Three simple characters that tell a reader who maintains a project, how much of its code is written by AI, and how closely a human checks the AI's work.
A project publishes its rating in an ACR.md file at the root of its repository. ACR A2b, for example, means:
| Position | Measures | A2b |
|---|---|---|
| 1 | Maintainer Expertise: how experienced the person responsible for the code is (in a solo project, you; on a team, whoever approves changes), A (expert) to E (non-programmer) |
A Expert |
| 2 | AI Share: how much of the code AI wrote, 0 (none) to 4 (76–100%) |
2 26–50% |
| 3 | Oversight: how closely a person checked the AI's work, a (verified) to e (unchecked) |
b Reviewed |
The rating doesn't judge whether using AI is good or bad. It tells readers how the code was made, so they can decide what it means for them.
- Get your rating: answer three questions and copy the
ACR.mdand README badge. - Add
ACR.mdto the root of your repository. - Check the file with the validator.
- Add the GitHub Action to check it on every push and pull request.
- Ask to be listed in Rated Projects on the site.
Not sure which levels fit? The rating examples cover common setups, and the FAQ covers the edge cases.
- aicoderating.com: the website and the spec.
- action: a GitHub Action that checks
ACR.mdagainst the spec.
The spec is a draft, and feedback now shapes version 1.0. Open an issue to suggest a change or ask a question, or read the contributing guide. Everyone taking part is expected to follow the code of conduct.
The spec is licensed under CC BY 4.0, and the code for the site and the Action under MIT.