Skip to main content
Skills are installable packages that give an Operator OS agent new capabilities. A skill is a self-contained bundle — prompt fragments, tool definitions, reference documents, or any combination — that the agent loads at startup and uses automatically. Where MCP connects to external servers, Skills are installed locally into the agent’s workspace and travel with it. They are the recommended way to share reusable agent behaviors across deployments.

​
ClawHub registry

ClawHub is the default skills marketplace for Operator OS. It hosts community-contributed and officially-maintained skill packages indexed by capability. The ClawHub registry is enabled by default. No API key is required to search or install public skills.

​
Discovering and installing a skill

1

Search for a skill

Use the CLI to find skills matching a capability you need:
operator skills search "github integration"
The command fans out the query to all configured registries, merges results ranked by relevance score, and prints a table of matching slugs, display names, summaries, and versions.The agent can also search autonomously using the built-in find_skills tool:
find_skills(query="github integration", limit=5)
2

Review the skill details

Note the slug from the search results. You can inspect the skill’s summary and version in the search output before committing to installation.
3

Install the skill

Install the skill into the current workspace:
operator skills install <slug>
For example:
operator skills install github
To pin a specific version:
operator skills install github --version 1.2.0
The agent can also install skills autonomously using the install_skill tool:
install_skill(slug="github", registry="clawhub")
4

Restart the agent

Skills are loaded at agent startup. Restart the agent process for a newly installed skill to take effect.
operator restart

​
Where skills are stored

Skills are installed into the skills/ subdirectory of the agent’s workspace: 
{workspace}/
└── skills/
    ├── github/
    │   └── SKILL.md
    ├── docker-compose/
    │   └── SKILL.md
    └── postgres/
        └── SKILL.md
Each skill lives in its own directory named by slug. The primary file is SKILL.md, which contains the skill’s instructions, tool definitions, and any supporting content the agent needs.
Because skills are stored in the workspace directory, they are portable. Moving or syncing the workspace to another machine carries all installed skills with it.

​
Registry configuration

Registry settings live under tools.skills.registries in config.json. 
{
  "tools": {
    "skills": {
      "registries": {
        "clawhub": {
          "enabled": true,
          "base_url": "https://clawhub.ai",
          "search_path": "/api/v1/search",
          "skills_path": "/api/v1/skills",
          "download_path": "/api/v1/download"
        }
      }
    }
  }
}

​
ClawHub fields

​
registries.clawhub.enabled
boolean
default:"true"
Enable or disable the ClawHub registry. When false, ClawHub is not queried during search or install.
​
registries.clawhub.base_url
string
default:"https://clawhub.ai"
Base URL of the ClawHub instance. Override this if you are running a self-hosted ClawHub mirror.
​
registries.clawhub.search_path
string
default:"/api/v1/search"
API path used for skill search queries.
​
registries.clawhub.skills_path
string
default:"/api/v1/skills"
API path used to fetch metadata for a specific skill by slug.
​
registries.clawhub.download_path
string
default:"/api/v1/download"
API path used to download skill archives during installation.

​
Custom registry

You can point Operator OS at any registry that implements the ClawHub API contract. This is useful for:
  • Private internal skill libraries
  • Air-gapped deployments that cannot reach the public internet
  • Testing a self-hosted ClawHub instance
Add a second registry key alongside clawhub: 
{
  "tools": {
    "skills": {
      "registries": {
        "clawhub": {
          "enabled": true,
          "base_url": "https://clawhub.ai",
          "search_path": "/api/v1/search",
          "skills_path": "/api/v1/skills",
          "download_path": "/api/v1/download"
        },
        "internal": {
          "enabled": true,
          "base_url": "https://skills.corp.example.com",
          "search_path": "/api/v1/search",
          "skills_path": "/api/v1/skills",
          "download_path": "/api/v1/download"
        }
      }
    }
  }
}
When multiple registries are enabled, search requests are fanned out concurrently (up to 2 at a time by default) and results are merged and ranked by score before being returned. Install requests are routed to the registry specified by the registry argument of install_skill.
The custom registry must implement the same REST API paths as ClawHub. There is currently no additional authentication mechanism beyond what you can control at the base_url level (e.g. a reverse proxy with basic auth).

​
How the agent uses skills

After installation, skills are loaded when the agent starts. The agent reads each SKILL.md file found in {workspace}/skills/*/ and incorporates its content — additional instructions, tool descriptions, reference data — into its context. If a skill contributes tool definitions, those tools are registered alongside the built-in tools and become available to the agent immediately. The agent can also discover and install skills on its own during a conversation. When a user asks it to perform a task it does not currently know how to do, it may call find_skills to search for relevant skills and then install_skill to extend its own capabilities.

Build docs developers (and LLMs) love

Get started for freeTalk to us