📄 中文摘要
随着智能代理系统的不断发展，构建复杂的代理系统已变得相当成熟。然而，验证这些系统的有效性却缺乏相应的严谨性。提出了一种综合框架，旨在对大规模语言模型（LLM）代理进行离线评估，以确保其在实际生产环境中的可靠性和性能。该框架不仅考虑了代理的功能性，还涵盖了安全性和可解释性等关键因素，从而为开发者提供了一个全面的评估工具，促进了 LLM 代理的实际应用和改进。
生产就绪的 LLM 代理：离线评估的综合框架
📄 English Summary
Production-Ready LLM Agents: A Comprehensive Framework for Offline Evaluation
The development of sophisticated agent systems has advanced significantly, yet the rigor in validating their effectiveness remains underdeveloped. A comprehensive framework for offline evaluation of large language model (LLM) agents is proposed, aimed at ensuring their reliability and performance in real-world production environments. This framework considers not only the functionality of the agents but also critical factors such as safety and interpretability. By providing developers with a thorough evaluation tool, it facilitates the practical application and enhancement of LLM agents.