WIth Skyfeed, I stopped trying to make complex one-size-fits-all all regex and just brute-forced it.

Are you using (?:^|\W) and (?:$|\W) instead of \b to get around the fact that in Skyfeed \b considers extended Latin characters to be word boundaries? Eg: /\bbcd\b/ will match the “bcd” in “âbcdẽ.”