Agent Team

What This Skill Does

The Agent Team skill for OpenClaw is a sophisticated multi-agent orchestration framework designed to manage and deploy specialized AI agents. Unlike standard single-agent interactions, this skill allows you to build a collaborative team of 'soul-defined' agents, each with unique identity configurations, specific models, and dedicated roles. Whether you are coding, writing, data-analyzing, or researching, Agent Team acts as the central hub to spawn specialized workers or engage them in long-term interactive chat sessions. It leverages OpenClaw’s workspace to maintain persistent agent personas, ensuring that every task is executed with the context and expertise required for that specific domain.

Installation

To integrate this agent management system into your local OpenClaw environment, use the primary installation command provided by ClawHub:

clawhub install openclaw/skills/skills/jiangzhiyu/agent-team

Once installed, you must ensure your system has write permissions to ~/.openclaw/workspace/agents/ as the tool relies on this directory to dynamically load your agent configurations (SOUL.md and config.json). After installation, you can verify the setup by running agent-team list to see the pre-installed team members.

Use Cases

1. Software Development Life Cycle: Utilize the 'coder' agent for initial implementation, then chain the 'reviewer' agent to perform quality assurance on the generated output before using the 'writer' agent to document the new features.
2. Academic or Market Research: Deploy the 'researcher' agent to conduct deep dives into specific topics, then pipe that raw output into the 'analyst' agent to synthesize findings into actionable charts or summaries.
3. Content Production: Use the 'writer' agent in interactive chat mode to maintain a consistent tone and style across a series of blog posts or technical reports.

Example Prompts

1. 'agent-team spawn coder "Create a Python function to parse JSON data from an API response and handle potential network timeouts."'
2. 'agent-team spawn researcher "Generate a comparative analysis table between current open-source LLMs and proprietary models."'
3. 'agent-team chat reviewer' (followed by) 'Review the code I just generated for logic errors and security vulnerabilities.'

Tips & Limitations

• Configuration Hierarchy: Always define your primary model in config.json. If a model fails, the system automatically falls back to the configured secondary model, so ensure both are valid API endpoints.
• Performance: For complex tasks, use the spawn command to isolate the environment, as this prevents context window overflow compared to long-running chat sessions.
• File System Limits: Since each agent stores its identity in local files, ensure no two agents share a name to avoid collision errors in the ~/.openclaw/workspace/agents/ directory.
• Model Selection: Choose your model based on the requirement; use qwen3-max for complex reasoning tasks, while qwen3-coder-next provides the most stable performance for coding-specific workloads.