Multi-Agent

The Multi-Agent skill lets your primary OpenClaw agent spawn, coordinate, and aggregate results from multiple specialized sub-agents running in parallel. Define agent roles — researcher, writer, critic, coder — each with their own soul.md personality fragment and skill set, then assign them sub-tasks from a larger goal. The orchestrator agent collects outputs, resolves conflicts, and synthesizes a final result. Communication between agents uses a shared message bus with role-based access, preventing agents from overwriting each other's work. Sub-agents can be stateful across multiple turns or stateless one-shot executors depending on task requirements. Resource limits (max tokens, max iterations, allowed tools) are set per sub-agent to prevent runaway costs. Task dependency graphs are supported — define which sub-agents must complete before others can start. Execution logs show each agent's reasoning chain for full transparency. Ideal for complex research, automated code review pipelines, and content production workflows that benefit from multiple independent perspectives combined into a single high-quality output.

Installation

clawhub install multi-agent

Install: clawhub install multi-agent