[redhale]

This is a cool idea!

But my question is: why not just use HTML+CSS?

This is what I do, and it works amazingly well in my experience. LLMs are really good at HTML, and they can generate a clickable prototype in a few seconds.

And the benefit of HTML is that it can actually resemble the thing I'm going after. It's still throw-away code -- when it comes time to build, this HTML is usually used in screenshot form as I might use a quick and dirty Figma mock-up.

Going through ASCII feels like forcing the LLM to go through a translation layer that is going to make things more complicated (like needing 6 skills to be able to do it, for instance).