create-subagent

What This Skill Does

The create-subagent skill acts as a powerful orchestrator for the OpenClaw ecosystem, allowing users to spawn and manage specialized AI agents (SubAgents) to handle complex, multi-step workflows. Instead of overloading your main AI agent with every detail, you can delegate specific domains—like code development, research, writing, or data analysis—to dedicated agents designed for those tasks. This skill provides a structured framework for defining a system prompt (task), assigning a unique label, and choosing an execution mode (run or session) to ensure optimal task handling and lifecycle management.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/aning35/create-subagent

Ensure that your openclaw-config.json file is configured correctly to permit agent spawning. After modifying your configuration, please restart the OpenClaw Gateway to apply the changes. The skill is maintained within the openclaw/skills repository and authored by aning35.

Use Cases

• Software Engineering: Deploy a 'dev-agent' to handle codebase refactoring, bug fixes, or scaffold new features.
• Research and Intelligence: Use a 'research-agent' to browse the web, aggregate information, and summarize complex topics into concise reports.
• Content Creation: Utilize a 'writer-agent' to draft technical documentation, blog posts, or emails while maintaining specific tone requirements.
• Data Operations: Invoke a 'data-agent' to clean datasets, perform statistical analysis, or generate visualizations via Python scripts.

Example Prompts

1. "I need to build a new microservice in Node.js, can you spin up a dev-agent to help me with the architecture?"
2. "Please create a research-agent to find the latest trends in serverless architecture and summarize them for my presentation."
3. "Show me all my currently active SubAgents and tell me which one is handling the data analysis task."

Tips & Limitations

• Session Management: Remember that 'session' mode requires thread binding support, which is not currently available in all channels like Webchat. Stick to 'run' mode for those environments.
• Lifecycle: SubAgents are automatically archived 60 minutes after their task completion to keep your workspace clean.
• Concurrency: You are limited to 8 concurrent SubAgents by default; monitor your active tasks using /subagents list to avoid hitting capacity limits.
• Steering: If a SubAgent is going off-track, use the /subagents steer command to provide corrective feedback without needing to kill and restart the instance.