Hi,
Based on recent discussions, it sounds like we probably need a
discussion in the framework of a GR to understand where we stand
regarding AI-assisted contributions to Debian.
So, here is an attempt at a proposal.
I think this is a very good proposal. It does not ban things on arbitrary basis. It follows community precedent on use of non-free tools in the development process. And it gives enough guidance to reduce annoying contributions.
The only thing I'd consider is establishing a process for providing even more and even more specific guidance to people using AI tools in certain Debian areas. Like, establishing a team that would provide base instruction files for AI Agents describing properties of a good AI contribution, like providing a CAUDE.md file saying things like "minimise changed code lines", "make added code readable and easy to understand", "if the package has a test suite add a test covering the added feature", "avoid adding new dependencies", "make sure that all package tests pass after changes" ...
A collection of best-practices rules (that would also be good for human contributors to know) in a machine-readable format would be a good way to improve the baseline quality of contributions.
-- Best regards,
Aigars Mahinovs