Skills Know Their Neighbors: Cluster-Contrastive Capability Pages for Skill Retrieval

2026-08-05

AI Summary

This work addresses the challenge of retrieval and routing errors in large-scale skill repositories, where semantically similar skills often exhibit insufficient textual distinction. To mitigate this, the authors propose Capability Pages—a structured representation that formalizes a skill’s capability as its executable region through offline compilation, yielding a contrastive encoding comprising positive triggers, negative boundaries, and discriminative anchors. Notably, negative boundaries are explicitly introduced to model mutual exclusivity among skills, effectively reducing ambiguity-induced retrieval errors without requiring modifications to online models. The approach integrates clustering-based contrastive learning, a two-stage retrieval pipeline (recall followed by rejection), and cross-lingual transfer. Evaluated on SRA-Bench, it achieves a 2.94-point gain in Recall@10 and a 3.62-point improvement in end-to-end task success rate; on the Chinese SSL-SkillDiscovery benchmark, it attains an MRR@50 of 73.07%.

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 stating which similar requests should be routed elsewhere. We formalize a skill's capability as its \emph{executable region}, the set of queries it can solve, and view its document as a lossy observation of that region. This view exposes a document-imposed component of retrieval error that cannot be removed by improving the retriever alone. We therefore propose \emph{Capability Pages}, cluster-contrastive skill representations containing a positive trigger $\Tpos$, a negative boundary $\Tneg$, and a discriminative body $B$. An offline compiler compares neighboring skills to write these fields. At inference time, the index uses $\Tpos$ and $B$ for candidate recall, while the router uses $\Tneg$ to reject confusable alternatives. On SRA-Bench, which contains 26{,}262 skills and 5{,}400 questions from six datasets, Capability Pages improve Recall@10 for all five tested retrievers, with a mean gain of $2.94$ points. Adding $\Tneg$ to candidate cards improves end-to-end task success by $3.62$ points on average across four executors and six datasets. A transfer evaluation on Chinese SSL-SkillDiscovery reaches $73.07\%$ MRR@50 using the same encoder across conditions. Capability Pages require no modification to the online models; they improve routing by rewriting the offline skill library.

Problem

• skill retrieval
• executable region
• retrieval error
• capability representation
• large language model agents

Innovation

• Capability Pages
• cluster-contrastive
• executable region
• skill retrieval
• negative boundary