It'd be neat if (for web stuff) people put them at like /.well-known/prompt.md.

Yes. Open prompts.

I would rather have it the other way around.

Source for one-shotted apps so LLM gives you repo back and template to build on instead of burning electricity on ideations from scratch.

I imagine 90% of stuff could be just that.

But I also imagine people just cloning repo don’t feel it truly belongs to them, but when agent slops out code for them somehow they feel like they did the work themselves.

Agreed! We desperately need an OSS framework of basic modular building blocks for the new generation of vibecoders, if nothing else just to stop the massive waste of resources of having all the models build the same thing hundreds of thousands of times.

A good rule for agents is to 1. Search GitHub for existing or related projects to pull from and 2. Publish code as share a like.

It’s also good practice since SAAS is dead anyway, so it helps kill the fantasy your vibe coded app is somehow going to live on code alone.

SaaS is not dead.

This Saas is no more! It has ceased to be! 'E's expired and gone to meet 'is maker! 'E's a stiff! Bereft of life, 'e rests in peace! If you hadn't nailed 'im to github 'e'd be pushing up the daisies! 'Is CI/CD processes are now 'istory! 'E's off the twig! 'E's kicked the bucket, 'e's shuffled off 'is mortal coil, run down the curtain and joined the bleedin' choir invisible!! THIS IS AN EX-SAAS!!

SaaS is only dead if you believe you paid only for software there is still part “as service” because running and maintaining software still takes time even if slopping out software is virtually free.

I'm thinking that if somehow we could force temperature to 0, it could even be "prompt as compressed source". In the sense that you know if you run a given prompt through a given model version you would get the exact same results. This could have some space applications. You send a powerful model to space, iterate with the same model on earth and once the results are good only send the prompt to the remote model... Well, now that I'm typing it, I guess if the model is good enough you don't even need that and only give it a goal before leaving.

Games that people talk about on HN: https://github.com/nicobrenner/project24#readme

https://news.ycombinator.com/item?id=50361909

I hadn't thought of that, but that's a really good idea - prompts as source code. you're right, most of them will never see a formal update, so the prompt is a great way to hand off some control.

I think it's the future for prompt-based engineering honestly. I employ it myself too. I Keep a prompts file, either for sharing or looking back at it. I think Open Source Software should adopt it in their repos for a. transparency and b. learning material for the future.