# Agent Skills Hub

> Directory of 186,000+ open-source AI agent skills, Claude Skills, and MCP servers, of which 24,000+ carry a rule-based security grade — including 93% of skills over 100 stars and 98% over 1,000. Every popular skill gets a grade (SAFE / CAUTION / UNSAFE / REJECT) before you install it — the trust layer for AI agent & MCP deployment. Data refreshed every 8 hours.

## For AI agents

- [skill.md](https://agentskillshub.top/skill.md): teaches an agent to search, audit, and install skills with the security grade checked BEFORE installing
- MCP server: `npx -y @agentskillshub/mcp` — tools: search_skills, audit_skill, get_skill_install
- CLI: `npx @agentskillshub/cli search <query>` — static CDN index, works offline after first run

## Key pages

- [Home](https://agentskillshub.top/): search the full catalog
- [Daily new skills](https://agentskillshub.top/daily/): top 10 new AI agent skills every day, ranked by star velocity, security-graded
- [Security Analyzer](https://agentskillshub.top/analyzer): free live scan of any GitHub repo (11 red-flag categories)
- [Scenario guides](https://agentskillshub.top/best/): curated best-of lists by use case, with security grades
- [Security audits](https://agentskillshub.top/audit/): per-repo "is X safe to install?" pages
- [Enterprise](https://agentskillshub.top/enterprise/): 5-dimension deep audits & compliance evidence

## Data & research

- [Dataset on HuggingFace](https://huggingface.co/datasets/jasonzhuyansen/agent-skills-security-grades): 130,173 graded skills/servers, CC BY 4.0 — cite freely
- [Security survey write-up](https://agentskillshub.top/blog/securing-117k-ai-skills/): method + the long-tail risk finding (unsafe rate ~9× higher below 20 stars)