Orchestrate multiple Claude Code agents using Openclaw Skills to work in parallel on complex software projects via automated worktrees and review chains.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claude-swarm
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-swarm 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 Swarm?
Claude Swarm is a high-performance orchestration layer designed for developers who need to scale their coding output through parallel processing. By leveraging Openclaw Skills, this tool automates the entire lifecycle of multiple Claude Code agents working simultaneously across isolated git worktrees. It provides a structured framework for managing complex tasks, from initial planning and endorsement gates to automated multi-round reviews and final branch integration.
This skill transforms the Claude Code CLI into a collaborative swarm, intelligently assigning roles like architects for high-level design and builders for feature implementation. By using Openclaw Skills to manage these agents, developers can ensure that parallel work remains organized, quality-checked, and conflict-free, significantly reducing the time required for large-scale refactors or feature rollouts.
Claude Swarm Use Cases
- Executing large-scale codebase refactors across multiple modules simultaneously.
- Developing multiple independent features in parallel branches without manual context switching.
- Implementing automated review and fix cycles to maintain high code quality across a project.
- Managing complex merge operations and conflict resolutions using Opus-level intelligence.
- Coordinating a fleet of agents for batch processing of documentation or test suite generation.
How Claude Swarm Works
- Define the project roadmap by breaking work into discrete tasks within a JSON configuration file, assigning specific roles to each task.
- Pass tasks through a mandatory endorsement gate, a safety feature of Openclaw Skills that prevents unauthorized or runaway agent execution.
- Spawn agents into isolated git worktrees and dedicated tmux sessions to ensure true parallel execution without file system interference.
- Monitor progress via automated watchers that handle real-time notifications and trigger secondary review agents upon task completion.
- Integrate all validated branches into the main codebase using an automated integrator that resolves merge conflicts and runs final quality checks.
Claude Swarm Setup
- Copy the scripts/ and config/ directories from this skill to your local workspace.
- Initialize your configuration:
cp config/swarm.conf.example config/swarm.confand adjust your environment variables. - Ensure the following dependencies are installed: bash 4+, tmux, git, gh, jq, and the Claude Code CLI.
- Define your tasks in a JSON file and launch the swarm:
bash scripts/spawn-batch.sh "/path/to/project" "batch-name" "Batch Description" /tmp/tasks.json
Claude Swarm Data Schema & Taxonomy
| Component | Description | Data Structure |
|---|---|---|
| Task Manifest | Defines agent IDs, prompts, and roles | JSON Array in /tmp/tasks.json |
| Duty Table | Maps roles to specific Claude models (Opus/Sonnet) | config/duty-table.json |
| Swarm Config | Environment settings and notification hooks | config/swarm.conf |
| Agent Workspace | Isolated environment for active coding | Git Worktrees at <project>-worktrees/<task-id>/ |
| Process Tracking | Real-time monitoring of agent output | Named tmux sessions (claude-<task-id>) |
Claude Swarm Advanced Features
- Intelligent model fallback logic that switches from Opus to Sonnet or Haiku during rate limit events.
- Integrated notification system supporting Webhooks for Slack/Discord and dedicated Telegram bot alerts.
- Automated review chains that perform up to three rounds of self-correction and bug fixing before pushing code.
- Non-interactive execution mode using the
--permission-mode bypassPermissionsflag for seamless automation. - Centralized integration watcher that manages complex multi-branch merges and conflict resolution automatically.
SKILL.md
Loading
Related Openclaw Skills
Arxiv Papers Search
A specialized automation tool for searching, retrieving metadata, and downloading academic papers from the arXiv preprint server.
File Indexer
A lightning-fast filesystem catalog using SQLite to index file metadata for instant lookups by name, size, and date.
302.AI API Integration Assistant
A specialized developer tool for discovering and integrating over 1400 AI services from 302.AI directly into any codebase.
Codex Swarm
A high-performance orchestration framework for spawning parallel OpenAI Codex agents to handle complex, multi-task coding projects simultaneously.
Daily Mood
An intelligent emotional companion that delivers personalized morning and evening messages tailored to your specific mood and language preference.
Career News
A multi-source news aggregator that delivers tailored morning briefs across multiple professional fields to keep experts ahead of their industry.
Featured*
