review

What problem does it solve? Code reviews often miss architectural violations because reviewers must manually classify files, trace dependencies, and remember every bad-practice rule. This Skill automates that groundwork by running a deterministic dependency-graph analysis first, then checking each changed file against the relevant skill's registry.yaml of patterns and bad practices, producing consistent, actionable findings. ## Core Features & Use Cases - Graph-Based Pre-Analysis: Runs a graph CLI once per review to classify every changed node (controller, use case, entity, SDK hook), resolve direct and transitive dependencies, and compute blast radius before reading any code. - Registry-Driven Rule Matching: Matches implementations against each skill's registry.yaml patterns and bad practices, plus cross-cutting critical rules like enum-over-string in Zod schemas and avoiding unnecessary as casts. - Structured Findings Output: Produces a fixed report with Status (APPROVED or CHANGES_REQUESTED), severity-ranked findings with file:line locations, required refactors, newly discovered bad practices, and a coverage section proving every layer was traced. - Use Case: After implementing a feature spanning a React route, SDK hook, controller, and use case, run this review to verify schema reuse, dependency direction, and contract consistency across the transitive downstream before merging. ## Quick Start Ask the AI to review the current diff against the base branch using the review workflow and report any bad practices with severity levels.