New challenges posed in many areas of AI research represent a departure from domain-specific applications-To systems that can more effectively cope with larger and more uncertain domains. Such knowledge intensive applications require the easy and efficient utilization of tremendous amounts of knowledge. The magnitude of knowledge and often stringent response constraints that characterize such applications poses a computationally prohibitive search problem. A proposed technique for addressing this problem is a parallel search technique known as marker-passing. Past work in markerpassers has shown they often return too much information, becoming a bottleneck of the system in which they are embedded. This paper presents the design of a flexible marker-passing mechanism embedded in a commonsense reasoning model, RABIT (Reasoning About Beliefs In Time), which overcomes this difficulty. The unique design we present avoids traditional drawbacks in markerpassing implementations by emphasizing search over inference as the goal of the marker-passing process. This marker-passing design is powerful due to its separation of the marker-passing process from the knowledge contained in the network itselL thus allowing for its potential use not only in the area of commonsense reasoning, but also in many other domains, including, but not limited to, natural language processing, general-purpose planning, and robot navigation.