SkillSeam：审计智能体技能集合的六项原则
原标题：SkillSeam: Six Principles for Auditing Agent Skill Collections
AI 摘要
该论文提出 SkillSeam，一种用于审计智能体技能集合的方法，将六项集合级设计原则（持久性梯度、系统一致性、制度门控、正交覆盖、流程、粒度纪律）分别映射到具体失败机制与可观测指标。通过从单一封闭技能系统生成字节差异的 L0–L6 扰动阶梯，实验测得展平持久性层级使加载技能 token 增加 60%、悬空锚点使总 token 增加 64%、同义别名使非规范路由上升并翻转一半匹配复述对、重叠通道使所有权冲突上升、平淡触发词使路由冲突上升并膨胀加载技能 token 3.7 倍、粒度错配造成最大准确率下降。作者发布了字节差异变体、任务切片、汇总结果和一页设计清单，供其他技能系统测量相同失败通道。
正文节选
SkillSeam: Six Principles for Auditing Agent Skill Collections Abstract A folder of competent skills is not yet a reliable system. Skills rarely fail alone; they fail at the seams of a collection. As an agent’s skill library grows, procedures compete for attention, aliases double-load, boundaries blur, and poorly sized skills turn routing errors into task failures. We introduce SkillSeam, a method that audits the relationships through which individual skill files become a system. It maps each co