> The best experiences I can remember were when I was building games prototypes in XNA. I was only using the basics of the framework and writing a lot of the engine myself.

This happens to me as well, it's easy to keep churning out code when it's coming from my current bag of knowledge. My flow gets wrecked when I try to improve my code, because I have to stop writing the way I'm currently writing things the way I want to. When I mean "improve" code, it has to attain a certain metric and would usually entail giving benefits to other people working on the same codebase. Do I want the code to be more readable? more performant? more modular? more abstracted? would it depend-on/use another library?

> What type of work do you do that regularly gets you in this state?

I find that I go back to a stop-start fashion of working when I don't know have a list of small wins I need to do to finish a big feature or "improving" code. Without that list, it leads me to procrastination.

So the first type of work I do? I bust out my pen and I get a piece of paper. Then I write down a small checklist of classes/methods I need to write, or small things I need to research (with a timebox).