I think Opus 5.5 works even better with recording tutorial videos of your real application / ui, using agent written playwright e2e tests. I’ve been working on this workflow for couple of months and built a service around it (screenci.com), so I have some hands-on experience with it.

Instead of using Choice with criteria "space_0", "space_1"..., it could be even more elegant to use criteria names like "", " "... which could be directly inserted into response.