技能检索新方法：Capability Pages 提升代理技能路由性能
原标题：Skills Know Their Neighbors: Cluster-Contrastive Capability Pages for Skill Retrieval
AI 摘要
arXiv 上发布了一篇关于技能检索的研究论文，提出了一种名为 Capability Pages 的集群对比技能表示方法，通过包含正触发器、负边界和判别性主体来提升大型语言模型代理的技能检索性能。在 SRA-Bench 基准测试中，该方法使五种检索器的 Recall@10 平均提升 2.94 点，端到端任务成功率平均提升 3.62 点，并在中文 SSL-SkillDiscovery 上达到 73.07% 的 MRR@50。该方法无需修改在线模型，仅通过重写离线技能库即可改进路由。
正文节选
Computer Science > Information Retrieval Title:Skills Know Their Neighbors: Cluster-Contrastive Capability Pages for Skill Retrieval View PDF HTML (experimental) Abstract:As skill libraries grow, large language model agents must retrieve reusable skills from candidates that often share the same topic and vocabulary but implement different capabilities. Retrieval is limited not only by the scorer but also by the text being scored: a document may describe what a skill does without stat