The closest implementation matching Claude Code's skills system — same prompt format, same marketplace, same folders, just using CLI instead of tools.

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

Found this useful? Follow @nummanthinks for more AI tooling!

OpenSkills brings Anthropic's skills system to all AI coding agents (Claude Code, Cursor, Windsurf, Aider).

For Claude Code users:

• Install skills from any GitHub repo, not just the marketplace
• Share skills across multiple agents
• Version control your skills in your repo

For other agents (Cursor, Windsurf, Aider):

• Get Claude Code's skills system universally
• Access Anthropic's marketplace skills via GitHub
• Use progressive disclosure (load skills on demand)

OpenSkills replicates Claude Code's skills system with 100% compatibility:

✅ Same prompt format — <available_skills> XML with skill tags ✅ Same marketplace — Install from anthropics/skills ✅ Same folders — Uses .claude/skills/ by default ✅ Same SKILL.md format — YAML frontmatter + markdown instructions ✅ Same progressive disclosure — Load skills on demand, not upfront

Only difference: Claude Code uses Skill tool, OpenSkills uses openskills read <name> CLI command.

Advanced: Use --universal flag to install to .agent/skills/ for Claude Code + other agents sharing one AGENTS.md.

npm i -g openskills

# Install from Anthropic's marketplace (interactive selection, default: project)
openskills install anthropics/skills

# Or install from any GitHub repo
openskills install your-org/custom-skills

openskills sync

Done! Your agent now has skills with the same <available_skills> format as Claude Code.

When you use Claude Code with skills installed, Claude's system prompt includes:

<skills_instructions>
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 skills using this tool with the skill name only (no arguments)
- When you invoke a skill, you will see <command-message>The "{name}" skill is loading</command-message>
- The skill's prompt will expand and provide detailed instructions

Important:
- Only use skills listed in <available_skills> below
- Do not invoke a skill that is already running
</skills_instructions>

<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>plugin</location>
</skill>

<skill>
<name>xlsx</name>
<description>Comprehensive spreadsheet creation, editing, and analysis with support for formulas, formatting, data analysis...</description>
<location>plugin</location>
</skill>
</available_skills>

How Claude uses it:

1. User asks: "Extract data from this PDF"
2. Claude scans <available_skills> → finds "pdf" skill
3. Claude invokes: Skill("pdf")
4. SKILL.md content loads with detailed instructions
5. Claude follows instructions to complete task

OpenSkills generates the exact same <available_skills> XML in your AGENTS.md:

<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: Bash("openskills read <skill-name>")
- 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>

<skill>
<name>xlsx</name>
<description>Comprehensive spreadsheet creation, editing, and analysis with support for formulas, formatting, data analysis...</description>
<location>project</location>
</skill>

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

</skills_system>

How agents use it:

1. User asks: "Extract data from this PDF"
2. Agent scans <available_skills> → finds "pdf" skill
3. Agent invokes: Bash("openskills read pdf")
4. SKILL.md content is output to agent's context
5. Agent follows instructions to complete task

Everything is identical except the invocation method.

Both use the exact same format:

---
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 the extract_text.py script in scripts/
3. For bundled resources, use the base directory provided in the skill output
4. ...

[Detailed instructions that Claude/agent follows]

Progressive disclosure: The full instructions load only when the skill is invoked, keeping your agent's context clean.

MCP (Model Context Protocol) is Anthropic's protocol for connecting AI to external tools and data sources. It's great for:

• Database connections
• API integrations
• Real-time data fetching
• External service integration

Skills (SKILL.md format) are different — they're for:

• Specialized workflows (PDF manipulation, spreadsheet editing)
• Bundled resources (scripts, templates, references)
• Progressive disclosure (load instructions only when needed)
• Static, reusable patterns

Why not implement skills via MCP?

1. Skills are static instructions, not dynamic tools MCP is for server-client connections. Skills are markdown files with instructions.

2. No server needed Skills are just files. MCP requires running servers.

3. Universal compatibility CLI works with any agent (Claude Code, Cursor, Windsurf, Aider). MCP requires MCP support.

4. Follows Anthropic's design Anthropic created skills as SKILL.md files, not MCP servers. We're implementing their spec.

5. Simpler for users openskills install anthropics/skills vs "configure MCP server, set up authentication, manage server lifecycle"

MCP and skills solve different problems. OpenSkills implements Anthropic's skills spec (SKILL.md format) the way it was designed — as progressively-loaded markdown instructions.

You can use both Claude Code plugins and OpenSkills project skills together:

In your <available_skills> list:

<skill>
<name>pdf</name>
<description>...</description>
<location>plugin</location>  <!-- Claude Code marketplace -->
</skill>

<skill>
<name>custom-skill</name>
<description>...</description>
<location>project</location>  <!-- OpenSkills from GitHub -->
</skill>

They coexist perfectly. Claude invokes marketplace plugins via Skill tool, OpenSkills skills via CLI. No conflicts.

Problem: If you use Claude Code + other agents (Cursor, Windsurf, Aider) with one AGENTS.md, installing to .claude/skills/ can create duplicates with Claude Code's marketplace plugins.

Solution: Use --universal to install to .agent/skills/ instead:

openskills install anthropics/skills --universal

This installs skills to .agent/skills/ which:

• ✅ Works with all agents via AGENTS.md
• ✅ Doesn't conflict with Claude Code's native marketplace plugins
• ✅ Keeps Claude Code's <available_skills> separate from AGENTS.md skills

When to use:

• ✅ You use Claude Code + Cursor/Windsurf/Aider with one AGENTS.md
• ✅ You want to avoid duplicate skill definitions
• ✅ You prefer .agent/ for infrastructure (keeps .claude/ for Claude Code only)

When not to use:

• ❌ You only use Claude Code (default .claude/skills/ is fine)
• ❌ You only use non-Claude agents (default .claude/skills/ is fine)

Priority order: OpenSkills searches 4 locations in priority order:

1. ./.agent/skills/ (project universal)
2. ~/.agent/skills/ (global universal)
3. ./.claude/skills/ (project)
4. ~/.claude/skills/ (global)

Skills with same name only appear once (highest priority wins).

openskills install <source> [options]  # Install from GitHub (interactive)
openskills sync [-y]                   # Update AGENTS.md (interactive)
openskills list                        # Show installed skills
openskills read <name>                 # Load skill (for agents)
openskills manage                      # Remove skills (interactive)
openskills remove <name>               # Remove specific skill

• --global — Install globally to ~/.claude/skills (default: project install)
• --universal — Install to .agent/skills/ instead of .claude/skills/ (advanced)
• -y — Skip interactive selection (for scripts/CI)

Default (recommended):

openskills install anthropics/skills
# → Installs to ./.claude/skills (project, gitignored)

Global install:

openskills install anthropics/skills --global
# → Installs to ~/.claude/skills (shared across projects)

Universal mode (advanced):

openskills install anthropics/skills --universal
# → Installs to ./.agent/skills (for Claude Code + other agents)

All commands use beautiful TUI by default:

Install:

openskills install anthropics/skills
# → Checkbox to select which skills to install
# → Shows skill name, description, size
# → All checked by default

Sync:

openskills sync
# → Checkbox to select which skills to include in AGENTS.md
# → Pre-selects skills already in AGENTS.md
# → Empty selection removes skills section

Manage:

openskills manage
# → Checkbox to select which skills to remove
# → Nothing checked by default (safe)

From Anthropic's skills repository:

• xlsx — Spreadsheet creation, editing, formulas, data analysis
• docx — Document creation with tracked changes and comments
• pdf — PDF manipulation (extract, merge, split, forms)
• pptx — Presentation creation and editing
• canvas-design — Create posters and visual designs
• mcp-builder — Build Model Context Protocol servers
• skill-creator — Detailed guide for authoring skills

Browse all: github.com/anthropics/skills

my-skill/
└── SKILL.md
    ---
    name: my-skill
    description: What this does and when to use it
    ---

    # Instructions in imperative form

    When the user asks you to X, do Y...

my-skill/
├── SKILL.md
├── references/
│   └── api-docs.md      # Supporting documentation
├── scripts/
│   └── process.py       # Helper scripts
└── assets/
    └── template.json    # Templates, configs

In your SKILL.md, reference resources:

1. Read the API documentation in references/api-docs.md
2. Run the process.py script from scripts/
3. Use the template from assets/template.json

The agent sees the base directory when loading the skill:

Loading: my-skill
Base directory: /path/to/.claude/skills/my-skill

[SKILL.md content]

1. Push to GitHub: your-username/my-skill
2. Users install with: openskills install your-username/my-skill

Use Anthropic's skill-creator for detailed guidance:

openskills install anthropics/skills
openskills read skill-creator

This loads comprehensive instructions on:

• Writing effective skill descriptions
• Structuring instructions for agents
• Using bundled resources
• Testing and iteration

• Node.js 20.6+ (for ora dependency)
• Git (for cloning repositories)

Apache 2.0

Implements Anthropic's Agent Skills specification.

Not affiliated with Anthropic. Claude, Claude Code, and Agent Skills are trademarks of Anthropic, PBC.