Autonomous underwater vehicle (AUV) missions are challenging due to limited or no communication with the vehicle and uncertainty about what may be encountered during the mission. The vehicle has limited battery power and memory to store datasets and does not know how much of these resources its actions will consume. Managing the uncertainty by building contingency plans (as in previous work) is infeasible owing to the large number of contingencies. Purely online planning is also difficult because of restricted computation. We propose a mixture of off-line planning and on-line plan repair, presenting an approach for deleting parts of a plan when resources are tight, or stitching new plan fragments into an existing plan when additional resources are available. We discuss this novel approach using a simulated AUV mission domain.