MirrorBench: An Extensible Framework to Evaluate User-Proxy Agents for Human-Likeness - Explained Simply | ArXiv Explained