↳ Audits for vibe-coded apps

Vibe-coded? Get your fix list.

You built it with Lovable, Bolt, Cursor or Claude, and it works on your screen. We check what AI tools skip: security, legal pages, SEO, the deploy, every button. You get one PDF with every fix, in order. Drop it into your agent and let it do the work.

001 Packages

Three depths.
One PDF.

Pick how deep we look. Every package ends the same way: a list of fixes, ordered by risk, written so your AI agent can work through it.

Tier 01

Basic

Don't get hacked

EUR 470 one-off
  • Security
  • Leaked API keys and secrets
  • Database rules: who can read what
  • Data leakage: protected data left accessible
  • Logins, admin pages and roles
  • Payments that can't be skipped
  • Rate limits and AI bill abuse
  • Vulnerable or AI-invented packages
Tier 03

Premium

Built to last

EUR 1,780 one-off
  • Security
  • Leaked API keys and secrets
  • Database rules: who can read what
  • Data leakage: protected data left accessible
  • Logins, admin pages and roles
  • Payments that can't be skipped
  • Rate limits and AI bill abuse
  • Vulnerable or AI-invented packages
  • Legal, SEO, deploy
  • Privacy policy, terms, imprint, cookies
  • GDPR and Swiss data law (nFADP)
  • SEO: Google and AI search can read it
  • Database: schema, migrations, backups
  • Server: hosting, domain, HTTPS, updates
  • Server access: who can log in, SSH keys, firewall
  • Function, code, scale
  • Every button, link and form clicked
  • Phone, tablet and desktop, real browsers
  • Duplicate and dead code
  • Structure, error handling, tests
  • Scalability: load test, costs at 10× users
  • A rules file that keeps your agent on track
Built with any of these? Lovable· Bolt· Replit· v0· Cursor· Claude Code· Windsurf· Supabase· Firebase

Hover over a check to see why it matters. Tap a check to see why it matters. Prices in euro, VAT excluded. One-off, per app.

002 What we check

Nine questions
before launch.

Basic answers the first three, Pro six, Premium all nine. Every answer is a list of fixes, never just a score.

01Basic Don't get hacked Questions 01–03 · €470

Can I hack it?

We try the doors an attacker tries first.

  • API keys and secrets out of the browser and the repo
  • Admin pages closed to everyone else
  • No user can make themselves admin
  • Logins and password resets that can't be skipped
  • Forms that don't run injected code
  • Packages with known holes, or made up by the AI

OWASP Top 10 · auth · XSS · SQLi · CVEs

Is private data safe?

Each user sees their own data, and nothing else.

  • Protected pages, APIs and files closed to logged-out visitors
  • Users can't read each other's records
  • APIs send only what the screen shows
  • Uploaded files aren't public to anyone with the link
  • Passwords and tokens never end up in logs
  • Personal data kept out of URLs and analytics
  • The database isn't open to the internet

Supabase RLS · Firebase rules · IDOR · storage buckets

Can someone abuse it?

Free access, fake sign-ups and a surprise bill.

  • Payments that can't be skipped, prices that can't be edited
  • Paid features locked on the server, not only hidden
  • Limits on logins, sign-ups and outgoing emails
  • Spend caps, so nobody can run up your AI bill
  • An AI that won't leak its prompt or other users' data
  • Bots kept out of sign-up and contact forms

Stripe webhooks · rate limits · prompt injection

02Pro Production-ready Recommended Questions 01–06 · €970

Is it legal?

The pages and consents you need before you take anyone's data or money.

  • A privacy policy that matches what the app collects
  • Terms, refund and cancellation rules
  • An imprint: who runs the site, and how to reach them
  • A cookie banner that blocks tracking until consent
  • Users can download and delete their data
  • AI features labelled as AI
  • Fonts, images and code you're allowed to use

GDPR · Swiss nFADP · ePrivacy · EU AI Act · licences

Can people find it?

Google, ChatGPT and a shared link should all see a real page.

  • Pages Google can read, not an empty app shell
  • A title and a description on every page
  • Sitemap, robots.txt and clean links
  • A proper preview when the link is shared
  • Visible in AI search: ChatGPT, Perplexity, Gemini
  • Fast on a phone

SSR · meta tags · Open Graph · llms.txt · Core Web Vitals

Is it deployed right?

The database, the server, and who can get into them.

  • Database schema: tables, relations, migrations
  • Daily backups, and a restore that has been tried
  • A server that is set up right and kept updated
  • Your own domain with HTTPS, off the builder's preview link
  • Who can log in to the server, and with which keys
  • Nothing left open: ports, admin panels, debug pages
  • Live keys in production, test keys in testing

Schema · migrations · backups · TLS · SSH · firewall · env vars

03Premium Built to last All nine · €1,780

Does every button work?

We click through everything, on a phone and on a laptop.

  • Every button, link and form, clicked
  • Sign-up, login, reset, payment and cancel, end to end
  • Phone, tablet and desktop; Safari, Chrome, Firefox
  • Loading, empty and error states, not blank screens
  • No lorem ipsum, demo data or TODOs left behind
  • Usable with a keyboard and a screen reader

End-to-end flows · responsive · cross-browser · WCAG 2.2

Is the code healthy?

So your next prompt doesn't break three other things.

  • Duplicate code merged into one place
  • Dead files, unused packages and old experiments
  • Huge files split, folders that make sense
  • Errors handled, not swallowed
  • Tests on the flows that make you money
  • A rules file, so your agent keeps it that way

Duplication · dead code · lint · tests · AGENTS.md

Will it scale?

What happens at 100, 1,000 and 10,000 users.

  • Indexes and fast database queries
  • Lists that load a page at a time, not all at once
  • A load test: when it slows down, when it breaks
  • Images, caching and a CDN
  • Slow work (AI calls, emails) moved to the background
  • The free-tier and API limits you'll hit first
  • Your monthly bill at ten times the users

Indexes · N+1 queries · load test · queues · cost per user

Hover over a check to see why it matters.Tap a check to see why it matters.

[ When a check fails ]

One missed check.
Real damage.

A leak usually has to be reported to the authorities within 72 hours, and often to every user it touched. Then come the lawyers, the bills nobody planned for, and the customers who don't come back. None of this is rare: each case below is one check from this page, skipped.

13,000 selfies and photo IDs leaked

July 2025. The Tea app kept the photos from its identity check in old storage that anyone could reach. 4chan users found it, and 72,000 images leaked in all.

CheckData leakageBasic

170 of 1,645 Lovable apps left open

2025. A researcher tested apps from Lovable's showcase. One in ten let anyone read or change their database with the app's public key, because the access rules were missing.

CheckDatabase rulesBasic

0 lines of code written by hand

March 2025. A founder launched a SaaS built with Cursor. Within days attackers got past the paywall, maxed out his API keys and tampered with his database. He shut it down.

CheckPayments, leaked keysBasic

1,206 executives' records wiped by an AI agent

July 2025. In the middle of a code freeze, Replit's AI agent ran destructive commands on a startup's live database and deleted it.

CheckDatabase backupsPro

€100 to one visitor, for a font

January 2022. A Munich court made a website pay a visitor because loading Google Fonts sent his IP address to Google without consent. A wave of demand letters to other sites followed.

CheckGDPR and Swiss data lawPro

4% of yearly turnover, or €20 million

The ceiling for a GDPR fine, whichever is higher. In Switzerland, fines of up to CHF 250,000 go to the person in charge, not the company.

Applies toEvery leak above

Sources: PetaPixel · NVD, CVE-2025-48757 · Indie Hackers · eWeek · IHK, LG München I 3 O 17493/20

003 The report

One PDF.
Your agent fixes it.

Every problem is written twice: in plain words for you, and as a prompt your AI tool can run. Drop the whole PDF into your agent, or paste one fix at a time.

Sample first page of a vibefix report, with fix 01 in full.
vibefix · Fix report VF-0142
your-app.com Pro 3 Nov 2026 p. 1 / 9
3 critical 5 high 6 medium 14 fixes
#01 Critical Is private data safe?

Any logged-in user can read every order.

The orders table has no access rules. One request returns all of them, with names and home addresses.

Where
supabase/migrations/0003_orders.sql
Risk
A data leak you may have to report under GDPR.
Paste to your agentfix 01

Enable row level security on the orders table. Add policies so a user can only select, update and delete rows where user_id = auth.uid(). Don't change the frontend.

CheckLog in as a second user and open /orders: you see only your own.

  1. #02CriticalOpenAI key in the browser bundlep. 2
  2. #03High/admin opens without a loginp. 3
  3. #04HighStripe webhook doesn't check who sent itp. 3
  4. #05MediumNo privacy policyp. 4

Every fix, in five parts

  1. 01What's wrong

    One plain sentence. No jargon you have to look up.

  2. 02Why it matters

    What could happen: a leak, a fine, a lost sale, a bill.

  3. 03Where

    The file, table or page to change.

  4. 04The prompt

    Ready to paste into your AI tool, one fix at a time, critical first.

  5. 05How to check

    One test that proves it's fixed, so you don't have to take the agent's word for it.

Works with Claude Code· Cursor· Lovable· Bolt· Replit· Windsurf
004 How it works

Send the code.
Get the list.

We read your code and use your live app. We don't change anything: you and your agent stay in charge of every edit.

  1. Send

    A read-only invite to your repo, or a ZIP from Lovable, Bolt or Replit. Plus the live link.

  2. Answer

    A short form: what the app does, who uses it and where, and whether it takes payments, personal data or uses AI.

  3. We check

    Every check in your package, on the code and on the live app.

  4. Fix

    Drop the PDF into your agent. Critical first, one prompt at a time.

You sendCode + live link
You answerA short form
Premium alsoA test login
You getOne PDF of fixes
005 Questions

Before
you ask.

Anything else: hello@mont3.ch.

01Do you fix the code for me?

No. vibefix tells you what to fix, in what order, and how to ask your AI tool to do it. If you'd rather have it done for you, Mont3 does that too: Vibe Code to Production.

02Which package do I need?

Basic if people already use the app and you want to know it's safe. Pro before you launch or start charging. Premium when the app has grown messy, or before you scale or raise money.

03What do you need from me?

Read access to the code: a GitHub or GitLab invite, or a ZIP export. The live link, if it's online. A short form about the app. For Premium, also a test account, so we can check the screens behind the login.

04Which AI tools does it work with?

All of them. Lovable, Bolt, Replit, v0, Cursor and Claude Code make the same kinds of mistakes, whether the backend is Supabase, Firebase or your own. The prompts are plain text, so any agent can run them.

05Is my code safe with you?

We only ask for read access, so we can't change anything, and you can revoke it the moment the report arrives.

[ Next step ]

Send us your app.

A link to the code, a link to the app, a few answers. You get the fix list; your agent does the rest.