What is this?
A Chinese-language static analysis document collection dissecting the source code of Claude Code, after its npm package accidentally shipped unminified source maps revealing the full TypeScript codebase.
A Chinese-language static analysis document collection dissecting the source code of Claude Code, after its npm package accidentally shipped unminified source maps revealing the full TypeScript codebase.Is this a tool I run, or something I read?It is a set of markdown chapters and appendices, not executable software — there is nothing to install or run, only documents to read on GitHub.
It is a set of markdown chapters and appendices, not executable software — there is nothing to install or run, only documents to read on GitHub.How do I go through the analysis?Open the README's chapter index and click into the markdown file for the topic you want, from architecture and security to Memory, Skills, MCP, Sandbox and UI components.
Open the README's chapter index and click into the markdown file for the topic you want, from architecture and security to Memory, Skills, MCP, Sandbox and UI components.Do I need any tools or an API key to use this?No, it is a static documentation repository with no code dependencies, package manager, or API key requirement.
No, it is a static documentation repository with no code dependencies, package manager, or API key requirement.Is this officially sanctioned by Anthropic?No — the authors state it is for academic/technical research only, forbid commercial use, and will remove content if Anthropic objects; all rights to Claude Code remain with Anthropic.
No — the authors state it is for academic/technical research only, forbid commercial use, and will remove content if Anthropic objects; all rights to Claude Code remain with Anthropic.Who would want to read this?Developers and security researchers curious about how a production AI coding agent is actually built internally — its memory system, sandboxing, tool-call/MCP plumbing, and TUI components.
Developers and security researchers curious about how a production AI coding agent is actually built internally — its memory system, sandboxing, tool-call/MCP plumbing, and TUI components.How big was the leaked source code?According to the report, the leaked Claude Code package exposed its complete TypeScript source: 1,902 source files totaling 513,237 lines of code.
According to the report, the leaked Claude Code package exposed its complete TypeScript source: 1,902 source files totaling 513,237 lines of code.What topics does the analysis actually cover?Eighteen chapters plus appendices covering program architecture, user-data collection and security risks, Agent Memory, Skills, Tool Call mechanics, MCP integration, Sandbox isolation, context/prompt management, multi-agent logic, session persistence, and a full breakdown of the TUI component tree.
Eighteen chapters plus appendices covering program architecture, user-data collection and security risks, Agent Memory, Skills, Tool Call mechanics, MCP integration, Sandbox isolation, context/prompt management, multi-agent logic, session persistence, and a full breakdown of the TUI component tree.Does it compare Claude Code to other coding agents?Yes, chapter 16 and an appendix compare Claude Code's architecture and capabilities against Codex, Gemini CLI, Aider, and Cursor.
Yes, chapter 16 and an appendix compare Claude Code's architecture and capabilities against Codex, Gemini CLI, Aider, and Cursor.Are the claims backed by actual code, or just guesses?The authors say the analysis is based on static reading of the leaked src/ directory, with a dedicated chapter (17) and file-tree appendix cross-referencing claims to source evidence and public documentation.
The authors say the analysis is based on static reading of the leaked src/ directory, with a dedicated chapter (17) and file-tree appendix cross-referencing claims to source evidence and public documentation.What are similar projects?how-claude-code-works, CodeMason, qwen-code, claude-code-book, ai-agent-deep-dive, learn-coding-agent, Hy-Memory, lean-ctx
how-claude-code-works, CodeMason, qwen-code, claude-code-book, ai-agent-deep-dive, learn-coding-agent, Hy-Memory, lean-ctxWhich topics is it in?source-code-analysis, security-research, static-analysis, agent-memory, sandbox, academic-research, reverse-engineering, anthropic, china, documentation, agents, coding-agent
source-code-analysis, security-research, static-analysis, agent-memory, sandbox, academic-research, reverse-engineering, anthropic, china, documentation, agents, coding-agent