SkillsBench: Benchmarking LLM Agent Skills
Large language model agents are increasingly equipped with skills: structured packages of procedural knowledge that tell an agent how to carry out a particular kind of task. SkillsBench measures how much these skills actually help.
The benchmark spans 87 tasks across 8 domains and evaluates when, and by how much, curated skills improve agent performance on diverse real-world tasks. It is a large community effort, and I contributed as one of its co-authors.