git blame suggests that this hasn't ever matched on word boundaries (it didn't in August 2012, when it was first written -- see rPHU45f7bd6a).

I think the current behavior is fairly reasonable, and there isn't too much of a use case for linking half a word, although I can't immediately come up with cases where this would trigger a false positive -- for example, c[d[0]] is valid/common in C-like languages, but I don't think there's a valid/common way to follow that with a word character, and [[ isn't common in C-like languages.

Since there doesn't seem to be much of a driving use case here (and this isn't, as far as I can tell, a behavioral change) I'm inclined to leave it as-is, I'd be open to reconsidering this if you have a really cool use case for half-word links.