Cursor and Claude Code app security audit
Cursor and Claude Code write whatever stack you ask for, so there is no platform scan watching your data rules. This check signs in as two of your users on the live app and tries to read, change and delete each other's data, then shows you the request, the response and the fix. Five working days after access.
What gets checked on your app
- Your stack, whatever it is Next.js with Supabase, Postgres behind an ORM, Firebase: we read the authorisation code and data rules you actually have, then test them with two real sessions.
- What the AI reviewers leave to you Claude Code's /security-review and Cursor's Bugbot review code changes. Neither signs in as two of your users and asks for the other's records. That is the test this adds.
- Secrets in history and in the bundle Keys committed in earlier history, values the agent could read in .env, and anything a NEXT_PUBLIC_ or VITE_ prefix shipped to the browser.
- Routes and server actions Endpoints that trust an id from the request, admin functions hidden only in the UI, and Stripe webhooks that skip the signature check.
What you receive
- The request sent and the response received for every check
- A short video of any request where one user gets another user's data
- Findings ranked by what a stranger could do, each with the fix
- A one-page note of what was tested and what was left out
- One recheck after you apply the priority fixes
- £449 off the fix invoice if the fix starts within 14 days
See a sample report from a real test, or watch the recording of user A reading user B's note.
Questions
Don't Bugbot or /security-review already cover this?
They review code you show them. They cannot tell from code alone whose data a request is allowed to return in your live app. The two-account test checks the running system.
Do you need my editor or AI account?
No. We need the live URL, read-only access to the repository, and your written permission. We create our own two test users.
What if you find nothing?
The report still shows each request and response proving the check ran. If a check in scope has no request and response, you do not pay.
A vibe coding security checklist for Lovable, Bolt, v0 and more
The method is free. The check is the run, the evidence and the recheck. It is not a penetration test and does not certify the app as safe to launch.
Book before you launch
At checkout, add the app URL and the date you plan to turn on payments or let real users in. After payment, send read-only repository access and the signed permission. Run by David Salami. Full scope on the 5-Day Two-Account Check.