agent-review

Run after finishing any coding task to catch the 35 specific failure modes AI agents commit (dead code, hallucinated APIs, drive-by refactors, silent catches, spec drift). Use this skill when the user is wrapping up a code change, just before declaring "done", or whenever they want a review of the latest diff. Triggers on phrases like "review my changes", "before I commit", "check what I just did", "lint my diff", "is this clean", or any moment you've just made code edits across files.