A learned model that generates task-specific modifications to a base execution harness to improve agent performance.