🛡️ Leakproof

Your AI built the app.
Who checked the locks?

Apps built with Lovable, Bolt, Cursor, v0 and Replit ship fast, and they ship the same security holes: keys in the browser, Supabase tables anyone can read, server actions anyone can call. Leakproof finds them and sends you the pull requests that fix them, for a fixed price.

The holes AI-built apps keep shipping

CRITICALTables without RLS

A Supabase table created without Row Level Security can be read and rewritten by anyone holding the anon key, and that key is in your frontend.

CRITICALSecret keys in the bundle

NEXT_PUBLIC_OPENAI_API_KEY, VITE_STRIPE_SECRET: anything with a public prefix is shipped to every visitor. Someone else then pays your AI bill.

CRITICALAllow-all rules

using (true) policies and Firebase if true rules look like security but let everyone in.

HIGHEndpoints that trust everyone

Server actions and API routes are public URLs. If they don't check who's calling and whether the record is theirs, anyone can curl them.

HIGHUser A sees user B's data

Change an id in the URL, get someone else's invoice. Insecure direct object references are the most common logic hole in generated CRUD code.

HIGHNo rate limits

Unthrottled sign-up, password reset and AI endpoints become an open relay for spam and a surprise bill.

Free scanner: one command, nothing to install

Open source, runs locally, and your code never leaves your machine. It catches the mechanical holes above in seconds.

npx github:beingRealFrank/leakproof

Also as a GitHub Action that runs on every pull request. Source and docs on GitHub.

A scanner can't see your live database settings, your business logic, or whether user A can read user B's records. That's what the audit is for.

Fixed-price audits

Quick Audit

$149 one-time

For a pre-launch sanity check.

  • Full read of your repo by our review agents, beyond the free scanner's patterns
  • Every finding reproduced and checked by a human before you see it
  • Written report: what's wrong, how bad, how to fix it, in priority order
  • Delivered within 3 business days
Request a Quick Audit

Bigger codebase or a deadline? Ask for a quote. You pay only after we've confirmed the scope and price in writing.

How it works

  1. Tell us about the app with the request form: stack, repo size, what you want checked.
  2. We confirm scope and price in writing within one business day. Nothing is tested until you agree.
  3. You grant read-only access: invite a GitHub collaborator with read access (or send a zip), and optionally a staging URL and test accounts.
  4. You get the report and, for Audit + Fix PRs, pull requests against your repo. You merge what you want.

How the work is done, plainly

AI agents do most of the reading, testing and fix-writing; that's why it's fast and affordable. A human, Frank Gilroy, reviews every finding and every pull request before it reaches you, and anything an agent can't reproduce doesn't go in your report. We only test systems you own and authorize, only against staging or read-only, and never modify or delete your data.

FAQ

Is this a penetration test or a compliance certificate?

No. It's a focused code-and-configuration audit for the holes AI-built apps actually have. If you need a SOC 2 or a formal pentest letter, we're not that.

What stacks do you cover?

Best on JavaScript/TypeScript (Next.js, React + Vite, Remix, SvelteKit, Express) with Supabase, Firebase, Prisma or Drizzle, and on Python (FastAPI, Flask, Django). Ask if you're on something else.

What happens to my code?

It's used only for your audit and deleted when the engagement ends. Our access is read-only; we remove ourselves as a collaborator when done.

What if you find nothing?

Then you get a report that says so, listing exactly what was checked, which is useful in its own right for investors, customers and your own sleep.