Workflow:Openclaw Openclaw Plugin And Skill Extension
| Knowledge Sources | |
|---|---|
| Domains | Extensibility, Plugin_System, Agent_Tools |
| Last Updated | 2026-02-06 12:00 GMT |
Overview
End-to-end process for extending OpenClaw capabilities by installing plugins (new channels, tools, services) and creating or installing skills (agent instructions and prompt-based abilities).
Description
This workflow covers both extension mechanisms in OpenClaw: plugins and skills. Plugins are npm packages that register new channels, tools, RPC methods, or services at the gateway level. Skills are markdown-based instruction files that extend the agent's prompt with new capabilities. Together they allow users to add functionality like voice calling, new messaging platforms, custom tools, and specialized knowledge without modifying the core codebase.
Key outputs:
- Installed and configured plugins providing new gateway capabilities
- Custom or community skills loaded into agent sessions
- Extended tool palette available to the agent during conversations
- New channels or services registered with the gateway
Usage
Execute this workflow when the built-in capabilities of OpenClaw do not cover a specific need. Use plugins when adding a new channel (e.g., Microsoft Teams, Matrix), a new service (e.g., voice calls), or custom RPC endpoints. Use skills when extending the agent's knowledge or adding prompt-based abilities (e.g., code review patterns, domain expertise).
Execution Steps
Step 1: Identify the Extension Type
Determine whether the desired functionality requires a plugin (gateway-level code) or a skill (agent-level prompt instructions). Plugins run as Node.js code with access to the gateway API. Skills are markdown files loaded into the agent context.
Key considerations:
- Plugins: for new channels, custom tools with code execution, background services
- Skills: for prompt engineering, agent behavior customization, knowledge injection
- Some features need both (a plugin provides the tool, a skill provides instructions)
- Check ClawHub for existing community skills before creating new ones
Step 2: Install a Plugin
Install the plugin package using the CLI or by adding it to the configuration. Plugins are npm packages that follow the OpenClaw plugin manifest format. Official extensions live in the extensions/ directory of the monorepo.
What happens:
- Plugin package downloaded and installed (npm install --omit=dev in plugin directory)
- Plugin manifest (openclaw.plugin.json) validated for required fields
- Plugin entry point loaded and registered with the gateway
- New channels, tools, or RPC methods become available
- Gateway restart required for plugin changes to take effect
Step 3: Install or Create a Skill
Install a skill from ClawHub or create a custom skill in the workspace skills directory. Skills are SKILL.md files with YAML frontmatter (name, description) and markdown body containing agent instructions.
What happens:
- ClawHub skills installed to ~/.openclaw/skills/ (managed) or workspace skills/
- Custom skills created as SKILL.md in ~/.openclaw/workspace/skills/<skill-name>/
- Skills auto-discovered on gateway restart (no config entry needed for workspace skills)
- Skill eligibility checked against requirements (binary dependencies, env vars, config flags)
- Eligible skills injected into agent system prompt
Step 4: Configure the Extension
Set any required configuration for the installed plugin or skill, including API keys, environment variables, and feature flags. Configuration lives in the main config file under the appropriate section.
Key considerations:
- Plugin config: plugins.entries.<id>.config in openclaw.json
- Skill config: skills.entries.<skill-name> with enabled flag, env overrides, API keys
- Environment variables can be set per-skill via skills.entries.<name>.env
- Plugin channels need their own channel configuration section
Step 5: Verify the Extension
Confirm the plugin or skill is loaded and functioning. List installed plugins and skills, check that new tools appear in the agent's tool list, and test the new capability with a real message.
What happens:
- Plugin list shows installed and enabled plugins
- Skill list shows eligible and loaded skills
- New tools visible in agent tool inventory
- Test message exercises the new capability
- Logs checked for any loading errors or missing dependencies