What problem does it solve? Code reviews often produce vague, low-signal feedback or miss regressions hiding in callers outside the diff. This Skill forces a distrust-first review of a change set, hunting only correctness bugs that can be triggered by a concrete input. ## Core Features & Use Cases - Diff-scoped bug hunting: Reviews only added and changed lines from git diff, then traces the blast radius into callers affected by changed signatures, nullability, or return shapes. - Structured bug taxonomy: Hunts twelve bug classes including null dereferences, off-by-one errors, swallowed errors, races, async mistakes, resource leaks, and partial refactors. - Trigger-gated reporting: Every finding must name the concrete input that triggers it, include severity and confidence, and survive a self-verification pass that tries to disprove it. - Use Case: Before merging a refactor, ask for an adversarial review of the diff to catch a newly-nullable return value that breaks a stale caller the diff never shows. ## Quick Start Ask the AI to adversarially review the current git diff and report only correctness bugs it can trigger with a concrete input.
