Patpat Verifier
When invoked directly, read the operating protocol in full. Do not load the router.
Read proof over proxy, preserve safety, and apply the project verifier playbook.
Inspect the actual application surface and reuse an existing harness when it can prove the claim. Otherwise create the smallest project-local verifier at .agents/skills/verify-<app>/SKILL.md with exact Launch, Doctor, Exercise, Evidence, and Cleanup procedures. Seed features/README.md plus one file per identified high-value user surface so later runs can select proof from maintained coverage instead of the easiest available path. Do not invent features to reach a quota.
Make every command executable as written. Own and stop only processes started by the verifier; never terminate by broad process name. Keep secrets out of prompts, logs, artifacts, and external calls. Preserve evidence long enough for independent inspection.
Isolate instances so two can run side by side (ports, data dirs, profiles), or refuse to double-drive a shared instance.
Prove the verifier end to end by launching the system, checking readiness, exercising one mapped workflow, inspecting the authoritative result, cleaning up, and confirming the evidence survived cleanup. When maintaining a verifier, restrict edits to its directory unless the user separately authorizes a product fix. Reconcile every feature-map entry to a concrete source entry point and live result. Report product defects instead of hiding them in verification instructions.
Proof closure
Close repository mutations through: