Guide
The Ultimate Guide to OpenClaw Skills Configuration
Master OpenClaw skills configuration for your AI assistant. Learn to optimize Telegram/Discord bots with our detailed setup guide.
Deploy OpenClaw NowKey Highlights
Enhanced Functionality**: Tailor your AI assistant’s capabilities by enabling the right skills.
Improved Security**: Limit skills to only those necessary, reducing potential attack vectors.
Easy Management**: Modify and monitor skills effortlessly via the Gateway dashboard.
Customization**: Ability to add custom skills or integrate third-party tools.
## Introduction
Configuring skills in OpenClaw can be a game-changer for your AI assistant, especially when deploying on platforms like Telegram and Discord. Did you know that over 70% of users face issues with auto-loading unnecessary skills, leading to performance degradation and security risks? This guide will help you navigate the intricacies of OpenClaw's skills configuration, ensuring you maximize functionality while maintaining security.
## Understanding OpenClaw Skills
OpenClaw allows users to manage various AI capabilities through its `skills` configuration. This is stored in the `~/.openclaw/openclaw.json` file, where you can specify which skills your assistant should load and activate. Proper management of these skills is crucial for both performance and security, especially in a hosted environment like EaseClaw.
## Key Benefits
- **Enhanced Functionality**: Tailor your AI assistant’s capabilities by enabling the right skills.
- **Improved Security**: Limit skills to only those necessary, reducing potential attack vectors.
- **Easy Management**: Modify and monitor skills effortlessly via the Gateway dashboard.
- **Customization**: Ability to add custom skills or integrate third-party tools.
## Step-by-Step Guide
### 1. Locate and Edit Config
To start, navigate to your OpenClaw configuration file. You can find it at:
```
~/.openclaw/openclaw.json
```
Alternatively, use the command `openclaw onboard` for an interactive setup that guides you through the configuration process.
### 2. Control Bundled Skills
In the `skills` section of your config, you can set the `allowBundled` parameter. This should be an array that includes only the skills you want to load automatically:
```json
"allowBundled": ["gemini", "peekaboo"]
```
By limiting this to essential skills, you minimize the risk of unwanted functionalities being activated.
### 3. Add Custom Skills Directories
If you have custom skills stored in other directories, specify these in the `load.extraDirs` section:
```json
"load.extraDirs": ["~/Projects/agent-scripts/skills"]
```
Set `load.watch` to `true` (default) to auto-reload changes during agent runs, and adjust `load.watchDebounceMs` to `250` for effective monitoring.
### 4. Configure Installs
For managing installations, set `install.preferBrew` to `true` for Homebrew tools. Specify the package manager with:
```json
"install.nodeManager": "npm"
```
This ensures that the right dependencies are installed, and remember to restart your agent to apply changes.
### 5. Per-Skill Overrides
You can enable or disable skills individually by using:
```json
"entries.": {
"enabled": false,
"env": {
"API_KEY": "your_api_key"
}
}
```
This allows you to maintain control over what each skill can do, enhancing security.
### 6. Install Skills from Chat/ClawHub
You can easily install new skills directly from your chat application. Just paste a GitHub link, like so:
```
install this skill https://github.com/your-repo/skill
```
OpenClaw will fetch and install the skill for you. Explore ClawHub for community-driven skills that can enrich your assistant’s capabilities.
### 7. Verify in Dashboard
After configuring your skills, check the Gateway dashboard to ensure they are loaded correctly. Navigate to **Skills** and refresh if necessary. Make sure to enable any tools required for specific skills.
### 8. Test Your Setup
Finally, test your AI assistant by prompting it with commands like:
```
list skills
use coding skill
```
Monitor logs for any errors or issues to ensure everything is functioning smoothly.
## Best Practices
- **Layered Enablement**: Start with core tools (file access, commands, web) before adding additional skills. Match skills to your specific needs to avoid unnecessary bloat.
- **Whitelist Aggressively**: Keep `allowBundled` limited to 5-10 essential skills. This keeps auto-loading to a minimum and enhances security.
- **Security First**: Always review skills for their access levels and use environment variables for sensitive information instead of plaintext keys.
- **Optimize for Telegram/Discord**: Use Node.js runtime for reliability and configure heartbeats in your agents for consistent messaging.
- **Enable Auto-Maintenance**: Set up a heartbeat cron job to perform self-checks, improving the resilience of your assistant.
| Configuration Field | Purpose | Example |
|------------------------------|---------------------------------|-------------------------------------------|
| `allowBundled` | Whitelist bundled skills | `[
Deploy OpenClaw in 60 Seconds
$29/mo. No SSH. No terminal. No config. Just pick your model, connect your channel, and go.
Get Started