A meta-agent skill that orchestrates complex macro tasks by decomposing them into subtasks performed by autonomous, specialized sub-agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-orchestrator
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-orchestrator using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Orchestrator?
The Agent Orchestrator is a powerful meta-skill designed to handle sophisticated projects that exceed the capacity of a single agent. By acting as a central coordinator, it breaks down high-level objectives into parallelizable subtasks, dynamically generates dedicated SKILL.md files for specialized sub-agents, and manages their entire lifecycle from creation to dissolution. This approach within Openclaw Skills ensures that complex work is handled with modularity and precision.
This skill leverages a robust file-based communication protocol, allowing multiple agents to work in isolation while maintaining a clear path for data consolidation. It is particularly effective for large-scale engineering, research, and content creation tasks where clear boundaries and explicit handoffs are required to maintain quality and coherence.
Agent Orchestrator Use Cases
- Orchestrating large-scale market research by delegating data collection, analysis, and writing to specialized agents.
- Managing complex software development features by spawning separate agents for implementation, unit testing, and documentation.
- Coordinating multi-stage content pipelines where drafting and quality assurance review happen in parallel agent environments.
- Automating high-volume data processing tasks across multiple autonomous workspaces using Openclaw Skills.
How Agent Orchestrator Works
- Task Decomposition: The meta-agent analyzes the macro task to identify success criteria and breaks it into independent, parallelizable subtasks.
- Agent Generation: It creates dedicated workspaces for each subtask, including a dynamically generated SKILL.md that defines the agent's specific role.
- Agent Dispatch: The orchestrator initializes agents by writing instructions to their inbox and spawning them via the Task tool.
- Checkpoint Monitoring: It periodically checks a status.json file within each agent's workspace to track progress from pending to completed.
- Consolidation: Once tasks are finished, the orchestrator collects outputs from individual outboxes, validates them, and merges them into a final deliverable.
- Dissolution: After a final summary is generated, the meta-agent archives or cleans up the temporary sub-agent workspaces.
Agent Orchestrator Setup
To begin using the orchestrator within the Openclaw Skills framework, ensure your environment has the necessary agent creation scripts.
# Create a sub-agent workspace
python3 scripts/create_agent.py <agent-name> --workspace <path>
# Dissolve and archive agents after task completion
python3 scripts/dissolve_agents.py --workspace <path> --archive
Agent Orchestrator Data Schema & Taxonomy
The skill follows a strict directory and file-based communication protocol for data organization:
| File/Folder | Purpose |
|---|---|
inbox/ |
Read-only for the sub-agent; contains the initial instructions and input files. |
outbox/ |
Write-only for the sub-agent; stores the completed deliverables for the orchestrator to collect. |
workspace/ |
The sub-agent's private area for processing and temporary file storage. |
status.json |
A state-tracking file containing keys like state (pending, running, completed, failed) and timestamps. |
SKILL.md |
The core definition file generated by the orchestrator to guide the sub-agent's behavior. |
Agent Orchestrator Advanced Features
- Dynamic SKILL.md generation for real-time specialization of autonomous sub-agents.
- Dependency graph management for handling sequential subtasks that require specific handoff orders.
- Automated output validation against success criteria defined during the decomposition phase.
- Integration with pre-built agent templates for research, analysis, and quality assurance within Openclaw Skills.
- Scalable architecture allowing for the management of dozens of parallel agents simultaneously.
SKILL.md
Loading
Related Openclaw Skills
Browser Auto Download
An intelligent browser-based automation tool that handles dynamic web content and multi-step download navigation where standard CLI tools fail.
OpenClaw Research Tool
A powerful CLI tool that provides AI agents with live web access and deep reasoning capabilities via OpenRouter.
Huckleberry Baby Tracker
A comprehensive AI agent skill that enables natural language tracking of baby sleep, feeding, diapers, and growth via the Huckleberry app API.
Apple Media Control
A comprehensive command-line integration for managing Apple TV, HomePod, and AirPlay devices using the pyatv library.
Communication
A sophisticated skill for Openclaw Skills that enables AI agents to synthesize deep context and emotional subtext to draft thoughtful, intentional responses.
z-article-card
Converts long articles into a series of 3:4 ratio pagination cards for social media sharing.
Featured*
