What problem does it solve? Catching structural defects before code lands is hard: SQL injection, race conditions, unsafe LLM output handling, and enum gaps slip through manual review. This Skill runs a structured pre-landing review of your branch diff so real problems surface with file:line citations before merge. ## Core Features & Use Cases - Two-pass checklist review: Pass 1 flags critical issues (SQL safety, race conditions, LLM trust boundaries, shell injection, enum completeness); Pass 2 covers informational findings like async/sync mixing, type coercion, and CI/CD pipeline issues. - Adversarial multi-model review: Dispatches a Claude adversarial subagent plus optional Codex passes (adversarial challenge and structured review with a P1 gate on diffs over 200 lines), then synthesizes cross-model findings. - Fix-First workflow: Mechanical fixes (dead code, N+1 queries, magic numbers) are auto-applied; judgment calls (security, race conditions, large fixes) are batched into a single decision question. - Greptile triage and design review: Classifies Greptile bot comments as valid, fixed, or false positive with evidence-based replies, and runs a frontend design checklist when the diff touches UI files. - Use Case: Before merging a feature branch, ask for a pre-landing review; the Skill diffs against the base branch, runs all passes, auto-fixes mechanical issues, and reports remaining findings with recommended fixes. ## Quick Start Ask the assistant to review this PR against the base branch before merging.
