Learning paths

Start Here

Three guided reading paths through the guides and essays on this site. Pick one, read in order, and check off what you finish — progress stays in your browser.

New to Claude Code

0 of 7

From first install to a tool that fits your hand. Set up the CLI, learn the moves that matter, then decide how it stacks up against the alternative.

  1. Claude Code CLI 設定 2026：5 分鐘快速上手

    Install the CLI, authenticate, and finish your first working session in about five minutes.

  2. Claude Code Cheat Sheet

    Keep this open in a tab: every command, shortcut, and flag on a single page.

  3. Claude Code 指南

    The complete reference. Skim the basics now, come back for the deep sections as you need them.

  4. AGENTS.md 模式：哪些真正改變了代理行為

    Your instructions file is the highest-leverage config you have. These patterns actually change behavior.

  5. Claude Code Hooks 教學：從零打造 5 個生產級 Hook

    Build five working hooks from scratch. This is the step from using the tool to shaping it.

  6. 為 Claude Code 打造自訂技能：完整教學指南

    Package your own workflows as skills that switch on exactly when they are needed.

  7. Claude Code 對決 Codex CLI 2026：決策參考指南

    You know the tool now. See how it compares to its main rival, and when each one wins.

Agentic Engineering

0 of 7

How to build, run, and trust AI agents that do real engineering work: the harness, the economics, the failure modes, and where the discipline is heading.

  1. 證據關卡

    Start with the working discipline: never accept “it should work” from an agent. Require proof.

  2. Agent 架構

    The complete playbook for building an agent harness: skills, hooks, memory, subagents, and orchestration.

  3. 複利工程：讓程式碼庫加速而非衰退

    Why harness work pays compound interest: each feature you ship makes the next one cheaper.

  4. 迴圈工程：在驗證成本低廉之處，迴圈才能取勝

    The rule that decides what to automate: loops win only where checking the result is cheap.

  5. AI 代理在無人監督下運行時究竟會出什麼問題

    Seven ways agents fail when nobody is watching, and the signal that catches each one early.

  6. Agent 操作員手冊：監督看不見的工作

    Running agents is a role of its own. What supervision looks like when you cannot read every line.

  7. 代理取代的是審查者，而非審查本身

    Where this is heading: the reviewer role is dissolving while the review itself moves into the pipeline.

iOS with Agents

0 of 7

Ship Apple-platform apps with AI agents doing the heavy lifting, from wiring Claude Code into Xcode to the on-device models your app can ship.

  1. iOS Agent 開發

    The end-to-end playbook for building iOS apps with agents. Start here and branch out.

  2. 兩個MCP伺服器讓Claude Code變成iOS建置系統

    Wire Claude Code into Xcode so it can build, run tests, and read crash logs on its own.

  3. Apple 開發中的 Hooks：拯救專案的模式

    An agent working near your signing keys needs guardrails. Four patterns that bound the damage.

  4. Xcode 27 全面走向 Agent

    What Apple ships natively: Xcode 27's built-in agents, and where they fit next to your CLI.

  5. Foundation Models 代理工作流程：應用程式內 vs 工具型 LLM

    Two AIs touch your app: the agent writing the code and the model shipping inside it. Keep their jobs straight.

  6. Foundation Models 裝置端LLM：Tool 協定

    Put Apple's on-device model to work inside your app, starting with the Tool protocol.

  7. iOS 27 中的 Foundation Models：工具呼叫控制

    Control how the on-device model uses tools, with iOS 27's calling modes and built-in readers.