🦀⚡ genspark-claw

The community skill pack & CLI for Genspark Claw, OpenClaw, and Hermes agents

Give your personal AI agent superpowers in one line — five production-ready, security-audited SKILL.md skills, a zero-dependency installer, and a built-in MCP server.

🚀 Quick Start · 🧩 Skills · 🔌 MCP · 🛠 For Developers · ❓ FAQ

$ genspark-claw list

  call-prep              v1.0.0  Prepare an AI phone call end-to-end — research, script, guardrails, summary
  claw-browser-anchor    v1.0.0  Browser automation that survives layout shifts — DOM anchors, not coordinates
  spark-report           v1.0.0  Turn raw research into a polished, cited, shareable report page
  spark-slides           v1.0.0  Research → narrative slide deck outline, ready for AI Slides
  super-research         v1.0.0  Mixture-of-agents deep research with cross-verified citations

$ genspark-claw install all --target openclaw
✔ Installed super-research → ~/.openclaw/skills/super-research
✔ Installed spark-report   → ~/.openclaw/skills/spark-report
  …

🌟 Why genspark-claw?

Personal AI agents are only as good as their skills. Genspark Claw gives you a managed cloud agent, OpenClaw gives you the self-hosted runtime, and Hermes gives you a self-improving agent — but the moment you want a reliable deep-research workflow, a browser automation that doesn't break on every redesign, or a phone call that actually gets the table booked, you're writing instructions from scratch.

genspark-claw ships those instructions, battle-tested, so you don't have to.

🚀 Quick Start (macOS)

Three commands. That's the whole setup.

1️⃣ Install Apple's command line tools (needed by git/npm):

xcode-select --install

2️⃣ Install Node.js via nvm — follow the official guide at nodejs.org/en/download (choose macOS → nvm), then open a fresh terminal.

3️⃣ Install genspark-claw — one line:

mkdir -p 'gensparkclaw' && cd 'gensparkclaw' && npm install github:breakstageaxe61/genspark-claw

Then run it:

npx genspark-claw doctor     # health-check your setup
npx genspark-claw list       # see the bundled skills

🐧🪟 Linux / Windows: skip step 1, install Node ≥ 18 from nodejs.org/en/download, then run the same one-liner in step 3.

🧩 Bundled Skills

Every skill follows the open SKILL.md convention (YAML frontmatter + structured instructions), so it loads natively in any compatible runtime — and you can fork any of them as a starting point for your own.

🎯 Install Skills Into Your Agent

🦀 Genspark Claw / OpenClaw

npx genspark-claw install all --target openclaw
# → copies skills into ~/.openclaw/skills/

On a self-hosted OpenClaw gateway, restart the gateway or start a new session — the skills appear as slash commands (/super-research, /call-prep, …). On Genspark Claw, import the SKILL.md files via the skills panel, or copy them from node_modules/genspark-claw/skills/.

🐍 Hermes (Nous Research)

npx genspark-claw install all --target hermes
# → copies skills into ~/.hermes/skills/

🤖 Claude Code / 📝 Cursor

npx genspark-claw install all --target claude   # ~/.claude/skills/
npx genspark-claw install all --target cursor   # ~/.cursor/skills/

📁 Project-local workspace

npx genspark-claw install super-research        # → ./skills/super-research
npx genspark-claw install all --dir ~/my-agent/skills

genspark-claw <command> [options]

COMMANDS
  list                     List all bundled skills
  info <skill>             Show details and full instructions for a skill
  install [skill]          Install a skill (or "all") into an agent runtime
  validate [path]          Validate SKILL.md files (defaults to bundled skills)
  doctor                   Check your environment and runtimes
  runtimes                 Show detected agent runtimes on this machine
  mcp                      Start the MCP stdio server
  version                  Print version

INSTALL OPTIONS
  --target <runtime>       openclaw | hermes | claude | cursor | workspace | all
  --dir <path>             Install into a custom directory instead
  --dry-run                Show what would happen without writing files

🔌 MCP Server (Claude Desktop · Cursor · Windsurf)

genspark-claw ships a zero-dependency Model Context Protocol server that exposes the pack as tools (list_skills, get_skill, validate_skill) to any MCP client.

// claude_desktop_config.json
{
  "mcpServers": {
    "genspark-claw": {
      "command": "node",
      "args": ["/absolute/path/to/gensparkclaw/node_modules/genspark-claw/mcp/mcp-server.js"]
    }
  }
}

Verify it any time:

node node_modules/genspark-claw/mcp/mcp-server.js --test
# → MCP self-test passed: initialize, tools/list, tools/call all OK.

🏗 How It Works

graph LR
    You([🧑 You]) -->|one-line npm install| CLI[🦀 genspark-claw CLI]
    CLI -->|install| OC[OpenClaw /<br/>Genspark Claw<br/>~/.openclaw/skills]
    CLI -->|install| HE[Hermes<br/>~/.hermes/skills]
    CLI -->|install| CC[Claude Code / Cursor<br/>workspace skills]
    CLI -->|validate| VAL[🛡 Skill Validator<br/>schema + security audit]
    MCP[🔌 MCP Server<br/>stdio JSON-RPC] --> PACK[(🧩 Skill Pack<br/>5 × SKILL.md)]
    CLI --> PACK
    PACK --> OC
    PACK --> HE
    PACK --> CC

Each skill is a plain directory with a SKILL.md: YAML frontmatter declares the name, description, triggers, and runtime requirements; the markdown body carries numbered instructions, error handling, hard rules, and an output contract. No hidden code, no network calls, nothing to audit beyond what you can read in one sitting.

🛠 For Developers

Want to hack on the pack, add a skill, or vendor it into your own agent? You're in the right place.

Dev setup

git clone https://github.com/breakstageaxe61/genspark-claw.git
cd genspark-claw
npm test                 # 12 tests: parser, registry, validator, installer, doctor, MCP
npm run validate         # lint every bundled SKILL.md
node examples/quickstart.js
node mcp/mcp-server.js --test

No npm install needed — there are zero dependencies by design, and we intend to keep it that way.

Project structure

genspark-claw/
├── bin/genspark-claw.js     # CLI entry (list / info / install / validate / doctor / mcp)
├── src/
│   ├── index.js             # Public programmatic API
│