A parser that would walk all web document elements, exercise keyboard access functions, then report back on what is or is not accessible via keyboard, touch, or click.
I think that would actually be the best way to implement it. A bookmarklet would be nice, however I don't see that fitting within the technical constraints of size limits.