CoMA: Compositional Human Motion Generation with Multi-modal Agents - View it on GitHub