What problem does it solve? After implementing a plan, code often drifts from what was specified — unplanned changes sneak in, planned items get skipped, and security or pattern violations go unnoticed until they compound. This Skill systematically compares actual implementation work against the original plan to catch drift, dangerous decisions, and convention violations before they become expensive. ## Core Features & Use Cases - Plan Drift Detection: Uses parallel sub-agents to verify each planned change matches the actual code, flagging MISSING, DRIFT, and EXTRA (unplanned) changes via git diff analysis. - Safety & Quality Scan: Checks changed files for security risks (injection, hardcoded secrets), performance issues (N+1 queries), reliability gaps, and data safety problems, plus pattern consistency against sibling files. - Structured Findings Report: Produces a severity-ranked report (CRITICAL/WARNING/OBSERVATION) with per-dimension verdicts, an overall APPROVED/NEEDS ATTENTION/REJECTED verdict, and concrete fix options with tradeoffs. - Interactive Triage: Walks through each finding to apply fixes, skip, or record recurring issues as lessons in context/foundation/lessons.md; supports resuming from a saved report. - Use Case: After finishing phase 3 of a multi-phase feature plan, run the review to confirm the phase's changes match the plan, automated success criteria pass, and no unplanned endpoints were added — then triage the two warnings it finds. ## Quick Start Ask the AI to review the implementation against the plan at context/changes/my-feature/plan.md and triage any findings.
