微软这篇论文把技能库的坑量出来了：307次失败里大多数是技能“帮倒忙”。做Agent的值得看看。
微软和同事的新论文（arXiv:2608.11888）评估了技能库对智能体的实际影响。他们对比技能引导与匹配参考运行，将307个智能体失败归因于加载的技能，其中125个为功能失败、182个为效率回归。失败很少来自无关技能，而看似相关的技能会促使智能体错误实现或遗漏任务要求。成本回归并非由提示长度解释，最大来源是过度验证（67例），其次是重型实现管线（30例）。技能将验证清单变成强制性工作。
Very interesting new paper from Microsoft and colleagues. (bookmark it) Skill libraries are used i...
Very interesting new paper from Microsoft and colleagues. (bookmark it) Skill libraries are used in every major harness on the assumption that more guidance is free. This work measures what a bad skill actually costs you. They attribute 307 agent failures to specific loaded skills, 125 functional failures and 182 efficiency regressions, by comparing each skill-guided run against a matched reference run that solves the same task. The failures rarely come from irrelevant skills. Seemingly relevant skills push the agent to incorrectly implement or omit something the task required. Cost regressions are not explained by prompt length either. The largest source is excessive verification at 67 cases, followed by heavy implementation pipelines at 30 cases. It turns out that skills quietly turn validation checklists into mandatory work. Paper: arxiv.org/abs/2608.11888 Track more trending AI papers in our academy: academy.dair.ai 💬 0 🔄 0 ❤️ 1 👀 618 📊 1 ⚡