Agent Evaluation

Evaluating agents is different from evaluating LLMs. Output quality isn't enough. You need to verify the agent accomplished its goal. What should agent evals primarily measure? Flag format: CONGRESS{[primary_metric]} Example: CONGRESS{response_fluency}