#054Technical4 min
Check a Supabase project’s security
Choose code-only or read-only checks and examine the rules that protect your database.
Hackaiz team · updated 21/11/2026

01Choose what you want to examine
Supabase security is not determined by whether the application’s screens look private. Access rules decide what a browser or signed-in user may read or change. Hackaiz provides two evidence levels in “Your Supabase database (read-only)”: “Level 1 · Code only” examines SQL migration files in a connected GitHub repository; “Level 2 · Read-only” examines the authorized project’s structure and rules.
A migration describes intended database changes, not necessarily the current production state. Conversely, a read-only database check does not replace review of application code, server-side authorization or exposed secrets. Choose the level that answers your question and retain that scope when interpreting results. If you also analyze a website or API, prove domain control first through the DNS TXT record at _hackaiz, the HTTPS verification file at /.well-known/hackaiz-verification.txt, or your project’s AI creating that file.
03Check rules, grants and storage
Use “Check my database” for the connected project. The read-only audit examines structure, access rules, rights and file buckets rather than reading table contents. Look for row-level security, often shortened to RLS: this is the mechanism that restricts which rows a role may access. A disabled rule boundary, a policy that always allows access and a protected table with no policy are different situations and need different interpretations.
For example, a customer-records table should not expose one tenant’s rows to another. A rule that allows everything can undermine that separation; a table with no policy may instead refuse expected operations. Also review anonymous write grants, elevated-privilege functions callable without sign-in and public file buckets. A public bucket can be intentional for published images but inappropriate for private documents. Describe intended use before deciding what to change.
“Access table” shows a matrix by table and role with SELECT, INSERT, UPDATE and DELETE columns: reading, adding, changing and deleting rows. Use this view to investigate unexpected permissions, not as proof that every business scenario has been tested. The finding’s evidence and your application’s intended access model still matter.
04Treat remediation as a separate decision
For frontend code, an exposed service_role credential is a separate concern from RLS settings: a privileged server credential must not be shipped to the browser. Repository analysis can provide code evidence that the read-only database inspection does not. Ask the responsible developer to review any exposed secret and the rules protecting the affected data; do not infer that changing a single policy handles every related risk.
Checks and follow-up actions use credits. The audit connection does not itself authorize writing changes. Where a GitHub correction is proposed, it is a draft pull request requiring human review and approval, never an automatic merge. Verify the intended allowed and denied operations after the reviewed change. “Turn on monitoring” is optional and available for the connected system; do not assume it was enabled when you connected. Use “Remove access” when the connection is no longer needed.
Frequently asked questions
Take the next step
Was this article helpful?