I'm still new to Construct, so my suggestions may be short sighted.
- More Python integration. Currently exposure to the core engine is hit or miss. By more, I guess I mean full exposure. If I can "point and click" to it, I should be able to just python it.
- Speaking of Python, the current editor is kind of fumbly. Tabs for example sometimes work, sometimes change focus in the window. Sometimes the tab brackets don't work. Copy-pasting acts weird at times. The "intellisense" (helper text) isn't very complete or helpful, usually. It's not any one thing, it's just not as polished as it probably could be. Right now I use IDLE and copy-paste. But, then there's copy-paste issues at times (python could be less nazi about indentation in general, but what do you expect from a language that has an optional end statement operator).
- Networking. Nothing more than a layer or "wrapper" is really required, but some native support to connect and send packets around would be swell. PodSixNet works pretty good right now. Having integrated support would be faster, though. Object Replication would be dreamy creamy, but you did ask for blue sky stuff. This is the stuff end users would be interested in. Making a network object and telling it to be broadcast, and it internally be taken care of.
RakNet does all of this, quite simply too. I know the licensing might be an issue for an opensource project, but that should be between the game creator and Jenkins, not Construct and Jenkins (RakNet is free for most people -- only when the dough rolls in does he ask for a cut).
- Sprite Indexing and or easier/more obvious instance selection. Instancing in Construct is rather "masked." I don't know why, as it's extremely vital to select the thing you're trying to modify. For Python too!
- Astar module? I have it working in python, sure, but again, more integration means faster computations (compiled C++ vs. python...) It's a pretty easy thing. You guys even have a "node" editor built into some modules, so laying custom paths down would probably be easy, even for 1.0. Pre-generated pathing nodes would be a must too (and would probably be a lot faster in C than again, Python). Not that python is dog slow, it's pretty quick, but more speed is more speed! *grunt, grunt*
- Speaking of Astar, that's probably too low level anyway. Why not just go for sky blue and do AI actors? It's a pretty important part of a game, these days.
Hm, I guess this is more of a 1.0 wishlist than a 2.0 one. Most of this stuff I could see be attainable for 1.0, or well, should be anyway.
Like I said, I'm new still. Can't think outside the box yet =)