Learn › After you ship

A security checklist for vibe-coded apps

An app built with AI has the same security problems as any new app, and they cluster in a few places: secrets left in the code, pages that forget to check who is asking, packages with known flaws, and input that is trusted without checking. This checklist covers nine of them, most serious first. You can work through it without reading much code.

The checklist

1. No secrets in the code

A secret is a password or key that lets your app talk to another service: your database, your payment provider, your email sender. It belongs in your host’s settings, never in a file in the repository. Search your code for the names of the services you use and look for long random strings next to them.

If you find one, removing it is not enough. Anyone who saw the repository may have it. Create a new key in that service and delete the old one.

2. Every private page checks who is asking

The most common serious flaw in a new app is a page or an address that returns data without checking the visitor is signed in, or that checks they are signed in and not that the data is theirs. Sign in as one test user and try to open another test user’s order, profile or file by changing the number in the address.

3. The database has its own rules

If your app talks to its database directly from the browser, the database needs rules about who can read which rows. Many hosted databases call this row-level security, and it is often switched off by default. Check that it is on for every table that holds customer data.

4. No package with a known vulnerability

Your app is built on packages written by other people. Flaws in them are published in public lists. A scanner compares your packages against those lists and tells you which to upgrade.

5. Packages are reasonably current

A package several major versions behind is harder to patch when a flaw is found. You do not need the newest of everything. You do need not to be years behind.

6. Input is checked on the server

Anything a visitor types can be hostile. The check has to happen on your server, because a check in the browser can be skipped. Look for forms that save straight to the database.

7. Errors do not leak details

A crash page that shows code, file paths or a database error tells an attacker how your app is built. Visitors should see a plain message, and you should get the detail privately.

8. Sign-in is not home-made

Use your framework’s or your provider’s sign-in. Passwords stored by hand, tokens that never expire and password-reset links that can be guessed are where home-made sign-in goes wrong.

9. Somebody finds out when something breaks

A security problem you never hear about stays open. You need to know when your app crashes, when a deploy fails and when a new flaw is published in something you use.

Which of these can be automated

CheckCan a tool do it?
Secrets in the codeYes. Bugstack scans for them every night
Packages with known vulnerabilitiesYes. Bugstack scans every four hours, and opens the upgrade as a pull request for npm packages
Out-of-date packagesYes. Bugstack scans every day
Missing sign-in and ownership checksPartly. A security audit looks for them and tries to prove each one
Database rulesCheck these yourself in your database’s settings
Finding out when something breaksYes. This is what Bugstack is for

Bugstack does autonomous software maintenance: it watches your app and your repository and opens fixes as pull requests. Two limits to know: when it finds a secret it tells you and gives you a checklist, and it never rotates the secret itself. And its automatic upgrade pull requests are for npm packages today; Python, Ruby and Go packages are scanned and reported.

Start with an audit

A Bugstack security audit reads your repository, plans which areas to check, and tries to prove each problem before it reports it, so the list you get is short and real. Fixes for what it finds start on their own and arrive as pull requests. A workspace’s first audit is on us, up to $50 of credit.

Questions

What is the most common security mistake in AI-built apps?

Two are worth checking first because they are serious and easy to miss: a secret committed to the repository, and a page that returns one user’s data to another.

I found a key in my code. What now?

Treat it as stolen. Create a new key in the service it belongs to, put the new one in your host’s settings, delete the old key in the service, then remove it from the code.

Does passing this checklist mean my app is secure?

It means the most common problems are closed. Security is ongoing: new flaws are published every week, which is why the scanning has to keep running.

See it work on your own code.

$100 a month

Includes $100 of usage credit every month. Start with a 14-day trial: $50 of credit and a first security audit on us. A card is required.