What problem does it solve? Code and document reviews often grade summaries instead of actual acceptance criteria, wave through unverifiable claims, or merge work that was never content-verified. This Skill enforces an evidence-based review loop where every criterion gets a verdict, an evidence tier, and a falsifiability check before anything merges. ## Core Features & Use Cases - Per-criterion rubric: Each acceptance criterion is graded by index with a verdict (met, not met, cannot check), an evidence tier, and a falsifiability statement — never a single PR-level approval. - Adversarial pass and anti-patterns: Actively attempts to falsify the two weakest criteria and checks six named failure patterns such as checks that have never been seen to fail. - Feedback loop and verified merge: Sends findings back to the same build session, re-reviews on the new head, and squash-merges only after every criterion is disposed of, with a content diff proving the landing. - Use Case: A lead agent reviews a PR for ticket LIAB-123, finds two criteria unverifiable, hands them back with specific evidence requests, re-reviews the fix, then squash-merges and confirms the landing with git diff against origin/main. ## Quick Start Ask the agent to review the open PR for your ticket against its acceptance criteria and run the merge once every criterion passes.
