What problem does it solve? Reviewing diffs by hand is slow and inconsistent, and generic AI reviews often miss repo-specific conventions or invent noise. This Skill runs a structured, multi-phase code review of pending changes against the repository's own rules (AGENTS.md, .claude/rules/) and returns a prioritized, deduplicated findings report. ## Core Features & Use Cases - Scoped diff ingestion: Review a branch, staged changes, a file path, or a custom base ref, with worktree and untracked-file pre-flight checks. - Specialist reviewer routing: Dispatches parallel reviewer agents for correctness, conventions, tests, maintainability, concurrency, security, performance, and more based on the diff's risk surface. - Normalization and confidence gating: Deduplicates findings by fingerprint, boosts cross-reviewer agreement, filters low-confidence noise, and separates pre-existing issues from new ones. - Actionable output: Produces a P0–P3 prioritized markdown report, files issue tickets for blockers, and offers safe auto-fixes with a build/lint exit gate. - Use Case: Before merging a branch that fixes a chat streaming race condition, run the skill to catch a missed lock release on cancellation, get it filed as a ticket, and verify the fix with a green build. ## Quick Start Ask the AI to review my current branch changes against main and report any critical issues before I merge.
