Score goal-state pairs and output a JudgeVerdict with done status and next_action_hint.
npx skills add https://github.com/CleanExpo/Pi-Dev-Ops --skill tao-judge
Please help me install this Agent Skill. Skill: tao-judge Source: https://github.com/CleanExpo/Pi-Dev-Ops/tree/main/skills/tao-judge Command: npx skills add https://github.com/CleanExpo/Pi-Dev-Ops --skill tao-judge
This Skill turns subjective “goal achieved” judgments into a repeatable, measurable termination decision for an agent loop, preventing wasted iterations and stalled runs.
done, reason, score (0..1), and a next_action_hint to guide continuation.GOAL_MET.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.
High-intent search queries and answers about installing and using this skill.
To stop an autonomous agent loop when a goal is met, you apply a termination gating mechanism that scores the goal-state pair and returns a structured verdict with a done status. This prevents wasted iterations by only terminating when the evaluator explicitly reports the goal is met.
Termination gating in agent orchestration is the process of evaluating whether an autonomous agent has sufficiently completed a specified goal. It scores the current workspace state into a measurable 0..1 scalar to decide whether the iterative worker loop should stop or continue.
You score autonomous agent performance during a TAO run by applying an evaluator to score the goal-state pair at judge-checkpoints. This produces a measurable 0..1 score and a next_action_hint based on evaluator reasoning to guide continuation.
Yes, you can evaluate the current workspace state without running the full agent loop by performing a one-shot evaluation pass. This allows you to score the goal-state pair independently of the iterative worker steps.
You prevent wasted iterations in an autonomous agent worker loop by using a structured JudgeVerdict to gate termination. The evaluator returns a done status, a score, and a next_action_hint, ensuring the loop only stops when the goal is explicitly met.
BuiDommAutomate performance optimization workflows with evaluator-gated goals and durable OMX artifacts.
Sounder25Detect agent reasoning loops from transcripts and tool outputs.
GamezarOrchestrate multi-step task implementation with LLM-as-Judge verification.
smykla-skalskiCoordinate multi-agent orchestration sessions by delegating tasks to worker agents.