What problem does it solve? Reviewing pull requests consistently is hard: reviewers must trace changes back to user story acceptance criteria, verify tests actually exercise real infrastructure instead of mocks, and catch security issues like SQL injection or LLM trust boundary violations. This Skill automates that entire review cycle and produces a structured APPROVE or REQUEST CHANGES verdict. ## Core Features & Use Cases - User Story Traceability: Extracts the US ID from the branch name, reads acceptance criteria and required tests from USER-STORIES.md, and verifies each item against the diff. - Real-Evidence Test Enforcement: Runs unit, integration, agent, and e2e pytest layers, blocking approval when runtime behavior is only validated with mocks instead of real PostgreSQL, Docker, Neo4j, or HTTP services. - Two-Pass Code Quality Review: Applies a critical pass (SQL safety, race conditions, LLM output trust, shell injection) and an informational pass (async/sync mixing, dead code, test gaps), auto-fixing mechanical issues. - Use Case: A developer opens PR #42 implementing US-055. Invoke the skill to fetch the diff, run the affected test layers, lint with ruff and mypy, check README co-changes, and receive a verdict with inline comments; on approval the user story is automatically marked DONE. ## Quick Start Ask the assistant to review PR 42 with the approve-pr skill and issue a verdict.