create-subagent

What This Skill Does

The create-subagent skill for OpenClaw empowers users to dynamically spawn, monitor, and control secondary AI agents to perform specialized tasks. By acting as a management layer, this skill allows for the delegation of complex workflows—ranging from software development and data analysis to deep research and creative writing—without cluttering the primary conversation context. It manages the full lifecycle of these agents, from initial instruction-setting and model selection to session monitoring and termination, ensuring you maintain a clean and efficient workspace.

Installation

To install this skill, run the following command in your terminal within the OpenClaw environment: clawhub install openclaw/skills/skills/aning35/create-subagent Ensure your openclaw-config.json file is correctly configured to permit subagent spawning, specifically setting the allowAgents list and defining maxConcurrent limits. You must restart the OpenClaw Gateway after updating your configuration for the changes to take effect.

Use Cases

• Software Development: Spawn a dedicated 'dev-agent' to handle architectural design, bug fixing, and unit testing while you focus on high-level feature requirements.
• Information Synthesis: Deploy a 'research-agent' to traverse multiple sources, extract relevant data, and compile structured reports automatically.
• Content Production: Use a 'writer-agent' to draft documentation, technical blog posts, or email templates with specialized tone and structure constraints.
• Complex Data Processing: Delegate heavy data cleaning and statistical analysis tasks to a 'data-agent' that handles Python script generation and execution.

Example Prompts

1. "I need to analyze some logs. Create a data-agent for me and point it to the file in /projects/data/logs."
2. "Can you set up a developer subagent to refactor the authentication module using our existing React and Node.js patterns?"
3. "Please launch a research-agent to find the latest trends in edge computing for 2024 and summarize them into a markdown document."

Tips & Limitations

• Execution Mode: Use run for transient, one-off tasks. Use session for long-running processes, though note that session mode requires thread binding which may not be supported on all web interfaces.
• Resource Limits: The system enforces a default limit of 8 concurrent agents and a max spawn depth of 2. Plan your hierarchy accordingly.
• Cleanup: All subagents have an automatic 60-minute inactivity timeout. Explicitly use /subagents kill to free up resources immediately if a task finishes early.
• Monitoring: Always utilize /subagents log to verify if your agent has encountered blockers, as agents operate autonomously until they signal completion.