zjunlp/SkillNet

What it solves

SkillNet provides an infrastructure layer for AI agent skills, treating them as reusable software assets. It prevents agents from having to rebuild the same capabilities from scratch by allowing developers to discover, install, and orchestrate existing skills.

How it works

SkillNet operates as a library and SDK that manages the lifecycle of agent skills. It provides a public skill library for searching (via keyword or semantic intent) and downloading skills from GitHub. It can automatically generate structured skill packages from prompts, documents, or execution traces. The system also includes evaluation tools to score skills on safety, cost, and maintainability, and a graph analysis tool to map dependencies and handoff relationships between skills.

Who it’s for

AI agent developers and researchers who want to reuse existing capabilities rather than coding them manually, and those integrating skill marketplaces into agent runtimes like Claude Code, OpenClaw, or JiuwenClaw.

Highlights

• Skill Discovery & Installation: Search a public library of 500K+ GitHub skills and install them directly into local workspaces.
• Automated Authoring: Create portable skill packages from repositories, PDFs, or prompts.
• Quality Evaluation: Score skills for completeness, executability, and safety before deployment.
• Scenario Orchestration: Select scene-specific skills and generate execution-ready prompts for downstream agents.
• Broad Integration: Compatible with Model Context Protocol (MCP) and various agent frameworks.