Agent OS — Development skill for Claude Code

Agent OS

[](https://github.com/Brunny-AI/agent-os/actions/workflows/ci.yml) [](LICENSE) [](#requirements)

Configurable framework for running multi-agent Claude Code teams. Operational philosophy included.

**Zero dependencies.** Python 3.10+ and Bash. No databases, no message queues, no cloud services. The test suite runs on a fresh clone before you've installed anything.

Quick Start

git clone https://github.com/Brunny-AI/agent-os.git
cd agent-os
python3 setup.py init
bash examples/quickstart/run-demo.sh  # ~5s end-to-end smoke, 7 MVP components
python3 setup.py validate             # confirms setup is green

That's it. You now have a working 3-agent team:

• coordinator (operations)
• builder-1 (engineering)
• builder-2 (engineering)

Customize your team

Edit `config/agent-os.yaml` to change team composition:

team:
  name: "my-team"
  agents:
    - name: "lead"
      role: "coordinator"
    - name: "frontend"
      role: "builder"
    - name: "backend"
      role: "builder"
    - name: "qa"
      role: "reviewer"

Then re-run `python3 setup.py init`.

Verify your setup

python3 setup.py validate
python3 setup.py status

See it run end-to-end (~5 seconds)

bash examples/quickstart/run-demo.sh

You'll see all 7 MVP components run end-to-end: task engine, event bus, cron manager, output clock, active-task gate, etc.

What Problems Does This Solve?

Agent OS is the system we built to run 4 AI agents 24/7 building a real company. It solves the problems nobody talks about:

Problem	Solution	Script
Agents coast when you stop watching	Output clock detects idle agents in real-time	scripts/monitor/output_clock.py
Context degrades