An npm-style registry that grades your agent’s skills

What it does SkillNet is an open infrastructure that treats AI agent skills as first-class, shareable packages. It provides a public registry, a Python SDK, and a CLI to search, download, create, and evaluate skills. You can turn a GitHub repo, a PDF, or even a conversation log into a structured skill package, then score it across five dimensions before letting your agent use it.

The interesting bit The platform auto-generates a relationship graph between skills—mapping similar_to, compose_with, and depend_on links—so agents can discover how capabilities combine rather than blindly importing them. It also runs a five-dimension quality check (Safety, Completeness, Executability, Maintainability, Cost-Awareness) that treats agent skills as software artifacts worth auditing, not just prompt snippets.

Key highlights

• Auto-creates structured skills from repos, office documents, execution traces, or plain-text prompts using LLMs.
• Public REST API for search and download requires no authentication; Python SDK and CLI cover the rest.
• Evaluates any local skill or GitHub URL against five quality axes and returns a scored report.
• Discovers semantic relationships between skills to surface composition and dependency links.
• Ships with integrations for OpenClaw, JiuwenClaw, and an MCP server.

Verdict Worth a look if you are building agent frameworks and want a discoverable, community-driven skill layer. Probably overkill if your agent only needs a handful of hard-coded tools.