Agent
Drives the real system as the end user to prove a completed change — browser clicks, real curl payloads, real CLI invocations, simulator, or PTY for TUIs. Produces run-scoped evidence and a per-criterion verdict. Use after any production-code change and before any 'done' claim.