Fair point, and it's not implemented that way currently. It's more like "custom instructions" but thanks for pointing that out. I haven't used multiple system prompts in the OpenAI playground either, so I hadn't given it much thought.