Skills for your agent, sold as markdown files

What it does genspark-claw is a community skill pack plus installer for personal AI agents. Each “skill” is a directory containing a SKILL.md — YAML frontmatter plus numbered instructions — covering deep research, report generation, slide outlines, browser automation, and AI phone-call prep. The CLI copies them into the right folder for whatever runtime you use, and a bundled MCP server exposes the pack as tools to Claude Desktop, Cursor, or Windsurf.

The interesting bit The skills are pure prose, not code — no hidden logic, no network calls, nothing to audit beyond what you can read in one sitting. The installer itself is zero-dependency Node (stdlib only), and a built-in validator flags curl | bash patterns, base64 payloads, and destructive commands before a skill reaches your agent.

Key highlights

• Five skills: super-research (mixture-of-agents research with cross-verified citations), spark-report, spark-slides, claw-browser-anchor (DOM anchors instead of screen coordinates), and call-prep
• One SKILL.md standard loads natively in Genspark Claw, OpenClaw, Hermes, Claude Code, and Cursor
• Zero runtime dependencies — no supply-chain surface, instant cold starts
• 12 tests, CI across Node 18/20/22 on Ubuntu, macOS, and Windows
• Programmatic API and a self-testing MCP stdio server included

Caveats

• “Production-ready” here means well-written instructions, not software that executes anything — the value depends entirely on how well your agent follows markdown.
• The README’s claims of being “battle-tested” and “security-audited” aren’t backed by external evidence; the audit is the built-in validator.

Verdict Worth a look if you run OpenClaw, Hermes, or Genspark Claw and want a curated starting set of agent workflows instead of writing prompts from scratch. If your agent already has its skills dialed in, this is mostly a nice template for the SKILL.md convention.