Adversarial Execution — an execution-time review gate: two independent strong models must both prove your work actually works before it's marked done. A Claude Code / Codex skill. - View it on GitHub