Robotic Skill Acquisition via Instruction Augmentation with Vision-Language Models