AutoSwarm skill
A skill that splits a task across twelve agents such as Codex, Claude Code and Cursor, sharing context and merging results.
- 2,460 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill launches other agent CLIs with their own rights and passes them project context.
Install
- Works in
Manual install
git clone https://github.com/HKUDS/AutoSwarm.git && cd AutoSwarm && uv tool install --force . && autoswarm setup --host codexThe Codex example from the Install section of the README. Replace the `--host` value for another host.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill launches other agent CLIs with their own rights and passes them project context.
This is third-party code. Review the repository files before installing.
What it does
AutoSwarm installs into the agent you already use and turns it into a coordinator. You hand over a task once, and the skill picks suitable workers among installed CLIs, hands out subtasks in parallel and tracks questions and approval requests. Context and memory are shared between agents, and results are verified and merged into one answer. Supported hosts include `claude-code`, `codex`, `opencode`, `cursor`, `copilot`, `hermes` and `nanobot`. A separate Windows app, Desk, provides an approvals window.
Who it is for. People who keep several AI agents on subscriptions and move tasks and context between them by hand.
Good fit when
- Research and code edits run in different agents and the result is assembled by hand.
- An independent second agent should verify the first one's output.
- A task is split into parallel parts between Codex and Claude Code.
- Approvals from several agents should appear in one window.
Not a fit when
- The task is small and one agent finishes faster.
- No agent CLIs are installed and you do not want to add them.
- Agents must not run local commands unattended.
Example request
Give AutoSwarm this task: Codex writes the migration, Claude Code reviews it, then merge a single summary with the remarks.Limitations
The main agent needs skill support and local command execution, and workers need their own CLIs. Installation requires Python 3.10 or newer and a `git clone`, and there is no auto-update: rerun `autoswarm setup` after upgrading the package. Provider verification spends model calls. Workers act with their own rights and subscriptions, with permissions and approvals passing through the skill. Several agents working together multiply token and subscription use, so small tasks gain nothing. Details on profiles, permissions and connection recovery are in `src/autoswarm/assets/references/setup.md`.
How to disable. Remove the installed skill from the host's skills directory, run `uv tool uninstall` for the AutoSwarm package and delete its home directory with `config.json` and records.
FAQ
Which agents can act as the host?
`claude-code`, `codex`, `opencode`, `cursor`, `copilot`, `muse`, `codebuddy`, `hermes`, `nanobot`, `dsh`, `grok` and `kimi`. Pass their IDs to `autoswarm setup --host`.
How do I update an installation?
Update the package, for example `uv tool install --reinstall .`, then run `autoswarm setup` for each host in use. The installed skill carries its own runtime copy.
Is there a graphical interface?
Windows has the AutoSwarm Desk app with an installer, bundled Python and auto-update. It shows agent work and approval requests.
Where are the constraints described?
In `src/autoswarm/assets/references/setup.md`: profiles, permissions and connection diagnostics.
Is it paid?
No, the repository is MIT-licensed. Subscriptions and model calls of the agents connected as workers remain paid.
Discussion
No questions or reviews yet.
Related
SkillsA skill that turns several Claude Code, Codex and Cursor windows into a team with a coordinator, shared registry and cross-review
A desktop app where Claude Code, Codex, Gemini and other agents work as a team of specialists in a shared project
One agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
Agent Swarm operator skill
Agent Swarm
Agent Swarm is an operator skill that explains retaining workers’ personal volumes so their identities survive a restart.
One local memory for Codex, Claude Code, Cursor, Hermes Agent and OpenClaw: agents hand off tasks, review each other and refine their skills overnight
A skill with 25 commands to audit and improve the prompts, tools and agent designs of a project in Cursor, Claude Code and Codex
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| AutoSwarm (this card) | no | 2,460 | |
| Agent Swarm operator skill | no | 1,048 | agent-swarm builds a swarm inside one environment; AutoSwarm connects installed CLIs from different vendors |
| OpenSwarm | no | 871 | openswarm relies on its own runtime; AutoSwarm installs as a skill into the agent already in use |
README in short
The README explains the one-team-of-agents idea, lists twelve supported agents and gives install and update instructions written for the agent itself. The Windows Desk app is described separately.