I want to put a question to this community, based on something that happened to me this week.

I built a small stock/ETF quote app (FastAPI + Vue, single Docker image), MIT-licensed, public source, tested and running in production on my own Unraid server. I wrote a CA template for it following the selfhosters

templating guide and submitted it to their community template repo - for free, because I thought others might find it useful.

The PR was closed within one minute of the first comment, unreviewed, as "pure AI slop": https://github.com/selfhosters/unRAID-CA-templates/pull/684

Why? Because my PR description said it was drafted with Claude, and my commits carry Co-Authored-By: Claude trailers. Nobody looked at the source, the template, or the app. The second maintainer confirmed it: the trailers themselves were the reason - "just a trust thing."

Here's my problem with that, and I say this as someone who has written software for over 30 years, with a GitHub account dating back to 2009: those trailers are disclosure. I design, review, and test everything I ship.

If I had quietly stripped the attribution, the PR would have passed. A policy that rejects disclosed AI assistance while waving through hidden AI assistance doesn't protect anyone from bad software — it just teaches

developers to stop being transparent. That's the opposite of trust.

I'm not arguing anyone must accept AI-assisted contributions. Review them. Test them. Reject them on merit. That's what maintaining means. What I'm questioning is rejection by label, unseen — and I hope that as more of

the software this community runs on is built with AI in the loop (it already is, whether disclosed or not), we find a better answer than this.

I've written about the episode in more detail on Reddit: https://www.reddit.com/r/unRAID/comments/1uxy2d7/rejected_from_a_ca_template_repo_because_my/

The app itself will come to CA through my own template repo via the official submission portal, so this isn't a request for help — it's a question about what kind of contribution culture we want. I'm genuinely

interested in how other developers and users here see it.

Anyone who says "pure AI slop" isn't going to take the time to actually look at something since their response is on par with 5 year old being told no. There's no actual evaluation being done, just a gut response of irrational hatred.

30 minutes ago, whipdancer said:

Anyone who says "pure AI slop" isn't going to take the time to actually look at something since their response is on par with 5 year old being told no. There's no actual evaluation being done, just a gut response of irrational hatred.

Im actually 7.

The selfhosters repo is a private entity unaffiliated with unraid. If you don’t like their rules make your own repo ?

Yes — and as I wrote in the post, I'm already on that route. But step back and look at what that means: even more effort to bring one small free app to Unraid.

The hurdles should be as low as possible, not higher every year.

And in my opinion, this is the real issue: the classic maintainer system is outdated. Relying on a single volunteer's gut feeling, energy and mood to decide what the community gets is a bottleneck — the maintainer in question even admitted on the PR that it was about effort and lost interest, not quality. These days the heavy lifting should be automated: a service that checks defined criteria — does it build, is the source public and licensed, does the image pass a security scan, is the template valid — and gives a rating.

Humans then only look at the edge cases instead of playing god over every submission. The official CA submission portal

already works exactly in that direction (automated validation and scanning). That's the model — not a private repo where disclosed AI assistance gets you closed unseen.

7 minutes ago, ReginaldBull said:

Relying on a single volunteer's gut feeling, energy and mood to decide what the community gets is a bottleneck — the maintainer in question even admitted on the PR that it was about effort and lost interest, not quality.

That's why selfhsoters is winding down.

Why did you even try to submit to a third party repo? Make your own and take responsibility for your app instead of trying to conveniently make it someone else's problem.

3 hours ago, ReginaldBull said:

Here's my problem with that, and I say this as someone who has written software for over 30 years, with a GitHub account dating back to 2009: those trailers are disclosure. I design, review, and test everything I ship.

"Conventional" development provides some implicit trust, if you don't understand what you're doing at least some you're not getting your app working in the first place.

With AI-backed development that's all out the window, anyone can get something that "works" but without the author being necessarily capable of reviewing the output, and that has led to countless scandals about serious issues/vulnerabilities in new apps. So with AI-assisted development trust goes out the window too regardless of how much experience you have since your experience probably doesn't encompass everything a full stack app involves these days. You might have 30 years experience in backend dev, but the frontend the AI generated for you can have massive issues you wouldn't recognise even if you "review" it because that's not your field and you essentially have 0 years of experience on that particular subject, and they'll go through.

Also don't use AI to respond...

Edited July 16Jul 16 by Kilrah

So with AI-assisted development trust goes out the window too regardless

Time is over for conventional programming. I agree if you argue 30 years means nothing these days. But these days we have tools like Fable and these tools are really capable of doing their job quite well. Better than 90% of conventional programmers. You just have to push them in the right direction. However, in my case, they blocked me because they read the word "Claude" - that’s what I'm complaining about.

> but without the author being necessarily capable of reviewing the output

I'm capable of reviewing the code, but honestly, I don't do it anymore — those days are over too.

I've defined a set of skills that lay out how I want the code written, plus a bunch of hooks that run security checks and so on.

I develop with Claude, and Codex reviews the code. From time to time, especially before a release, I ask Fable to run a security scan, check the tests, and so forth.

That's it. That's how development works today, and how it'll work for the next few years.

3 hours ago, ReginaldBull said:

I'm capable of reviewing the code, but honestly, I don't do it anymore — those days are over too.

Herein lies the problem, the community has spoketh, they want humans, who are capable of reading/writing code, doing due diligence.

You aren't seeing the forest for the trees.

Nobody is saying you cant use claude to assist in development. But when you wont respect people's time, they wont respect you.

Edited July 16Jul 16 by MowMdown

I don't like "AI assisted development" too, even it is hip these days.

AI may help you to find certain functions or structures to use, but I never would let it write full code for me.

It would take longer to check it than to write it myself.

And I have seen a lot of "AI-generated-mistakes" already.

So, if the community rules are "no AI", don't complain, just obey.

(And yes, "30 years" is nothing, you youngster 🤣 I did not count really, but I guess, I am over 50yrs now already... or maybe a little less...)

On 7/16/2026 at 8:23 PM, MowMdown said:

But when you wont respect people's time, they wont respect you.

You say I didn't respect people's time???? It's the opposite - they didn't respect what I had to offer. AI-supported or not - makes no difference.

I was my time, my app, I offered it to the community, and the repo maintainer canceled my PR only because he read "Claude" in the description.

This is what I call disrespectful. The repo-maintainer acts God-like and that’s not OK!

You claim the initial reviewer did not look at the repo, because the comment, and action happened in the same minute. You do realize there is a window of more than an hour from when you opened the PR to the action was taken, right. That's a lot of time to review. I have never commented that I start reviewing anything, the review and action just arrives.

In a issue reply you asked for a policy, which was present, it doesn't explicitly call out the PR content, but the nature of the policy clearly states that there is some requirements for AI, which did not make it into the PR at all (did your agent miss it?). I find the lack of acknowledgement for the present policy disrespectful, it probably also added to the decision that was made.

We may act god-like, it's our repo, it's not the single way in to CA, It's not the decider for apps to be added to CA. It is not affiliated with Limetech nor the maintainer of the Community Applications plugin. This style of CA repo was (and maybe even still is) unprecedented, as there was no established repo that accepted "outsiders" (as in non-established persons in the ecosystem) to add applications, so additional screening needs to take place, as the blast-radius for getting the repo removed/banned affects quite a few different style of apps.

On 7/16/2026 at 2:26 PM, Kilrah said:

Also don't use AI to respond...

A significant portion of this forum's members aren't native English speakers.

Using AI to structure and clarify their thoughts before posting isn't gaming anything -- it's a communication tool that lets people with deep technical knowledge participate more effectively than their English fluency alone would allow.

The question isn't whether AI was involved.

It's whether the content is accurate, relevant and useful.

Those are separable questions regardless of how the text was assembled.

This response was formatted with AI assistance.

7 hours ago, Lolight said:

A significant portion of this forum's members aren't native English speakers.

Using AI to structure and clarify their thoughts before posting isn't gaming anything -- it's a communication tool that lets people with deep technical knowledge participate more effectively than their English fluency alone would allow.

The question isn't whether AI was involved.

It's whether the content is accurate, relevant and useful.

Those are separable questions regardless of how the text was assembled.

This response was formatted with AI assistance.

There is a difference between using AI to act as a translator and using AI to generate an answer to a question. If the person proposing the question wanted an answer from a clanker, they would have just asked the clanker.

Translation tools have existed long before AI and are perfectly capable converting text between languages. AI isnt needed for this.

If you want to bounce ideas off of an LLM, you should still format said thoughts into your own words and feelings and not just lazily post the LLM generated text as a formal response to a question. Lazy and wastes everyone's time.

This response was NOT formatted with AI assistance nor was AI used in the process.

17 hours ago, MowMdown said:

There is a difference between using AI to act as a translator and using AI to generate an answer to a question.

The translator distinction is fair as far as it goes.

But AI assistance for non-native speakers isn't just translation -- it's helping someone express nuanced technical thinking in natural English rather than stilted literal output.

Those are meaningfully different tools producing meaningfully different results.

On laziness -- the relevant measure is whether the content is accurate, relevant and useful.

How it was assembled is a process question not a quality question.

Accurate and helpful content isn't lazy regardless of what tools assisted its production.

The disclosure at the end of my post was transparency, not confession.

That's exactly what ReginaldBull was arguing for in this thread -- and what got him turned away from the template repo in the first place.

59 minutes ago, Lolight said:

How it was assembled is a process question not a quality question.

In theory yes but the process is known to commonly lead to low quality, and people to often not take the time to verify and fix that.

If someone had to write something themselves they will have had no choice than to put in the effort to write it. When they use AI "just post the output, it saves me time and effort" becomes possible and tempting, and experience shows people often do that. You might not but the fact that people do means that when AI use is obvious there will always be an amount of suspicion and lower trust.

5 hours ago, Lolight said:

The translator distinction is fair as far as it goes.

But AI assistance for non-native speakers isn't just translation -- it's helping someone express nuanced technical thinking in natural English rather than stilted literal output.

Those are meaningfully different tools producing meaningfully different results.

On laziness -- the relevant measure is whether the content is accurate, relevant and useful.

How it was assembled is a process question not a quality question.

Accurate and helpful content isn't lazy regardless of what tools assisted its production.

The disclosure at the end of my post was transparency, not confession.

That's exactly what ReginaldBull was arguing for in this thread -- and what got him turned away from the template repo in the first place.

It doesn't matter what the reasoning is, posting the prompted result without any self-formatting and revision just tells me you took no effort to formulate a response.

Using LLM for assistance is fine, just dont post the raw result because if that's the answer I wanted, I would have just done that myself.

LLMs are not accurate and their output needs refinement and correcting multiple times.

That's a more precise framing from both of you and worth acknowledging.

The probabilistic argument is fair -- unverified AI output is a real problem and the temptation to skip verification exists.

That's a legitimate quality concern.

But there's a difference between "AI output often has quality problems therefore disclose and verify" and "AI formatting is obvious therefore suspect."

The first is a reasonable quality standard.

The second is what's being applied here -- style as a proxy for quality rather than actual evaluation of the content.

The disclosure at the end of my post was exactly what MowMdown is now describing as acceptable -- AI used for assistance, result revised and verified, transparently disclosed.

If that's the standard then ReginaldBull should never have been turned away from the repo.

11 minutes ago, Lolight said:

If that's the standard then ReginaldBull should never have been turned away from the repo.

The repo has a rule about it, you may or may not agree with it but it's there and submissions need to follow it. If you don't agree then you don't submit. If you do anyway don't be surprised if you get refused.

Even more so when OP demonstrated his lack of care/concern by asking where the rules were and being pointed to them, showing he didn't even bother with the basics.

Edited July 22Jul 22 by Kilrah