Check your own app — free, in about a minute.
The free mini-scan checks your live homepage for TLS/HTTPS basics, response security headers, cookie flags, exposed sensitive files, Firebase/Supabase backend rules, and the WordPress plugin versions your site exposes. You get a score and the finding titles — no account, no card.
Run the free mini-scan →Supabase gives AI-built apps a real Postgres database, an API generated from that database, and a key you paste into the browser. That design is safe exactly as long as one setting holds: Row Level Security. With RLS off — or on but without the right policies — the anon key in your client code is a skeleton key, because PostgREST will happily serve your tables to anyone who asks.
Agents building on Supabase optimize for "the query works." RLS is where "works" and "works only for the right person" diverge. Walk this list before you launch.
☐ 1. RLS is enabled on every table in an exposed schema.
Not the tables you remember — every table. In the Supabase dashboard (Table Editor → table → RLS) or via pg_tables / the rls_enabled flag, confirm each table in public has RLS on. Tables created quickly during an AI build session are the ones most likely to have it off. A table with RLS disabled is readable and writable by the anon role through the API.
☐ 2. Every operation you use has a policy — select, insert, update, delete. RLS with no policy for an operation denies that operation, which is why half-finished policy sets "work" in testing: reads succeed (a policy exists), and nobody noticed updates are governed by a different policy — or by none you intended. List the policies per table and read them as an attacker: which rows can a stranger see, change, or delete?
☐ 3. No policy says using (true) to the world.
A permissive policy is the Supabase version of an open Firebase rule. The pattern that ships is create policy "public read" on t for select using (true) — added to make a feed load, never revisited. If a table holds anything user-specific, the policy should compare the row's owner column to auth.uid().
☐ 4. Ownership is enforced by auth.uid(), not by a client-supplied value.
Policies that trust a user_id sent in the request body can be satisfied by anyone who sets that field to someone else's ID. The check belongs in the policy, against the caller's JWT identity.
☐ 5. The service_role key is nowhere near the client.
The service role bypasses RLS entirely — that's its purpose, server-side. If it appears in your browser bundle, a shipped .env, or a mobile app, RLS no longer matters: whoever extracts it owns the database. It belongs in server-only environment variables. (Our scanner looks for exactly this class of leak in served JS — Stripe live keys, AWS keys, private keys, and tokens; see the link below.)
☐ 6. Storage buckets match the tables' posture. Public buckets serve files to anyone with the URL. If users upload anything private — documents, images tied to accounts — the bucket should be private with storage policies scoped like your table policies. The bucket tab is a different screen from the table tab; that's why it gets missed.
☐ 7. You tested as a stranger, not as yourself. Log out. Take the anon key from your own site's network traffic and query your API the way an outsider would: list the tables PostgREST exposes, try reading the ones that matter. If rows come back, that's your answer — no console setting outranks an observed response.
Your Supabase project URL and anon key are in your client code by design, so an outside-in check can do precisely what step 7 does: PreShipIt's scan reads the PostgREST root metadata, and if tables are listed, runs one count-only check — zero rows downloaded. Tables readable by the anon key are a High finding; a metadata-only response is a Low. Nothing about your data leaves the database in the process.
Related: Bolt Exposed API Keys — how keys end up in shipped code, and which ones are actually secret. Exposed .env File Check — the five-minute version of step 5 for any stack. Is Your Vibe-Coded App Safe to Ship? — where database rules fit in the full list of holes AI builders ship.
Not sure whether your tables answer to strangers? Run the free mini-scan at preshipit.com/free-scan for the outside-in basics — the full scan runs the Supabase and Firebase exposure probes against your own project, counts only.
The PreShipIt Brief — a short weekly email for people shipping vibe-coded apps: the week's important vulnerabilities and launch-safety checks, in plain English.
New and free. Double opt-in: we email you a confirmation link before anything else arrives. Unsubscribe any time.
Check your own app — free, in about a minute.
The free mini-scan checks your live homepage for TLS/HTTPS basics, response security headers, cookie flags, exposed sensitive files, Firebase/Supabase backend rules, and the WordPress plugin versions your site exposes. You get a score and the finding titles — no account, no card.
Run the free mini-scan →← Back home · Run a free mini-scan · Scan my app — $29 · See a sample report