BlogFlowQA
How to test a vibe-coded app before launch
How to test a vibe-coded app in an afternoon, before real users arrive. Name the journeys that must work, walk them by hand, try a second account and bad input, then turn the journeys into tests that run on every change.
This guide shows how to test a vibe-coded app before you put it in front of users, whether you built it with Lovable, Bolt, Replit, v0, Cursor or Claude Code. At the end you will have a written list of three to five journeys that must work, a record of walking each one by hand, a short list of bugs found with a second account and bad input, and tests that walk the same journeys on every change. The first four steps need a browser, two email addresses and an afternoon, and no tool.
How to test a vibe-coded app: check what you have not seen#
You have seen the screens you asked for, usually while signed in as yourself, with data you entered. You have probably not seen what a stranger sees: an empty account, a wrong password, a second customer's data one address away. The steps below look in those places.
1. Name the three to five journeys that must work#
A journey starts where a real person starts and ends at something they came for. Write each one as a sentence with an outcome you can see on the page.
- "A new person signs up, confirms their email and lands on an empty dashboard."
- "A signed-in person creates an invoice and sees it in the list."
- "A person on the free plan pays and sees the paid plan on their account page."
Stop at five. If a journey broke on launch day and you would not notice for a week, it is not on this list.
2. Walk each journey by hand, once, as a stranger#
Open a private window, so nothing from your own sign-in carries over. Use an email address the app has never seen. Use the deployed address you will launch on, not the builder's preview.
Walk every journey from the first page to the outcome and write down each place you hesitated. Open the confirmation email on a phone and tap the link there. Sign out, sign in again and check that what you made is still there. Ask for a password reset and finish it.
3. Test with two accounts#
This step looks for one customer reading or changing another customer's data. The pages can look right while the rules behind them let anyone through.
Create account A and account B in two different browsers. Then try each of these as B.
- Copy the address of a page that shows A's record and open it.
- Change the number or ID in an address of your own record to a neighbouring one.
- Search, filter and export, and look for anything of A's in the results.
- Open your browser's developer tools on the Network tab, load a list page and read what came back, looking for rows that are not B's.
- If the app has roles, open an admin page's address as an ordinary member.
Every one of these should end in "not found" or "you do not have access". Anything else is a bug to fix before launch.
4. Try bad input and empty states#
Check the cases you did not describe when you asked for the app.
| Try this | Where | A good result |
|---|---|---|
| Submit with every field blank | Each form | A message beside the field, nothing saved |
Paste 5,000 characters, an emoji and <b>bold< |
Text fields | Saved and shown as plain text, layout intact |
| Enter a negative number, zero and a date in the past | Amounts and dates | Refused with a reason |
| Click the submit button twice quickly | Create and pay | One record, one charge |
| Open a brand new account | Every list page | An empty state that says what to do next, not a blank page or a spinner |
| Use a declined test card | Checkout | A clear message and an unchanged plan |
| Narrow the window to phone width | Every journey | Nothing cut off, every button reachable |
Write each failure down with the steps that led to it, what you expected and what you saw. Add a screenshot and the console error when you hand the bug back to your coding agent.
5. Turn the journeys into tests that run on every change#
The next prompt you give the builder can change a page you did not mention, and you will not walk five journeys by hand after every prompt.
Turn each journey into a test that a browser replays without you. Whichever tool you use, hold each test to three rules.
- It starts on a fresh browser with a new account, not your own sign-in.
- It ends by checking an outcome on the page, such as the invoice in the list, not only that a page loaded.
- It goes red when you break the thing it protects. Hide the invoice list once and watch it fail.
You can write the tests in Playwright, ask your coding agent to write them, or record them in a tool. Run them after every deploy and once a night.
Where FlowQA fits#
FlowQA does step 5 for you. You give it your app's address, the AI agent explores the site and proposes a plan of journeys, and you cross off the ones you do not want. It then generates a test for each journey and keeps a test only once it has passed twice on fresh sessions and failed on a page broken the way its checks describe. That shows a test can fail, not that it checks the right outcome, so read each check against the sentence you wrote in step 1.
The free plan covers 1 project, up to 100 tests with 10 AI-generated a month, 1,000 cloud test runs a month, 1 schedule and a CI trigger.
FlowQA does not replace steps 2 to 4. It walks the journeys you approve; it does not look for one customer's data in another's account, and it does not judge whether an empty state reads well. Do those by hand before launch, then start free and let the tests hold what you checked.
FlowQA explores your site, writes the tests, keeps each one only once it fails on a page broken the way its checks describe, and runs them after every change.
See FlowQA