Best practices when extending dplyr?

As a little context, I'd opened a previous topic on doing pairwise operations in dplyr which turned into the package pwiser. pwiser is just a slight modification on dplyr::across() so I'd hoped it would "just work" with dbplyr, and the other dplyr backends, but this turned out not to be the case (brshallo/pwiser#8). However I'd wondered to myself whether there were some "best practices" I could/should have followed that would have come with this benefit of having pwiser work with backends "out of the box" (or with minimal additional effort) along with other benefits, like being more robust to potential future updates to dplyr (that I'm sure pwiser is vulnerable to).

P.s. I am not actively developing pwiser as dplyover can do the same thing and much more.