Crafting inputs and context to steer LLM outputs toward desired behavior and quality.