I'll start with the analytic problem. Suppose in a language, /kon-ie/ → konue. That resembles but is distinct from what you are (apparently) interested in. This is just plain old rounding harmony (or backing harmony, or rounding and backing harmony), which in this case affects diphthongs. If a language has a contrast between diphthongs and glide-vowel sequences, this can be easily sorted out, but such a contrast is rare. Lacking a contrast, you might then call on phonetic intuition ("it sounds like a glide"), which doesn't or shouldn't cut it in the scientific world. You might build an argument around durational measurements, since a canonical glide is typically shorter than a canonical high vowel, although nobody has, so far, done a reasonably extensive survey of duration patterns of glides and vowels in 'comparable' positions, so we don't have an established threshold value for glide versus vowel. Perhaps one might argue for a particular analysis of a high vocoid plus vowel based on prosody, if you can establish the syllabification [kon.we], or [ko.nue] where the second syllable is heavy. That, in principle, should not be too hard to do (e.g. look for which syllable attracts stress, if you're lucky enough to have such a stress diagnostic).
The other problem is that surface [konwe] could derive from intermediate konue, thus surface arguments about the output don't establish what vocoid-harmony does (unless you specifically are restricting yourself to IO relations in OT). If the underlying question is about the status of glides in vowel harmony systems (as targets, in particular), then a surface [j ~ w] alternation isn't sufficient to establish that a glide as opposed to a vowel was the target.
Those caveats aside, having been on the lookout for such a thing for decades, I've never encountered any example. Long-distant vowel-consonant harmony is quite uncommon, so the non-attestation of the kind of pattern you describe may result from the fact that vowel-to-consonant assimilations typically only apply between adjacent segments.