This work is about the metatheory of actions, and here we address the problem of what a good domain description for reasoning about actions should look like. We state some postulates concerning this sore spot, which establishes the notion of a modular domain description. We point out the problems that arise when modularity is violated and propose algorithms to overcome them.