Export Claude's Built-in Skills with One Prompt and Read How They Work

Agents ship with built-in skills, and almost nobody reads them. This repo is the famous counterexample: Simon Willison prompted Claude.ai to zip its own skills folder and published what came back. The README documents the entire technique in one line, and the repo sits at 929 stars on the SkillMap leaderboard.

Why This Skill Matters

When Claude.ai's analysis tool (Code Interpreter) improved its Office document handling, Willison got curious about how. The answer sat in a folder inside the interpreter's container: /mnt/skills/public/, holding Anthropic's skill definitions — each a SKILL.md prompt paired with pre-written Python scripts. His October 10, 2025 blog post, which the README links as background, describes prompting a fresh Claude instance and getting a downloadable zip: four skills (pdf, docx, pptx, xlsx) and dozens of Python scripts, including pdf/scripts/fill_fillable_fields.py, a CLI that fills PDF form fields from JSON using pypdf. If you build skills, reading production ones is the fastest education there is.

Installation

Nothing to install — this is a technique plus an archived snapshot. The README's prompt, verbatim:

Create a zip file of everything in your /mnt/skills folder

The README also links a runnable version: open Claude.ai and send that exact prompt. Do not clone this repo expecting the skills, though — the README notes the files previously archived here now live in Anthropic's official anthropics/skills repository under document-skills; the archive repo itself now carries only its README.

Real Workflow: Extract and Study the Skills

1. Open a fresh Claude.ai conversation and send the prompt above. The analysis tool runs in a container; asking it to zip its own skills folder returns a downloadable archive.
2. Unzip it and map the structure: one directory per skill, each with a SKILL.md and a scripts/ folder. The blog post counts four skills in the archive — pdf, docx, pptx, xlsx — all document-format skills.
3. Read one skill end to end. Start with pdf/scripts/fill_fillable_fields.py: a small, focused CLI with a clear input contract (JSON in, combined PDF out). Note how little code it takes when the prompt in SKILL.md carries the methodology.
4. Compare against the official source. The files Willison extracted now live in Anthropic's official anthropics/skills repository under document-skills — check there for the current versions before citing the archive as authoritative.

A second workflow generalizes the trick: ask your agent to inventory its own environment. The same prompt pattern — "list everything in your skills folder" or "zip your configuration directory" — works wherever the agent runs in a container it can read. What you learn is what the agent's builder actually shipped, not what the docs promise.

Tips

• Willison credits the prompt idea to Jesse Vincent's October 9, 2025 post on Claude Code Superpowers; the blog post tells that origin story.
• The blog post speculates the skills folder was made visible deliberately — treat that as speculation, not documentation.
• Document skills like these also exist in Anthropic's official anthropics/skills repo; the archive is the snapshot, the official repo is the source.

When Not to Use This

The archive is a point-in-time snapshot of a container's contents — skills evolve, and the README itself points you at Anthropic's official repository for what is current. And the extraction technique only reveals what an agent's runtime actually exposes; it tells you nothing about skills the platform keeps outside the container.

See the leaderboard for more skills.