claude-code-book

Why we included this project

This is a serious open technical reference for anyone who wants to stop treating Claude Code as a black box and see how a production-grade agent harness is actually built. The book skips prompt tricks and API recipes in favor of walking through the core subsystems, from the conversation loop and tool system to the permission pipeline, context compression, memory, hooks, subagent scheduling, and MCP integration, explaining the reasoning behind each design choice. Its 139 architecture diagrams isolate reusable patterns rather than vendor-specific details, so people building agents on LangChain, AutoGen, or from scratch get transferable mental models. Engineering leads who want a shared vocabulary for evaluating agent frameworks will also find it a rare, concrete study in agent-architecture thinking.