Hacker News Reader

Comment by lionkor | Hacker News Reader

Lol the AGENTS.md:

> NO Co-Authored-By: Claude trailer. > NO " Generated with Claude Code" footer in PR bodies, commit messages, issue comments, or anywhere else.

I actually have a much better idea how to use AI responsibly: Simply don't use Claude to completely vibe code AND COMMIT your code.

There is no doubt in my mind that this is vibe-coded, in the bad sense, because of this line. There is no workflow that lets Claude write code AND commit, without PRs, that includes review by a human.

This is not the HN I signed up for. I don't mind AI use, responsible AI use, but zero oversight has not worked and will not work.

View on HN

Gosh, It's so hard to be excited about any software project nowadays, because I cannot know how much effort was actually put into it. For all I know the author merely wrote a few prompts and paid Anthropic for the tokens. Good job I guess?

I think simply stating the extent of AI use openly is a good start. For example, this is what I've done on my hobby project[0]. I feel like an AI use disclosure page or document helps, especially when you didn't vibe-code but still used AI in a somewhat responsible way (for review, for writing tests, writing CI pipeline stuff, etc.).

[0]: https://libls.org/ai-use

Having a cool idea, executing it and showcasing it to the world is not good anymore?

For serendipity still matters.

"I actually have a much better idea how to use AI responsibly: Simply don't use Claude to completely vibe code AND COMMIT your code."

I usually open a new chat in the ide and ask it to "generate a commit message", then i copy paste it.

This means that if my code was written 20% by me and 80% by Google Antigravity, and Claude then adds a Co-Authored-By-line, then i think it's wrong. All it did was generate a commit message.

That's fair, but that workflow is amateurish at best anyway. You're free to do that, if that's how you work, or if it all gets squashed in the end regardless, but a more sustainable workflow is to make committing itself an explicit part of self-review.

Here's what I do:

`git commit -p`, which basically shows you chunks of changes and you can confirm whether you want to stage those changes. This forces you to briefly review your changes and allows you to select lines or chunks that you don't want to commit (yet).

Then, you have a lot of context, so then its much easier to compose a commit message. Your editor auto-opens at the end of that, so you type in WHY you made the changes (and if you don't know, at least put WHAT you did), save and close the editor, and that's one commit.

If you skip all of this, the commits are less likely to help you later when you need them.

> This means that if my code was written 20% by me and 80% by Google Antigravity, and Claude then adds a Co-Authored-By-line, then i think it's wrong. All it did was generate a commit message.

But do you do make sure that the commit (and other related documentation updates) properly gives Antigravity the 80% credit, yes?

I'm already at the point of assuming any recently started project is at least part vibe coded, possibly mostly, unless it says otherwise, and my maximum level of being impressed is adjusted accordingly. Even if it says otherwise I might be cautious, as taking a cynical outlook is increasingly turning out to be a useful way of dealing with the world…

Not sure you can go by that alone. It’s a solo project so it’s common to not use PRs in that case.

People review in different ways, sometimes before the commit or at a different cadence.

I get the concern, but I think it would be easy to tell if it’s responsible conclusively just by browsing around the code a bit. If it’s a mess, it’ll be obvious.

It is NOT a solo project if you use Claude AND Antigravity, then it's a 3 people project lol

Note that it's also possible to review changes afterwards. Many (human) teams have been working like this.

I agree that many vibe-coders are likely to skip reviews completely, but it doesn't have to be like this.

Reply on news.ycombinator.com