Add an AI Policy

[FBruzzesi]

Opening as an issue to track progress, following the Discord thread discussion.

---

Important

The goal is to set sensible guardrails that protect maintainer time without categorically rejecting AI-assisted work, while keeping human contributors and human review as the priority.

---

Proposed measures

• Mandatory AI disclosure on PRs. Requires adding a PR-template field asking whether any AI tool or model was used, and which one. The point is transparency.
• Responsibility clause in CONTRIBUTING.md. Add wording along the lines of:

  You are responsible for every line you submit, regardless of how it was produced. Disclosure does not transfer responsibility to the tool.

• Add an AGENTS.md (under discussion)
• Ship a user-facing SKILL.md files. Examples from FastAPI: library-skills.io and FastAPI skills
• AI-assisted PR review as a supplement to, and never a replacement for, human review. Open questions are:
  • Cost: Do any vendors offer a free allowance for open source projects worth investigating?
    • coderabbit Pro plan is free for Open Source projects
    • qodo also free for OSS
  • If individual maintainers use AI personally for review, we could add a review.md describing what to look for and how to structure the review.
•  Author responsiveness as the bar: PRs where the author does not engage with review feedback are closed without extended discussion. This is reactive, not preventive: we cannot screen for it upfront, but we can be explicit that PRs where the author does not respond to review feedback will be closed without extended discussion, whether or not AI was used.

---

Additional references:

• Thread originating PR: feat: add Selector.__xor__ and Expr.__xor__ for symmetric difference #3621
• Related human-contributor improvements: chore: Adopt uv as package manager #3608, feat: Introduce testing.constructors module and a pytest plugin #3552
• A community-maintained Survey of OSS AI policies: melissawm/open-source-ai-contribution-policies
• Polars contributing guide: https://docs.pola.rs/development/contributing/
• AGENTS.md standard: https://agents.md/
• Library skills pattern: https://library-skills.io/

[dangotbanned]

• Polars contributing guide: docs.pola.rs/development/contributing

I think we should steal this part from First-time contributions

You must post a screenshot of you successfully running the test suite (...), locally on your machine (not the CI).

Not impossible for a bot to do - but it is definitely an easy temperature check for us to see if that PR has a screenshot 🙂.

The polars ci adds a first-contribution tag, so that's another thing to make it easy for us to see when we might need a closer look.
If you follow that link, there are PRs that still ignore this request that look like AI to me

[dangotbanned]

@MarcoGorelli how would feel about adding a new label like suggest close, so we can avoid any close/reopen whiplash moving forward 🫣?

I'm happy to wait for a second opinion before making a call

[MarcoGorelli]

yup sure