📄 中文摘要
大多数迁移在开始之前就失败了，因为人们并不真正了解系统的功能。遗留系统的失败往往不是由于语法或框架，而是因为其行为缺乏文档支持且难以理解。在使用代理进行开发时，这种理解的缺失变得更加关键。针对跨技术栈迁移的情况，现有系统需要迁移到不同的技术栈，通常面临的最大挑战是对当前系统知识的不完整。Copilot、Codex、Claude Code 等工具使得可以互动地探索代码库，从而帮助开发者更好地理解和迁移现有系统。
基于大型语言模型的代码库分析以支持迁移：Codex、Claude 和 VS Code 代理的比较
📄 English Summary
LLM-Assisted Codebase Analysis for Migration: Comparing Codex, Claude, and VS Code Agents
Most migrations fail before they even begin due to a lack of understanding of the system's functionality. Legacy systems typically do not fail because of syntax or frameworks, but rather due to undocumented behavior and poor comprehension. This issue becomes even more critical when development is conducted with agents. In cross-stack migration scenarios, where a system must be transitioned to a different technology stack, the greatest challenge often lies in incomplete knowledge of the current system. Tools like Copilot, Codex, and Claude Code enable interactive exploration of codebases, assisting developers in better understanding and migrating existing systems.