Introduction
Helps you and your AI assistant write browser tests that match how your team already works.
Cue helps you write browser tests — and helps an AI assistant write them the way your team already writes them.
If you have ever asked an AI to write a test and got back something that looked right but used a button that doesn’t exist, or ignored every convention in your project, this is built for that problem.
The two things it does
- Checks your tests automatically — A set of rules about what a good test looks like here. They run every time you save and every time you push, so a test that breaks them fails the build — not the code review.
- Learns what your app looks like — It visits your app, writes down what’s actually on each page, and remembers. So nobody has to guess what a button is called.
Why guessing is the expensive part
An AI has never seen your app. Ask it for the “log in” button and it will invent
something plausible — #login-btn, maybe, or .submit.
Plausible is the problem. The test doesn’t fail with “I made this up.” It fails like your app is broken. So somebody spends an afternoon debugging a login page that works perfectly.
Cue fixes this by writing down what’s really there. When something isn’t written down, it says “I don’t know, go and look” instead of guessing. That refusal is the whole point.
Try it
npx @wiluszdamian/cue init
It looks at your project, shows you exactly what it plans to write, and waits for you to say yes. Nothing happens until you agree.
- Set it up — Five minutes, start to finish.
- How it all fits together — The idea behind it, in plain terms.
What it isn’t
- Not a Playwright tutorial. Playwright’s own docs are excellent and kept up to date by the people who build it. Cue points you at them.
- Not a replacement for your process. It has opinions about tests. It has none about your roadmap, your tickets, or how you plan features.
- Not magic. It won’t write good tests for a feature nobody can explain. It will stop an assistant from inventing details it doesn’t have.
Everything here
The site in site/ renders these same files, in this order; on GitHub, this
index is the way around them.
Getting started
- Set it up — five minutes, start to finish.
- Install for your assistant — Claude Code, Cursor, Codex and the rest.
- Teach it about your app — so nobody has to guess a selector.
- Write your first test
Guides
- How it all fits together
- Writing tests
- Finding elements
- When advice conflicts — which source wins, and why.
- Keeping it current
- Working as a team
- Does it actually work?
Reference
- All the rules — generated from
rules/constitution.yaml. - Who decides what — generated from
rules/ownership.yaml.
Help