SkillNet: Create, Evaluate, and Connect AI Skills

A consortium of researchers from institutions including Zhejiang University and Alibaba Group has unveiled SkillNet, a groundbreaking open infrastructure designed to solve a fundamental bottleneck in AI agent development: the lack of systematic skill accumulation. Current AI agents, while capable of tool invocation and complex task execution, often operate in isolation, repeatedly 'reinventing the wheel' instead of building upon proven strategies. SkillNet addresses this by providing a unified platform to create, evaluate, and organize AI skills at scale, formalizing them as evolving, composable assets. This represents a significant shift from viewing agent capabilities as ephemeral experiences to treating them as durable, transferable knowledge.

The core of SkillNet is a unified ontology that integrates skills from heterogeneous sources and establishes rich relational connections between them. The infrastructure includes a massive repository of over 200,000 skills, an interactive platform, and a versatile Python toolkit for developers. Skills are rigorously evaluated across five dimensions: Safety, Completeness, Executability, Maintainability, and Cost-awareness. In experimental evaluations on challenging benchmarks like ALFWorld, WebShop, and ScienceWorld, SkillNet demonstrated substantial performance gains, improving average rewards by 40% and reducing the number of execution steps by 30% across multiple backbone AI models. This provides a robust foundation for the next generation of AI agents, enabling them to achieve true mastery through cumulative learning rather than isolated problem-solving.