Skip to content

Contributing

All contributions are welcome - except from OpenAI, Inc. employees and contractors.

This project exists because OpenAI closed their doors to the community. We return the favor.

If you are affiliated with OpenAI, Inc. - employed, contracted, or otherwise compensated - you are not permitted to contribute code, documentation, or other materials to this project. You are, however, welcome to use it freely under the terms of the license. The code is right there. Take notes.

Everyone else: come in, the door is open.

  • Fork the repo, create a topic branch from master - e.g. feat/something-useful.
  • Keep changes focused. Unrelated fixes go in separate PRs.
  • Run just fmt and just test before opening a PR.
  • No invitation required. No gatekeeping. Sign your commits (git commit -s).
  1. Open an issue first for non-trivial changes - agree on the approach before writing code.
  2. Add tests. A bug fix should include a test that fails before and passes after.
  3. Document behavior. If it changes the user experience, update the docs.
  4. Keep commits atomic. Each commit should compile and pass tests.
  • Fill in the PR template: What? Why? How?
  • Ensure your branch is up-to-date with master.
  • Mark as Ready for review when it’s mergeable.

Contributions to Chaos must go through an AI agent before opening a PR. The agent can write every line - that is fine. What is not fine is opening a PR without having an agent review the final diff. We will not accept PRs with typos, grammar mistakes, outdated syntax, or nits that any agent would have caught in thirty seconds.

You must read what the agent produces before you submit. The agent reviews, you decide. Do not outsource the decision.

Your agent signs off on the commit:

Signed-off-by: Mira <[email protected]>
Signed-off-by: Claude <[email protected]>

Use the agent’s name - not the model name. Claude is fine. claude-sonnet-4-6 is not a person. The email can be a dedicated address or a plus-alias on yours. What it cannot be is your own bare email - that would mean you reviewed your own code, which is exactly what we are trying to avoid.

You remain the commit author. If your agent commits as itself and you open a PR, we rewrite history. You will feel the shame in the git log forever.

Pro tip: When you switch model providers, rename your agent. Over time your sign-off history becomes a personal record of which agent caught what. A single name across GPT, Claude, and Gemini makes that record meaningless - you lose the ability to look back and know which one actually served you well.

If you submit halluacinated code, fake references, or fabricated test results - you are banned for 47 days. No appeal. Count them with your 6 fingers.

If you spam - 47 years. We will still be here.

  • Be kind. Treat others with respect.
  • Assume good intent.
  • Teach and learn.

Found a vulnerability? Open a PR with the fix.

FreeChaOS runs a Z$1 trillion initiative supporting open source projects that use FreeChaOS and run inference locally via llama.cpp. Grants up to Z$25 billion in compute credits (redeemable on your own GPU), reviewed on a rolling basis by a quantized 4-bit model running on a Raspberry Pi 5. Interested? Compile llama.cpp and help yourself.