For Supabase apps built with Lovable, Bolt, Replit, v0 and Cursor
RLS on isn’t RLS right.
Supabase’s Security Advisor tells you when a table has no rules. This reads the rules you did write and tells you whether they actually stop the wrong person: rules any signed-in user passes, users who can edit their own plan or credits, update checks with a hole in them. Plus the SQL to fix each one. Free. It runs in your browser, so your schema never leaves your machine.
Your rules, read back to you in plain English.
In your project folder, run this and paste the file, or paste the files in supabase/migrations:
supabase db dump -f schema.sqlTo check storage rules too, add --schema public,storage. No dump handy? The sample shows what a report looks like.
The mistakes behind most Supabase leaks.
- 2.1
Row Level Security left off
Any table in the public schema created without ENABLE ROW LEVEL SECURITY. Anyone with your anon key can read it, and usually write to it.
- 2.2
Rules that always say yes
USING (true) and WITH CHECK (true) on reads, edits and deletes, for anyone or for every signed-in user.
- 2.3
Signed in is not the same as owner
Rules like auth.role() = 'authenticated' or auth.uid() is not null, which let every account reach every row.
- 2.4
Edits that go too far
UPDATE rules that check ownership before the edit but not after it, and own-row edits on tables with columns like role, is_admin, plan or credits.
- 2.5
Views and grants that skip the rules
Views without security_invoker, materialized views in the API, and insert, update or delete granted to anon where nothing else stops it.
- 2.6
Functions and files
SECURITY DEFINER functions the API can call with no sign-in check or no fixed search_path, and storage policies that make files public.
SeverityCritical and High mean someone can likely reach data that is not theirs. Check manually means it might be fine, and only you know. A Note is worth knowing, not worth worrying about.
What a checker cannot see.
A clean report is a good sign, not proof. The only real proof is signing in as one user and failing to reach another’s data.
- 1It reads rules. It cannot run them, so it cannot tell whether auth.uid() = id compares against the right column.
- 2Policies that call your own functions, like is_team_member(team_id), are taken on trust.
- 3Settings made in the dashboard and never written to SQL are invisible to it. A full supabase db dump catches most of them.
- 4Keys in your frontend, sign-in settings and Edge Functions are out of its reach. The launch checklist covers those.
Want a human to prove it with a two-account test?
A Leak Check signs in as one user and tries to reach another’s data, files and functions, then shows you exactly what came back. 48 hours, $490, credited in full to a Sprint booked within 30 days.