SkillRep：智能体技能检索的表示设计空间
原标题：SkillRep: Characterizing and Understanding the Representation Design Space of Agent Skill Retrieval
AI 摘要
论文提出 SkillRep，将编码器微调、查询侧与文档侧扩展、背景投影和词法通道五种表示选择组合为一个检索器，用于从大规模技能库中为 LLM 智能体检索技能。在 SRA-Bench 上，使用 0.6B 编码器时 MRR@10 比同规模最强检索器提升 7.02 分（官方技能池）和 8.79 分（自建 111,247 技能池 UniPool），并在 SkillsBench 上表现最佳。研究还通过背景、未言明词汇和稀释三个量度量请求与技能之间的差距，解释各模块为何相互竞争或互补。
正文节选
SkillRep: Characterizing and Understanding the Representation Design Space of Agent Skill Retrieval Abstract. Retrieving the right skill from a large library is essential for LLM agents, yet requests and skills are written for different purposes and in different words. Existing retrievers narrow this request–skill gap with one or two representation choices each, so it is unknown whether their gains hold in combination. No prior work measures the gap itself, so it is also unclear which part of it