Beginner — Getting Started
This guide is written for people who are not full-time developers but have at least dipped into a terminal and written code alongside an AI before. The goal is to get you all the way to a working extension. If you get stuck along the way, keep 06-troubleshooting.md open in another tab and check the matching section.
Who this guide was written for
Section titled “Who this guide was written for”A good number of people who pick up this base are not full-time developers, and that shaped the tone of the guide. There is one prerequisite to flag, though.
You have opened a terminal at least once and have written code with an AI before. Commands get written out one line at a time, but this guide does not teach what a terminal is, what git is, or why Node is needed. Those are not the places this manual covers — keep a free primer open in a side tab (a YouTube intro to git and the terminal, MDN’s Node.js docs, Anthropic’s Claude Code quickstart) and lean on it as you go.
A few assumptions follow from that.
- You recognize terminal, git, and Node by name and have a rough sense of what each does. Commands are still written out one line at a time on top of that.
- The first run does not need an external service like Supabase or your own server. The default backend mode is
console-log, which means the extension prints to the browser console and nothing else has to be wired up. - The code itself gets written by an AI (Claude Code in this guide). You only need to know what kind of extension you want to build.
If a section already feels obvious, skim it and move on. It still helps to glance at each step once so you know where to come back when something does not behave the way you expect.
If unfamiliar words start piling up step after step, the parts this manual is actually about — the safety net for places where AI slips, the recovery flow, the scenario-driven checks — get harder to reach. Those are the heart of this base. If the entry steps keep blocking you, it is faster to pause, pick up the primer for a session, and come back.
| Step | What you do | Rough time |
|---|---|---|
| 1 01-prerequisites.md | Install Node, git, Chrome, and Claude Code | 20–40 min if it is your first time |
| 2 02-clone-and-install.md | Get the base and run pnpm install | 5–10 min |
| 3 03-first-run.md | Type “start extension development” and watch the first run in console-log mode | 15–30 min |
| 4 04-real-backend.md | (Optional) Wire up Supabase or your own server | 30 min – 1 hr |
| 5 05-customize-design.md | (Optional) Repaint the extension in your design tone — Figma URL auto-sync or a one-word tone | 15–30 min |
| 6 06-troubleshooting.md | A list of the most common stuck points | Only when you need it |
| 7 07-automated-safety-net.md | Understand the safety net the base runs for you | 10 min |
| 8 08-when-ai-gets-stuck.md | Unblock the workflow itself when the AI loops or you lose track | Only when you need it |
| 9 09-distribution.md | (Optional) Get your extension into your users’ hands | 30 min – 2 hr |
| 10 10-staying-up-to-date.md | (Optional) Pull base updates into your fork | 10–30 min per cycle |
If Node, git, and the MV3 manifest already feel familiar, the Expert manual covers the same base in a shorter pace.
When you need help
Section titled “When you need help”The first place to look is 06-troubleshooting.md. Most of the things that go wrong on a first run are covered in there.
If that does not match your symptom, copy the error message verbatim into the Claude Code window and ask “how do I fix this?”. It will usually give you a sensible next step to try.
If you are still stuck after that, open a GitHub issue or post in your team’s Slack with three pieces of information bundled together: where you got stuck (which file, which command), the full error message, and your environment (operating system, Node version, Chrome version).
The first time through is the slowest part. Once your environment is set up, the next extension you build skips step 1 entirely, and the whole loop feels noticeably shorter.
Where the manual ends and the base begins
Section titled “Where the manual ends and the base begins”The steps above cover what the base handles for you — the safety net, context recovery, scenario-driven checks, the licensee update channel. The manual itself stays free; keep it open in a side tab as you go.
The base code and the .claude/ bundle (agents, skills, hooks) ship soon as a paid license, Solo single-seat plus a year of the update channel. Pricing and the registration channel will be posted on this page and on the landing page once they land. The moment you read through the manual and start wondering “where does the actual code that runs all this automation live?”, that is the point where you would pick up the license and fork the base.