refactor(shadcn-lint): agentic linting cc @millions-afternoon-47582 @hundreds-twilight-32833 Please let me know if you have concerns, I’m eager to merge this, we can revert easily if needed (well our agents can 😅) I’m a bit excited, because of the consequences: • the direct consequence, this seems easier for agents, so putting in place this linting will help agents fix stuff the right way, not only catch it, but also not fix it with a million special variables all over the place • the indirect consequence: LLM abilities are at the point where UI auto-testing is unbelievably good, even for a large surface (this PR had 137+ screenshots taken), so I’d love to move forward to automating this. I find this just so cool: OpenHands UI Audit from PR 17435 • 137+ screenshots of visuals • and, it found pre-existing issues!

This is very cool! What kicked off that PR audit with all the screenshots? Was that just Astra? I did some testing recently with screenshots but didn't have much success and would love to revisit it.

I was amazed it worked out so well too! It went like: • prompted to test UI • it made 3 screenshots and said “smoke” UI test done 😂 (“smoke-testing” is very GPT specific, 5.4-6 at least) • prompted to UI test the entire PR • it did 137 screenshots It actively used the qa-changes skill, this is also relevant, I believe. It does insist on testing main branch vs PR branch, so it implies strongly the entire PR

Which skill is the result of combining that one with the one from poteto/pstack

So I’d say: • it's promptable • LLM capabilities are there, at long running full UI testing of an app • Astra can do it

Oh oh I see

So that, regardless of PR, it can test the entire user-facing functionality