Claude Code Source Analysis Orange Book

Claude Code: Source Code Analysis

**Dissecting Anthropic's AI Engineering Decisions from 510,000 Lines of Leaked Code**

Orange Book Series | April 2026 | By Huashu

About This Book

On March 31, 2026, Anthropic accidentally shipped the complete Claude Code source code inside an npm package. 1,902 TypeScript files, 510,000 lines of code, every design decision laid bare.

This book is not a source code reading journal. It dissects the **design decisions** behind Claude Code -- one of the world's most sophisticated AI coding tools. Why does search use grep instead of a vector database? Why does the memory system only store preferences, not code? Why does Auto mode run a second AI for safety reviews? Why Bun over Node.js?

Every choice has a reason. Those reasons are more valuable than the code itself, because they are transferable.

Downloads

Version	File	Pages
English	Claude-Code-Source-Code-Analysis-v260411.pdf	~80
Chinese	Claude-Code-Source-Analysis-zh-v260411.pdf	~80

💡 Download the PDF for the best reading experience. GitHub's online preview may not render properly.

Table of Contents

Part I -- The Big Picture

• Ch 01 A Billion-Dollar Blueprint
• Ch 02 Architecture Overview: The Skeleton of the Harness

Part II -- Core Systems

• Ch 03 System Prompt Engineering
• Ch 04 The Tool System: 4 Primitives, 59 Tools
• Ch 05 The Permission System: Trust by Design
• Ch 06 The Memory System: Remember Preferences, Forget Code
• Ch 07 Context Management: Surviving Long Conversations
• Ch 08 Search: Why grep Beats RAG

Part III -- Advanced Architecture

• Ch 09 Multi-Agent Architecture: Run Like a Company
• Ch 10 The Future in Feature Fl