tao-judge

What problem does it solve?

This Skill turns subjective “goal achieved” judgments into a repeatable, measurable termination decision for an agent loop, preventing wasted iterations and stalled runs.

Core Features & Use Cases

• Goal-state termination gating: Scores a goal-state pair into a single scalar and decides whether the loop should stop.
• Structured verdict output: Returns a JudgeVerdict with done, reason, score (0..1), and a next_action_hint to guide continuation.
• Autoresearch-aligned scoring: Uses the score as the primary scalar and terminates only when the evaluator explicitly reports GOAL_MET.
• Operational checkpoint & one-shot evaluation: Supports being called at a judge-checkpoint after worker steps or as a one-shot scoring pass without running the full loop.

Quick Start

Ask your agent system to call tao-judge as a goal-completion evaluator at each judge-checkpoint to decide when to stop or continue the TAO loop.