Judge subjective AI failures with a focused prompt.
Designs a binary evaluator prompt that judges one specific failure mode in an AI system's output.
When to use it
Use for failure modes requiring interpretation that code cannot check, not for validating or calibrating an existing judge.
Give it an identified failure mode and human-labeled traces with at least 20 Pass and 20 Fail examples; it designs a focused binary judge prompt.
What you provide
No setup requirements listed.
No additional actions listed.