What problem does it solve? Teams delegating implementation work to AI agents, external providers, or other engineers need an independent final review that does not trust implementer summaries or green CI alone. This Skill enforces a strict, provider-neutral audit of a fixed pull request head before any merge or release authority is exercised. ## Core Features & Use Cases - Provenance-Checked Audit: Records base/head SHAs, implementer identity, verifier identity, and reviewed artifacts, rejecting reviews where implementer and verifier identities collide. - Evidence-Based Verification: Reproduces every required command, records exit codes and outputs, and performs user-observable checks when behavior changes. - Bounded Correction Loop: Sends at most one evidence-based correction to a failing provider, then quarantines and decomposes the work instead of silently repairing it. - Use Case: After Jules or a local agent opens a PR and CI passes, run this review to verify the diff matches the approved scope, confirm no secrets or unrelated changes slipped in, and produce a durable pass/fail/blocked record for the human authority gate. ## Quick Start Use the GatekeeperReview skill to independently audit the current pull request against its approved specification and produce an evidence-backed decision record.
