概述

评估 Agent 性能是迭代优化的基础。

核心指标

实现

class AgentEvaluator:
    def evaluate(self, agent, test_cases):
        results = []
        for case in test_cases:
            result = {
                "task": case.name,
                "completed": agent.run(case.input) == case.expected,
                "steps": agent.last_run_steps,
                "tools_used": agent.last_tools
            }
            results.append(result)
        return self.summarize(results)