What problem does it solve? After non-trivial code edits, teams need a consistent second-model review before committing or shipping, but manually invoking reviewer CLIs, bundling diffs, and validating results is repetitive and error-prone. This Skill standardizes that closeout review into one structured helper invocation. ## Core Features & Use Cases - Structured Review Helper: Bundles a diff, calls one selected engine (Codex by default, or Claude, Droid, Copilot), validates the structured result, and exits nonzero when actionable findings exist. - Flexible Targeting: Reviews dirty local changes, branch diffs against a base ref, or single commits, with optional prompt files and evidence datasets for extra context. - Review Panels: Runs multiple reviewers against one frozen bundle with per-engine model and thinking/effort settings. - Use Case: After fixing a PR branch, run the helper with --mode branch --base origin/main in parallel with focused tests, verify each finding against the real code, fix accepted issues, and rerun until the review exits clean. ## Quick Start Ask the agent to run an autoreview on the current branch against origin/main using the default Codex engine and report any accepted findings before committing.
