The isInRemoteFrame check here is trying to crawl the frame tree, but if you're not even in the document yet, which is a totally valid way to use an IntersectionObserver, it'll crash because we can't tell if you're in a remote frame. The lower level IO API uses ExecutionContext for this, so it always grabs the frame's document instead.

From C++ we'll need to figure out how to make this work, but looking at the owner document of the element you're observing doesn't work because that document might have no frame (ex. <template> or document.implementation.createHTMLDocument()).

estark@ is trying to use ElementVisibilityObserver to observe password inputs, for now I suggest we just don't observe ones that originate from frameless documents since those are rare, but we need to figure out how to fix the ElementVisbilityObserver API itself here.