SkillApt：基于反事实证据学习何时激活智能体技能
原标题：SkillApt: Learning When to Activate Agent Skills from Counterfactual Evidence
AI 摘要
论文提出 SkillApt，一个检索后激活控制器，用于判断已检索到的 Skill 是否应在当前执行状态下真正加载。它通过匹配的 WITH/WITHOUT 执行构建每个 Skill 的反事实证据，并聚合相似历史状态的结果做出 LOAD/ABSTAIN 决策。在 SRA-Bench 上，SkillApt-E 保持与 BM25 Top-1 相同的准确率（0.838），同时将激活率从 100% 降至 31.5%，平均 token 用量减少 74.3%。作者主张检索与激活应视为两个独立决策。
正文节选
SkillApt: Learning When to Activate Agent Skills from Counterfactual Evidence Abstract Large language model agents increasingly retrieve reusable Skills and inject them into the active context. However, a retrieved Skill can be relevant yet unnecessary, costly, or even harmful in the current execution state. We present SkillApt, a post-retrieval activation controller that decides whether a retrieved Skill should actually be loaded. SkillApt builds per-Skill evidence from matched WITH/WITHOUT exe