GCPC：基于检查清单的智能体技能轨迹部分信用评估
原标题：Grounded Checklist Partial Credit for Agent Skill Trajectories
AI 摘要
本文提出了一种名为GCPC（Grounded Checklist Partial Credit）的智能体轨迹评估方法，由人类定义可复用规则，LLM根据任务指令和官方验证器生成任务特定的检查清单，并基于执行日志证据逐项评分，缺失证据时弃权。在包含4,455条轨迹的SkillsBench评估中，GCPC比整体判断更能区分官方PASS和FAIL结果（AUC 0.689 vs 0.619），并在96条轨迹的人工评估中更贴近人类判断。该方法还揭示了pass@1隐藏的技能影响，并成功迁移到Terminal-Bench和SWE-bench。
正文节选
assertionroute \tl_set:Ne\generatedroutegeneratedroute \tl_set:Ne\judgeroutejudgeroute Grounded Checklist Partial Credit for Agent Skill Trajectories Abstract. Language-model agents increasingly tackle long-horizon tasks in interactive environments, yet their evaluation commonly relies on task-level success rates by reducing an entire execution trajectory to whether the task passes an official verifier. This binary score hides partial progress and is particularly limited for procedural agent ski