Adversarial multi-model code review with dual Optimizer/Skeptic agents across Sonnet and Opus. Finds issues, challenges its own findings, and only auto-fixes what survives cross-model consensus.