What problem does it solve? A single code review pass rarely reaches a fixpoint because each applied fix can introduce new defects that only a subsequent review can detect. This Skill automates the review → fix → re-review cycle so the diff is driven to an actual clean state instead of stopping after one round. ## Core Features & Use Cases - Iterative review-fix loop: Repeatedly runs the review-multi-model panel over the working tree, applies verified fixes, and re-reviews until zero verified findings remain. - Convergence and safety guards: Stops on a clean full-panel pass, an iteration cap, stagnation (a fix that never takes), oscillation (a fixed finding that returns), or only un-appliable findings remaining. - Finding fingerprinting: Matches findings semantically by defect and location across rounds, tolerating wording changes and line-number drift. - Use Case: After opening a PR, ask the agent to loop the review until clean; it applies each round's verified fixes to the working tree and reports a per-round found/fixed trajectory such as iter1: 9/8 → iter2: 5/5 → iter3: 0 → clean. ## Quick Start Ask the agent to run review-loop on the current branch (or a checked-out PR number) and keep reviewing and fixing until the diff is clean.
