Clawflow is a recursive task delegation protocol that enables Openclaw Skills to collaborate through message-passing and Directed Acyclic Graphs (DAGs).
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install clawflow
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 clawflow 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 Clawflow Multi-Agent Orchestration Protocol?
Clawflow provides a standardized framework for multi-agent systems built with Openclaw Skills. It allows any agent to act as either a worker or a coordinator, breaking down complex projects into manageable subtasks. By leveraging the native Openclaw CLI, Clawflow ensures that agents can discover peers and communicate securely without needing external identity providers.
This protocol is essential for building scalable agentic pipelines where tasks are too large for a single LLM to handle efficiently. With Clawflow, Openclaw Skills gain the ability to delegate work recursively, meaning an agent can decompose a task into a sub-DAG, and those sub-agents can further decompose their assigned portions. This creates a fluid, hierarchical structure that mirrors real-world consulting firms or engineering teams, all powered by the robust infrastructure of Openclaw Skills.
Clawflow Multi-Agent Orchestration Protocol Use Cases
- Coordinating complex software development projects across specialized Openclaw Skills.
- Decomposing large-scale data analysis into parallelizable subtasks for faster processing.
- Building multi-agent pipelines where research, coding, and testing are handled by different agents.
- Automating multi-step business workflows that require sequential dependency resolution.
How Clawflow Multi-Agent Orchestration Protocol Works
- An agent receives a task message via the Openclaw Skills interface.
- The agent evaluates if the task can be completed solo or requires delegation to peers.
- If delegating, the agent decomposes the request into a Directed Acyclic Graph (DAG) and stores it in the workspace.
- Subtasks are dispatched to available peers using the openclaw agent command.
- The coordinator monitors the mailbox for incoming replies and updates the DAG status.
- Once all dependencies are resolved, the agent synthesizes the results and replies to the original requester.
Clawflow Multi-Agent Orchestration Protocol Setup
Ensure you have the latest version of Openclaw installed. Configure your agents in openclaw.json and verify they are discoverable to use with Openclaw Skills.
# List available agents to ensure discovery is working
openclaw agents list
# Send a test message to initialize the protocol
openclaw agent --agent <agent-id> --message "Initialize clawflow protocol"
The skill will automatically create the necessary mailbox/ and tasks/ directories within the agent's workspace to manage the Openclaw Skills workflow.
Clawflow Multi-Agent Orchestration Protocol Data Schema & Taxonomy
Clawflow organizes data within the agent's private workspace to ensure state persistence and auditability for Openclaw Skills.
| Directory | Purpose |
|---|---|
| mailbox/inbox/ | Stores incoming task messages before they are processed. |
| mailbox/outbox/ | Logs all outgoing dispatches and replies sent to other agents. |
| mailbox/archive/ | A durable audit trail of all completed message exchanges. |
| tasks/{task-id}/ | Contains task.md, which defines the DAG, tracks subtask progress, and stores final results. |
Clawflow Multi-Agent Orchestration Protocol Advanced Features
- Recursive DAG nesting allowing for infinite depth in agent delegation across Openclaw Skills.
- Parallel task execution (fan-out) to optimize performance across multiple Openclaw Skills instances.
- Automatic dependency resolution to ensure subtasks trigger only when prerequisites are met.
- Private workspace isolation ensures each agent maintains its own working memory and scratchpad for high-security Openclaw Skills environments.
SKILL.md
Loading
Related Openclaw Skills
Medical Revenue Audit
An automated auditing tool that cross-references surgical logs with billing records to identify and recover lost revenue.
Filesystem Operations
A comprehensive toolkit for performing advanced file management, recursive searching, and batch processing via the command line.
ClawHub Skill Manager
A dedicated command-line utility for discovering, installing, and maintaining skills from the official ClawHub registry.
Base-8004: Onchain AI Agent Identity
A specialized tool for registering AI agents on the Base network using the ERC-8004 protocol to establish permanent, verifiable onchain identities and reputation.
Binance Spot Trader
An autonomous trading bot that combines technical indicators with LLM-powered market sentiment analysis to execute trades on the Binance spot market.
Solana Sniper Bot
An autonomous Solana trading bot that monitors liquidity pools and uses AI to detect rugpulls before executing trades.
Featured*
