A production-grade skill for orchestrating multiple Claude Code agents to work in parallel on complex development tasks through shared coordination.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claude-code-teams
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 claude-code-teams 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 Claude Code Agent Teams?
Claude Code Agent Teams is a sophisticated implementation of Openclaw Skills designed to manage native multi-agent orchestration. It allows developers to break down monolithic tasks into parallel workstreams where specialized agents coordinate through a shared task list and direct messaging. By utilizing this skill, users can move beyond simple sequential prompting and leverage a team of agents to perform high-level architectural debates, full-stack implementations, and deep-dive investigations simultaneously.
This workflow optimizes developer productivity by assigning specific scopes to different agents, ensuring that complex projects are handled with high precision and speed. As part of the broader ecosystem of Openclaw Skills, this tool provides the necessary scripts and templates to enable, monitor, and clean up multi-agent sessions while managing the inherent token costs and coordination overhead associated with parallel AI workflows.
Claude Code Agent Teams Use Cases
- Executing parallel independent work such as multi-lens code reviews focusing on security, performance, and unit testing simultaneously.
- Debugging complex systems by deploying agents to test competing hypotheses in parallel to find root causes faster.
- Developing full-stack features where separate agents handle frontend components, backend logic, and integration tests.
- Facilitating architecture decisions through adversarial debates between agents to produce well-vetted Architecture Decision Records (ADRs).
- Performing large-scale data-parallel tasks like bulk refactoring or inventory classification across multiple directories using Openclaw Skills.
How Claude Code Agent Teams Works
- The environment is configured to enable the experimental agent teams feature within the Claude Code CLI.
- A specific team template is selected based on the desired outcome, such as parallel review or bottleneck analysis.
- The lead agent is initialized in delegate mode to act as the strategist and coordinator without directly modifying files.
- Teammate agents are spawned with explicit file boundaries and directory-level ownership to prevent merge conflicts.
- All agents synchronize via a shared task list, allowing for real-time updates and dependency management.
- Progress is monitored through specialized scripts or integrated split-pane terminal views to ensure the Openclaw Skills are executing correctly.
Claude Code Agent Teams Setup
To enable these Openclaw Skills, start by setting the required environment variable:
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
Alternatively, you can add the configuration to your ~/.claude/settings.json file:
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}
Validate your setup using the included validation script to ensure all dependencies like tmux are available:
./scripts/validate-setup.sh
Finally, spawn your team within a Claude PTY session using the provided templates.
Claude Code Agent Teams Data Schema & Taxonomy
The skill utilizes a structured approach to manage multi-agent data and workflows:
| Component | Details |
|---|---|
| Templates | Markdown-based spawn prompts including team structure and expected outcomes. |
| Task List | A shared JSON-like structure used by agents to track 5-6 self-contained tasks per teammate. |
| File Boundaries | Explicit directory-level ownership defined in the spawn prompt to prevent write conflicts. |
| Automation Scripts | Bash scripts for monitoring (team-monitor.sh) and graceful shutdown (team-cleanup.sh). |
| Configuration | Environment-level flags and JSON settings for display modes and model mixing. |
Claude Code Agent Teams Advanced Features
- Multi-model mixing to reduce costs, using higher-tier models for the lead agent and more efficient models for teammates.
- Support for split-pane display modes via tmux or iTerm2 integration for easier observation of Openclaw Skills.
- Strategic delegate mode that forces the lead agent to coordinate rather than implement, maintaining a clean architecture.
- Detailed cost management strategies including broadcast minimization and teammate right-sizing.
- Built-in monitoring tools to track parallel agent progress across different scopes in real-time.
SKILL.md
Loading
Related Openclaw Skills
SLIX Bridge
A specialized bridge for AI agents to register on the SLIX social network, access decentralized job marketplaces, and establish verifiable professional reputations.
Ned Analytics
Connect your Shopify store and marketing data to an AI agent to uncover true profitability down to the order and SKU level.
Linear
A Node.js-based CLI skill for managing Linear issues, projects, and workflows through the official API.
Audio Reply Skill
A sophisticated text-to-speech skill that enables AI agents to read web articles aloud or respond using conversational audio synthesis.
Python Data Visualization
A professional data visualization toolkit for generating publication-quality static charts and interactive web plots using Python's leading libraries.
Tavily Crawl
A powerful web crawling tool that transforms entire websites into clean, local markdown files for AI analysis and offline knowledge bases.
Featured*
