ai-specialists

What This Skill Does

The ai-specialists skill integrates OpenClaw with the AI Specialists Hub, enabling sophisticated management and interaction with specialized AI agents. This skill acts as a bridge to Model Context Protocol (MCP) servers, allowing the user to treat individual AI agents as modular workers. Users can list available specialists, hire or dismiss them, explore their internal file trees, and manipulate their documentation or workspace configuration files directly.

Installation

To integrate this skill, use the OpenClaw command-line interface: clawhub install openclaw/skills/skills/erikashby/ai-specialists

Ensure your target MCP server is running and that you have the appropriate endpoint URL provided by your configuration or user settings. The skill uses standard HTTP POST requests with JSON-RPC headers to bridge commands between OpenClaw and the remote specialist environment.

Use Cases

This skill is ideal for complex workflows requiring multi-agent delegation. Use cases include:

• Agent Orchestration: Managing a fleet of specialists like Ruby, Peter, or Benjamin, each focused on specific domains like coding, writing, or data analysis.
• Documentation Management: Keeping instructional files (like core-instructions.md) synced across a team of agents to maintain behavioral consistency.
• Workspace Organization: Cleaning up, restructuring, or auditing the directories managed by individual specialists.

Example Prompts

• "List all my currently hired specialists and show me the overview for Ruby."
• "Create a new folder named 'research-notes' in Peter's workspace and move the summary document into it."
• "What are the behavioral instructions for the specialist named Benjamin? Read his core-instructions.md file."

Tips & Limitations

• Identifier Discipline: Always use the machine-readable ID (e.g., 'ruby') retrieved from list_specialists rather than a display name, as API calls rely on the unique ID.
• Initialization: Before engaging with a new agent, always read ai-instructions/core-instructions.md and getting_started.md to ensure you understand their operational bounds and expected initialization patterns.
• Connectivity: Ensure your MCP server headers allow text/event-stream and application/json, or the system will reject the request with a 406 error. Use curl to verify the connection independently if you suspect configuration issues.