claude-code-book

This repository is a freely readable Chinese-language technical book, approximately 420,000 Chinese characters across 15 chapters and four appendices, dedicated to dissecting the internal architecture of Claude Code rather than teaching users how to prompt it. Written in Chinese with an English translation available, it examines how Claude Code functions as an Agent Harness, covering its conversation loop built on an async-generator main loop, a four-stage permission pipeline with speculative classifier racing on a 2-second Promise.race, four-level progressive context compression strategies, a hook system spanning 26 lifecycle events, MCP integration across eight transport protocols, and a Fork mechanism for subagent context inheritance. The book includes 139 Mermaid architecture diagrams, a glossary of 100 bilingual terms, and a catalog of 89 feature flags across 13 categories. A standout concrete detail is the startup optimization analysis showing latency reduction from 160ms to 65ms through lazy loading. Architects, senior engineers building their own agent frameworks with tools like LangChain or AutoGen, and Claude Code power users seeking to understand design intent rather than surface-level usage are the primary intended audience.

Lo que la gente pregunta sobre claude-code-book