This paper explores the role of spontaneous retrieval in prospective memory, in an agent implemented in the Soar cognitive architecture. At goal initiation time, spreading activation causes the goal to be the most activated element in long-term memory, at which point it is spontaneously retrieved into working memory and pursued. We show that goal encoding specificity increases prospective memory performance, while a lengthier retention interval decreases performance if the percepts are differentially presented; both trends qualitatively resemble results described in psychology literature. However, a large space of possible spontaneous retrieval implementations remain unexplored, and much work remains to be done before spontaneous retrieval in a cognitive architecture can be fully understood.