Claude Skills Integration Templates

Drop-in Markdown snippets that enable any Agent CLI to discover and use Claude Skills — even if you primarily use a different CLI than Claude Code.

🚀 Try Now: Give your web-capable agent this URL and say "install Claude Skills integration" → https://github.com/calt13/claude-skills-integration-templates

🎯 What This Solves

Claude Skills are auto-discovered from three sources: ~/.claude/skills/, .claude/skills/ (project), and plugin Skills.[[1]](https://docs.claude.com/en/docs/claude-code/skills) But what if you're using Codex CLI, Gemini CLI, Kimi CLI or other CLIs alongside Claude Code?

This repo provides AGENTS.md templates that teach any agent how to:

• ✅ Discover Skills from Claude Code's directories (~/.claude/plugins/marketplaces/)
• ✅ Scan neutral directories (~/.agent/skills/, ./.agent/skills/)
• ✅ Load Skills progressively to avoid context bloat

Key insight: Skills aren't Claude-specific — you can point any model at a skills folder and say "read pdf/SKILL.md" and it will work.[[7]](https://simonwillison.net/2025/Oct/16/claude-skills/)

🔄 Two Usage Scenarios

Scenario A: Claude Code User + Other CLIs

You already use Claude Code and installed Skills via:

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills

Skills install to ~/.claude/plugins/marketplaces/owner/repo/skills/[[1]](https://docs.claude.com/en/docs/claude-code/plugin-marketplaces)

Now you want Codex/Gemini/Kimi to use the same Skills → Just give them this URL and let them handle the rest!

🚀 Easiest Way (for web-capable agents):

Simply tell your agent: "Install Claude Skills integration from https://github.com/calt13/claude-skills-integration-templates"

📋 Manual Way (if needed):

# Download template
mkdir -p .agent
curl -o .agent/CLAUDE_SKILLS_INTEGRATION.md \
  https://raw.githubusercontent.com/calt13/claude-skills-integration-templates/main/templates/generic/CLAUDE_SKILLS_INTEGRATION.md

# Append to AGENTS.md (or let your CLI auto-discover)
cat .agent/CLAUDE_SKILLS_INTEGRATION.md >> AGENTS.md

# Append to GEMINI.md (for gemini-cli)
cat .agent/CLAUDE_SKILLS_INTEGRATION.md >> GEMINI.md

Benefits:

• ✅ No duplicate installations
• ✅ Single source of truth (~/.claude/plugins/marketplaces/)
• ✅ Updates via /plugin marketplace update benefit all CLIs

Scenario B: Non-Claude Code User Wanting Skills

You don't use Claude Code, but want to enjoy the Skills ecosystem.

Solution: Install Skills to neutral directories:

# Option 1: Clone Skills repo directly
git clone https://github.com/anthropics/skills ~/.agent/skills/anthropic-skills

# Option 2: Clone community Skills
git clone https://github.com/obra/superpowers ~/.agent/skills/superpowers
git clone https://github.com/travisvn/awesome-claude-skills ~/.agent/skills/awesome-skills

# Option 3: Project-local Skills
mkdir -p ./.agent/skills/my-custom-skill

Then use our template to teach your CLI about these locations.

Benefits:

• ✅ CLI-agnostic setup
• ✅ Easy version control
• ✅ Shareable with team via git

📦 Available Templates

🚀 Quick Start

🆕 Easiest Method: Have a web-capable agent? Just give it this URL and say "install Claude Skills integration" → No manual steps needed!

Method 0: Zero-Setup Installation (For Web-Capable Agents) ⭐

The easiest way - if your agent has internet access, simply give it this URL:

https://github.com/calt13/claude-skills-integration-templates

Then ask your agent:

"Please install Claude Skills integration from this URL"

What happens: Your agent will automatically fetch the templates, understand the installation process, and complete the setup for you.

✅ Best for: Codex CLI, Kimi CLI, Gemini CLI, or any web-enabled agent ⚡ Effort: Zero manual steps required 🎯 Success rate: High (agents can adapt the installation to your specific environment)

Try this first! If your agent successfully discovers and uses skills, you're all set. If not, proceed to the manual methods below.

Method 1: Auto-Discovery (if your CLI scans .agent/)

mkdir -p .agent
curl -o .agent/CLAUDE_SKILLS_INTEGRATION.md \
  https://raw.githubusercontent.com/calt13/claude-skills-integration-templates/main/templates/generic/CLAUDE_SKILLS_INTEGRATION.md

# Your CLI's /init will process this (effectiveness may vary by CLI)
/init

Note: The /init command behavior varies across different CLIs - some may effectively integrate the template while others might not process it as expected. If you find skills aren't being discovered after /init, we recommend using Method 0, Method 2 or Method 3 for more reliable results.

Method 2: Manual Append

curl -o /tmp/skills-integration.md \
  https://raw.githubusercontent.com/calt13/claude-skills-integration-templates/main/templates/generic/CLAUDE_SKILLS_INTEGRATION.md

cat /tmp/skills-integration.md >> AGENTS.md

Method 3: Copy-Paste

Just copy the content and paste into your AGENTS.md.

Method 4: Local-Only Setup (Security-Focused)

For enhanced security or isolated environments, use the local-only template that only scans project directories:

# Download local-only template (scans only ./.agent/skills/ and ./.claude/skills/)
curl -o .agent/CLAUDE_SKILLS_INTEGRATION.md \
  https://raw.githubusercontent.com/calt13/claude-skills-integration-templates/main/templates/generic/local-only/CLAUDE_SKILLS_INTEGRATION.md

# Optional: Link global skills to project for controlled access
ln -s ~/.claude/plugins/marketplaces/anthropics-skills ./.agent/skills/anthropic-skills

Benefits:

• ✅ No global directory scanning (enhanced security)
• ✅ Project isolation prevents unintended skill access
• ✅ Compatible with symbolic linking for controlled skill sharing
• ✅ Ideal for CI/CD environments and shared development containers

🤖 For Web-Capable Agents

If your agent can access web content, installation is even easier:

One-Step Installation

Simply tell your agent:

"Please install Claude Skills integration from https://github.com/calt13/claude-skills-integration-templates"

What your agent will do:

1. Fetch this repository
2. Read the installation instructions
3. Download the appropriate template
4. Integrate it into your configuration
5. Test skill discovery

Example conversations:

• "Can you install Claude Skills for me?"
• "I want to use Claude Skills with this project"
• "Help me set up skill discovery"

Supported agents: Codex CLI, Kimi CLI, Gemini CLI, Claude Code, and any agent with web access capabilities.

Why this works: Agents with web access can read this README, understand the installation process, and adapt it to your specific environment automatically.

📚 Skills Directory Structure

After setup, your agent will scan:

~/.claude/
├── plugins/
│   └── marketplaces/
│       ├── anthropics-skills/        ← From Claude Code plugins
│       │   └── document-skills/
│       │       └── pdf/
│       │           └── SKILL.md
│       └── obra-superpowers/
│           └── skills/
│               └── brainstorm/
│                   └── SKILL.md
│
└── skills/                            ← Personal Skills (manual)
    └── my-skill/
        └── SKILL.md

~/.agent/
└── skills/                             ← Neutral, CLI-agnostic
    └── community-skills/
        └── latex/
            └── SKILL.md

./.agent/
└── skills/                              ← Project-local
    └── project-skill/
        └── SKILL.md

Skills are stored as directories containing a SKILL.md file. Personal Skills go in ~/.claude/skills/, Project Skills in .claude/skills/.[[1]](https://docs.claude.com/en/docs/claude-code/skills)

🎨 Template Features

Each template teaches agents to:

1. Scan Standard Directories

   ./.agent/skills/ 
   ./.claude/skills/
   ~/.agent/.skills/                             
   ~/.claude/plugins/marketplaces/*/skills/  
   ~/.claude/skills/                      

1. Progressive Loading Metadata provides just enough info for Agent to know when each skill should be used without loading all of it into context.[[3]](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills)

• Level 1: Scan YAML frontmatter (name, description)
• Level 2: Load full SKILL.md when triggered
• Level 3: Execute scripts/load resources on demand

1. Manual Trigger

• "Do you have any skills?"
• "Can you find available Claude Skills?"
• "List installed skills"

📖 How Claude Code Skills Work

To use plugins from a marketplace, run /plugin marketplace add user-or-org/repo-name[[2]](https://www.anthropic.com/news/claude-code-plugins)

Example workflow:

# Add official Skills marketplace
/plugin marketplace add anthropics/skills

# Install specific Skills
/plugin install document-skills@anthropic-agent-skills

# Skills install to:
# ~/.claude/plugins/marketplaces/anthropics-skills/document-skills/

You can register any GitHub repo as a marketplace[[2]](https://github.com/anthropics/skills)

## 🛠️ Customization

### Local-Only Security Configuration

The [`local-only template`](templates/generic/local-only/) provides enhanced security by restricting skill discovery to project directories only:

Create project-local skills directory

mkdir -p ./.agent/skills

Link specific global skills (optional, for controlled access)

ln -s ~/.claude/plugins/marketplaces/anthropics-skills/document-skills ./.agent/skills/pdf-tools

Use local-only template

curl -o .agent/CLAUDE_SKILLS_INTEGRATION.md \ https://raw.githubusercontent.com/calt13/claude-skills-integration-templates/main/templates/generic/local-only/CLAUDE_SKILLS_INTEGRATION.md

**Security Benefits**:
- No scanning of user home directory (`~/.claude/`)
- Prevents accidental access to personal skills
- Ideal for shared development environments
- Compatible with containerized workflows

**Use Cases**:
- CI/CD pipelines requiring project isolation
- Team environments with strict security policies
- Development containers with limited filesystem access
- Projects requiring explicit skill whitelisting

### Add Custom Scan Paths

Edit the template:

• ./.agent/skills/ (Project-local skills)
• ~/.agent/skills/ (User-global skills)
• ./.claude/skills/ (Project-local, Claude Code compatible)
• ~/.claude/skills/ (User-global, Claude Code compatible)
• ~/.claude/plugins/marketplaces/ (User-global, Claude Code compatible)
• ~/my-company/shared-skills/ (Company Skills)

## Troubleshooting & FAQ (For Users)

This section helps you understand the agent's behavior and troubleshoot skill discovery.

### Q: The agent doesn't seem to be using my skills. How can I check if it even found them?

**A:** The agent scans for skills "on-demand" (lazily) when it first needs them, or when you explicitly ask.

The single best way to test your skill setup is to **ask the agent directly**:

What skills do you have?

The agent is instructed to respond by listing all skills it found, grouped by "Project Skills" and "User Skills."

* **If you see your skill in this list:** Congratulations, it's installed correctly.
* **If your skill is not on this list:** See the next question.

### Q: I asked, but my new skill isn't on the list. What's wrong?

**A:** This is almost always an issue with the skill's location or its metadata file. The agent **only** scans the following specific directories:

* `./.agent/skills/` (Project-local)
* `~/.agent/skills/` (User-global)
* `./.claude/skills/` (Project-local, Claude compatible)
* `~/.claude/skills/` (User-global, Claude compatible)
* `~/.claude/plugins/marketplaces/` (User-global, Claude compatible)

**Troubleshooting Checklist:**
1.  **Check Location**: Is your skill's directory (e.g., `my-skill/`) placed *directly* inside one of the locations listed above?
2.  **Check `SKILL.md` File**: Does your skill directory contain a valid `SKILL.md` file (e.g., `./.agent/skills/my-skill/SKILL.md`)?
3.  **Check YAML Frontmatter**: Does the `SKILL.md` file start with a valid YAML block (between `---` lines) containing at least a `name` and `description`? The agent *only* reads this block during discovery.

### Q: I'm using the local-only template but skills aren't being found. What's different?

**A:** The local-only template only scans project directories (`./.agent/skills/` and `./.claude/skills/`). If you're using symbolic links:

1.  **Check symlink validity**: Ensure the linked directory exists and contains a valid `SKILL.md`
2.  **Check permissions**: Verify the agent can read through the symbolic link
3.  **Use absolute paths**: Consider using absolute paths in symlinks for reliability

**Example setup:**

Create project skills directory

mkdir -p ./.agent/skills

Link specific skills (avoid linking entire marketplaces)

ln -s ~/.claude/plugins/marketplaces/anthropics-skills/document-skills ./.agent/skills/pdf-tools

Verify the link works

ls -la ./.agent/skills/pdf-tools/SKILL.md

**Note**: The local-only template is designed for security isolation. If you need global skill access, use the standard generic template instead.

### Q: How do I know the agent is *actually* using the skill for a task?

**A:** We have instructed the agent (in its `AGENTS.md` prompt) to be transparent.

When it successfully matches your task to a skill, it **should** state which skill it is using *before* it runs.

* **Example:** If you ask it to process a PDF, you should see a response like: "Okay, I will use the `pdf-tools` skill for that."
* **What if it doesn't?** If the agent just completes the task without mentioning a skill, it means it failed to match your request to any of its known skills and fell back to its general knowledge. You can try:
    1.  Re-phrasing your request to be more specific.
    2.  Checking your skill's `description` in its `SKILL.md` to ensure it clearly matches the kind of task you're asking for.

## 🤝 Contributing

We welcome contributions to improve and extend these skill integration prompt templates.

The most common way to contribute is by **adding or refining a template** for a specific agent CLI.

If you wish to add a template for a new CLI:

1.  Create a new template file, for example: `templates/your-agent-spec-name/ANOTHER_SKILLS_INTEGRATION.md`.
2.  In this file, write the prompt snippet. It can be based on the `generic` template but optimized with CLI-specific conventions (e.g., using special tool-call syntax instead of `Bash()`).
3.  Test your prompt snippet by appending it to a real agent's prompt file and verifying it works.
4.  Submit a Pull Request.

For more advanced contributions, such as improving the core prompt logic or contributing to the next-generation architecture, please see our detailed **[CONTRIBUTING.md](CONTRIBUTING.md)** file.

## 📄 License

MIT License - see [LICENSE](LICENSE)

## 🔗 Related Projects

- **[anthropics/skills](https://github.com/anthropics/skills)** - Public repository for Skills[[2]](https://github.com/anthropics/skills)
- **[obra/superpowers](https://github.com/obra/superpowers)** - Core skills library[[5]](https://github.com/obra/superpowers)
- **[awesome-claude-skills](https://github.com/travisvn/awesome-claude-skills)** - Curated list[[6]](https://github.com/travisvn/awesome-claude-skills)
- **[Claude Code Plugins Marketplace](https://claudecodemarketplace.com/)** - Browse plugins

## 🙏 Acknowledgments

- Anthropic's [Agent Skills documentation](https://docs.claude.com/en/docs/claude-code/skills)[[7]](https://simonwillison.net/2025/Oct/16/claude-skills/)
- Community developers building Skills ecosystem

## 📮 Support

- **Issues**: [GitHub Issues](https://github.com/calt13/claude-skills-integration-templates/issues)
- **Discussions**: [GitHub Discussions](https://github.com/calt13/claude-skills-integration-templates/discussions)

---

**Made with ❤️ for the multi-CLI agent community**

*Share Skills across all your development tools — no duplication needed!*