evaluator

An auto-evaluation and hallucination guard that scores an agent's output against a task-derived rubric, checks every claim for support, and returns an accept / revise / reject verdict with specific fixes. It derives the rubric from the task, tests factual grounding against provided context, probes e