Anthropic open-sourced the skills that power Claude's document features. 16 skills.
Most are examples. But 4 of them, PDF, DOCX, PPTX, XLSX, are the actual production code running behind Claude's "create a document" feature. The 3 worth installing right now: → MCP Builder. Step-by-step guide for building Model Context Protocol servers. API research, tool naming, auth patterns, error handling. → Webapp Testing. Playwright-based toolkit with a decision tree. Server lifecycle manager. Screenshots, logs, the works. → Skill Creator. A skill for building skills. Draft, test, benchmark, optimize triggering. If you're building your own, start here. They are not Claude-only. It's an open standard: OpenClaw, Cursor, GitHub Copilot, and the same SKILL format you can deploy on any agent. And the architecture is smart. The agent only loads skill name + description into context. Full instructions load only when needed. Linked files load only for edge cases. You're not burning tokens on skills you're not using. How to install (2 minutes): Claude Code: > /plugin install document-skills@anthropic-agent-skills OpenClaw: > git clone anthropics/skills → ~/.openclaw/skills/ > symlink what you need → restart gateway
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
Anthropic open-sourced the skills that power Claude's document features. 16 skills.
Most are examples. But 4 of them, PDF, DOCX, PPTX, XLSX, are the actual production code running behind Claude's "create a document" feature.
The 3 worth installing right now:
→ MCP Builder. Step-by-step guide for building Model Context Protocol servers. API research, tool naming, auth patterns, error handling.
→ Webapp Testing. Playwright-based toolkit with a decision tree. Server lifecycle manager. Screenshots, logs, the works.
→ Skill Creator. A skill for building skills. Draft, test, benchmark, optimize triggering. If you're building your own, start here.
They are not Claude-only. It's an open standard: OpenClaw, Cursor, GitHub Copilot, and the same SKILL format you can deploy on any agent.
And the architecture is smart. The agent only loads skill name + description into context.
Full instructions load only when needed. Linked files load only for edge cases. You're not burning tokens on skills you're not using.
How to install (2 minutes):
Claude Code:
> /plugin install document-skills@anthropic-agent-skills
OpenClaw:
> git clone anthropics/skills → ~/.openclaw/skills/
> symlink what you need → restart gateway