Guide · Agent skill
How to use the Claude Skills meta-skill to build agent skills
Updated · written with AI from the project's own documentation and checked against our catalogue data
The claude-skills meta-skill is an agent skill that teaches an AI agent how to build other agent skills. You give it docs, specs, API notes or code, and it helps organize that material into a SKILL.md entrypoint with triggers, boundaries and examples, plus optional references, scripts and templates. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what to check before you rely on the result.
Claude Skills Meta-Skill2025Emma/vibe-coding-cn · MIT Download the skill Quick start
- 1Download the ZIP. Download the claude-skills ZIP from this page.
- 2Install the folder. Unzip it into ~/.claude/skills/ for all projects, or into .claude/skills/ in one project. In the Claude apps, upload it under Settings, then Capabilities, then Skills.
- 3Prepare your source material. Gather the docs, specs, code samples or existing SKILL.md you want to work from. Note the common failure modes if you have them.
- 4Ask for a skill. Describe the task in plain words, such as creating a skill from your docs or refactoring a skill that misfires.
- 5Scaffold the folder. Run create-skill.sh with --full or --minimal to generate the skill skeleton.
- 6Validate. Run validate-skill.sh, then again with --strict, and fix any frontmatter, name or section problems it reports.
- 7Review before sharing. Check the copied content, the commands and any factual claims before you use the skill.
What an agent skill is and what this one does
An agent skill is a folder that gives an AI agent instructions and resources for a specific kind of work. Its main file is SKILL.md, which starts with YAML frontmatter containing a name and a description. The description tells the agent when the skill applies.
This meta-skill does not do domain work itself. It builds domain skills. Its SKILL.md recommends a layout with SKILL.md as the required entrypoint, an optional references/ folder with a references/index.md navigation file, an optional scripts/ folder and an optional assets/ folder for templates.
- Creates a new skill from docs, specs or repositories
- Refactors an existing skill that is too long, unclear, inconsistent or misfires
- Designs activation triggers and boundaries (what the skill will not do)
- Moves long text into references/ and extracts a short Quick Reference
- Adds a quality checklist and a validator
Install the claude-skills skill
Download the ZIP of the claude-skills skill from this page. You can install it in one of two ways depending on how you use Claude.
For Claude Code, unzip the folder into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to make it available only in that project. For the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
- Personal (all projects): unzip into ~/.claude/skills/
- Single project: unzip into .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check what your computer needs to run the scripts
The skill's instructions are plain text, so the agent can read them anywhere. The bundled scripts are shell scripts, so running them needs a bash shell and standard Unix tools. The source names awk and basename specifically.
The skill runs scripts and does not use the network. Read the scripts before you run them, and run them only in folders you control.
- Requires: bash shell to run the scripts
- Requires: standard Unix tools used by the scripts (awk, basename)
- Runs scripts: yes
- Uses network: no
Trigger the skill with natural requests
You do not need special syntax. Describe the task in plain words that match the skill's triggers. The skill is meant to activate when you want to create a skill from material, refactor a weak skill, design activation rules, pull a Quick Reference out of long material, split content into references, or add a quality gate.
If you give it material, include the source and say what you want the result to be. The skill asks 1-3 questions when required inputs are missing.
- Create a Claude skill from this API documentation and make it reusable by my team
- Refactor this SKILL.md, it is too long and sometimes triggers when it should not
- Split this long internal guide into references/ with an index
- Add a quality checklist and validator to my existing skill
Scaffold and validate a new skill with the scripts
The skill ships two shell scripts. create-skill.sh generates a new skill folder with a skeleton, either full or minimal. validate-skill.sh checks the structure of a skill folder, and a strict mode enforces the spec more fully. Run these from the repository root or from the skills/claude-skills/ folder, adjusting the paths.
Templates are in assets/template-minimal.md and assets/template-complete.md. Use the minimal one for small domains and the complete one for complex, production-grade domains.
- Basic validation first, then --strict before delivery
- Name must match ^[a-z][a-z0-9-]*$ and should match the directory name
- Uncertain claims should carry a verification path rather than be stated as fact
./skills/claude-skills/scripts/create-skill.sh my-skill --full --output skills
./skills/claude-skills/scripts/validate-skill.sh skills/my-skill
./skills/claude-skills/scripts/validate-skill.sh skills/my-skill --strictReview the output before you use it
The skill asks the agent to structure whatever source content it receives, and it may copy that content into the new skill. Review the result before you share or deploy it, especially if the source is sensitive.
The skill does not verify external facts. Its own rules say uncertain claims must be flagged, not confirmed. Check version-specific details, commands and prices against the official documentation yourself. The source lists these official Claude resources: support.claude.com articles on what skills are, using skills in Claude, and creating custom skills, plus docs.claude.com API skills guide.
- Read the scripts before running them
- Check every command and file name the new skill contains
- Verify facts against the official links listed in the skill
What people use it for
- Packaging a team's internal API documentation as a reusable agent skill
- Shortening a bloated SKILL.md by moving long text into references/ and rewriting the Quick Reference
- Creating a consistent skill template and quality checklist for several new skills
- Adding activation boundaries to a skill that triggers on the wrong requests
Frequently asked questions
How do I install the claude-skills skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ to use it in all projects, or into .claude/skills/ inside one project. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
What does the claude-skills skill require?
It needs a bash shell to run its scripts, plus standard Unix tools such as awk and basename. It runs scripts and does not use the network.
Is the claude-skills skill free to use?
The source lists the licence as MIT. The source does not mention pricing, so check the listing on this page for any cost details.
Does this skill create a skill automatically from my files?
It guides the agent to extract patterns from docs, specs or code and organize them into SKILL.md with references, scripts and assets. You should still review the output, because the skill may copy source content into the new skill and does not verify external facts.
Why does my skill fail validation?
Common causes listed in the skill are a name that does not match ^[a-z][a-z0-9-]*$ or the directory name, a description without concrete trigger keywords, or missing sections such as When to Use This Skill and Not For / Boundaries. Fix those and run validate-skill.sh again, adding --strict for the full check.
How does this compare with writing a SKILL.md by hand or using other tools?
The source does not compare this skill with other tools. You can write a SKILL.md by hand using the included minimal or complete templates, and this skill adds a scaffold generator, a validator and a checklist to that process.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are a developer or maintainer who needs to turn internal docs, API knowledge or a bloated existing skill into a consistent, reusable agent skill. Pick something else if you only want to use an existing domain skill, or if you need the agent to verify facts or handle sensitive material without a careful review.
Download the skill