What problem does it solve? Manual code review of local changes is slow and error-prone, and single-pass AI reviews produce false positives. This Skill fans out one reviewer agent per dimension over your git diff, then adversarially verifies each finding so only confirmed defects reach you. ## Core Features & Use Cases - Multi-Dimension Review: Spawns one code-reviewer finder per dimension (correctness, security, performance, maintainability) over the working-tree and staged diff, falling back to the last commit when the tree is clean. - Adversarial Verification: Pipelines every finding through an independent skeptic agent that tries to refute it against the actual code, cutting false positives before reporting. - Ranked Verdicts: Returns confirmed findings sorted by severity, with Correctness and Security marked as blockers and Performance/Maintainability as warnings. - Use Case: After finishing a substantive change, invoke the skill before declaring the work done; it reviews the diff, reports blockers first, and you fix confirmed Correctness/Security findings in the same turn. ## Quick Start Ask the AI to run the review-changes skill on the current working-tree changes and report any confirmed blockers.
