Skip to content

README.md

Latest commit

 

History

History
200 lines (139 loc) · 5.84 KB

README.md

File metadata and controls

200 lines (139 loc) · 5.84 KB

SkillsHub

Centralized skills directory for AI agent teams. Sync, update, and create agentskills.io skills across all your agents.

How It Works

  1. Your team's skills live in a GitHub repo (single source of truth)
  2. skillshub sync pulls them to ~/.agents/skills/ where all agents find them natively
  3. Agents can update and create skills from conversations via the MCP server
  4. Changes are committed and pushed — teammates get them on their next sync

Quick Start

Requirements: Python 3.11+, Git, and access to your team's skills GitHub repo.

Install

pip install git+https://github.com/airopshq/skillshub.git

Connect to your team's skills repo

skillshub init https://github.com/your-org/skills.git

This clones the repo and syncs all skills to your local agent directories.

Configure your agent

Claude Code (one command):

skillshub setup claude-code

For other agents (OpenClaw, Cowork, etc.), see Agent Setup Details below.

Setting Up a New Skills Repo

If you're starting from scratch for your team:

1. Create the repo

gh repo create your-org/skills --private
cd skills
mkdir skills && touch skills/.gitkeep
git add . && git commit -m "Initial commit"
git push --set-upstream origin main

2. Organize by team (optional)

For larger orgs, organize skills into folders:

your-org/skills/
├── engineering/
│   └── skills/
│       ├── deploy-staging/SKILL.md
│       └── code-review/SKILL.md
├── sales/
│   └── skills/
│       └── call-prep/SKILL.md
└── company-wide/
    └── skills/
        └── security-questionnaire/SKILL.md

Teammates subscribe to the folders they need:

# Engineering
skillshub init https://github.com/org/skills.git --path engineering/skills --path company-wide/skills

# Sales
skillshub init https://github.com/org/skills.git --path sales/skills --path company-wide/skills

# Everything
skillshub init https://github.com/org/skills.git

3. Add your first skill

skillshub create my-first-skill
# Edit ~/.skillshub/repo/skills/my-first-skill/SKILL.md
skillshub push ~/.skillshub/repo/skills/my-first-skill

Or ask your agent: "create a skill called my-first-skill that does X"

Agent Setup Details

Claude Code

skillshub setup claude-code

This adds the MCP server and SessionStart hook automatically. Start a new session, run /mcp — should show skillshub · ✔ connected.

How it works: Skills sync to ~/.agents/skills/ on session start. Claude activates them natively (slash commands, auto-activation). Write-back goes through the MCP update_skill/create_skill tools.

OpenClaw

OpenClaw runs remotely — all setup is done through chat.

  1. Tell it: "Install skillshub: pip install git+https://github.com/airopshq/skillshub.git"
  2. Tell it: "Run skillshub init https://github.com/your-org/skills.git"
  3. Tell it: "Add an MCP server: name skillshub, command skillshub, args ["mcp"], transport stdio"
  4. Optionally: "Run skillshub sync every 5 minutes to keep skills fresh"

How it works: Skills sync to the host filesystem. OpenClaw picks them up natively. Write-back goes through MCP.

Claude Cowork

Cowork runs in a sandboxed VM — it can't run CLI commands on your machine. It accesses skills entirely through MCP.

skillshub setup cowork

This adds the MCP server to claude_desktop_config.json automatically. Restart Claude Desktop to activate.

How it works: Cowork uses MCP tools (list_skills, get_skill) to discover and read skills. Write-back goes through update_skill/create_skill. No filesystem sync needed — MCP reads directly from the repo.

Other Agents (Cursor, Copilot, Gemini CLI, etc.)

Most agents scan ~/.agents/skills/ natively. Just run skillshub sync and skills appear. For write-back, configure the MCP server using your agent's MCP config format.

Usage

Sync latest skills

skillshub sync

Automatic in Claude Code (SessionStart hook). Manual for other agents.

Update a skill from a conversation

Tell your agent: "update the deploy skill to also run smoke tests" — calls update_skill MCP tool, commits to GitHub.

Create a skill from a conversation

Tell your agent: "create a skill called lint-check that runs our linting pipeline" — calls create_skill MCP tool.

Push a skill from the CLI

skillshub push ./my-skill

View history

skillshub list                     # List all skills
skillshub log                      # Recent changes
skillshub log my-skill             # History for one skill
skillshub diff my-skill            # Latest change
skillshub rollback my-skill HEAD~1 # Revert

Requirements

  • Python 3.11+
  • Git (for repo operations)
  • GitHub access to your team's skills repo (HTTPS or SSH)

CLI Reference

Command Description
skillshub init <url> Clone repo and configure. Use --path to subscribe to specific folders
skillshub sync Pull latest and distribute to agent directories
skillshub push <dir> Push a local skill to the repo
skillshub list List all skills
skillshub log [skill] Show version history
skillshub diff <skill> Show changes between versions
skillshub rollback <skill> <ref> Restore a previous version
skillshub create <name> Scaffold a new skill
skillshub mcp Start the MCP server (stdio)

Development

For contributors working on skillshub itself:

git clone https://github.com/airopshq/skillshub.git
cd skillshub
uv sync              # Install dependencies
uv run skillshub     # Run from source

See VISION.md for the full product vision and architecture.