SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks
Application Category
📝 Abstract
Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Software Engineering to +51.9pp for Healthcare) and 16 of 84 tasks show negative deltas. Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming. Focused Skills with 2--3 modules outperform comprehensive documentation, and smaller models with Skills can match larger models without them.
Problem
Research questions and friction points this paper is trying to address.
Agent Skills
benchmarking
LLM agents
procedural knowledge
task performance
Innovation
Methods, ideas, or system contributions that make the work stand out.
SkillsBench
Agent Skills
LLM agents
Skill evaluation
Procedural knowledge
🔎 Similar Papers
No similar papers found.
