Goal-directed behavior in dynamic environments requires holding multiple action goals in working memory (WM) while remaining flexible enough to update plans when unexpected distractions occur. To investigate how WM capacity and prioritization constrain replanning after interruption, we developed a dynamic navigation working memory paradigm modeled after the arcade game Snake, in which participants encoded 1, 2, or 4 target spatial locations and navigated to capture them. On half of the trials, an unexpected distraction displaced the agent during navigation, requiring participants to replan their trajectory relative to their new spatial position. Without distraction, participants spontaneously prioritized targets based on spatial proximity. Following a displacing distraction, participants flexibly updated their plans to prioritize the target closest to their new position, but this replanning ability declined as WM load increased and was effectively constrained to updating a single active goal in working memory.