openai/openai-agents-python ? reverse-engineered prompt

I want to build a system using Python that lets me create and manage a team of AI agents. The idea is to have a main 'assistant' agent that I can give high-level goals to. This main agent should then be able to delegate specific tasks to other, more specialized agents.

For example, I'd want a 'coder' agent that can work with files on my computer inside a safe sandbox. It should be able to read through a project, run commands, and even apply code patches to fix bugs or add features. I also want a 'researcher' agent. Importantly, I need a "human in the loop" feature so I can approve the agents' plans before they execute them. It would be great if I could also interact with the main assistant via voice. Let's use the new OpenAI agents SDK for this and make sure it can connect to various LLM providers, not just OpenAI.