🦞 ClawSkills

Every agent skill on GitHub — in one install.

Comprehensive · Offline · One-command · Cross-platform

We crawled 842 GitHub repos, ClawHub, OpenClaw, and 10 awesome lists — extracted every SKILL.md and SOUL.md — and packaged them into one searchable index. Install once, access all of them.

125K skills total:

• 11,389 agent skills — from 842 repos across .claude/, .cursor/, .agents/, .openclaw/, .codex/
• 90K research skills — from arXiv, PMC, PLOS, Nature, eLife
• 24K tech skills — from GitHub, StackOverflow, web sources

Quick Start

pip install clawskills-rai
clawskills bundle-install --domain claw     # one-time: download skill index
clawskills install docker-essentials        # search + install → .claude/skills/

Works with Claude Code, Cursor, OpenClaw, Codex, and any --target path.

Search Skills

Find skills across the full 125K index:

clawskills skill-search "kubernetes" --top 5
clawskills skill-search "CRISPR" --domain research --top 3

Browse & Install

Preview matches before installing:

clawskills install "react" --list          # browse matches
clawskills install "react" --pick 2        # install 2nd result
clawskills install coding-agent            # install best match directly

Real-World Example

Here's what a real installed skill looks like — docker-essentials from Pr-E/openclaw-master-skills:

# Docker Essentials

Essential Docker commands for container and image management.

## Container Lifecycle
docker run -d --name web -p 8080:80 nginx     # run in background
docker exec -it web bash                       # attach to running container
docker stop web && docker rm web               # stop and remove

## Images
docker build -t myapp:latest .                 # build from Dockerfile
docker images --filter dangling=true           # list unused images
docker system prune -af                        # clean up everything

This file was installed to .claude/skills/docker-essentials/SKILL.md — Claude Code picks it up automatically.

Full Workflow: Setting Up a New Project

Browse, pick, and install multiple skills in under a minute:

1. Browse — clawskills install "docker" --list to see what's available
2. Install — clawskills install docker-essentials writes to .claude/skills/
3. Add more — clawskills install react-expert for React patterns
4. Done — your agent picks up the new skills on next run

What's Inside

11,389 agent skills from 842 repos, 93% domain-classified. Updated 2026-04-04.

Domain	Skills	Examples
AI & Agents	4,265	coding agents, prompt engineering, RAG, MCP servers
Web & Apps	3,877	React, Vue, REST APIs, Notion, Discord
DevTools	731	Docker, CI/CD, GitHub Actions, testing
Data	548	SQL, scraping, API integrations
Security	300	auth, auditing, triage
Linux	261	CLI tools, cron, SSH
Cloud	166	AWS, Azure, Vercel, Terraform
Programming	164	Python, Go, architecture patterns

Also available: 90K research + 24K tech skills

clawskills bundle-install --auto          # auto-detect project type
clawskills skill-search "CRISPR gene editing" --domain research --top 5

Research: arXiv (3.5K), PLOS journals (62K across 7 sub-journals), eLife (941), other academic (24K). Tech: Linux (7.5K), web (6K), programming (4K), devtools (2.2K), security (1.2K), and more.

Where do the skills come from?

Agent skills (11,389 after dedup)

Source	Skills	How
ClawHub	4,293	Search API + registry page extraction
Awesome lists	5,028	10 curated community lists
GitHub code search	1,135	19 queries across 6 agent platforms
Repo mining (842 repos)	933	SKILL.md / SOUL.md from tracked repos

Research skills (90,084) — PLOS, arXiv, eLife, PMC, Springer, CrossRef

Tech skills (24,515) — GitHub repos, StackOverflow, Tavily/Baidu/Zhihu/XHS

Top contributing repos

Repo	Files	Type
clawdbot/skills	7,546	agent-templates
openclaw/skills	7,546	archive
TravisLeeeeee/awesome-openclaw-personas	404	agent-templates
Pr-E/openclaw-master-skills	363	agent-templates
secondsky/claude-skills	208	skill-collection

Build the index yourself

export GITHUB_TOKEN=<YOUR_GITHUB_TOKEN>
clawskills claw-full-build --max-skills 50000 --max-repos 1000

4-tier pipeline (~40 min): ClawHub API → repo discovery → mine files → deduplicate.

Generate new skills with LLM

cp .env.example .env  # set OPENAI_API_KEY + OPENAI_BASE_URL
clawskills capture "React Server Components@20"

Installation

pip install clawskills-rai

No API key needed for searching and installing.

From source / environment variables

git clone https://github.com/LabRAI/ClawSkills.git && cd ClawSkills
pip install -e ".[dev]"

Variable	Needed for
GITHUB_TOKEN	claw-full-build (higher API rate limits)
OPENAI_API_KEY	capture (LLM skill generation)
HF_ENDPOINT	bundle-install in China (https://hf-mirror.com)

FAQ

Do I need API keys? — No. Install and search work fully offline.

Does it work offline? — Yes. Everything is local SQLite after bundle-install.

install vs skill-search? — install writes SKILL.md into your agent directory. skill-search just searches.

CLI reference

Command	What It Does
install "<query>"	Search + install a skill
install "<query>" --list	Browse matches
skill-search "<query>"	Search the full index
claw-full-build	Build index from scratch
bundle-install --domain claw	Download pre-built index
capture "<topic>@N"	Generate skills with LLM

Contributing

Open an issue first for larger changes.

License

MIT — Responsible AI Lab @ Florida State University

Authors: Tianming Sha · Yue Zhao · Lichao Sun · Yushun Dong