arxivcs.SEcs.AI2026-07-01
Skills Are Not Islands: Measuring Dependency and Risk in Agent Skill Supply Chains
Changguo Jia, Tianqi Zhao, Runzhi He, Minghui Zhou
Agent skills package reusable operational knowledge for Large Language Model (LLM) agents, yet as they grow in scope, they become dependency-bearing artifacts whose identities, versions, and provenance remain implicit. This opacity already causes duplicated dependencies and incon…