动手学 CS146S 中文

★ 1.5kGetting Started

#course#chinese#assignments

Overview

A Chinese edition of Stanford's CS146S 'Vibe Coding Together' course (1.5k+ stars), focused on software engineering in the AI era: as LLMs reshape the traditional software development lifecycle, how the next generation of engineers can use AI to become 10x more productive. It argues that development has shifted from 0-1 code writing to an iterative loop of plan → AI-generate → modify → repeat, and uses hands-on assignments plus guest lectures from industry pioneers (Claude Code, Warp, Vercel, a16z) to teach AI-assisted development, automated testing, smart documentation and vulnerability detection.

Who it's for

Programmers with roughly CS111-level experience who want to keep up with AI-assisted development, plus self-learners who treat Stanford courses as a resource.

What you'll learn

01Course positioning: software engineering in the AI era
The course answers one key question: how next-generation engineers should leverage LLMs to become 10x more productive and prepare for future careers. It argues AI tools not only dramatically boost developer productivity but also make software engineering more accessible to a broader audience, and it demonstrates how development has shifted from 0-1 coding into an iterative loop of plan → AI-generate → modify → repeat.
02Grading and prerequisites: you learn by doing
The prerequisite is programming experience roughly at CS111 level (CS221/229 knowledge recommended). Grades are 80% final project, 15% weekly assignments and 5% class participation — a strongly hands-on design: you don't learn by listening, you learn by building. The course advises budgeting about 10-12 hours per week.
03A ten-week outline: from LLM fundamentals to post-agent deployment
The syllabus spans ten weeks: coding LLMs and prompt engineering, anatomy of coding agents (including MCP), AI IDEs, coding agent patterns, the modern terminal (Warp), AI testing and security, modern software support (code review), automated UI and app building, life after agents (observability and incident response), and the future of AI software engineering. Each week ships readings, slides (with Chinese versions) and guest lectures.
04Assignments and toolchain: quality assignments published every Sunday
The project actively maintains high-quality assignments and releases new ones every Sunday; published ones cover LLM prompting practice, first steps in the AI IDE, building a custom MCP server, coding with Claude Code, agentic development with Warp, secure AI coding, code-review reps and multi-stack web app builds. The repo also curates a CLI tool list and a comparison of Chinese coding platforms (GLM-4.6, Kimi K2, Doubao-Seed-Code, DeepSeek-Coder, Qwen3-Coder, etc.).
05How to use it: a self-study audit path
Even if you can't attend Stanford, you can audit it weekly on your own: for each topic, read the materials, watch the (bilingual) slides and guest lectures, then do the assignment. The FAQ also answers the language question (examples lean on Python/JavaScript; the point is modern development practices, not any specific language) and whether AI tool experience is needed (no — it starts from the basics).

Practice tips

Audit it week by week: one topic per week — read the materials, flip through the bilingual slides and guest talks, then verify yourself with that week's assignment.
Week 2 is all about MCP — build a custom MCP server for the assignment and you'll have a transformed understanding of how agents 'use tools'.
Map the tools you already use (Claude Code / Codex / Cursor / Warp, etc.) onto the syllabus and run one small experiment per week in your own project.
Form a study group or join the course chat, and review each other's weekly assignments — right on theme for Week 7, 'code review'.