Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Supabase RLS Review

    1

    This skill walks your agent through a careful review of a project you own and stops before anything touches production.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Supabase RLS Review

    Example session with this skill installed

    Before we launch, review my Supabase project todo-ai for RLS and policy mistakes. I own it. The anon key is in our React frontend. I ran your catalog query in the SQL Editor and saved the result as check_result.csv.

    • Read your context and instructions
    • Compiled the supabase rls review

    Summary

    Fictional demo project

    todo-ai, reviewed from an exported catalog query (no keys used). 3 findings: 0 Critical, 2 High, 1 Medium, 0 Low. No write tests were performed.

    SeverityObjectRuleWhat it means for you
    Highpublic.todosrls_disabledRLS is off. Anyone holding the anon key from your frontend can read and change every todo through the REST API, depending on grants.
    Highpublic.profilespolicy_true (read_every_profile)The policy condition is literally true, so every user can read every profile.
    Mediumstorage.todo-attachmentspublic_bucketAnyone with a file URL can download attachments. Fine for public images, not for private uploads.

    Draft fix (review on staging, not applied)

    BEGIN;
ALTER TABLE public.todos ENABLE ROW LEVEL SECURITY;
CREATE POLICY "todos: owner can read"   ON public.todos FOR SELECT TO authenticated USING ((select auth.uid()) = user_id);
CREATE POLICY "todos: owner can insert" ON public.todos FOR INSERT TO authenticated WITH CHECK ((select auth.uid()) = user_id);
CREATE POLICY "todos: owner can update" ON public.todos FOR UPDATE TO authenticated USING ((select auth.uid()) = user_id) WITH CHECK ((select auth.uid()) = user_id);
CREATE POLICY "todos: owner can delete" ON public.todos FOR DELETE TO authenticated USING ((select auth.uid()) = user_id);
DROP POLICY "read_every_profile" ON public.profiles;
CREATE POLICY "profiles: owner can read" ON public.profiles FOR SELECT TO authenticated USING ((select auth.uid()) = id);
ROLLBACK; -- change to COMMIT only after review and the isolation test

    Before I finalize this: is todos.user_id the ownership column, and should other signed-in users see any part of a profile (for example a display name)? If yes, we move those fields to a separate public view.

    Next steps

    1. Export current policies (select * from pg_policies where schemaname in ('public','storage')) so you can roll back.
    2. Apply the draft on a staging branch and run reference/isolation-test.sql against public.todos. Every block must print PASS.
    3. Decide whether todo-attachments should be private; if so, use the per-user folder pattern from policy-patterns.md.
    4. Re-export the catalog query and rebuild the report to confirm the findings are gone.

    This is a configuration review, not a penetration test. An empty report would not prove the app is safe.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Apps built quickly with Lovable, Bolt, v0 or by hand often ship the Supabase anon key in the frontend while Row Level Security is off, or on with a policy that is just true. This skill walks your agent through a careful review of a project you own and stops before anything touches production.

    What your agent does with it

    1. Installs the open-source supabase-rls-check CLI (MIT, pinned to v0.2.0), or registers its MCP server.
    2. Prints a read-only catalog query for you to run in the Supabase SQL Editor. It reads pg_class, pg_policies, pg_proc, pg_extension and storage.buckets, never table rows. No service_role key or database password is needed.
    3. Builds an HTML/Markdown report and a fix.sql draft wrapped in BEGIN … ROLLBACK.
    4. Explains each finding as "who can read or change what" before showing any SQL.
    5. Replaces the placeholder policy with a pattern that fits your data model: owner-only rows, team membership, public read-only catalog, per-user Storage folders, or a locked-down SECURITY DEFINER helper. It asks which column holds ownership instead of guessing.
    6. Proves the fix on staging with a two-user isolation test (plain SQL, or a pgTAP version for supabase test db in CI).
    7. Re-runs the check and lists anything left as accepted risk.

    Rules checked

    RLS disabled, true policies, anon/public policies on tables that look personal or payment-related, SECURITY DEFINER functions the anon role can execute, SECURITY DEFINER functions without a fixed search_path, views that bypass RLS, views that expose auth.users to the API, policies that trust user-editable user_metadata, auth.uid() calls re-evaluated per row (performance), public Storage buckets, extensions installed in public. With your consent and the public anon key, an optional probe checks which tables answer an anonymous read and whether a service_role JWT appears in your site's JavaScript.

    Ground rules built in

    • Only projects you own or have written permission to test.
    • Never asks for the service_role key or a database password.
    • Read-only. Fixes are drafts you apply on staging first.
    • This is a configuration review, not a penetration test. Zero findings does not prove an app is safe.

    What's inside

    SKILL.md plus reference/rules.md, policy-patterns.md, isolation-test.sql, pgtap-rls-test.sql and staging-checklist.md. The plain-SQL isolation test was run on PostgreSQL: correct owner policies pass, and a using (true) policy or a table with RLS off fails as expected.

    Read-only audit only, not a pen-test. User must manually run a SQL export query. Requires staging environment for fix verification. Cannot bypass RLS or access projects without permission.

    Open docs

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed today

    Accounts
    Supabase
    Software
    Python 3.11+pipxsupabase-rls-check (open source, MIT)

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables

    Allowed Hosts

    supabase.com

    Terminal: installs and runs the supabase-rls-check CLI. Read/Write files: reads the exported catalog CSV and writes report.html, report.md and fix.sql locally. Network: pipx install from GitHub, and the optional anonymous probe against your own project. Environment variable: SUPABASE_ANON_KEY for that optional probe only. The skill never asks for the service_role key and never writes to your database.

    Listedtoday

    What's inside

    Frequently Asked Questions

    No. The review runs on a read-only catalog query you paste into the Supabase SQL Editor and export as CSV. The optional live probe uses only the public anon key your app already ships, read from an environment variable.

    No. The skill never runs INSERT, UPDATE, DELETE or DDL against your project. It writes a fix.sql draft wrapped in BEGIN ... ROLLBACK that you review and apply to a staging copy first, then prove with a two-user isolation test.

    No. It is a configuration review of RLS, policies, functions, views and Storage buckets. It only reviews projects you own or have written permission to test, and an empty report does not prove the app is safe.

    New in Data & Databases