How to deal with fully LLM-generated merge requests

[Daniel Ramos]

Hola Uri,

Thank you for answering with your name and for reading it all.

Linus said it this month, when the kernel dropped its mandatory AI tag: "We're not going to make stupid rules that say 'you can't use tools for X'", because understanding "has very little to do with what tools they use".
https://www.theregister.com/a/5303140
By the rule of "you never wrote it, a machine generated it", every commit would have to be written in binary, since nobody reads what the compiler generated either.

On the style: formal, structured text is how most of the academic and engineering world writes.
I studied that way and finished my electrical engineering degree with a 3.5 GPA; WES evaluated it as a full Canadian engineering degree.
I am also autistic, and plain, explicit structure is how I communicate best, with or without any tool.
If a text is harder to read, the same tools can make it easier: a translator, a summary, an LLM, or simply asking.
Tutorials can be informal, and I learned a lot from them in the 90s and still do when I need a new skill quickly.
Scientific and professional work cannot be made less tiring by making it less exact.
It is better to use a translation tool, LLM or not, than to pretend to know it all, and better to ask the community than to set aside what someone wrote because a tool helped them write it.

On AGENTS.md, you are right that it helps people who use agents honestly, and I follow it wherever a project has one.
It does nothing for those who remove it or never disclose, and those are the people a tool rule is written for.

On respecting policies: I read SlimeVR's "Use of AI" section, quoted it back in the one issue I opened there (SlimeVR-Server #1946, 11 October, still unanswered), and sent no code, because your project asks for that.
Since my long message was hard to read, here are your own project's numbers in two lines, from GitHub's search, bots left out.
Of the 388 pull requests opened in SlimeVR in 2026, 7 disclose AI help, under 2%; 2 were merged after a normal review, 4 were closed and 1 is open, and 1 of the 105 pull requests open today discloses it.
That is the same picture as KDE's 1.4%: no flood, and where the code was read, it was merged.
I ask for the same respect here: judge the work, the tests and the numbers.
No one in this thread has disputed any of them yet.

Gracias,
Daniel Campos Ramos
Sparky Stereo OS