SkillNet: An Open Infrastructure for AI Skill Management

A new study introduces SkillNet, an open-source infrastructure designed to address fragmentation in the development of AI agents. The goal is to provide a unified mechanism for the accumulation and transfer of skills, allowing agents to avoid "reinventing the wheel" and capitalize on previously learned strategies.

SkillNet organizes skills within a unified ontology, facilitating the creation of skills from heterogeneous sources, establishing relational connections, and enabling multi-dimensional evaluation in terms of safety, completeness, executability, maintainability, and cost. The infrastructure includes a repository of over 200,000 skills, an interactive platform, and a Python toolkit.

Experimental evaluations on ALFWorld, WebShop, and ScienceWorld demonstrate that SkillNet significantly enhances agent performance, increasing average rewards by 40% and reducing execution steps by 30% across multiple backbone models. By formalizing skills as evolving, composable assets, SkillNet aims to provide a solid foundation for the transition from transient experience to durable mastery.