destination: compat-artifacts
 
   upload-coverage:
     docker:

tests/claude_code/_pr_gate_unit_tests/test_circleci_pr_gate_wiring.py

@@ -113,6 +113,49 @@ def test_pr_gate_job_exports_proxy_env_used_by_tests(
     assert "LITELLM_PROXY_API_KEY" in commands
 
 
+def test_pr_gate_job_persists_compat_result_artifacts(
+    circleci_config: dict,
+) -> None:
+    """The conftest writes per-cell tagged-union JSON + the per-provider
+    rate-limit summary to paths controlled by COMPAT_RESULTS_PATH /
+    COMPAT_RATE_LIMIT_SUMMARY_PATH. The PR gate must (1) point both env
+    vars at a known directory, and (2) `store_artifacts` that directory
+    so reviewers can pull the breakdown when a red gate needs triage.
+
+    Without this, the conftest silently writes the artifacts to the
+    working directory and no CI step persists them.
+    """
+    job = circleci_config["jobs"][JOB_NAME]
+    commands = "\n".join(_job_step_runs(job))
+    assert "COMPAT_RESULTS_PATH" in commands, (
+        "PR gate must override COMPAT_RESULTS_PATH so the compat JSON "
+        "lands in a directory we explicitly persist below."
+    )
+    assert "COMPAT_RATE_LIMIT_SUMMARY_PATH" in commands, (
+        "PR gate must override COMPAT_RATE_LIMIT_SUMMARY_PATH so the "
+        "per-provider rate-limit summary lands in a persisted directory."
+    )
+    # The chosen directory must be wired into a store_artifacts step.
+    store_artifacts_paths: list[str] = []
+    for step in job.get("steps", []):
+        if not isinstance(step, dict):
+            continue
+        sa = step.get("store_artifacts")
+        if isinstance(sa, dict) and isinstance(sa.get("path"), str):
+            store_artifacts_paths.append(sa["path"])
+    assert store_artifacts_paths, (
+        "PR gate must declare at least one store_artifacts step so the "
+        "compat-results.json / compat-rate-limit-summary.json can be "
+        "downloaded from the CircleCI artifact browser."
+    )
+    # And the export must point at that directory (any of them), so the
+    # conftest actually writes inside the persisted tree.
+    assert any(path in commands for path in store_artifacts_paths), (
+        "PR gate exports COMPAT_RESULTS_PATH but not into any directory "
+        f"declared by store_artifacts (declared: {store_artifacts_paths})."
+    )
+
+
 def test_existing_proxy_e2e_anthropic_job_unchanged(circleci_config: dict) -> None:
     """No regression to the existing `proxy_e2e_anthropic_messages_tests`
     job (acceptance criterion). We don't lock its full body, but we do