Reading 摘要總目錄
CS146S 的 9 週共 45 篇 reading（W8、W10 原 syllabus 沒列 reading）。每篇 200-400 字繁中摘要，含核心論點 / 關鍵概念 / 對 CS146S 的意義 / 對 vibe coder 的 takeaway / 原文連結。
統計
| 狀態 | 篇數 | 說明 |
|---|---|---|
| ✅ Complete | 43 | 抓到原文後做的繁中摘要 |
| 🟡 Best-effort | 1 | 原文取得受限（access code），用主題脈絡做高層摘要 |
| 🔴 Failed | 1 | 原文無法抓取（X/Twitter 需登入），只放連結 + disclaimer |
來源類型分布
| 類型 | 篇數 | 範例 |
|---|---|---|
| Blog post | 33 | Anthropic, Stytch, Cloudflare, Splunk, OWASP, … |
| GitHub README | 6 | MCP servers, MCP SDK, Awesome Claude Agents, Super Claude, ACE-FCA, O3 prompt |
| YouTube | 3 | Karpathy Deep Dive, Anthropic Prompt Engineering, AI Code Review Lessons |
| 2 | OpenAI Codex use cases, Anthropic uses Claude Code | |
| arXiv | 1 | AI-Assisted Assessment of Coding Practices |
| Twitter/X | 1 | How FAANG Vibe Codes（failed） |
Week 1：Introduction to Coding LLMs and AI Development
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 1 | Deep Dive into LLMs | YouTube | 🟡 高層摘要 |
| 2 | Prompt Engineering Overview | Blog (Google Cloud) | ✅ |
| 3 | Prompt Engineering Guide | Blog (promptingguide.ai) | ✅ |
| 4 | AI Prompt Engineering: A Deep Dive | YouTube (Anthropic) | 🟡 高層摘要 |
| 5 | How OpenAI Uses Codex | ✅ |
Week 2：The Anatomy of Coding Agents (MCP)
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 6 | MCP Introduction | Blog (Stytch) | ✅ |
| 7 | Sample MCP Server Implementations | GitHub | ✅ |
| 8 | MCP Server Authentication | Blog (Cloudflare) | ✅ |
| 9 | MCP Server SDK (TypeScript) | GitHub | ✅ |
| 10 | MCP Registry | Blog | ✅ |
| 11 | MCP Food-for-Thought | Blog | ✅ |
Week 3：The AI IDE
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 12 | Specs Are the New Source Code | Blog (Ravi Mehta) | ✅ |
| 13 | How Long Contexts Fail | Blog | ✅ |
| 14 | Devin: Coding Agents 101 | Blog (Cognition) | ✅ |
| 15 | Getting AI to Work In Complex Codebases (ACE-FCA) | GitHub | ✅ |
| 16 | How FAANG Vibe Codes | Twitter/X | 🔴 無法抓取 |
| 17 | Writing Effective Tools for Agents | Blog (Anthropic) | ✅ |
Week 4：Coding Agent Patterns (Claude Code)
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 18 | How Anthropic Uses Claude Code | ✅ | |
| 19 | Claude Code Best Practices | Blog (Anthropic) | ✅ |
| 20 | Awesome Claude Agents | GitHub | ✅ |
| 21 | Super Claude Framework | GitHub | ✅ |
| 22 | Good Context, Good Code | Blog (StockApp) | 🟡 高層摘要 |
| 23 | Peeking Under the Hood of Claude Code | Medium | ✅ |
Week 5：The Modern Terminal (Warp)
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 24 | Warp University | Blog | ✅ |
| 25 | Warp vs Claude Code | Blog | ✅ |
| 26 | How Warp Uses Warp to Build Warp | Notion | ✅ |
Week 6：AI Testing and Security
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 27 | SAST vs DAST | Blog (Splunk) | ✅ |
| 28 | Copilot RCE via Prompt Injection | Blog | ✅ |
| 29 | Finding Vulnerabilities Using Claude Code and OpenAI Codex | Blog (Semgrep) | ✅ |
| 30 | Agentic AI Threats: Identity Spoofing | Blog (Palo Alto Unit 42) | ✅ |
| 31 | OWASP Top Ten | Blog (OWASP) | ✅ |
| 32 | Context Rot: Degradation in AI Context Windows | Blog (Chroma) | ✅ |
| 33 | Vulnerability Prompt Analysis with O3 (CVE-2025-37899) | GitHub | ✅ |
Week 7：Modern Software Support (AI Code Review)
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 34 | Code Reviews: Just Do It | Blog (Coding Horror) | ✅ |
| 35 | How to Review Code Effectively (GitHub Staff Engineer) | Blog (GitHub) | ✅ |
| 36 | AI-Assisted Assessment of Coding Practices in Modern Code Review | arXiv | ✅ |
| 37 | AI Code Review Implementation Best Practices | Blog (Graphite) | ✅ |
| 38 | Code Review Essentials for Software Teams | Blog | ✅ |
| 39 | Lessons from millions of AI code reviews | YouTube (Graphite) | ✅ |
Week 8：Automated UI and App Building
本週原 syllabus 沒列 reading list。延伸推薦資源見 W8 講義。
Week 9：Agents Post-Deployment (SRE / Observability)
| # | 標題 | 來源類型 | 狀態 |
|---|---|---|---|
| 40 | Introduction to Site Reliability Engineering | Blog (Google SRE Book) | ✅ |
| 41 | Observability Basics You Should Know | Blog (last9.io) | ✅ |
| 42 | Kubernetes Troubleshooting with AI | Blog (Resolve) | ✅ |
| 43 | Your New Autonomous Teammate | Blog (Resolve) | ✅ |
| 44 | Role of Multi Agent Systems in Making Software Engineers AI-native | Blog (Resolve) | ✅ |
| 45 | Top 5 Benefits of Agentic AI in On-call Engineering | Blog (Resolve) | ✅ |
Week 10：What’s Next for AI Software Engineering
本週原 syllabus 沒列 reading list。延伸推薦資源見 W10 講義。
高優先閱讀清單（給時間有限的讀者）
如果只有時間讀 5-10 篇，建議這份精選：
- How OpenAI Uses Codex — 最 actionable 的 production case study
- MCP Introduction — vibe coder 必須建立的 MCP mental model
- MCP Food-for-Thought — 寫 MCP server 的設計品味
- Specs Are the New Source Code — 理解 PRD 在 agent 時代的角色
- Writing Effective Tools for Agents — Anthropic 官方的 tool design 心法
- How Anthropic Uses Claude Code — Anthropic 內部使用真實樣貌
- Claude Code Best Practices — Anthropic 官方 best practice
- How Long Contexts Fail — 4 種 context failure mode + fix
- Lessons from millions of AI code reviews — Graphite Diamond 的 52% action rate 定義 AI reviewer 達成 human parity
- Introduction to Site Reliability Engineering — Google SRE Book 經典開頭
延伸主題群
把跨週相關 reading 整在一起：
- MCP 完整系列（W2 全 6 篇）— 從 protocol 介紹 → server 實作 → auth → registry → 設計反思
- Claude Code 系列（W4 全 6 篇 + W2 #6 MCP servers）— Claude Code 的內部實作、用法、生態系
- AI security 系列（W6 全 7 篇）— 從 SAST/DAST 基礎 → prompt injection 攻擊 → 真實 CVE 案例 → 防禦
- Agent observability 系列（W9 全 6 篇）— 從 SRE 入門 → observability 三柱 → Resolve.ai 的 AI-native incident response
- Code review 系列（W7 全 6 篇）— 從人類 review 哲學 → AI reviewer 實證 → Graphite Diamond