SkillCorpus: Consolidating and Evaluating the Open Skill Ecosystem for Real-World LLM Agents
Computation and Language
Summary
The authors studied reusable skill files called SKILL.md, which help large language model agents perform tasks. They found many such skills online but noticed they were scattered, repetitive, and varied in quality. To fix this, they created SkillCorpus, a system that collects, organizes, and evaluates these skills, grouping them by type and quality. They tested SkillCorpus on several benchmarks and showed it improves agent performance noticeably. Their work is the first to clearly show when a curated collection of these skills helps agents and when it does not.
Authors
Yanze Wang, Pengfei Yao, Tianyi Sun, Chuanrui Hu, Yan Xiao, Yunyun Han, Jun Sun, Yafeng Deng
Abstract
Agent skills, SKILL.md files that package reusable procedural knowledge for an LLM agent, are a popular mechanism for extending agent capabilities. Public repositories now host them in large and growing numbers, yet these artifacts are fragmented, redundant, and uneven in quality, and their value in practice is unclear. A core question remains open, namely how to consolidate this open-source SKILL.md ecosystem into a single usable corpus, and what bounds its benefit on real-world agent tasks. We present SkillCorpus, a framework that aggregates, curates, matches, and evaluates the open skill ecosystem at scale. It filters ~821,000 crawled skills through a multi-stage pipeline into 96,401 skills organised by a 16-class taxonomy and three quality facets (utility, robustness, safety), and pairs them with a fine-tuned retrieval-and-selection stack that matches task-relevant skills. We evaluate end-to-end across three benchmarks (SkillsBench, GDPVal, QwenClawBench), two harnesses, and two open backbones with a frontier robustness check. Integrating SkillCorpus yields consistent gains across all three benchmarks, largest on SkillsBench (+7.5 pp). An operational analysis traces the gains to a coverage boundary and a harness boundary. SkillCorpus is, to our knowledge, the first end-to-end account of when a curated, retrieval-served community corpus improves real agent tasks, and where it does not. The dataset, models, and code will be released upon acceptance.