A wealth of developmental research suggests that preschoolers are capable of reporting, imitating, and performing sequential actions they engage in routinely. However, few studies have explored the developmental and cognitive mechanisms required for learning how to perform such routines. A previous computational model of routines argued that a representation of task contexts underlying routines could change flexibly. This position was supported by the empirical evidence that if adults are interrupted in the course of a routine, they make fewer errors if they are interrupted just before the selection of context-dependent action than if they are interrupted earlier. Another computational model examined how efficiently adults learned to perform routines and suggested the relationship of learning efficacy with executive functions. The present study aimed to examine whether the above-mentioned models and evidence from adults can be extended to preschoolers by using an experimental task, in which children were required to play the role of a baker and repeatedly make toast for either a cat or mouse, with momentary distractions. Experiment 1 showed that earlier interruption tended to cause older children to produce more branch point errors than interruption immediately before the branch points, whereas younger children tended to be vulnerable to both interruptions. Further, across 2 experiments, this study showed that the developmental differences in how young children represent task contexts were associated with their executive functions. These findings indicate that the representational flexibility of task contexts underlies children's performance of repeated sequential actions and its association with executive functions.