What problem does it solve? Code review in AI-assisted workflows often trusts the implementer's report instead of the actual diff, letting spec drift, boundary violations, and missing test evidence slip through. This Skill performs adversarial, task-local review that verifies implementations against approved requirements, design, and mechanical evidence before a task is accepted as complete. ## Core Features & Use Cases - Mechanical Verification: Runs the project's test suite, greps for placeholder markers (TBD/TODO/FIXME) and hardcoded secrets, checks boundary scope against the task's declared _Boundary:_, and audits test evidence including RED-phase output for behavior changes. - Spec and Design Alignment: Confirms acceptance criteria coverage, requirements alignment by section number, design-mandated structures, and boundary commitments, rejecting silent substitutions or hidden cross-boundary coupling. - Structured Verdicts: Returns APPROVED or REJECTED with severity-graded findings (Critical, Important, Suggestion, FYI), mandatory remediation steps, and captures human corrections as durable learnings. - Use Case: After an implementer marks a task READY_FOR_REVIEW, run the review to inspect the actual git diff, verify tests genuinely prove the required behavior, and reject the task if it spills outside its approved boundary. ## Quick Start Review task 3.2 of my current spec against its requirements, design, and boundary constraints and give me a verdict.
