[Feature Request] Establish AI policy for this repository

[mbuechse]

OpenStack (or rather, OpenInfra) does have an AI policy.

I myself also think it's important to clearly mark which contributions have been generated by AI or assisted by AI.

I'm even inclined not to review such content, because I need to address a human being with my review, and if that human being doesn't understand the code in question, they can only relay my messages to the AI, which seems like a waste of my time. On the other hand, if the human being has used AI but does understand the code quite well, that would be a different case, but how can I reliably discern the two cases?

[fkr]

Thanks @mbuechse for raising this. I think, we should (try) to establish this for the overall SCS Community and not just the repo - however this being currently the most active repo I think, it makes sense to start here and see whether we can draft something that the overall Org would like to adopt. I've so far not really read into this topic, I'll do that over the next week.
Are there others here that already have a solid proposition?

[berendt]

We use https://osism.cloud/de/ki-manifest.

[gerbsen]

The linux kernel uses this: https://github.com/torvalds/linux/blob/master/Documentation/process/coding-assistants.rst Matplotlib has now established this: https://matplotlib.org/devdocs/devel/contribute.html#restrictions-on-generative-ai-usage

These sentences look copy/pastable:

In particular, it is also strictly forbidden to post AI generated content to issues or PRs via automated tooling such as bots or agents. We may ban such users and/or report them to GitHub.

curl is also very outspoken on this topic: https://curl.se/dev/contribute.html#on-ai-use-in-curl

So what do we want? Here are some things I think are useful:

• explainability from the contributor
• fitting license to redistribute contribution
• nothing auto-generated // no agentic contributions
• human in the loop (how to prove?)
• documentation of what tools were used

[mbuechse]

Also of relevance (particularly the "Why not LLMs?" section): https://codeberg.org/small-hack/open-slopware

[jklare]

Let's try to keep this lean for now. What the Linux kernel does seems very reasonable to me.
Given that the number of people actively working on the SCS project is limited, we should actively welcome every new contributor and not exclude any, just because they are using new technologies. Having said that, we should still make sure that:

• all contributions are following legal requirements
• contributors take full responsibility for the code they contribute (no matter if they used paper and pencil, an IDE with autocomplete, Claude Code or any mixture of those to create it)
• the tools used are mentioned, if they generated a significant amount of the code

Apart from those minimal requirements, I think we should wait for "the Problem" to actually appear in our repository, before we solve it.