Traditionally, plan recognition is defined as the process of matching a set of observations from a user to a library of possible plans in a domain, indicating the possible candidate plans of the user. We have developed a system which allows updates to the plan library during plan recognition and which amends the list of candidate plans, without having to restart the plan recognition process. This work builds on H. Kautz's model (1987) of plan recognition and applies truth maintenance. We describe a graphic-oriented tool which we have developed for users of our system. We comment on the insights which we have gained, about the kinds of users and the kinds of modifications which we might want to accommodate. We emphasize the significance of the system, allowing incomplete or incorrect libraries to be amended. We summarize the challenges which remain, both for interface design and plan recognition. Finally, we elaborate on the potential use of our system for a variety of artificial intelligence applications including knowledge based systems, cooperative AI and natural language understanding.ETX