<p>Navigating daily tasks relies on working memory to recall past information, manage current goals, and plan future actions. A key factor influencing the dynamic allocation and updating of cognitive resources is an individual’s state relative to their goals. For instance, when following a mental shopping list, decisions about which item to prioritize depend largely on the person’s location. In real-world scenarios, agent states are dynamic and often disrupted by urgent distractions or important interruptions. How people adapt to such perturbations remains unclear because studies of distraction resilience in working memory typically fix agent states. To address this, we developed a working memory paradigm inspired by the arcade game <i>Snake</i>, simulating agent movements in a dynamic environment. Participants (<i>N</i> = 50) controlled a snake in a rectangular field to locate memorized targets (apples) and earn points. Each trial required encoding 1, 2, or 4 apple locations, followed by memory-guided navigation to capture all apples. In half of the trials, distractions (the sudden appearance of additional target grapes) required participants to deviate from their initial plans and collect grapes before resuming the search for hidden apples. Without distractions, participants prioritized nearby targets, using proximity as a cue for memory allocation. When distractions perturbed the agent’s position, participants flexibly redistributed resources to prioritize targets nearer the updated position. This flexibility declined with higher memory loads, and critically, reliance on working memory following distraction was limited to a single item regardless of load. These findings reveal dynamic working memory redistribution as a mechanism that enables flexible but constrained resilience to distraction in dynamic environments.</p>