Andrew Ng × Anthropic · Official Course Deep Dive
Based on the Agent Skills course by Andrew Ng and Anthropic on DeepLearning.AI. We provide source-level analysis of Claude Code, Harness Engineering methodology research, and comparative study of MCP and Context Engineering approaches across Anthropic, OpenAI, and independent builders.
Claude Code is Anthropic's official CLI tool, built on ~512,000 lines of TypeScript across 1,884 files. We break down the architecture from a system-wide view down to eight core subsystems.
Complete breakdown of 6 subsystems: Entry Layer, Query Engine, Tool System, Command System, Permission System, and Multi-Agent Coordination. Based on 1,884 TypeScript files.
Read more →The complete course library is currently available in Chinese — including Agent Skills L0-L8 lessons, Harness Engineering methodology, and the Harness architecture book (10 chapters + 2 appendices).
Browse (Chinese) →The Agent Skills course — taught by Elie Schoppik — covers how to build reusable Skills for Claude, combine them with MCP and Subagents, and apply them across Claude.ai, Claude Code, Claude API, and the Agent SDK. English lesson-by-lesson notes are being translated from the Chinese originals. Below is what's already mapped out.
Agent Skills Dev publishes source-level analysis of Claude Code and engineering methodology research on building AI agents — not generic intro tutorials. Every piece is grounded in first-hand source code or primary references. We currently publish primarily in Chinese (34 pages) with English translations expanding from the Claude Code architecture piece outward. Follow along if you build AI agents for a living.