In this paper we demonstrate how to use action systems to specify and refine self-adaptive systems. Focusing on the special case in which there are a fixed set of possible configurations, we use a recently-introduced synchronization operator to achieve separation of concerns between adaptation policy and steady-state system execution. The use of action systems permits the application of standard techniques for refinement to construct correct implementations from higher-level models. We illustrate the application of this idea to the problem of coping with dynamically-varying user requirements.