This paper presents GHOST, a combinatorial optimization solver an RTS AI developer can use as a blackbox to solve any problems encoded by a constraint satisfaction/optimization problem. We show a way to model three very different RTS problems by a constraint satisfaction/optimization problem, each problem belonging to a specific level of abstraction, and test our solver on them, using StarCraft as a testbed. For the three problems (the target selection problem, the wall-ion problem and the build order planning problem), GHOST shows very good results computed within some tens of milliseconds. Game AI,