Your AI skills,
everywhere
at once

You prompt naturally. Your installed skills activate automatically — same result, every project, every time.

Claude Code
> build me a testimonial carousel for the homepage
 
skills used:
css-carousel scroll-snap layout, no JS
design-tokens your spacing, colors, fonts
smooth-scroll anchor offsets, active nav
accessibility
Claude Code
> ship this to staging and let QA know
 
skills used:
deploy staging pipeline, env checks
notify-slack posts to #qa-releases
env-config secret injection, vault refs
rollback
Claude Code
> review this PR before I merge
 
skills used:
security-review OWASP checks, auth patterns
code-standards your lint rules, naming conventions
test-coverage flags untested paths
perf-audit
metaskill is a manager for your AI instructions. One reference file catalogs where they all live. Six commands keep them organized, synced, and installable — across every project, device, and team member.
Get Started Wait, what are skills?
Start here

First: what are skills, prompts, and agents?

AI coding tools (Claude Code, Cursor, Windsurf, Copilot) can be extended with custom instruction files. These are just text files that tell the AI how to do specific tasks — your way. There are three types:

S Skill
A text file that teaches your AI a specific capability. Think of it as a cheat sheet you give to an assistant so they know exactly how your team does things.
e.g. css-carousel.md — "Build a pure-CSS carousel with these patterns..."
P Prompt
A structured template for a specific task. Like a recipe card — it defines the inputs, steps, and expected output format so the AI gives consistent results.
e.g. /review — "Analyze this PR for security issues and..."
A Agent
A specialized AI worker with its own instructions and personality. Agents can use skills, follow prompts, and work on their own — like a team member with a specific role.
e.g. review-agent — reviews every pull request automatically
You already get this
A Figma library
References shared design components stored in team files. One library keeps everything consistent across projects.
=
Same idea, for AI tools
library.yaml
References skills, prompts, and agents stored in repos. One file keeps all your AI instructions consistent across projects.

What actually changes

The before and after of having a central place for your AI instructions.

Without metaskill
  • Copy-paste instruction files between projects manually
  • Every project has a slightly different version of the same skill
  • New team members ask "where are the skills?" and get no answer
  • New laptop? Start setting up your AI tools from scratch
  • You improve a skill in one project — the other 9 copies go stale
  • Your best AI tooling is locked inside one project folder
With metaskill
  • One command installs skills from wherever they live
  • Every copy stays in sync with a single source of truth
  • New team members run /library use * and they're fully set up
  • New laptop? One command, all your AI tools are back
  • Improve a skill anywhere, push it back to the source in one step
  • Your best AI tooling is available in every project, on every device
Who is this for

Built for anyone who
works with AI tools

If you're using AI coding tools across multiple projects and you've started building custom instructions, this is for you — no matter your role.

⌨

Engineers

You've got custom skills scattered across 5+ projects. You want them everywhere without copy-pasting.

⚙

Tech Leads

You want your whole team using the same reviewed, approved AI instructions — not everyone's random personal version.

✎

Designers

You use AI to scaffold components, generate tokens, or audit interfaces. You want those workflows to stay consistent across projects.

⚖

Platform Teams

You're distributing approved AI capabilities across your org. You need a system, not a shared Google Doc.

The Problem

Your AI instructions are
everywhere. And nowhere.

You've built great skills and prompts. But across multiple projects, they're scattered, duplicated, and impossible to keep in sync. Sound familiar?

01 — Duplication

Copy-pasted between projects

That css-carousel skill exists in 12 different versions across your projects. Each slightly different. None is the "official" one.

02 — Drift

Always out of sync

You improved a skill last week — but only in one project. The other 9 copies are already outdated.

03 — No shared access

"Where are the skills?"

New team member joins. They ask where the AI instructions are. Nobody has a clear answer. Everyone has their own version.

04 — Device sprawl

Every machine needs setup

Your laptop, your work Mac, your cloud sandbox. Each one needs the same skills installed. Each one is a manual process.

How it works

One reference file.
That's the whole trick.

Your library.yaml file doesn't contain your skills — it points to them. Like a playlist doesn't contain music — it references where the songs live. Your library references where your skills, prompts, and agents live.

Sources can be local folders (just for you), private repos (shared with your team), or public repos (available to anyone). You control who has access by controlling who has access to the repo. When you run a command, metaskill reads this file, fetches what's needed, and places it exactly where your AI tools expect.

library.yaml your single source of truth
# Where to install things defaults: skills_dir: .claude/skills agents_dir: .claude/agents commands_dir: .claude/commands # Team skill (private repo) skills: meta-agent: source: github.com/team/agentics path: skills/meta-agent private: true # Public skill (anyone can pull) css-carousel: source: github.com/community/frontend path: skills/css-carousel # Agents: autonomous workers agents: review-agent: source: github.com/team/agentics path: agents/review private: true # Prompts: structured templates prompts: orchestrator: source: /local/path/prompts path: orchestrator.md
// pointers, not copies

It's a catalog, not a container

The library doesn't store your skills — it references where they live. Your actual files stay in their repos or folders. The library is just a list of where to find them.

// no app to install

It's just a text file

The entire system is a markdown file that your AI tool reads and follows. No software to install, no server to run, no accounts to create. Just text that your AI understands.

// keep your edge private

Private by default

Sharing is just repo access. Keep skills in a local folder for yourself, a private repo for your team, or a public repo for the community. No special permissions layer — Git handles it.

// improve anywhere, sync everywhere

Two-way sync

Improve a skill on your laptop. Push the change back to the source. Every other machine and team member picks it up on next sync.

Workflow

Build. Catalog. Install. Sync.

Four steps from scattered files to a coordinated system. Works the same whether you're working solo or in a team of fifty.

01

Build

Create skills, prompts, and agents where you naturally work — inside your actual projects.

02

Catalog

Run /library add to register it. This just adds a pointer — no files are copied.

03

Install

Run /library use on any machine to pull and install from the source instantly.

04

Sync & improve

Skills are live. Improve them anywhere, push changes back with /library push.

Commands

Six commands. That's it.

Type these inside Claude Code (or any AI tool that supports skills). That's the entire interface.

/library add

Register a new skill, prompt, or agent by telling it where the source lives. This adds a line to your library — nothing is installed yet.

/library use

Install items onto your machine. Fetches from the source and places files where your AI tool can find them.

/library push

Made improvements to an installed skill? Push your changes back to the source so everyone benefits.

/library list

See everything in your library — what's registered, where it comes from, and whether it's installed.

/library search

Find items by name, type, or source. Use wildcards like meta-* to match multiple items at once.

/library sync

Pull the latest versions from all sources. Keep every machine and team member up to date.

Stop copy-pasting skills.
Start managing them.

Three steps. Takes about 30 seconds. After this, you'll have the /library command available in Claude Code.

install metaskill
# Step 1: Download metaskill
$ git clone https://github.com/fliptation/metaskill.git
 
# Step 2: Copy the skill files into Claude Code's skill directory
$ mkdir -p ~/.claude/skills/library/cookbooks
$ cp metaskill/SKILL.md ~/.claude/skills/library/
$ cp metaskill/cookbooks/*.md ~/.claude/skills/library/cookbooks/
 
# Step 3: Create your library (includes 5 bundled skills)
$ cp metaskill/library.yaml.example ~/.claude/library.yaml
 
# Done! Open Claude Code and type /library list

Your library comes with 5 ready-to-use frontend skills: css-carousel, design-tokens, bem, smooth-hash-scrolling, and css-variables. Run /library use css-carousel to install one.

View on GitHub

Inspired by

indiedevdan

The idea of organizing AI skills as structured, reusable files was inspired by indiedevdan's video on AI agent skills.