What problem does it solve? Reviewing pull requests consumes scarce human attention, and shallow reviews either miss real defects or bury reviewers in noise. This Skill reviews a barwise PR the way the codebase demands: it verifies the PR body's claims instead of inferring them, checks the failure modes agent-authored changes actually exhibit, and hands the human a defensible recommendation with exactly what was and was not checked. ## Core Features & Use Cases - Claim verification: Ranks the PR body's claims, tests the load-bearing one first, re-runs every number behind its command, and watches new gates go red on a planted defect before trusting them. - Barwise-specific invariant checklist: Runs checklist.md, the invariants CI cannot check, grouped by what the diff touches (core changes, new types, copies, tests, specs, skills, dependencies). - Ranked, actionable findings: Each finding cites file and line, a concrete failure scenario, the authority it violates, and a fix, with CONFIRMED, PLAUSIBLE, and REFUTED states earned by evidence. - Use Case: Ask it to review PR #409 before merging. It reads the body as claims, checks out the head commit, re-runs the claimed test counts, applies the checklist groups that trigger, and returns a five-part review: recommendation, reading order with risk peak, cleared regions with their tests, ranked findings, and a non-empty Not checked section. ## Quick Start Review pull request #409 on this repository and tell me whether it is ready to merge, without posting anything to GitHub.
