Some questions about LLMs

1

So I recently tried to submit an APP to Flathub, and the PR is closed as “ai-slop”.

Well I don’t have anything to complain about, rules are rules, and I missed the line that says “Applications containing AI-generated or AI-assisted code, documentation, or any other content are not allowed” Here

For me, the app is not heavily relied on AI code, it just use AI to review the codes and find mistake/improvements, and I consider that as the learning progress when writing code.

The project first started off as a fork of Motrix, the well-known internel download manager, which is left non-maintained for 3yrs, what I hope, is that to give it a second life, so I rewrote it with new version of framework, and using rust to re-write the download engine (aria2).

It really fells disappointed that people simply says it is AI “slop” when they see commitment record saying “Accept suggestions from AI”. While I DO understand why ppl might just feel unsafe, or goofy about that, but AI code review are no longer something that unreliable. I personally also don’t just accept ANY suggestions from AI, and the ratio of AI code in this project is just 100% small…

In this thread I see ppl in the community talking about this…

If you are proud of what your development has resulted in, you can be proud of the process that lead to it as well.

And I am absolutely proud of what my development has resulted in.

As you can see, I am not trying to hide that I am using LLM to assist coding, it is a progress of learning. What I am trying to say, is that LLM assisted codes are not necessarily “slop”, for my project, it is not a “work for me” project, it is constantly maintained (issues usually resolve in a few days), and it is well structured. I would say it is a second life to a discontinued project, and I’m trying to reach to more users.

So does it really make sense that Flathub simply rejects all codes related to AI, even it is well maintained? I really doubt that…

P.S. Sorry if my English is bad, its not my native language

2

P.S. If you wonder why there are so many commits with title “Accept suggestions from AI”, it is because the AI reviewer I use keep giving me new suggestions, e.g. feat(ui): Flyout tray panel by YueMiyuki · Pull Request #103 · YueMiyuki/Risuko · GitHub

3

Some obvious giveaways of LLM usage outside using review tools:

  • verbose commit messages like feat(ui): refine flyout UI - consistent border-radius, no menu bar gap, fix macOS activation on hide or expand task file cleanup for generated sidecars and update task name/path normalization mixed with simple “Bump version”
  • the fact that since April your repository has gained ~80000 lines of Rust, including from-scratch implementation of BitTorrent stack, or both Gnutella and ADC protocol having been merged on the single day
  • em-dashes in code comments; comments on internal functions no one would bother documenting
  • zero TODO or FIXME comments whatsoever

Your mention of the thread asking for clarification what is slop is at very best misleading, because over there, it’s explained that there’s no reason to hide commit history full of mistakes and fixes, and your app is not that, even if you are proud about it.

Strict anti-LLM wording is there exactly to prevent this sort of lawyering you are attempting here.

4

For the first point, I used AI to help fix a git command mistake, and that is why it is the only unverified commit , and about the ~80000 lines of Rust, it is the initial commit when I started off the project, merging in one day does not mean it is written in one day.

Comments are written by SWE (Free agent on Devin, which is the only model I use to automate docs) so it pass coderabbit ai docs coverage check, e.g:

CleanShot 2026-06-11 at 00.05.02@2x
CleanShot 2026-06-11 at 00.05.02@2x1804×828 80.6 KB

And finally for your information, I do have TODO in the project, they are gone now just because features are getting stale over time. The last TODO is TOFU for ssh client, in this commit

5

AND in early versions, e.g. here, before rebranding to “Risuko”, I also patch issues from original repo, with todo list like this locally (see the PR comment)

6

Single mega commits like your “initial” one from April aren’t inspiring confidence it’s not vibe-coded, and the fact various LLMs tools alter your commit messages directly fall into “we don’t allow LLM” rule. Right now, I don’t think anyone on the team would be comfortable publishing this.

7

Alright, thanks for answering.

Won’t argue anything, you guys are just getting your job done.

Maybe one day I can come back with more users and the team can be more comfortable publishing the app

Thanks :slight_smile: