Claude Code Report

This report analyzes the core source code of Claude Code's Agent section, aiming to quickly learn the structure and some technical details of industrial-grade MAS (Multi-Agent System). Notice: Please credit the source when reposting. Thank you!

[中文](./README.zh.md) | [English](./README.md)

---

Claude Code Source Analysis

Reverse-engineering the Claude Code Agent architecture using nanobot as a reference.

Table of Contents

• Ch 00 — Global Flow Diagram
• Ch 01 — Project Overview
• Ch 02 — Tech Stack
• Ch 03 — Directory Structure
• Ch 04 — Startup Flow
• Ch 05 — LLM Streaming Call Details
• Ch 06 — Tool Invocation System
• Ch 07 — Streaming Execution Engine (disabled by default)
• Ch 08 — Thinking / Reasoning Mode
• Ch 09 — Task System
• Ch 10 — Skill System
• Ch 11 — Compact: Conversation Compression
• Ch 12 — Memory System
• Ch 13 — Context Management
• Ch 14 — Hooks Extension
• Ch 15 — Sandbox & Permission System
• Ch 16 — MAS toward Harness Engineering
• Ch 17 — MCP Integration
• Ch 18 — Summary: Claude Code Architecture Overview

---

Ch 00 — Global Flow Diagram

Chapter references: [Ch04](#ch-04--startup-flow) · [Ch05](#ch-05--llm-streaming-call-details) · [Ch06](#ch-06--tool-invocation-system) · [Ch07](#ch-07--streaming-execution-engine-disabled-by-default) · [Ch08](#ch-08--thinking--reasoning-mode) · [Ch09](#ch-09--task-s