SkillNet, the open-source agent skill platform from Zhejiang University’s NLP lab (zjunlp), now indexes over 500,000 searchable AI agent skills and has shipped native integrations for OpenClaw and Model Context Protocol (MCP), according to the project’s GitHub repository. The platform treats agent skills as first-class, shareable packages with semantic search, automated creation, and five-dimension quality scoring.
What SkillNet Does
The platform works as a package manager for agent capabilities. Developers can search the skill index via keyword or semantic vector search, install skills from GitHub with a single command, and auto-create new skills from repositories, PDFs, or conversation logs, as MarkTechPost detailed in a hands-on walkthrough.
Each skill gets a multi-dimensional quality evaluation covering correctness, robustness, documentation, and reusability. Skills are organized in a semantic relationship graph that maps dependencies and composability. The technical report, published in March 2026, covers the system architecture, automated creation pipeline, and evaluation methodology.
Agent Framework Integrations
SkillNet shipped its OpenClaw integration in February 2026, available as a built-in skill that lets agents automatically search, download, create, evaluate, and analyze skills without manual configuration, per the project repository. The MCP server integration, contributed by CycleChain, launched in March 2026. JiuwenClaw added native SkillNet support in late March as its built-in skill marketplace.
The platform has been tested on ALFWorld, WebShop, and ScienceWorld benchmarks with reproduction scripts available in the repository.
Why Skill Portability Matters
As agent frameworks proliferate, the same capability often gets rebuilt independently for each ecosystem. SkillNet’s approach, treating skills as versioned, searchable, and installable packages across frameworks, addresses the duplication directly. The 500,000-skill index gives new agent projects a starting library instead of a blank page. The open question is adoption: whether agent framework maintainers will converge on a shared skill format, or whether SkillNet becomes one of several competing registries.