ClawTeam-OpenClaw CLI
A CLI agent swarm: a leader spawns workers in separate git worktrees, and they share tasks and exchange messages via files.
- 1,475 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The leader launches worker agents on its own, and each runs commands in its worktree with your user permissions.
- For OpenClaw you must allow `clawteam` in the approvals list, so grant it by absolute path and do not switch the mode to `full`.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/clawteam-openclawDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The leader launches worker agents on its own, and each runs commands in its worktree with your user permissions.
- For OpenClaw you must allow `clawteam` in the approvals list, so grant it by absolute path and do not switch the mode to `full`.
This is third-party code. Review the repository files before installing.
What it does
ClawTeam-OpenClaw is a fork of HKUDS/ClawTeam tuned for the OpenClaw agent. The idea is that the agent assembles the team, not a human writing orchestration code. The leader calls `clawteam spawn`, and each worker gets its own git worktree, a tmux session or subprocess and an identity. Workers pick up tasks with `clawteam task list`, mark progress and message each other with `clawteam inbox send`. Task dependencies are set with `--blocked-by`, and a dependent task unblocks itself. You follow progress on the kanban board `clawteam board serve` or a tiled tmux view. State lives in the `~/.clawteam/` directory, the transport is file-based with optional ZeroMQ for cross-machine exchange. `clawteam launch <template>` brings up a ready team from a TOML template, and you can write your own. The fork's differences from upstream, per its author: `openclaw` as the default agent, per-agent session isolation, automatic setup of command execution approvals and hardened spawn backends. Version 0.3.0 added a cost panel by agent and model, retries with backoff, a circuit breaker and Hermes Agent support. It also works with Claude Code, Codex, Cursor and any CLI agent.
Who it is for. Developers who run OpenClaw or another CLI agent and want it to split a large task across parallel workers on its own.
Good fit when
- A large feature splits into backend, frontend and tests, and each part should run on its own branch without clashes in the working directory.
- You need a kanban board showing which tasks are taken, blocked and done.
- A research run launches several agents with different hypotheses and the leader gathers the results.
- You want to mix models in one team: the leader on one, workers on others.
- You need a view of token spend by agent and task.
Not a fit when
- The task is small and one agent solves it in one session, so a swarm adds only overhead.
- You need a proven production orchestrator with queues and support. The project is young, at version 0.3.0.
- You want the official version without OpenClaw changes: take upstream HKUDS/ClawTeam.
Example request
Build a web app. Use clawteam to split the work across several agents: backend, frontend and tests, then merge the results into main.Limitations
Install from a clone with `pip install -e .`: plain `pip install clawteam` installs upstream with `claude` as the default agent, and `npm install -g clawteam` installs an unrelated package of the same name. Python 3.10 or newer is required, tmux is needed for the tiled view and optional on Windows, where the subprocess backend works. Each worker runs a full agent with access to its worktree, so token spend grows with the number of agents. For OpenClaw the README asks you to set an allowlist of commands in execution approvals.
How to disable. Remove the skill from your agent's directory and run `pip uninstall clawteam`. Delete the `~/.clawteam/` directory and the created worktrees by hand.
FAQ
How does the fork differ from HKUDS/ClawTeam?
Per its author, OpenClaw is the default agent, with session isolation, automatic approvals setup and hardened spawn backends added. Upstream fixes are synced.
How do workers talk to each other?
Through inboxes in `~/.clawteam/`: `clawteam inbox send` and broadcast. The transport is file-based, and ZeroMQ is available across machines.
Does it work on Windows?
Yes, on Windows 10 and 11 the subprocess backend turns on automatically. For the tiled tmux view run it in WSL.
Which agents does it work with?
OpenClaw by default, plus Claude Code, Codex, Hermes Agent, nanobot, Cursor and any CLI agent.
Can I watch spend?
Yes, `clawteam board cost` shows tokens and cost by agent, model and task.
Discussion
No questions or reviews yet.
Related
CLI toolsOne runtime for Claude Code, Codex, OpenCode and more: a run counts as done only when checks the agent did not write pass
AgentOS runs a team of OpenClaw agents: workspaces, scheduled tasks, activity reports and human approvals
The clawctl CLI manages a fleet of AI agents on local machines over SSH and Ansible: install, start, upgrade and token tracking.
OpenClacky is an open-source AI agent in Ruby with 16 tools, any OpenAI-compatible key and extension through skills.
A compact OpenClaw alternative: a personal assistant on Claude Agent SDK with agents in isolated containers and messengers added via skills
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
README in short
The repository is presented as a fork of HKUDS/ClawTeam with deep OpenClaw integration. The README covers the swarm, team templates, a cost panel and install from a clone. MIT license.