This is a checklist you can run yourself in an afternoon. It is written for people who built an app with an AI builder and have never done a security review. Each item says what to check, why it matters, and how you know you passed.
Your data
- 1. Row Level Security (RLS) is on for every table with user data. RLS is the database's own set of rules about who may read or change each record. The database is where your app keeps its data, and a table is one list of records in it. Why: the Supabase documentation says to enable it on every table in an exposed schema. Pass: each table shows RLS enabled.
- 2. You can explain every policy in one sentence. A policy is one of those rules. Pass: no policy allows anonymous visitors to read private data.
- 3. The two-account test passes. Pass: account B cannot open account A's records. OWASP, a nonprofit that lists common security problems, ranks broken access control first in its 2021 list. That is one person seeing another person's records.
These three matter most. If you only have an hour, spend it here. Data rules are where the largest leaks in apps like yours come from, and where an AI builder's default is most likely to be too open.
Your keys
A key is a long password your app uses to prove who it is to another service.
- 4. No secret key in browser code. The browser is the program people use to open websites, such as Chrome or Safari, and browser code is what every visitor downloads. A secret key is one that must stay hidden, because it opens everything. Pass: a search of your page source for secret and service_role finds nothing.
- 5. No secret in a VITE_ or NEXT_PUBLIC_ variable. A variable is one named setting, and a name starting with either prefix is copied into the code every visitor downloads. Both documentation sets say those prefixes publish the value.
- 6. .env files are not in your repository. A .env file is a settings file where an app keeps its keys, and a repository is the folder that holds your app's code and its history. GitHub secret scanning can alert you if a credential is committed, which means saved into that history.
- 7. Test and live keys are separate. Pass: production (the live app) uses live keys, preview (a test copy) uses test keys.
Your sign-in
- 8. Password reset and email confirmation work end to end. Why: a broken reset flow locks out real users on day one.
- 9. Admin pages need an admin, not just a signed-in user. Pass: a normal test account cannot open them.
Your pages and forms
- 10. Open every page signed out. Pass: nothing private appears.
- 11. Try odd input in each form. Pass: no crashes, no raw code showing on the page.
Your safety net
- 12. You know how to replace a key (make a new one and turn the old one off) and how to put the previous version of your app back online. Pass: you have written down the steps.
How to score yourself
Mark each item pass, fail or not sure. Fix the fails first, starting with data and keys. Treat "not sure" as a fail until you have tested it, because an unchecked rule protects nobody. Run the list again after any large change from your AI builder, since new features can quietly change data rules.
What the built-in scans add
Lovable's documentation describes a Quick scan at publish and a deeper scan you can run on request, and notes neither replaces a professional review for sensitive data. Run them. They are a good second layer after your own checklist.
Why each item earns its place
The list is short on purpose. Items one to three cover who can read your data, which is where a leak does the most harm and where an AI builder's defaults are most likely to be too open. Items four to seven cover keys, because a single exposed secret can let a stranger spend your money or read your records. Items eight and nine cover the doors people use every day, sign-in and admin. Items ten and eleven catch what you only notice by using the app like a stranger. Item twelve is the one people skip: knowing how to recover is what turns a bad day into a short one.
Things the checklist does not try to cover
It does not cover legal duties such as privacy notices and rules about children's data, which depend on where you and your users are. It does not cover performance, backups or uptime. It does not test how your app behaves under attack. Those are real topics, and a professional should look at them if your app is more than a side project. What the list does is remove the cheap, common mistakes so that a reviewer's time goes to harder questions.
A realistic time budget
Data checks: about forty minutes the first time. Keys: twenty minutes. Sign-in and pages: thirty minutes. Safety net: fifteen minutes. Roughly two hours, then ten to twenty minutes per release afterwards. Most of the time goes into learning where things are in your dashboards, which you only do once.
Make it a habit, not a hurdle
Copy the twelve items into a note and tick them off with the date. Next release, you only re-check the items the change could have touched. Over a few releases the list becomes second nature, and the habit is worth more than any single review. If a check surprises you, write down what you found and how you fixed it; the pattern tells you where your builder tends to slip.
When to bring in a second opinion
If your app handles payments, health or financial details, children's data, or anything your users would be upset to see leaked, ask for an independent review. Nordik App Audit is an AI-performed audit of your app's code. It explains what it found in plain words and gives you a fix instruction for each finding. It is not a pentest (a test that tries to break into the running app), and like any review it can miss things. It supports every platform: any app can order, whatever it is built with. The current price and terms are on the Nordik home page. The free scan accepts apps from any platform and points you at where to look first. It is a starting point, not a verdict.