What problem does it solve? Code written by an AI session cannot be trusted to review itself, and rubber-stamp approvals let defects ship. This Skill brings in fresh reviewers who did not write the code, mandates them to find reasons to reject, and produces an honest signed verdict before a slice counts as done. ## Core Features & Use Cases - Independent multi-lens review: Spawns fresh subagent reviewers with distinct lenses (spec, correctness, seams, plus security and tests at DEEP depth) that read the source themselves and report findings with file:line evidence. - Severity-graded verdicts: Classifies findings as BLOCKER, MAJOR, or MINOR and issues SIGNED OFF, SIGNED OFF WITH CONDITIONS, or REJECTED verdicts, with a sweep that re-verifies open items from prior slices. - Repo-specific inspection sheets: Reads a per-repo REVIEW.md to toggle review passes (correctness, security, accessibility, data-safety) and apply custom severity bars and repo-specific checks. - Use Case: After building a feature slice against a blueprint doc, ask for a sign-off; the Skill detects the changed files, grades them against the slice's acceptance criteria, runs the tests, and returns a verdict with a punch list. ## Quick Start Ask the assistant to sign off on the slice that was just built so independent reviewers inspect it against the build doc and return a verdict.
