📄 中文摘要
当前的AI编码工具如Claude Code、Cursor和Copilot在理解代码库方面存在不足，主要是因为它们只关注单个文件，而忽视了整体结构和组件之间的依赖关系。现有的解决方案如GitNexus虽然通过图数据库来解决这一问题，但其复杂性和对服务器的依赖使得使用门槛较高。为了简化这一过程，提出了一种基于结构化Markdown和代码健康代理的解决方案，旨在更全面地映射代码及其相互关系，提升代码理解的效率和准确性。
我构建了运行在依赖图上的代码健康代理——无数据库，仅使用Markdown
📄 English Summary
I built code health agents that run on your dependency graph — no database, just Markdown
Current AI coding tools like Claude Code, Cursor, and Copilot struggle to fully understand codebases as they focus on individual files and overlook the broader context of component dependencies. Existing solutions such as GitNexus utilize a graph database to address this issue, but their complexity and server requirements create a high barrier to entry. A new approach is proposed that leverages structured Markdown and code health agents, aiming to provide a more comprehensive mapping of code and its interrelations, thereby enhancing the efficiency and accuracy of code understanding.