The green agent is designed to automatically evaluate and score the performance of a tested agent across generated tasks. During the evaluation process, it generates test tasks, interacts with the tested agent (purple agent), and evaluates its outputs. Structured scoring is provided solely as reference guidelines and does not directly influence the final scores. The assessment focuses on the tested agent’s performance in terms of LLM semantic reasoning, consistency, and explainability, producing quantitative evaluation results.