An **adversarial verification** skill for general-purpose agents—especially weaker models—that tries to *break* completed work before you report success, reducing skipped checks, hallucinated results, shallow execution, and context drift. - View it on GitHub