Docs (Proposal) : Add an AI contribution policy

[Jem256]

Background

Many open source projects have recently adopted policies for AI-assisted contributions in response to a growing volume of AI-generated PRs often targeting repos labelled beginner-friendly, which describes Polar. Since review bandwidth is a scarce resource, I think it's worth agreeing on a policy before this becomes a problem rather than after.

Proposal

Adopt a policy in the Bitcoin Core / ripgrep / uv lineage, AI tools are welcome, but with a human fully in the loop:

• AI-assisted coding is allowed but contributors remain solely responsible for what they submit.
• You should only open a PR if you know the language, could have written the code yourself, reviewed the code yourself, understand the surrounding code, and have actually run Polar with your changes.
• You must be able to explain your changes in your own words, no copying AI responses when replying to reviewers.
• No AI-generated issues, comments, or review replies.
• No autonomous-agent-driven PRs; violations may be closed without notice.
• Using AI to polish or translate your own writing as a non-native English speaker is fine, as long as it reflects your own voice.

Suggested implementation

1. Add AI_POLICY.md to docs (I can draft the doc upon alignment).
2. Add one line to CONTRIBUTING.md: "If you use AI tools while contributing, please read and follow the AI policy"

[Abdulkbk]

I agree. We should have an AI policy, and adopting an existing policy from the projects you listed and fine-tuning it to this project will absolutely make sense.

[sjh9714]

This is a good move, and the Bitcoin Core / ripgrep / uv lineage is the right one to follow. One data point that supports doing this before it's a problem (your words): I recently scanned 2,204 merged agent PRs across public repos, and the boundary-crossing rate on the long tail of small/beginner-friendly repos was about 8.6%, roughly double the 4.3% on 10k+ star projects — which lines up with your observation that these PRs disproportionately target beginner-friendly repos like Polar. (Methodology, if useful: https://github.com/sjh9714/mergewarden/blob/main/docs/study/what-2204-agent-prs-showed.md)

One practical note on enforceability: most clauses in this lineage ("you could have written it yourself", "explain in your own words") are honor-system, which is fine. But the "no autonomous-agent-driven PRs" clause is the one you can actually check cheaply — those PRs tend to arrive on cursor/**, codex/**, copilot/** branches or from *-swe-agent[bot] authors, so a lightweight branch/author signal can flag them for the extra scrutiny you're rationing review bandwidth for. Happy to share the exact patterns if helpful.