Agent Team Orchestrator

Agent Team Orchestrator

[](https://github.com/spacesky-cell/agent-team-orchestrator/actions/workflows/ci.yml) [](https://www.npmjs.com/package/@spacesky-cell/agent-team-orchestrator) [](LICENSE)

ATO is a local multi-agent task runner built on LangGraph. It decomposes a task into role-based work, runs ready branches in parallel, checkpoints progress, and puts mutating tools behind durable approvals. The same Python runtime is exposed through a CLI and an MCP stdio server.

[中文说明](README_CN.md) | [Quickstart](docs/QUICKSTART.md) | [MCP guide](docs/MCP_GUIDE.md) | [Architecture](docs/architecture.md)

Install

Requirements: Python 3.10+, Node.js 18+, and an LLM provider. The default provider reuses an authenticated [Claude Code CLI](https://docs.anthropic.com/en/docs/claude-code) installation.

npm install --global @spacesky-cell/agent-team-orchestrator
ato doctor

The npm package includes the ATO Python wheel. The first command that needs the core creates an isolated, versioned runtime and may download Python dependencies; npm installation itself runs no postinstall script and does not modify global Python. `ato --version` and help remain immediate.

`ato doctor` verifies the managed Python executable, `ato_core` version, packaged roles, project root, and Claude CLI availability. Advanced users can set `ATO_PYTHON` to a compatible Python that already contains `ato_core`.

Run A Task

ato roles
ato run "Review this repository and implement the highest-impact reliability fix" --detach
ato status 
ato audit 

Omit `--detach` to keep the CLI attached. When a mutating tool needs approval, the CLI shows the durable request ID and resumes the same LangGraph checkpoint after your decision.