Plainmend

For Lovable, Bolt, Cursor, Replit, and Supabase

Find the holes in your AI-built app. Fix them in plain English.

Plainmend reads the public page the way a visitor can. You get a short report and a fix prompt you can paste back into the tool that wrote the code. No security degree required.

The free scan is passive. Database reads and write probes wait until you verify ownership.

Example finding

The database key that skips every rule is in the JavaScript

A Supabase service_role key in the browser can read and write every row. The anon key is supposed to be public. This one is not.

The real report replaces this example with your site, a severity, and a prompt aimed at Cursor, Lovable, or Bolt.

Paste a URL

No account for the first scan. You confirm you are allowed to look at the site.

Read the public side

Scripts, headers, HTTPS, .env, .git, and source maps. Secrets are redacted in the report.

Prove ownership for the database

DNS, a meta tag, or a file. Only then do we ask whether the anon key can read or write tables.

What the free scan looks at

  • Stripe, OpenAI, AWS, GitHub, Slack, and SendGrid secrets in the page and its scripts.
  • A Supabase service_role key or sb_secret_ key. The anon key is noted, not treated as a leak.
  • Public /.env, /.git/config, /.git/HEAD, and JavaScript source maps.
  • CSP, HSTS, frame protection, nosniff, referrer policy, and permissions policy.
  • Whether the page actually loads over HTTPS.

What it will not do

  • No login guessing, no exploit payloads, no port scans.
  • No database request until the site is verified as yours.
  • The service_role key is never used, even if we find it. Finding it is the emergency. Using it would make things worse.
  • Row contents are not stored. A readable table is reported as a count.

Weekly scans and email alerts start at the Pro plan.

A one-page checklist, by email

Five checks worth doing before you send an AI-built app to real users. Signing up adds your email to the list.