What problem does it solve? Automated checks tell you whether code runs and passes rules, but not whether it solves the right problem or is well designed. This Skill performs a senior-engineer-style qualitative review of a branch's diff against its base, evaluating intent, architecture, and design, and returns actionable findings with severity, a verdict, and next actions. ## Core Features & Use Cases - Three-dimension review: Semantic analysis (does the code match the stated intent from a user story, ticket, or spec), architecture and design evaluation mapped to ISO/IEC 25010 characteristics (maintainability, reliability, security, performance, compatibility), and senior-style actionable feedback explaining the why behind each finding. - Severity-calibrated verdicts: Findings are classified as CRITICAL, MAJOR, MINOR, or SUGGESTION, producing an APPROVED, REJECTED, or INCOMPLETE verdict, with a configurable policy for fixing or justifying blocking findings. - Idempotent reports: Caches the review via a canonical fingerprint of the code plus the base commit, so an unchanged diff returns the existing report instead of re-reviewing; supports scoped modes like working-tree, path-limited scope, and blocking-only. - Use Case: Before opening a pull request, ask for a review of your feature branch against develop. The Skill inspects the full diff including uncommitted changes, checks it against the originating user story's acceptance criteria, flags a layering violation and an N+1 query with concrete suggestions, and writes the verdict report to docs/audits/code-review.md. ## Quick Start Ask the agent to review the code on your current branch against its base before creating the pull request.
