25 security checks for an app you built with AI
The full checklist from the reel, as one prompt. It makes your AI coding tool check your app item by item, show you where each one fails, and fix it.
How to use it
- Open your project in Claude Code (or any AI coding tool that can read your files).
- Copy the prompt and paste it in.
- Read what it reports before you let it fix anything. Ask it to explain any item you don’t understand.
The prompt
Check this app against the security list below. For each item, say pass or fail, and show me the file and line when it fails. Then fix what failed, one item at a time, and tell me what you changed. Don't mark anything as done without showing me the evidence. If an item doesn't apply to this app, say why. 1. Keep secret keys on the server. Anything in browser or app code can be read by a visitor. Keys that spend money or read data belong in server-side environment variables. 2. Keep secrets out of shared code. Add .env to .gitignore before the first commit. If a key was ever pushed, revoke it and make a new one. Deleting the file later does not un-leak it. 3. Lock down the database. Turn on per-row access rules (row level security in Supabase, security rules in Firebase) so the public key cannot read every table. 4. Use secure connections. Serve everything over HTTPS and redirect plain HTTP to it. 5. Block weak passwords. Check new passwords against a list of common ones and ask for a decent length. 6. Never store plain passwords. Store a slow, salted hash (Argon2 or bcrypt), or let an auth provider handle passwords. 7. Slow repeated login attempts. After a few wrong tries, add a growing delay or a short pause. Do not lock an account forever, or anyone can lock your users out. 8. Add two-step login. At the very least for admin accounts. 9. Expire password reset links. They should work once and stop working after a short time. 10. Make logout actually log out. End the session on the server, not just in the browser. 11. Delete test accounts. admin / admin123, test@test.com, demo users and sample data should not reach launch. 12. Lock the admin page. Check the admin role on the server for every admin page and action. Hiding the link is not a lock. 13. Show users only their own data. On every request, check that the record belongs to the signed-in user. Changing a number in the address bar should never open someone else’s order. 14. Check every input on the server. Checks in the browser can be skipped. Check type, length and range again on the server. 15. Stop comments running code. Escape anything users write before showing it, so a script tag shows up as text. 16. Limit file uploads. Allow only the file types you need, check the real contents, cap the size, and keep uploads private by default. 17. Set prices on the server. Never trust a total the browser sends. Work it out from your own price list. 18. Confirm payment, then unlock. Check with the payment provider (a signed webhook) before giving access. 19. Add rate limits. Cap requests per user on sign-up, login and anything that costs you money to run. 20. Cap my AI spending. Set a spending limit on the AI provider account and a usage limit per user. 21. Hide error details. Show visitors a plain message. Send the technical details to a private log. 22. Install security updates. Run the package manager’s audit and update anything with a known hole. 23. Alert me to strange logins. Log sign-ins and tell me about new devices and bursts of failed attempts. 24. Back up the database. Automatic, daily, stored somewhere else. 25. Test the backups. Restore one into a spare database to prove it works.
The 25 checks, and why each one matters
Secrets
- 1
Keep secret keys on the server. Anything in browser or app code can be read by a visitor. Keys that spend money or read data belong in server-side environment variables.
- 2
Keep secrets out of shared code. Add .env to .gitignore before the first commit. If a key was ever pushed, revoke it and make a new one. Deleting the file later does not un-leak it.
Data
- 3
Lock down the database. Turn on per-row access rules (row level security in Supabase, security rules in Firebase) so the public key cannot read every table.
- 4
Use secure connections. Serve everything over HTTPS and redirect plain HTTP to it.
Accounts
- 5
Block weak passwords. Check new passwords against a list of common ones and ask for a decent length.
- 6
Never store plain passwords. Store a slow, salted hash (Argon2 or bcrypt), or let an auth provider handle passwords.
- 7
Slow repeated login attempts. After a few wrong tries, add a growing delay or a short pause. Do not lock an account forever, or anyone can lock your users out.
- 8
Add two-step login. At the very least for admin accounts.
- 9
Expire password reset links. They should work once and stop working after a short time.
- 10
Make logout actually log out. End the session on the server, not just in the browser.
- 11
Delete test accounts. admin / admin123, test@test.com, demo users and sample data should not reach launch.
Who can see what
- 12
Lock the admin page. Check the admin role on the server for every admin page and action. Hiding the link is not a lock.
- 13
Show users only their own data. On every request, check that the record belongs to the signed-in user. Changing a number in the address bar should never open someone else’s order.
What people send you
- 14
Check every input on the server. Checks in the browser can be skipped. Check type, length and range again on the server.
- 15
Stop comments running code. Escape anything users write before showing it, so a script tag shows up as text.
- 16
Limit file uploads. Allow only the file types you need, check the real contents, cap the size, and keep uploads private by default.
Money
- 17
Set prices on the server. Never trust a total the browser sends. Work it out from your own price list.
- 18
Confirm payment, then unlock. Check with the payment provider (a signed webhook) before giving access.
- 19
Add rate limits. Cap requests per user on sign-up, login and anything that costs you money to run.
- 20
Cap my AI spending. Set a spending limit on the AI provider account and a usage limit per user.
When things go wrong
- 21
Hide error details. Show visitors a plain message. Send the technical details to a private log.
- 22
Install security updates. Run the package manager’s audit and update anything with a known hole.
- 23
Alert me to strange logins. Log sign-ins and tell me about new devices and bursts of failed attempts.
- 24
Back up the database. Automatic, daily, stored somewhere else.
- 25
Test the backups. Restore one into a spare database to prove it works.
And most importantly: make no mistakes. That one is a joke. It is why the other 25 exist.
This is a starting point, not a security audit
The checks come from standard guidance, mostly the OWASP Cheat Sheet Series. They catch common gaps; they do not prove an app is safe. If your app takes payments or holds sensitive data, have someone who does this for a living review it.