Adversarial Code Review

Adversarial multi-model code review that finds real defects, proves them, and fixes them. Every available model family (grok, codex, claude — including the one hosting this session) reviews the diff independently and blind; every finding is then cross-examined by a lane that did not raise it, so nothing survives on one model's say-so. Confirmed behavioral defects get a failing test before the fix and a passing one after; confirmed structural problems get fixed with the suite staying green as the proof behavior did not change. Use whenever the user asks for a hard, thorough, adversarial, multi-model, or cross-vendor review — "review my branch with multiple models", "tear this apart", "adversarial review", "have grok and codex check this", "find the real bugs and fix them with tests", "second opinion from another model on this diff" — or wants a review that ends in verified fixes rather than a list of comments. Prefer this over a single-model review whenever correctness matters enough to pay for several model f

context-labs Updated 958 repo stars

File contents

context-labs/clocked-in commit f7bd7989df

Frequently asked questions

npx get-skill add context-labs/adversarial-code-review