judge-verification

【开发与工程】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.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准