Chapter 1
What You'll Build (and Why It's Worth 90 Minutes)
A real deployed app, a real workflow, and the muscle memory to keep using both.
This is a follow-along guide. Open Cursor in one window, this document in another, and build with us. By the end you'll have a live, deployed app at your-app.vercel.app with:
- Magic-link sign-in (no passwords)
- Per-user todos: each user only sees their own
- Create, edit, complete, and delete
- Status filter: all / active / completed
- Mobile-friendly UI
The app is small on purpose. The point is the workflow, not the feature set. Once you've shipped this one, the same workflow scales to whatever you build next.
Time budget
| How you read it | Time |
|---|---|
| Read every prompt, run every verification step, push back on the AI when it cuts corners | ~90 minutes |
| Copy-paste, trust the agent, fix what breaks at the end | ~30 minutes (and you learn less) |
What you'll actually learn
The app is the artifact. The skills are the takeaway:
- How to switch between Opus and Sonnet for different tasks (and why that saves money)
- How to use Plan mode + the Build button so Opus plans and Sonnet executes in one click
- How to draft a
CLAUDE.mdthat the AI actually respects - How to use targeted
.cursor/rulesfor file-scoped guidance - How to install and use community skills from skills.sh
- How to plan a project at the right level: an overview
PLAN.mdsplit into phases, each with a ready-to-paste plan prompt that kicks off Plan mode for that phase - Why you open a fresh chat for each phase (and what happens when you don't)
- How to debug fast with screenshots dropped into the chat
- How to recover when the agent goes off the rails (it will, and that's fine)
If you only have 30 minutes, do sections 1 through 7 today and come back for the build steps tomorrow. The setup work compounds. The build steps are easier once your context layers are in place.



