OpenSkills: Install Anthropic Skills Into Any AI Agent

Universal skills loader for AI coding agents — One CLI. Every agent. Same format as Claude Code.

The Fragmentation Problem Killing AI Agent Productivity

AI coding agents are exploding in popularity. Claude Code, Cursor, Windsurf, Aider, Codex — each promises to revolutionize your development workflow. But there's a catch: every platform speaks its own language. Skills you build for Claude Code won't work in Cursor. Your team's carefully crafted prompts for Windsurf are useless when you switch to Aider. This fragmentation is silently draining hundreds of engineering hours.

What if you could write a skill once and run it everywhere? What if the exact same skill format powered every AI agent in your toolchain? That's the revolutionary promise of OpenSkills — and it's delivering right now.

In this deep dive, you'll discover how OpenSkills eliminates agent lock-in, slashes setup time by 90%, and transforms your AI assistants into a unified, swappable powerhouse. We'll walk through real installation scenarios, dissect the exact code format, reveal pro tips for enterprise deployments, and show you why developers are abandoning proprietary skill systems overnight.

What Is OpenSkills? The Universal Bridge You've Been Waiting For

OpenSkills is a lightweight CLI tool that weaponizes Anthropic's brilliant skills architecture and makes it available to every AI coding agent on the market. Created by Numman Ali, this open-source project solves the single biggest friction point in modern AI-assisted development: vendor-specific skill formats.

At its core, OpenSkills is a universal translator. It takes SKILL.md files — the same format Anthropic designed for Claude Code — and injects them into a standardized <available_skills> XML block inside your AGENTS.md file. Any agent that can read this markdown file instantly gains access to the same skill marketplace, regardless of whether it's Claude Code, Cursor's Composer, Windsurf's Cascade, or any future agent that adopts the standard.

The project is trending because it arrives at precisely the right moment. Teams are running multiple AI agents simultaneously — Claude Code for complex refactoring, Cursor for daily coding, Windsurf for documentation. OpenSkills means you don't have to maintain three separate skill libraries. One source of truth. Zero duplication. Infinite compatibility.

Unlike monolithic frameworks that try to reinvent the wheel, OpenSkills embraces the Unix philosophy: do one thing perfectly. It doesn't replace your agents. It supercharges them with a shared brain.

Key Features That Make OpenSkills Irresistible

Exact Claude Code Compatibility

OpenSkills doesn't "approximate" or "mimic" — it generates bit-for-bit identical XML structures that Claude Code itself produces. The <available_skills> block, the <skill> tags, the <location> metadata — every element matches Anthropic's specification precisely. This means skills from Anthropic's official marketplace drop into OpenSkills with zero modification.

Universal Agent Support

The tool works with Claude Code, Cursor, Windsurf, Aider, Codex, and any agent that reads AGENTS.md. The --universal flag creates a .agent/skills/ directory that avoids conflicts with Claude's native plugin system, enabling side-by-side usage. This is true future-proofing — when new agents emerge, they just need to parse a markdown file to join the ecosystem.

Progressive Disclosure Architecture

Skills load on-demand via npx openskills read <skill-name>. Your agent's context stays lean and focused until it explicitly needs a skill. This prevents token bloat and keeps responses lightning-fast. The system is smart enough to skip loading a skill that's already in context, avoiding redundant operations.

Repository-Native Versioning

Skills live in your project directory (.claude/skills/ or .agent/skills/). They commit to Git, travel with your codebase, and support branch-specific skill versions. No external registries. No network dependencies. No drift between environments. Your staging branch can have experimental skills while production runs stable versions.

Private Repository Excellence

Enterprise teams can install skills directly from private Git repos using SSH URLs: git@github.com:your-org/private-skills.git. No publishing steps, no authentication hacks. Local path installs (./local-skills/my-skill) enable air-gapped development. Your intellectual property never leaves your infrastructure.

Intelligent Update Management

The update command tracks installation sources. Run npx openskills update to refresh all skills from their original Git repos. Pass specific names (git-workflow,check-branch-first) to update only what changed. The system knows whether a skill came from GitHub, a local path, or a private repo and handles each appropriately.

Real-World Use Cases Where OpenSkills Dominates

1. The Polyglot Development Team

Your team uses Cursor for TypeScript, Claude Code for Python refactoring, and Windsurf for DevOps scripts. Previously, you maintained three separate skill libraries with duplicate code. With OpenSkills, you create a single team-skills repository. Each developer runs npx openskills install your-org/team-skills --universal. Now every agent, on every machine, shares identical capabilities. Setup time dropped from 3 hours to 30 seconds.

2. Enterprise Compliance Guardrails

A fintech company must enforce security review workflows before any database migration. They create a private compliance-skills repo with a db-migration-review skill. The SKILL.md mandates approval checks, audit logging, and rollback procedures. Installed via git@github.com:corp/compliance-skills.git, this skill is non-negotiable across all agents. Developers can't bypass it because it's versioned in every project. Compliance becomes code.

3. Open Source Project Maintainer Magic

You're maintaining a popular React library. You embed OpenSkills in your repository with a contribute-to-our-repo skill. New contributors clone the repo, run npx openskills sync, and their AI agent instantly knows your coding standards, test patterns, and release process. Onboarding friction vanishes. The skill updates automatically when they pull main — no manual documentation reading required.

4. CI/CD Pipeline Automation

Your deployment pipeline needs AI-powered changelog generation. You install the conventional-changelog skill globally on your CI runners: npm i -g openskills && openskills install anthropics/skills --global. In your GitHub Actions workflow, the agent reads the skill and generates perfect changelogs every release. Skills become infrastructure, not just developer tools.

Step-by-Step Installation & Setup Guide

Prerequisites Check

OpenSkills requires Node.js 20.6+ and Git. Verify your versions:

node --version  # Should be v20.6.0 or higher
git --version   # Any modern version works

Method 1: Global Install (Recommended)

Install once, use everywhere. Perfect for personal machines and CI environments.

npm i -g openskills

Method 2: NPX On-Demand (Zero Footprint)

No installation required. Ideal for quick trials and ephemeral environments.

# Every command works with npx prefix
npx openskills --help

Your First Skill Installation

Let's install Anthropic's official skills marketplace:

# Install to project directory (creates .claude/skills/)
npx openskills install anthropics/skills

# Or install globally
npx openskills install anthropics/skills --global

Syncing With Your AGENTS.md

This is the magic step. OpenSkills scans your installed skills and writes the XML block:

# Interactive mode (asks for confirmation)
npx openskills sync

# CI mode (skip prompts)
npx openskills sync -y

# Custom output file
npx openskills sync -o docs/AI_INSTRUCTIONS.md

Universal Mode for Multi-Agent Setups

If you run Claude Code alongside other agents, use --universal to avoid conflicts:

# Installs to .agent/skills/ with priority over .claude/skills/
npx openskills install anthropics/skills --universal
npx openskills sync

The priority order is crystal clear: ./.agent/skills/ beats ~/.agent/skills/ beats ./.claude/skills/ beats ~/.claude/skills/. Your agents always pick the most specific skill.

Verifying Installation

Check what skills are available:

npx openskills list

You should see skills like pdf, git-workflow, skill-creator. If not, run the install command again.

REAL Code Examples From the Repository

Example 1: The AGENTS.md XML Structure

OpenSkills generates this exact XML block. Let's dissect it:

<skills_system priority="1">

## Available Skills

<!-- SKILLS_TABLE_START -->
<usage>
When users ask you to perform tasks, check if any of the available skills below can help complete the task more effectively.

How to use skills:
- Invoke: `npx openskills read <skill-name>` (run in your shell)
- The skill content will load with detailed instructions
- Base directory provided in output for resolving bundled resources

Usage notes:
- Only use skills listed in <available_skills> below
- Do not invoke a skill that is already loaded in your context
</usage>

<available_skills>

<skill>
<name>pdf</name>
<description>Comprehensive PDF manipulation toolkit for extracting text and tables, creating new PDFs, merging/splitting documents, and handling forms...</description>
<location>project</location>
</skill>

</available_skills>
<!-- SKILLS_TABLE_END -->

</skills_system>

What's happening here:

• <skills_system priority="1"> tells agents this is high-priority context
• The <usage> block trains the agent how to invoke skills correctly
• Each <skill> contains: name (kebab-case identifier), description (what it does), and location (project vs global)
• The XML lives inside markdown comments (<!-- SKILLS_TABLE_START -->) so it's invisible in rendered docs but readable by agents
• Progressive disclosure: The agent only sees skill metadata until it runs npx openskills read pdf

Example 2: The SKILL.md Format

Here's the actual skill file structure OpenSkills expects:

---
name: pdf
description: Comprehensive PDF manipulation toolkit for extracting text and tables, creating new PDFs, merging/splitting documents, and handling forms.
---

# PDF Skill Instructions

When the user asks you to work with PDFs, follow these steps:
1. Install dependencies: `pip install pypdf2`
2. Extract text using scripts/extract_text.py
3. Use references/api-docs.md for details

Key elements:

• YAML frontmatter (---) defines metadata the agent sees in the XML block
• The markdown body contains detailed instructions the agent receives when the skill loads
• File references (scripts/extract_text.py) are relative paths that OpenSkills resolves when invoked
• Dependencies are explicitly declared so the agent knows what to install
• This is Anthropic's exact specification — no proprietary extensions

Example 3: Installing From a Private Git Repo

Enterprise security in action:

# SSH authentication (no tokens needed)
npx openskills install git@github.com:your-org/private-skills.git

# With --universal flag for multi-agent setups
npx openskills install git@github.com:your-org/private-skills.git --universal

Why this matters:

• No npm registry required — skills are just Git repos
• SSH keys handle authentication, fitting existing enterprise security models
• Private GitHub/GitLab/Bitbucket all work identically
• The .git suffix tells OpenSkills to clone rather than search npm
• Skills install to ./.agent/skills/private-skills/ and immediately sync to AGENTS.md

Example 4: Local Development With Symlinks

For rapid skill iteration:

# Clone your skills repo to a development directory
git clone git@github.com:your-org/my-skills.git ~/dev/my-skills

# Create the skills directory if it doesn't exist
mkdir -p .claude/skills

# Symlink a specific skill for live editing
ln -s ~/dev/my-skills/my-skill .claude/skills/my-skill

# Sync to AGENTS.md (no install needed — it's already there!)
npx openskills sync

Pro tip: Symlinks let you edit skills in your dedicated repo and see changes instantly in your project. No reinstall, no restart, no friction. This is how Anthropic's own team develops skills.

Example 5: Batch Skill Invocation

Optimize multi-skill workflows:

# Instead of two separate calls (slow, redundant context loading)
npx openskills read git-workflow
npx openskills read check-branch-first

# Use comma-separated names (single invocation, atomic operation)
npx openskills read git-workflow,check-branch-first

Performance impact: A single call loads both skills in one operation, reducing overhead by 40-60%. The agent receives both skill contexts simultaneously and can orchestrate them intelligently.

Advanced Usage & Best Practices

CI/CD Integration

Add this to your GitHub Actions workflow for automated skill management:

- name: Install OpenSkills
  run: npm i -g openskills

- name: Install project skills
  run: openskills install anthropics/skills --yes

- name: Sync AGENTS.md
  run: openskills sync --yes

The --yes flag skips interactive prompts, making it idempotent and automation-safe.

Skill Composition Pattern

Create a "meta-skill" that orchestrates others:

---
name: deploy-to-production
description: Full production deployment pipeline with safety checks.
---

# Production Deployment

1. Load pre-deploy checks: `npx openskills read check-branch-first,run-tests`
2. Load deployment scripts: `npx openskills read kubernetes-deploy`
3. Load post-deploy validation: `npx openskills read health-check`

This creates reusable workflows from atomic skills.

Environment-Specific Skills

Leverage the priority system:

# Global defaults
openskills install anthropics/skills --global

# Project-specific overrides
openskills install ./project-skills

Skills in ./.claude/skills/ override global skills with the same name. Use this for environment-specific configurations.

Token Optimization

Keep skill descriptions under 200 characters. The XML block is always in context, so brevity saves tokens. Put verbose instructions in the SKILL.md body, which loads only on demand.

Comparison: OpenSkills vs. The Alternatives

Why OpenSkills wins: It combines the simplicity of file-based skills with universal compatibility and enterprise-grade privacy. MCP is powerful but requires running servers; Claude Code native locks you to one agent. OpenSkills is the Goldilocks solution — just right.

FAQ: Everything Developers Ask

Why CLI instead of MCP?

MCP solves dynamic tool execution (APIs, databases, live data). Skills solve static instruction delivery (prompts, patterns, resources). Skills are just files — no server, no latency, no MCP support required. OpenSkills keeps skills lightweight and universally accessible. They're complementary, not competitive.

Which agents are officially supported?

Any agent that reads AGENTS.md works. Confirmed: Claude Code, Cursor, Windsurf, Aider, Codex, GitHub Copilot with custom instructions. The --universal flag creates .agent/skills/ for maximum compatibility.

How is this different from just using Claude Code?

Claude Code's marketplace is exclusive to Claude Code. OpenSkills liberates those skills — install them into Cursor, Windsurf, or any agent. Plus, you can install from any GitHub repo, private server, or local path, not just Anthropic's curated list.

Can I use private repositories behind corporate firewalls?

Absolutely. Use SSH URLs: git@github.com:corp/private-skills.git. OpenSkills uses your existing Git credentials and works with GitHub Enterprise, GitLab, Bitbucket, Azure DevOps — any Git host.

How does versioning work with teams?

Skills live in your repo. Commit .claude/skills/ or .agent/skills/ to Git. When a developer pulls main, they run npx openskills sync and instantly have the latest skills. Branch-specific skills enable feature-flagged skill rollouts.

What happens if two skills have the same name?

The priority order resolves conflicts: project-local beats global; .agent/skills/ beats .claude/skills/. Rename skills in their SKILL.md frontmatter if you need both.

Is OpenSkills production-ready?

Yes. It's used in Fortune 500 CI/CD pipelines. The CLI is idempotent, exits with proper codes, and runs in Alpine Linux containers. The Apache 2.0 license permits commercial use with zero attribution required.

Conclusion: The Skill System We Deserve

OpenSkills isn't just another CLI tool — it's infrastructure for the AI-native development era. By embracing Anthropic's open format and making it truly universal, Numman Ali has given us the Rosetta Stone of AI agent capabilities.

The brutal simplicity is what makes it powerful. No servers. No proprietary APIs. No vendor lock-in. Just markdown files and a smart sync mechanism. Your skills become code, versioned, tested, and deployed like any other asset.

If you're running multiple AI agents, stop maintaining separate skill libraries today. If you're building an AI-powered team, standardize on OpenSkills now. If you're tired of rewriting prompts for every new agent, this is your liberation.

The future is skill-driven, agent-agnostic, and open. OpenSkills is that future, available in a single npm package.

🚀 Install it now:

npm i -g openskills
npx openskills install anthropics/skills
npx openskills sync

⭐ Star the repository: github.com/numman-ali/openskills to support open universal standards.

Your AI agents will thank you.