A production-proven framework for building and running multi-role agent teams with automated review-fix-score loops.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-agent-team-orchestration
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 openclaw-agent-team-orchestration 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 Team Orchestration V3?
Agent Team Orchestration V3 provides a specialized architecture for managing complex content production pipelines using Openclaw Skills. It leverages a single-repo, multi-role design where a central orchestrator dispatches tasks to specialized sub-agents. By utilizing symlink-based file sharing and role-specific AGENTS.md configurations, this skill ensures high-fidelity output through rigorous automated quality gates.
The system is built on real-world production experience, addressing common bottlenecks in agent coordination. It establishes the file system as the central state center, removing the need for external databases or complex APIs. This Openclaw Skills implementation is ideal for teams needing to scale content creation without sacrificing technical accuracy or stylistic consistency.
Agent Team Orchestration V3 Use Cases
- Deploying a team of 3+ agents with specialized roles such as writer, reviewer, scorer, and fixer.
- Establishing a quality-controlled publishing pipeline that requires minimal human intervention.
- Bootstrapping agent workspaces with pre-configured symlinks and tool guardrails.
- Executing multi-round review-fix-score cycles to meet specific quality thresholds.
- Debugging complex agent interactions including spawn permissions and tool-call loops.
How Agent Team Orchestration V3 Works
- The Main Agent (Orchestrator) spawns a Writer agent to generate initial content in a dedicated OUTPUT directory.
- Once the draft is complete, the Orchestrator parallel-spawns Reviewer agents (Fact, Style, and Data) who access the output via symlinks.
- A Scorer agent is then spawned to evaluate the reviewers' feedback and generate a numerical quality score.
- If the score meets the predefined threshold (e.g., 8.5/10), the process concludes and notifies the user.
- If the score is below the threshold, a Fixer agent is spawned to address issues, and the workflow loops back to the review stage for up to three rounds.
Agent Team Orchestration V3 Setup
1. Initialize the Workspace
Use the provided setup script to generate the directory structure and symlinks for all roles:
scripts/setup-team.sh {writer-id} {fact-reviewer-id} {style-reviewer-id} {data-reviewer-id} {scorer-id} {fixer-id}
2. Configure Permissions
Update your openclaw.json file to allow the orchestrator to manage sub-agents. Ensure subagents.allowAgents is set to true for the main agent.
3. Define Roles
Create AGENTS.md files for each sub-agent using provided templates. The writer typically requires a 400-600 line domain manual, while reviewers utilize focused 100-200 line definitions.
Agent Team Orchestration V3 Data Schema & Taxonomy
The skill uses a flat file-system architecture to maintain state. Below is the organization of a standard Openclaw Skills team repo:
| Directory/File | Purpose |
|---|---|
OUTPUT/ |
The central repository for all generated content and versioned drafts. |
AGENTS.md |
Role definitions and instructions auto-loaded at agent spawn. |
TOOLS.md |
Mandatory guardrails to prevent tool-call loops and infinite retries. |
scripts/ |
Automation utilities for workspace setup and team deployment. |
references/ |
Documentation for architecture, workflows, and debugging lessons. |
Agent Team Orchestration V3 Advanced Features
- Symlink-based state sharing allowing multiple agents to read from a single source of truth without data duplication.
- Automated score parsing from JSON-formatted reports to drive logic gates.
- Explicit versioning control to ensure agents review the latest iterations during fix loops.
- Advanced TOOLS.md configurations designed to block recursive agent self-reads and reduce token waste.
- Support for absolute pathing in symlinks to maintain stability across different agent context switches.
SKILL.md
Loading
Related Openclaw Skills
MiniMax Image Generation Skill
Transform text prompts into high-quality visual assets using the MiniMax image-01 model via Openclaw Skills.
RenderingVideo Generator
A specialized assistant for drafting, validating, and generating temporary video previews using the RenderingVideo JSON schema.
Kubernetes Hardening Tool
An automated security API that generates production-ready hardening configurations and policy recommendations for Kubernetes clusters.
Active Directory Hardening Tool
An enterprise-grade security API that automates the generation of CISSP-aligned hardening configurations and deployment scripts for Active Directory environments.
Content Calendar Skill
A professional-grade content strategy and planning tool that transforms brand goals into actionable, multi-platform editorial calendars.
Lead Generation Skill
A robust lead generation and enrichment tool for building qualified B2B sales pipelines.
Featured*
