What problem does it solve? Standard code reviews often miss real bugs because reviewers grep instead of reading implementations, trust test names instead of assertions, and re-raise already-settled concerns. This Skill forces a structured adversarial review that surfaces logic errors, ignored errors, schema mismatches, and tests that pass with a buggy implementation. ## Core Features & Use Cases - Nine adversarial techniques: required upstream reading, CLAUDE.md compliance audit, adjacent comment checks, per-call traces, adversarial test critique, failure-mode enumeration, concrete walkthroughs, negative-space audit, and reference resolution. - Anti-pattern checklist: catches ignored errors, mutation of caller state, concurrency hazards, and tests that mock the failure mode they test, with a Go-specific guidance file loaded only for Go diffs. - Structured output: findings returned as JSONL with severity (P0–P3), category, file:line references, and honest confidence calibration. - Use Case: A PR with ten prior review rounds is about to merge. Run this Skill to trace every function call against its actual implementation, catching a predicate bug where an ignored error from a runtime resolver misclassifies fields. ## Quick Start Ask the AI to run an adversarial correctness review of the current branch or a specific PR before merging.
