Lovable
Supabase row-level security, what the public key can reach, edge function auth.
QA testing for apps built with AI
Move the lens over this app. That is what we do to yours.
Illustration: a billing dashboard that looks fine. Under an X-ray lens it shows four bugs: an API that answers without login, another company's invoices visible, a live secret key in the JavaScript, and a payment webhook that credits twice.
We test apps built with
The problem
Every fix can quietly break something else.
AI-built public apps exposing sensitive company data, out of about 380,000 scanned.
Red Access, via Axios (2026)of AI-generated code samples had a security flaw, across 100+ models.
Veracode GenAI Code Security Report (2025)exposed secrets and 2,000+ vulnerabilities in 5,600 vibe-coded apps.
Escape.tech (2025)CVEs a month traced to AI-written code, January to March 2026.
Georgia Tech SSLab, Vibe Security RadarPick one to see what your users would see.
The button is hidden, the route is not. Call the API or open /admin directly and it answers, signed in or not.
We call every route and API directly: signed out, and as the wrong user.
Your app, ready to test
Illustration: a QA pass over an app finds three bugs, in the database, webhooks and profile page, which are then fixed and covered by tests.
Press the button for the ten-second version.
How it works
Scope and fixed price, in writing.
A URL and two test accounts. No code needed.
Real browsers and phones, as a customer and as an attacker.
Every bug with steps, screenshot and a fix prompt.
Optional Playwright tests on every push.
Every tool cuts corners somewhere. This is where we look first.
Supabase row-level security, what the public key can reach, edge function auth.
Secrets in the browser, missing server-side validation, direct API calls.
Multi-file edits that quietly break other flows. We add regression tests.
Large refactors. We check every critical flow still works afterwards.
Empty, loading and error states, forms, accessibility and mobile.
Secrets handling, auth on every route, behaviour with real data.
Security rules: reads and writes across accounts and by signed-out users.
Launch Check report · acme-billing.app
2 critical2 high1 low31 passed
| ID | Severity | Finding |
|---|---|---|
| QA-0412 | Critical | Any signed-in user can read every customer's invoices |
| QA-0415 | Critical | Stripe webhook retry grants credits a second time |
| QA-0409 | High | Password reset link still works after it has been used |
| QA-0418 | High | Profile page crashes when the bio field is empty |
| QA-0421 | Low | Signup button cut off on iPhone SE |
| QA-0403 | Pass | Login, logout and session expiry on 4 browsers |
Sample report. App name and findings are illustrative.
QA-0412 in full: every finding comes like this.
GET /rest/v1/invoices · Chrome 131, Safari iOS 18
supabase.from("invoices").select("*")Only user B's invoices.
Invoices from every account.
Enable row-level security on the "invoices" table in Supabase. Add a SELECT policy so users can only read rows where user_id = auth.uid(). Add matching INSERT, UPDATE and DELETE policies. Do not change any other table. Show me the SQL you ran.
Sample from a Launch Check report. App and account names replaced.
Pricing
Not sure which one you need? Start with a Launch Check. If we find nothing serious, you get that in writing too.
$490fixed
Before you launch, demo to investors or open signups.
Takes 3-5 days.
Book a Launch Checkfrom$1,500
When every new prompt might break something that worked yesterday.
Takes 1-3 weeks.
Ask about Safety Netfrom$990/ month
For teams shipping weekly without a QA engineer.
Billed monthly.
Ask about QA Partneryears of software testing, for companies in
Testers, not developers marking their own work.
Paste-ready for Cursor, Lovable or Claude.
A URL and two test accounts to start.
Scope and cost in writing first.
More testers when the release is bigger.
Test data deleted when we finish.
Something else on your mind? Write to hello@qabasic.com.
Not for a Launch Check. We test the running app from the outside, the way users and attackers see it. A URL and test accounts are enough. For a Safety Net we work in your repository so the tests run in your CI.
No, that is exactly who we work with. The tool does not matter much, because we test behaviour, not style. We do know where each stack tends to cut corners, such as Supabase row-level security or unprotected API routes.
Every bug comes with reproduction steps and a fix prompt you can paste into your AI tool. If you would rather we fix them, we can scope that separately.
AI-written tests usually confirm the code it just wrote, with the same blind spots. We test from the outside with the intent to break things, and we check that every test fails when it should.
Three to five working days from the day we get access. You get the report and a walkthrough call at the end.
You do. Safety Net tests live in your own repository and run in your own CI. If you stop working with us, they keep running.
A Launch Check is $490, fixed. A Safety Net starts at $1,500 and QA Partner at $990 a month, depending on how many flows your app has. Either way you get a fixed price in writing before we start, within one business day of a short call.
Yes. We are happy to sign yours before we see anything, and test data is deleted when the engagement ends.
Prefer email? hello@qabasic.com