Cross-model deliberation — Claude proposes, Codex cross-examines, they debate, and the verdict gets executed. Two AI models arguing so your code doesn't have to.