Agents on the Bench: Large Language Model Based Multi Agent Framework for Trustworthy Digital Justice