I'm going for Castles and Chain Reaction, with a game where you solve Slitherlink puzzles on free-form graphs indirectly by making big-picture deductions (this corner is even; this one is odd; here's a loop for the Jordan Curve rule) to let your robot bulldozer sidekick build sand-castle walls by chaining local deductions together. This isn't quite as simple as I wanted, so I may be slightly over-scoping. But slitherlink has been my favorite logic puzzle for years and I've always wanted to do something with it. And I *think* I have a pretty good handle on how to structure the graph data. So I'm going to try it.
I haven't had a good first day: I came up with the idea, and have spent a bunch of time mulling over implementation strategies while doing the more repetitive farm work. But then we spent a whole chunk of the afternoon chasing down and catching a stubborn young steer, and now I'm beat. I put in about an hour and got some of the basic code structure in place, but I'm half falling asleep at the keyboard. So I'm going to bed now and I'll get up early and do some work first thing in the morning while my brain is fresh.