What problem does it solve? AI-generated code often completes 80% of a requirement while hiding the remaining 20% of defects: compilation failures, unhandled exceptions, regressions, and OWASP vulnerabilities. This Skill structures the review of AI-generated diffs so hidden defects are caught by the cheapest possible gate before human judgment is spent. ## Core Features & Use Cases - Ordered four-gate checklist: static checks (build, lint, types), test suite, security scanning (secrets, injection, SAST/SCA), then human or clean-context agent judgment, always in ascending cost order. - Provenance and audit rules: AI usage is recorded only in the commit message between braces, with no agent attribution in PR descriptions, code comments, ADRs, or Co-authored-by trailers. - Review summary output: produces a checklist report with residual risks and an explicit recommendation (approve, approve with reservations, or block), always requiring human validation before merge. - Use Case: Before approving a PR containing an agent-generated diff, run the checklist to confirm the build passes, tests cover the changed behavior, no secrets or injection vectors exist, and the diff scope matches the request. ## Quick Start Review this pull request diff using the pr-review-guard checklist and give me a recommendation before I merge.
