Merge pull request #41494 from BerriAI/litellm_auto_merge_price_sync

ci: auto-merge provider-info-sync PRs when CI, Greptile and Bugbot are clean

.circleci/scripts/classify_changes.sh

@@ -1,12 +1,14 @@
 #!/usr/bin/env bash
 set -uo pipefail
 
-category="${1:?usage: classify_changes.sh <backend|client|ui|provider-harness>}"
+category="${1:?usage: classify_changes.sh <backend|client|ui|provider-harness|cost-map-only>}"
 
 has_client=false
 has_backend=false
 has_ci=false
 has_provider_harness=false
+has_cost_map=false
+outside_cost_map_set=false
 while IFS= read -r file || [ -n "$file" ]; do
   [ -n "$file" ] || continue
   case "$file" in
@@ -20,9 +22,18 @@ while IFS= read -r file || [ -n "$file" ]; do
     .github/* | .circleci/*) has_ci=true; has_backend=true ;;
     *) has_backend=true ;;
   esac
+  case "$file" in
+    model_prices_and_context_window.json | litellm/model_prices_and_context_window_backup.json | model_prices_and_context_window.schema.json)
+      has_cost_map=true ;;
+    tests/test_litellm/* | tests/proxy_unit_tests/*) : ;;
+    *) outside_cost_map_set=true ;;
+  esac
 done
 
 case "$category" in
+  cost-map-only)
+    { [ "$has_cost_map" = true ] && [ "$outside_cost_map_set" = false ]; } && echo run || echo skip
+    ;;
   provider-harness)
     [ "$has_provider_harness" = true ] && echo run || echo skip
     ;;

.github/scripts/auto_merge_price_sync.py

@@ -0,0 +1,465 @@
+"""Auto-merge the provider-info-sync bot's cost-map pull requests.
+
+Evaluates every gate (author allowlist, cost-map-only diff, required and
+non-required checks, Greptile confidence, Bugbot review, human reviews) and
+merges with a merge commit when all of them hold. Every hold reason is
+logged; the process exits 0 on hold and 1 only on API or programming errors.
+``DRY_RUN=1`` prints the verdict without calling the merge endpoint.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+import subprocess
+import sys
+import time
+import urllib.error
+import urllib.request
+from collections.abc import Callable, Mapping, Sequence
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from typing import Final
+
+REPO_ROOT: Final = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+CLASSIFY_SCRIPT: Final = os.path.join(REPO_ROOT, ".circleci", "scripts", "classify_changes.sh")
+API_ROOT: Final = "https://api.github.com"
+CHANGED_FILE_CEILING: Final = 3000
+OK_CHECK_CONCLUSIONS: Final = frozenset({"success", "skipped", "neutral"})
+GREPTILE_LOGIN: Final = "greptile-apps[bot]"
+BUGBOT_LOGIN: Final = "cursor[bot]"
+GREPTILE_SCORE_RE: Final = re.compile(r"Confidence Score:\s*(\d)/5")
+BUGBOT_REVIEW_MARKER: Final = "<!-- BUGBOT_REVIEW -->"
+BUGBOT_STALE_MARKER: Final = "<!-- BUGBOT_REVIEW_STALE -->"
+BUGBOT_CLEAN: Final = "found no new issues"
+
+
+@dataclass(frozen=True, slots=True)
+class PullRequest:
+    number: int
+    title: str
+    author_login: str
+    state: str
+    draft: bool
+    mergeable: bool | None
+    mergeable_state: str
+    head_sha: str
+
+
+@dataclass(frozen=True, slots=True)
+class CheckRun:
+    name: str
+    status: str
+    conclusion: str | None
+
+
+@dataclass(frozen=True, slots=True)
+class CommitStatus:
+    context: str
+    state: str
+
+
+@dataclass(frozen=True, slots=True)
+class IssueComment:
+    author_login: str
+    body: str
+    updated_at: datetime
+
+
+@dataclass(frozen=True, slots=True)
+class Review:
+    author_login: str
+    state: str
+    body: str
+    commit_id: str
+    submitted_at: datetime
+
+
+@dataclass(frozen=True, slots=True)
+class Verdict:
+    merge: bool
+    reasons: tuple[str, ...]
+
+
+@dataclass(frozen=True, slots=True)
+class EvaluationInputs:
+    pr: PullRequest
+    changed_files: tuple[str, ...]
+    required_contexts: frozenset[str]
+    check_runs: tuple[CheckRun, ...]
+    statuses: tuple[CommitStatus, ...]
+    comments: tuple[IssueComment, ...]
+    reviews: tuple[Review, ...]
+    head_commit_date: datetime
+    self_check_name: str
+    author_allowlist: frozenset[str]
+
+
+def _is_bot_login(login: str) -> bool:
+    return login.lower().endswith("[bot]")
+
+
+def _classify(changed_files: Sequence[str]) -> str:
+    result: Final = subprocess.run(
+        ["bash", CLASSIFY_SCRIPT, "cost-map-only"],
+        input="\n".join(changed_files),
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    if result.returncode != 0:
+        return "error"
+    return result.stdout.strip()
+
+
+def evaluate(
+    inputs: EvaluationInputs,
+    *,
+    classify: Callable[[Sequence[str]], str] = _classify,
+) -> Verdict:
+    pr: Final = inputs.pr
+    reasons: list[str] = []
+
+    if pr.author_login.lower() not in {login.lower() for login in inputs.author_allowlist}:
+        reasons.append(f"author {pr.author_login!r} not in allowlist")
+    if pr.state != "open":
+        reasons.append("pr not open")
+    if pr.draft:
+        reasons.append("pr is a draft")
+    if pr.mergeable is None:
+        reasons.append("mergeability unknown")
+    elif not pr.mergeable:
+        reasons.append("pr not mergeable")
+    if pr.mergeable_state == "dirty":
+        reasons.append("pr has merge conflicts")
+
+    if len(inputs.changed_files) > CHANGED_FILE_CEILING:
+        reasons.append(f"changed file count {len(inputs.changed_files)} over {CHANGED_FILE_CEILING} ceiling")
+    else:
+        decision: Final = classify(inputs.changed_files)
+        if decision != "run":
+            reasons.append("changed files outside the cost-map-only set")
+
+    green_runs: Final = frozenset(run.name for run in inputs.check_runs if run.conclusion in OK_CHECK_CONCLUSIONS)
+    green_statuses: Final = frozenset(status.context for status in inputs.statuses if status.state == "success")
+    for context in sorted(inputs.required_contexts):
+        if context not in green_runs and context not in green_statuses:
+            reasons.append(f"required check {context!r} not green")
+    for run in inputs.check_runs:
+        if run.name == inputs.self_check_name:
+            continue
+        if run.status != "completed" or run.conclusion not in OK_CHECK_CONCLUSIONS:
+            reasons.append(f"check run {run.name!r} is {run.status}/{run.conclusion}")
+    for status in inputs.statuses:
+        if status.state != "success":
+            reasons.append(f"commit status {status.context!r} is {status.state}")
+
+    greptile: Final = tuple(
+        comment
+        for comment in inputs.comments
+        if comment.author_login == GREPTILE_LOGIN and GREPTILE_SCORE_RE.search(comment.body)
+    )
+    if not greptile:
+        reasons.append("greptile score not available")
+    else:
+        latest: Final = max(greptile, key=lambda comment: comment.updated_at)
+        match: Final = GREPTILE_SCORE_RE.search(latest.body)
+        score: Final = int(match.group(1)) if match else 0
+        if latest.updated_at < inputs.head_commit_date:
+            reasons.append("greptile score older than head commit")
+        elif score != 5:
+            reasons.append(f"greptile score {score}/5 below 5")
+
+    bugbot: Final = tuple(
+        review
+        for review in inputs.reviews
+        if review.author_login == BUGBOT_LOGIN
+        and BUGBOT_REVIEW_MARKER in review.body
+        and BUGBOT_STALE_MARKER not in review.body
+        and review.commit_id == pr.head_sha
+    )
+    if not bugbot:
+        reasons.append("bugbot review not available")
+    else:
+        latest_review: Final = max(bugbot, key=lambda review: review.submitted_at)
+        if BUGBOT_CLEAN not in latest_review.body:
+            reasons.append("bugbot reported issues")
+
+    latest_state_by_reviewer: Final[dict[str, str]] = {}
+    for review in sorted(inputs.reviews, key=lambda review: review.submitted_at):
+        if _is_bot_login(review.author_login):
+            continue
+        latest_state_by_reviewer[review.author_login] = review.state
+    for reviewer, state in latest_state_by_reviewer.items():
+        if state == "CHANGES_REQUESTED":
+            reasons.append(f"changes requested by {reviewer}")
+
+    return Verdict(merge=not reasons, reasons=tuple(reasons))
+
+
+def _request(token: str, method: str, path: str, body: Mapping[str, object] | None = None) -> object:
+    url: Final = path if path.startswith("http") else f"{API_ROOT}{path}"
+    data: Final = None if body is None else json.dumps(body).encode("utf-8")
+    request: Final = urllib.request.Request(
+        url,
+        data=data,
+        method=method,
+        headers={
+            "Accept": "application/vnd.github+json",
+            "Authorization": f"Bearer {token}",
+            "X-GitHub-Api-Version": "2022-11-28",
+        },
+    )
+    with urllib.request.urlopen(request) as response:
+        return json.loads(response.read().decode("utf-8"))
+
+
+def _request_allow_fail(
+    token: str, method: str, path: str, body: Mapping[str, object] | None = None
+) -> tuple[int, object | None]:
+    url: Final = path if path.startswith("http") else f"{API_ROOT}{path}"
+    data: Final = None if body is None else json.dumps(body).encode("utf-8")
+    request: Final = urllib.request.Request(
+        url,
+        data=data,
+        method=method,
+        headers={
+            "Accept": "application/vnd.github+json",
+            "Authorization": f"Bearer {token}",
+            "X-GitHub-Api-Version": "2022-11-28",
+        },
+    )
+    try:
+        with urllib.request.urlopen(request) as response:
+            return response.status, json.loads(response.read().decode("utf-8"))
+    except urllib.error.HTTPError as exc:
+        return exc.code, None
+
+
+def _items(payload: object, key: str | None = None) -> tuple[object, ...]:
+    source: Final = payload.get(key) if key and isinstance(payload, Mapping) else payload
+    if not isinstance(source, list):
+        return ()
+    return tuple(source)
+
+
+def _paginate(token: str, path: str, key: str | None = None) -> list[object]:
+    separator: Final = "&" if "?" in path else "?"
+    results: list[object] = []
+    for page in range(1, 10_000):
+        batch: Final = _items(_request(token, "GET", f"{path}{separator}per_page=100&page={page}"), key)
+        results.extend(batch)
+        if len(batch) < 100:
+            return results
+    return results
+
+
+def _text(value: object) -> str:
+    return value if isinstance(value, str) else ""
+
+
+def _int(value: object) -> int:
+    return value if isinstance(value, int) else 0
+
+
+def _bool(value: object) -> bool:
+    return value is True
+
+
+def _nested(value: object, *keys: str) -> object:
+    current: object = value
+    for key in keys:
+        if not isinstance(current, Mapping):
+            return None
+        current = current.get(key)
+    return current
+
+
+def _parse_time(value: object) -> datetime:
+    text: Final = _text(value)
+    if not text:
+        return datetime.min.replace(tzinfo=timezone.utc)
+    return datetime.fromisoformat(text.replace("Z", "+00:00"))
+
+
+def _load_pr(token: str, repo: str, number: int) -> PullRequest:
+    data: Final = _request(token, "GET", f"/repos/{repo}/pulls/{number}")
+    if not isinstance(data, Mapping):
+        raise RuntimeError(f"unexpected pull payload for #{number}")
+    return PullRequest(
+        number=number,
+        title=_text(data.get("title")),
+        author_login=_text(_nested(data, "user", "login")),
+        state=_text(data.get("state")),
+        draft=_bool(data.get("draft")),
+        mergeable=data.get("mergeable") if isinstance(data.get("mergeable"), bool) else None,
+        mergeable_state=_text(data.get("mergeable_state")),
+        head_sha=_text(_nested(data, "head", "sha")),
+    )
+
+
+def _list_candidate_prs(token: str, repo: str, base: str, allowlist: frozenset[str]) -> list[int]:
+    candidates: Final = _paginate(token, f"/repos/{repo}/pulls?state=open&base={base}")
+    return [
+        _int(item.get("number"))
+        for item in candidates
+        if isinstance(item, Mapping) and _text(_nested(item, "user", "login")).lower() in allowlist
+    ]
+
+
+def _changed_files(token: str, repo: str, number: int) -> tuple[str, ...]:
+    files: Final = _paginate(token, f"/repos/{repo}/pulls/{number}/files")
+    return tuple(_text(item.get("filename")) for item in files if isinstance(item, Mapping))
+
+
+def _required_contexts(token: str, repo: str, base: str) -> frozenset[str]:
+    payload: Final = _request(token, "GET", f"/repos/{repo}/rules/branches/{base}")
+    contexts: set[str] = set()
+    for rule in _items(payload):
+        if not isinstance(rule, Mapping) or rule.get("type") != "required_status_checks":
+            continue
+        checks: Final = _nested(rule, "parameters", "required_status_checks")
+        for check in _items(checks):
+            if isinstance(check, Mapping):
+                context: Final = _text(check.get("context"))
+                if context:
+                    contexts.add(context)
+    return frozenset(contexts)
+
+
+def _check_runs(token: str, repo: str, sha: str) -> tuple[CheckRun, ...]:
+    runs: Final = _paginate(token, f"/repos/{repo}/commits/{sha}/check-runs", key="check_runs")
+    return tuple(
+        CheckRun(
+            name=_text(item.get("name")),
+            status=_text(item.get("status")),
+            conclusion=item.get("conclusion") if isinstance(item.get("conclusion"), str) else None,
+        )
+        for item in runs
+        if isinstance(item, Mapping)
+    )
+
+
+def _statuses(token: str, repo: str, sha: str) -> tuple[CommitStatus, ...]:
+    payload: Final = _request(token, "GET", f"/repos/{repo}/commits/{sha}/status")
+    return tuple(
+        CommitStatus(context=_text(item.get("context")), state=_text(item.get("state")))
+        for item in _items(payload, "statuses")
+        if isinstance(item, Mapping)
+    )
+
+
+def _comments(token: str, repo: str, number: int) -> tuple[IssueComment, ...]:
+    comments: Final = _paginate(token, f"/repos/{repo}/issues/{number}/comments")
+    return tuple(
+        IssueComment(
+            author_login=_text(_nested(item, "user", "login")),
+            body=_text(item.get("body")),
+            updated_at=_parse_time(item.get("updated_at")),
+        )
+        for item in comments
+        if isinstance(item, Mapping)
+    )
+
+
+def _reviews(token: str, repo: str, number: int) -> tuple[Review, ...]:
+    reviews: Final = _paginate(token, f"/repos/{repo}/pulls/{number}/reviews")
+    return tuple(
+        Review(
+            author_login=_text(_nested(item, "user", "login")),
+            state=_text(item.get("state")),
+            body=_text(item.get("body")),
+            commit_id=_text(item.get("commit_id")),
+            submitted_at=_parse_time(item.get("submitted_at")),
+        )
+        for item in reviews
+        if isinstance(item, Mapping)
+    )
+
+
+def _head_commit_date(token: str, repo: str, number: int) -> datetime:
+    commits: Final = _paginate(token, f"/repos/{repo}/pulls/{number}/commits")
+    if not commits:
+        return datetime.min.replace(tzinfo=timezone.utc)
+    last: Final = commits[-1]
+    if not isinstance(last, Mapping):
+        return datetime.min.replace(tzinfo=timezone.utc)
+    return _parse_time(_nested(last, "commit", "committer", "date"))
+
+
+def _mergeable_or_refetch(token: str, repo: str, pr: PullRequest) -> PullRequest:
+    if pr.mergeable is not None:
+        return pr
+    time.sleep(5)
+    return _load_pr(token, repo, pr.number)
+
+
+def _gather_inputs(
+    token: str,
+    repo: str,
+    number: int,
+    base: str,
+    self_check_name: str,
+    allowlist: frozenset[str],
+) -> EvaluationInputs:
+    pr: Final = _mergeable_or_refetch(token, repo, _load_pr(token, repo, number))
+    return EvaluationInputs(
+        pr=pr,
+        changed_files=_changed_files(token, repo, number),
+        required_contexts=_required_contexts(token, repo, base),
+        check_runs=_check_runs(token, repo, pr.head_sha),
+        statuses=_statuses(token, repo, pr.head_sha),
+        comments=_comments(token, repo, number),
+        reviews=_reviews(token, repo, number),
+        head_commit_date=_head_commit_date(token, repo, number),
+        self_check_name=self_check_name,
+        author_allowlist=allowlist,
+    )
+
+
+def merge_request_body(pr: PullRequest) -> dict[str, str]:
+    return {"merge_method": "merge", "commit_title": f"{pr.title} (#{pr.number})", "sha": pr.head_sha}
+
+
+def _merge(token: str, repo: str, pr: PullRequest) -> None:
+    status, _ = _request_allow_fail(token, "PUT", f"/repos/{repo}/pulls/{pr.number}/merge", merge_request_body(pr))
+    if status in (200, 405, 409):
+        print(f"auto-merge-price-sync: PR #{pr.number} merge call returned {status}")
+        return
+    raise RuntimeError(f"merge call for PR #{pr.number} returned {status}")
+
+
+def main() -> int:
+    token: Final = os.environ.get("GH_TOKEN", "")
+    repo: Final = os.environ.get("REPO", "")
+    base: Final = os.environ.get("BASE_BRANCH", "main")
+    dry_run: Final = os.environ.get("DRY_RUN", "") != ""
+    self_check_name: Final = os.environ.get("SELF_CHECK_NAME", "auto-merge-price-sync")
+    allowlist: Final = frozenset(login.lower() for login in os.environ.get("PR_AUTHOR_ALLOWLIST", "").split() if login)
+    if not token:
+        print("auto-merge-price-sync: app credentials not configured")
+        return 0
+    if not repo:
+        print("auto-merge-price-sync: REPO not set", file=sys.stderr)
+        return 1
+
+    pr_number_env: Final = os.environ.get("PR_NUMBER", "")
+    candidates: Final = [int(pr_number_env)] if pr_number_env else _list_candidate_prs(token, repo, base, allowlist)
+    for number in candidates:
+        inputs: Final = _gather_inputs(token, repo, number, base, self_check_name, allowlist)
+        verdict: Final = evaluate(inputs)
+        for reason in verdict.reasons:
+            print(f"auto-merge-price-sync: PR #{number} hold: {reason}")
+        if not verdict.merge:
+            continue
+        print(f"auto-merge-price-sync: PR #{number} all gates green")
+        if dry_run:
+            print(f"auto-merge-price-sync: DRY_RUN merge suppressed for PR #{number}")
+            continue
+        _merge(token, repo, inputs.pr)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

.github/workflows/auto-merge-price-sync.yml

@@ -0,0 +1,61 @@
+name: auto-merge-price-sync
+
+on:
+  issue_comment:
+    types: [created, edited]
+  check_suite:
+    types: [completed]
+  status: {}
+  schedule:
+    - cron: "*/30 * * * *"
+  workflow_dispatch:
+    inputs:
+      pr-number:
+        description: "Evaluate only this PR number (empty = scan all open sync-bot PRs)"
+        required: false
+        default: ""
+
+permissions:
+  contents: read
+  pull-requests: read
+  checks: read
+  statuses: read
+
+concurrency:
+  group: auto-merge-price-sync
+  cancel-in-progress: false
+
+jobs:
+  auto-merge-price-sync:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    env:
+      PROVIDER_INFO_SYNC_APP_ID: ${{ secrets.PROVIDER_INFO_SYNC_APP_ID }}
+      PROVIDER_INFO_SYNC_APP_PRIVATE_KEY: ${{ secrets.PROVIDER_INFO_SYNC_APP_PRIVATE_KEY }}
+    steps:
+      - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Mint app token
+        id: app-token
+        if: ${{ env.PROVIDER_INFO_SYNC_APP_ID != '' && env.PROVIDER_INFO_SYNC_APP_PRIVATE_KEY != '' }}
+        uses: actions/create-github-app-token@bcd2ba49218906704ab6c1aa796996da409d3eb1 # v3.2.0
+        with:
+          app-id: ${{ secrets.PROVIDER_INFO_SYNC_APP_ID }}
+          private-key: ${{ secrets.PROVIDER_INFO_SYNC_APP_PRIVATE_KEY }}
+
+      - name: Auto-merge eligible sync PRs
+        env:
+          GH_TOKEN: ${{ steps.app-token.outputs.token }}
+          REPO: ${{ github.repository }}
+          PR_NUMBER: ${{ (github.event.issue.pull_request && github.event.issue.number) || github.event.inputs.pr-number || '' }}
+          BASE_BRANCH: main
+          PR_AUTHOR_ALLOWLIST: "berriai-litellm-provider-info-sync[bot]"
+          SELF_CHECK_NAME: auto-merge-price-sync
+        run: python3 .github/scripts/auto_merge_price_sync.py

tests/test_litellm/test_auto_merge_price_sync.py

@@ -0,0 +1,323 @@
+"""Tests for .github/scripts/auto_merge_price_sync.py.
+
+`evaluate` is pure: it takes the pull request plus the fetched facts and
+returns a Verdict, so each gate is exercised by building inputs where exactly
+one condition fails and asserting the matching hold reason. A merge verdict
+is the thing that spends an unreviewed merge, so the defaults below are the
+happy path that every case perturbs one part of.
+"""
+
+import importlib.util
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Final
+
+import pytest
+
+_REPO_ROOT = Path(__file__).resolve().parents[2]
+_MODULE_PATH = _REPO_ROOT / ".github" / "scripts" / "auto_merge_price_sync.py"
+_spec = importlib.util.spec_from_file_location("auto_merge_price_sync", _MODULE_PATH)
+merger = importlib.util.module_from_spec(_spec)
+sys.modules[_spec.name] = merger
+_spec.loader.exec_module(merger)
+
+HEAD_SHA: Final = "deadbeef" * 5
+HEAD_DATE: Final = datetime(2026, 1, 10, tzinfo=timezone.utc)
+ALLOWLIST: Final = frozenset({"berriai-litellm-provider-info-sync[bot]"})
+COST_MAP_FILES: Final = ("model_prices_and_context_window.json",)
+
+
+def _pr(**overrides: object) -> merger.PullRequest:
+    base: Final = {
+        "number": 1,
+        "title": "sync prices",
+        "author_login": "berriai-litellm-provider-info-sync[bot]",
+        "state": "open",
+        "draft": False,
+        "mergeable": True,
+        "mergeable_state": "clean",
+        "head_sha": HEAD_SHA,
+    }
+    return merger.PullRequest(**{**base, **overrides})
+
+
+def _greptile(score: int, updated_at: datetime) -> merger.IssueComment:
+    return merger.IssueComment(
+        author_login="greptile-apps[bot]",
+        body=f"Confidence Score: {score}/5",
+        updated_at=updated_at,
+    )
+
+
+def _bugbot(commit_id: str, body: str, submitted_at: datetime) -> merger.Review:
+    return merger.Review(
+        author_login="cursor[bot]",
+        state="COMMENTED",
+        body=body,
+        commit_id=commit_id,
+        submitted_at=submitted_at,
+    )
+
+
+def _inputs(**overrides: object) -> merger.EvaluationInputs:
+    base: Final = {
+        "pr": _pr(),
+        "changed_files": COST_MAP_FILES,
+        "required_contexts": frozenset({"build"}),
+        "check_runs": (merger.CheckRun(name="build", status="completed", conclusion="success"),),
+        "statuses": (),
+        "comments": (_greptile(5, datetime(2026, 1, 11, tzinfo=timezone.utc)),),
+        "reviews": (
+            _bugbot(
+                HEAD_SHA,
+                "<!-- BUGBOT_REVIEW --> cursor bugbot found no new issues",
+                datetime(2026, 1, 11, tzinfo=timezone.utc),
+            ),
+        ),
+        "head_commit_date": HEAD_DATE,
+        "self_check_name": "auto-merge-price-sync",
+        "author_allowlist": ALLOWLIST,
+    }
+    return merger.EvaluationInputs(**{**base, **overrides})
+
+
+def _evaluate(inputs: merger.EvaluationInputs) -> merger.Verdict:
+    return merger.evaluate(inputs, classify=lambda files: "run")
+
+
+def _holds(inputs: merger.EvaluationInputs, fragment: str) -> merger.Verdict:
+    verdict: Final = _evaluate(inputs)
+    assert not verdict.merge
+    assert any(fragment in reason for reason in verdict.reasons), verdict.reasons
+    return verdict
+
+
+def test_happy_path_merges() -> None:
+    verdict: Final = _evaluate(_inputs())
+    assert verdict.merge
+    assert verdict.reasons == ()
+
+
+def test_non_allowlisted_author_holds() -> None:
+    _holds(_inputs(pr=_pr(author_login="octocat")), "not in allowlist")
+
+
+def test_closed_pr_holds() -> None:
+    _holds(_inputs(pr=_pr(state="closed")), "pr not open")
+
+
+def test_draft_pr_holds() -> None:
+    _holds(_inputs(pr=_pr(draft=True)), "draft")
+
+
+def test_unmergeable_pr_holds() -> None:
+    _holds(_inputs(pr=_pr(mergeable=False)), "not mergeable")
+
+
+def test_dirty_pr_holds() -> None:
+    _holds(_inputs(pr=_pr(mergeable_state="dirty")), "merge conflicts")
+
+
+def test_non_cost_map_files_hold() -> None:
+    verdict: Final = merger.evaluate(_inputs(changed_files=("litellm/utils.py",)), classify=lambda files: "skip")
+    assert not verdict.merge
+    assert any("cost-map-only" in reason for reason in verdict.reasons)
+
+
+def test_required_context_missing_holds() -> None:
+    _holds(_inputs(check_runs=()), "required check 'build' not green")
+
+
+def test_required_context_via_commit_status_passes() -> None:
+    verdict: Final = _evaluate(
+        _inputs(
+            check_runs=(),
+            statuses=(merger.CommitStatus(context="build", state="success"),),
+        )
+    )
+    assert verdict.merge
+
+
+def test_failing_check_run_holds() -> None:
+    _holds(
+        _inputs(
+            check_runs=(
+                merger.CheckRun(name="build", status="completed", conclusion="success"),
+                merger.CheckRun(name="lint", status="completed", conclusion="failure"),
+            )
+        ),
+        "check run 'lint' is completed/failure",
+    )
+
+
+def test_in_progress_check_run_holds() -> None:
+    _holds(
+        _inputs(
+            check_runs=(
+                merger.CheckRun(name="build", status="completed", conclusion="success"),
+                merger.CheckRun(name="ui", status="in_progress", conclusion=None),
+            )
+        ),
+        "check run 'ui'",
+    )
+
+
+def test_own_check_run_is_ignored() -> None:
+    verdict: Final = _evaluate(
+        _inputs(
+            check_runs=(
+                merger.CheckRun(name="build", status="completed", conclusion="success"),
+                merger.CheckRun(name="auto-merge-price-sync", status="in_progress", conclusion=None),
+            )
+        )
+    )
+    assert verdict.merge
+
+
+def test_pending_commit_status_holds() -> None:
+    _holds(
+        _inputs(statuses=(merger.CommitStatus(context="codecov", state="pending"),)),
+        "commit status 'codecov' is pending",
+    )
+
+
+def test_greptile_missing_holds() -> None:
+    _holds(_inputs(comments=()), "greptile score not available")
+
+
+def test_greptile_four_of_five_holds() -> None:
+    _holds(
+        _inputs(comments=(_greptile(4, datetime(2026, 1, 11, tzinfo=timezone.utc)),)),
+        "greptile score 4/5",
+    )
+
+
+def test_greptile_older_than_head_holds() -> None:
+    _holds(
+        _inputs(comments=(_greptile(5, datetime(2026, 1, 9, tzinfo=timezone.utc)),)),
+        "older than head commit",
+    )
+
+
+def test_bugbot_missing_holds() -> None:
+    _holds(_inputs(reviews=()), "bugbot review not available")
+
+
+def test_bugbot_stale_marker_ignored() -> None:
+    _holds(
+        _inputs(
+            reviews=(
+                _bugbot(
+                    HEAD_SHA,
+                    "<!-- BUGBOT_REVIEW --><!-- BUGBOT_REVIEW_STALE --> cursor bugbot found no new issues",
+                    datetime(2026, 1, 11, tzinfo=timezone.utc),
+                ),
+            )
+        ),
+        "bugbot review not available",
+    )
+
+
+def test_bugbot_old_commit_ignored() -> None:
+    _holds(
+        _inputs(
+            reviews=(
+                _bugbot(
+                    "0" * 40,
+                    "<!-- BUGBOT_REVIEW --> cursor bugbot found no new issues",
+                    datetime(2026, 1, 11, tzinfo=timezone.utc),
+                ),
+            )
+        ),
+        "bugbot review not available",
+    )
+
+
+def test_bugbot_issues_found_holds() -> None:
+    _holds(
+        _inputs(
+            reviews=(
+                _bugbot(
+                    HEAD_SHA,
+                    "<!-- BUGBOT_REVIEW --> cursor bugbot found 2 new issues",
+                    datetime(2026, 1, 11, tzinfo=timezone.utc),
+                ),
+            )
+        ),
+        "bugbot reported issues",
+    )
+
+
+def test_changes_requested_holds() -> None:
+    _holds(
+        _inputs(
+            reviews=(
+                _bugbot(
+                    HEAD_SHA,
+                    "<!-- BUGBOT_REVIEW --> cursor bugbot found no new issues",
+                    datetime(2026, 1, 11, tzinfo=timezone.utc),
+                ),
+                merger.Review(
+                    author_login="human-reviewer",
+                    state="CHANGES_REQUESTED",
+                    body="",
+                    commit_id=HEAD_SHA,
+                    submitted_at=datetime(2026, 1, 12, tzinfo=timezone.utc),
+                ),
+            )
+        ),
+        "changes requested by human-reviewer",
+    )
+
+
+def test_superseded_changes_requested_merges() -> None:
+    verdict: Final = _evaluate(
+        _inputs(
+            reviews=(
+                _bugbot(
+                    HEAD_SHA,
+                    "<!-- BUGBOT_REVIEW --> cursor bugbot found no new issues",
+                    datetime(2026, 1, 12, tzinfo=timezone.utc),
+                ),
+                merger.Review(
+                    author_login="human-reviewer",
+                    state="CHANGES_REQUESTED",
+                    body="",
+                    commit_id=HEAD_SHA,
+                    submitted_at=datetime(2026, 1, 11, tzinfo=timezone.utc),
+                ),
+                merger.Review(
+                    author_login="human-reviewer",
+                    state="APPROVED",
+                    body="",
+                    commit_id=HEAD_SHA,
+                    submitted_at=datetime(2026, 1, 13, tzinfo=timezone.utc),
+                ),
+            )
+        )
+    )
+    assert verdict.merge
+
+
+def test_merge_request_pins_evaluated_head_sha() -> None:
+    body: Final = merger.merge_request_body(_pr(number=7, title="sync prices"))
+    assert body["sha"] == HEAD_SHA
+    assert body["merge_method"] == "merge"
+    assert body["commit_title"] == "sync prices (#7)"
+
+
+def test_classifier_cost_map_set_runs() -> None:
+    assert merger._classify(["model_prices_and_context_window.json", "tests/test_litellm/test_x.py"]) == "run"
+
+
+def test_classifier_backend_file_skips() -> None:
+    assert merger._classify(["model_prices_and_context_window.json", "litellm/main.py"]) == "skip"
+
+
+def test_main_without_token_logs_and_exits_zero(
+    monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
+) -> None:
+    monkeypatch.delenv("GH_TOKEN", raising=False)
+    assert merger.main() == 0
+    assert "app credentials not configured" in capsys.readouterr().out

tests/test_litellm/test_circleci_path_filter.py

@@ -87,6 +87,29 @@ CI = [".github/workflows/test-litellm-ui-unit.yml"]
         ("backend", BACKEND + CLIENT, "run"),
         ("client", BACKEND + CLIENT, "run"),
         ("ui", BACKEND + CLIENT, "run"),
+        ("cost-map-only", ["model_prices_and_context_window.json"], "run"),
+        ("cost-map-only", ["litellm/model_prices_and_context_window_backup.json"], "run"),
+        ("cost-map-only", ["model_prices_and_context_window.schema.json"], "run"),
+        (
+            "cost-map-only",
+            ["model_prices_and_context_window.json", "tests/test_litellm/test_x.py"],
+            "run",
+        ),
+        (
+            "cost-map-only",
+            ["model_prices_and_context_window.json", "tests/proxy_unit_tests/test_y.py"],
+            "run",
+        ),
+        (
+            "cost-map-only",
+            ["model_prices_and_context_window.json", "litellm/utils.py"],
+            "skip",
+        ),
+        ("cost-map-only", ["tests/test_litellm/test_x.py"], "skip"),
+        ("cost-map-only", ["model_prices_and_context_window.json", "docs/pricing.md"], "skip"),
+        ("cost-map-only", ["model_prices_and_context_window.json", "docs/foo.mdx"], "skip"),
+        ("cost-map-only", [], "skip"),
+        ("cost-map-only", DOCS, "skip"),
     ],
 )
 def test_classify_decisions(category: str, changed: list[str], expected: str) -> None: