What problem does it solve? Complex or long-running agent tasks often end with unverifiable completion claims, where work looks done but actually is not. This Skill forces a written task contract before work begins, so a verifier can later falsify premature success declarations. ## Core Features & Use Cases - False-Success Conditions: Requires listing at least three ways the task could appear successful while actually failing, such as green tests with altered assertions. - Evidence and Verification Ladder: Defines expected artifacts and a layered verification approach from static checks to human judgment, plus paired counter-metrics to catch Goodhart-style gaming. - Replan Budget: Caps retries per failure and mandates immediate re-planning when a whole layer of steps fails. - Use Case: Before delegating a multi-file refactor with irreversible deployment steps, write a contract specifying what evidence must exist at completion, then check each false-success condition independently at task close. ## Quick Start Ask the agent to write a task contract with false-success conditions, expected evidence, verification approach, and replan budget before starting the complex task.