{ "task": "(edit me) Describe the task this LLM-judge eval scores reference-pack against.", "output": "{{output-from-skill}}", "cases": [ { "name": "happy path", "criteria": "output satisfies the task" }, { "name": "edge case", "criteria": "output handles a corner input gracefully" }, { "name": "failure mode", "criteria": "output refuses gracefully on ambiguous input" } ] }