code-review

Trigger the workflow by asking to run code review, then choose a diff source such as uncommitted changes, staged changes, a branch diff, or a commit range. The Skill constructs the diff with git, launches parallel review layers, and presents triaged findings.

It uses three parallel layers: a Blind Hunter that sees only the diff, an Edge Case Hunter with project read access, and an Acceptance Auditor that checks the diff against a spec file. Findings from all layers are merged, deduplicated, and classified.

Yes, it runs in no-spec mode where the Acceptance Auditor layer is skipped. Findings are still reviewed by the Blind Hunter and Edge Case Hunter, but decision-needed classification is unavailable and nothing is persisted to a story file.

Each finding is classified as decision-needed, patch, defer, or dismiss. Patch findings can be applied automatically, decision-needed findings require your input, defer items are logged to a deferred-work file, and dismiss items are dropped as noise.

Yes, when a story key is known it updates the story file status to done or in-progress based on review outcome and syncs the matching entry in sprint-status.yaml. If no story key exists, only the story file is updated.

The failed layer name is recorded and the review proceeds with findings from the remaining layers. If all findings end up empty while layers failed, the Skill warns that the review may be incomplete instead of declaring a clean pass.