SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution

SkillsVote is a lifecycle-governance framework for managing Agent Skills, addressing the challenges of noise and redundancy in skill ecosystems. It utilizes a million-scale open-source corpus to ensure quality and verifiability, integrating structured skill folders for instructional context and decomposing execution trajectories into skill-linked subtasks. Experimental results on Terminal-Bench 2.0 and SWE-Bench Pro demonstrate that SkillsVote enhances agent performance through online evolution during task execution and offline transfer from curated skill libraries, making it a significant advancement for practitioners developing robust LLM-based agents.