I hope others weigh in as well, because I'm curious what others are thinking at this point.
Personally: I set out with Yesod to bring standard MVC backend server functionality to Haskell. Yesod for me is essentially feature complete. I don't have specific additions I'm looking to make. On the other hand, I'm _very_ hesitant to make breaking API changes, and adding new functionality into the core libraries of Yesod increases that probability. My recommendation for a while has been for new functionality to be added as new libraries that can iterate their designs.
I do have plenty of improvements I'd love to make to Yesod from a design perspective. As a simple example, I think overhauling the monad transformer approach to be more in line with the ReaderT pattern would be a good thing. And I've considered such changes in the past. But these kinds of changes seem to mostly appeal to a code cleanliness goal and involve lots of tedious changes for end users, so I've held off (in line with my stability goals).
That's the primary reason I haven't pushed new changes to Yesod in years.