Well, I actually use Unpoly, and I can assure you that LLMs don't get it, no matter how many pointers to the (excellent!) docs one includes.

Like, even just now, Claude Code with Opus 4-dot-latest, is absolutely convinced you need a bunch of fragile cascading Javascript listeners to dismiss a lower-level menu in case a dialog is opened, while the Unpoly docs, correctly and clearly, point out that 'shatter' exists for just that purpose.

And this is one of the use cases that I continue to highlight as the achilles heel of LLMs. I'm not holding it wrong: they're not reading it right.

ah, then I defer to your experience, i hope that it improves: unpoly is an excellent library

I have the same problem. I guess we will just have to train our own SLM with a carefully selected (unpolluted) training corpus.