The advantages of rule-based techniques are well known, but traditional expert system shells are unsuited to controller development. In this paper we clarify the differences between the controller development task and traditional applications of expert system shells, and describe a shell we have developed for rapid- prototyping development of controllers. To deal with the differences the shell uses fuzzy inference, a single inference step per time step, a special architecture for symbol grounding, and vector variables. We describe the script language, and two example applications. One is to control a simulation of a truck backing up, the other is as part of a sonar-sensing mobile robot control system.