Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Agent OS is a persistent multi-agent orchestration system for OpenClaw. It gives agents memory across sessions, routes tasks by capability, and tracks project state so work survives restarts. Multiple agents coordinate on a shared project without duplicating effort.
Unlike stateless agent frameworks, Agent OS persists memory and project state to disk, so agents accumulate knowledge and projects resume mid-execution without restarting from scratch.
When to use it
- Running a multi-phase software project across planning, design, and development agents
- Resuming a long-running agent task after a server restart
- Assigning research subtasks to a specialized research agent while a dev agent handles implementation
- Tracking which agent is blocked and why during complex project execution
- Building an agent pipeline where task B only starts after task A completes
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Build a new feature with planning, design, and development phases
- 1Register three agents with capabilities: research/planning, design/planning, development
- 2Call os.initialize() to set up the system
- 3Call os.runProject('Build a feature', ['planning', 'design', 'development'])
- 4TaskRouter decomposes the goal into sequential tasks and matches each to the best-fit agent
- 5Executor runs tasks one by one, persisting state to disk after each completion
Project completes at 100% progress with all task outputs saved; agents retain lessons learned for future runs