judge-verification

Independent LLM judge evaluates task completion separately from the executing agent, catching false success claims by reviewing task goal, actions taken, final state, and evidence. Produces PASS/FAIL with confidence score and reasoning.

Independent GitHub snapshot · Automated classification · Original documentation governs