How to Evaluate Custom Multi-Agent Systems with Your Benchmarks?

Hi! Thanks a lot for the awesome repo.
I'm interested in using your benchmarks to test some multi-agent frameworks, and I was wondering if you could provide some examples or guidance on how to build a multi-agent system and generate conversations that can be evaluated and fairly compared with the results in your paper https://arxiv.org/pdf/[2412.05449v1](https://arxiv.org/pdf/2412.05449v1) .
For example, how should I initiate system messages for the user and for different agents?

Thanks in advance for your help!