Three tracks

Shape how
the model
behaves

Claude Code gives you a dozen levers on model behavior. This course takes them one at a time. Each module ends with something running in your own project that day.

Start at module 00 See every module

A row of machined levers on a shared rail, each pressing at a different angle against one continuous sheet of paper.

On-ramp Use it, and meet the problems.

  1. 00 Install it and ask it something Start Claude Code inside a project you already have. 15 min
  2. 01 Let it change something Hand Claude Code a small task. 20 min
  3. 02 Why it forgot what you told it You can explain why an instruction you write gets weighed instead of obeyed. 15 min

The levers Each one works on its own.

  1. 03 Rules You can predict which instruction files load. 20 min
  2. 04 Your first command Ship a working slash command from a single file. 25 min
  3. 05 Skills Promote your command into a skill directory. 35 min
  4. 06 Settings and themes Write a custom theme, and predict which settings file wins when two of them disagree. 30 min
  5. 07 Shaping the model Say where any instruction you write ends up. 30 min
  6. 08 Subagents Write a subagent and state exactly what it inherits from the session that spawned it. 40 min
  7. 09 Hooks Write a hook that actually blocks. 50 min
  8. 10 MCP Add an MCP server to a project. 35 min
  9. 11 Permissions and the sandbox State why a hook is not a security boundary. 40 min

Packaging Only if you hand your work on.

  1. 11a Plan Mode Enter plan mode three ways. 30 min
  2. 11b Loops and Goals Pick the driver that keeps a session working (a condition. 45 min
  3. 12 Build the plugin Assemble the eight directories you already wrote into one installable plugin. 45 min

Build with Claude The Agent SDK: put the same loop inside your own app.

  1. 13 Your First Program You can install Node, run a file. 50 min
  2. 14 The Agent SDK You can run the Claude Code loop from your own JavaScript. 50 min
  3. 15 Build a Local Web App You can stand up a browser front end that talks to the Agent SDK through your own server. 60 min
  4. 16 Streaming and Activity You can stream an agent's answer into a browser as it is written. 50 min
  5. 17 Human-in-the-Loop You can route tool approvals. 55 min
  6. 18 Sessions, Checkpoints, and Durable State You can resume or fork a past session from the UI. 55 min
  7. 19 Give Your Agent Tools You can hand the agent a function from your own app as an in-process MCP tool. 50 min
  8. 20 Bring .claude Into Your App You can decide, per application. 40 min
  9. 21 Structured Application Output You can get a schema-validated object out of a multi-turn agent run. 50 min
  10. 22 Permissions Are Not a Sandbox You can lay out the seven layers between an agent. 55 min
  11. 23 The Workbench You can walk the workbench's accumulated tags end to end. 40 min
  12. 24 Ship It: Hosting, Cost, Observability You can size a host from the subprocess model. 60 min
  13. 25 Subagents You can justify each subagent you add with one of the four documented benefits. 45 min

Build the Chat Plugin One reader, no coding background, ends with a multi-agent chat app that is also a Claude Code plugin.

  1. 26 Start Here: Run the App You Will Build You can open a terminal. 50 min
  2. 27 What an Agent Is You can say what a model. 45 min
  3. 28 What a Plugin Is You can open a plugin folder. 45 min
  4. 29 One Agent That Stays Alive You can keep one agent's session alive across many messages. 55 min
  5. 30 The App's Claude Is Not Your Claude You can run the app on a machine full of personal hooks. 50 min
  6. 31 A Roster of Agents You can add an agent from the sidebar. 55 min
  7. 32 Agents That Message Each Other, in Turn You can let one agent send a message to another while the other is busy. 55 min
  8. 33 Caps That Only a Person Resets You can let two agents talk. 50 min
  9. 34 The Wiki, by Hand You can build a knowledge base that an agent maintains. 90 min
  10. 35 Skills That Keep the Wiki Honest You can give the agents a capture skill that routes a new fact to the right page. 50 min
  11. 36 Each Agent Keeps Its Own Notes You can hand two agents the same wiki. 55 min
  12. 37 Keep the Main Window Small You can add a reader agent that reads the whole record. 45 min
  13. 38 Hooks That Keep Agents Inside the Lines You can write a hook that puts an unanswered question in front of the agent at the exact prompt that touches it. 45 min
  14. 39 Approval as Data, Reach as Its Own Axis You can state the approval policy as one list. 45 min
  15. 40 Ship the Plugin You can run `claude --plugin-dir .` in a plain terminal. 45 min
  16. 41 The Whole App, Top to Bottom You can stand the whole application up with one command. 40 min
Rule one

Smallest artifact first. A rule is one line of markdown. A theme is {}. Each module starts at that minimum, runs it, then adds one field at a time.

Rule two

One new idea per module. Themes come before hooks because a wrong colour is a gentler place to learn precedence than a blocked tool call.

Two modules ship nothing

Rules and permissions ship nothing you can hand to a teammate. Capabilities are distributable. Guidance and policy are not.