harness-engineering-from-cc-to-ai-coding

Why we included this project

This is a full-length technical book, written in Chinese with an English preview, that works backward from the public Claude Code release to show how a production AI coding agent is actually built. It deliberately skips rehashing Anthropic's official docs and instead digs into the real implementation: the agent loop and tool orchestration, system and tool prompts, context compression and token budgets, permission and rule systems, hooks, multi-agent orchestration, and the skills system. Engineers building their own coding agents, agent frameworks, or tool-calling platforms will find patterns they can borrow directly. It also works as a study path for developers who want to understand Claude Code's internals before configuring or extending it, and for teams curious about turning source-level analysis into reusable engineering practice. The book ships as an mdbook site, so it is meant to be read rather than run.