diff --git a/LEARNING_GUIDE.md b/LEARNING_GUIDE.md index b96263374..a42645127 100644 --- a/LEARNING_GUIDE.md +++ b/LEARNING_GUIDE.md @@ -652,7 +652,7 @@ while True: - [OpenCode](https://opencode.ubitools.com/) — 开源的终端 AI Agent,可以参考 ### 相关仓库 -- [claude-code-sourcemap](https://github.com/anthropics/claude-code-sourcemap) — 你正在看的这个 +- [stevenchouai/claude-code-sourcemap](https://github.com/stevenchouai/claude-code-sourcemap) — 你正在看的这个非官方研究仓库 - [Prajwalsrinvas/claude-code-reverse-engineering](https://github.com/Prajwalsrinvas/claude-code-reverse-engineering) — 逆向工程分析 - [Yuyz0112/claude-code-reverse](https://github.com/Yuyz0112/claude-code-reverse) — 中文逆向分析项目 diff --git a/README.md b/README.md index ca7c8cdc8..ecedc7cdf 100644 --- a/README.md +++ b/README.md @@ -3,6 +3,10 @@ [![Tutorial Site](https://img.shields.io/badge/教程网站-在线阅读-blue?style=for-the-badge)](https://stevenchouai.github.io/claude-code-sourcemap/) [![GitHub Stars](https://img.shields.io/github/stars/stevenchouai/claude-code-sourcemap?style=for-the-badge)](https://github.com/stevenchouai/claude-code-sourcemap) +

+ Claude Code architecture source-map research flow +

+ > [!WARNING] > This repository is **unofficial** and is reconstructed from the public npm package and source map analysis, **for research purposes only**. > It does **not** represent the original internal development repository structure. diff --git a/assets/claude-code-architecture-map.svg b/assets/claude-code-architecture-map.svg new file mode 100644 index 000000000..3c12c231f --- /dev/null +++ b/assets/claude-code-architecture-map.svg @@ -0,0 +1,31 @@ + +Claude Code architecture map +A dark technical diagram for Steven Chou's AI agent engineering work. + + + + + + + + + + + +Claude Code Architecture Map +Unofficial source-map research: npm package → restored sources → readable agent runtime tutorial. + + npm packagecli.js + cli.js.map + source map restore4,756 files + runtime conceptsloop · tools · safety + tutorial website8 chapters + research notes + + + + + Agent loop mental model + model response → tool call → permission gate → shell/file operation → tool result → next model call + The repo turns minified runtime evidence into a learnable architecture map for AI coding agents. + +stevenchouai · trace over vibes +