This is a really neat concept. I can definitely see some interesting puzzles coming out of this idea. It was a good choice to use rounded hitboxes for everything, so that you could sort of fudge the corners if needed rather than getting stuck on 2 pixels in the direction you left behind.
The biggest quality-of-life change that I can see being made is making it more clear when you're able to interact with the X boxes. It could even be as simple as highlighting the box when the player comes into contact with them.