The cursor-shape property sets the shape of a caret, which addresses a reasonably common use case for text editing by allowing terminal-like styling for editable text.

But it makes little sense for carets used in other situations, particularly caret-based-navigation. The spec relies on font metrics and text shaping results to determine the size of a block caret and the width of an underscore, and these things are only available in text (or could come from the root).

Consider in particular this result for a chrome test with an image in content editable. https://source.chromium.org/chromium/chromium/src/+/main:third_party/blink/web_tests/platform/mac/editing/pasteboard/copy-standalone-image-expected.png

I propose the spec be updated to limit caret-shape to "editable elements" as defined for user-select. Even better, define all the caret properties as "must apply" to "editable elements" and may apply to other carets (like navigation or some future thing). The example above of an image would presumably fall back to "no next character".