Ask HN: How do you manage your prompts in agent workflows

I am working on AI Agents and have 6 agents that are engaging in a conversation. As I am working on the architecture, I noticed that managing the interdependencies between the prompts is annoying.

Are there best practices I don't know yet?