Supabase Security Audit

Audit a Supabase + Vercel project for RLS coverage, privilege escalation, cross-customer data leaks, anonymous exposure, magic-link flow correctness, and HTTP security headers — and apply hotfix templates when issues are found. Use whenever the user asks about security / RLS / audits, after any migration that touches profiles-like tables or auth, or before exposing a new customer-facing surface.

Run the audit, read the findings, apply hotfixes. The script is tuned for projects that use Supabase + Vercel and a profiles table with role in ('admin','agent','customer') (or similar), but its individual probes are independent — single-table projects benefit too.

When to Run

• User asks "are we secure?" / "security audit" / "RLS check".
• After any migration that touches profiles, RLS policies, or auth triggers.
• Before sending the first real magic link to a customer.
• After rotating credentials or changing Site URL / Redirect URLs.
• Before exposing a new surface (portal, CRM, admin tool) to real users.

Prerequisites

Credentials file at ~/.openclaw/credentials/supabase/credentials.env (override path with --cred) containing:

text

SUPABASE_URL=https://<ref>.supabase.co
SUPABASE_PROJECT_REF=<ref>
SUPABASE_ANON_KEY=...
SUPABASE_SERVICE_ROLE_KEY=...
SUPABASE_DB_PASSWORD=...

The pg Node package must be reachable. The script auto-discovers it from /tmp/sb-tools/node_modules/pg. If missing:

bash

mkdir -p /tmp/sb-tools && (cd /tmp/sb-tools && npm i pg)

Workflow

1. Run the audit

bash

node <skills-dir>/supabase-security-audit/scripts/audit.js \
  --probe-uid <existing-customer-uid> \
  --target-uid <another-customer-uid> \
  --site https://example.com

• --probe-uid / --target-uid are optional but unlock the live privilege-escalation probe and the cross-customer data-leak probe. Both probes run inside a transaction that is always rolled back — they do not mutate the database.
• --site adds a HEAD request to check security headers (CSP, HSTS, etc.).
• --public-tables overrides the default list checked for anonymous reads (defaults to profiles,deals,documents,invoices,inquiries,customer_events,tax_engagements).

Exit code is 0 if no critical findings, 1 if any.

2. Read the findings

Output is grouped: RLS coverage → anonymous exposure → UPDATE policies → live privilege escalation → cross-customer leaks → HTTP headers → summary.

Severity icons:

• 🚨 crit — exploitable now, fix today.
• 🟡 warn — best-practice gap or audit step skipped (e.g. no --probe-uid).
• ✅ ok — passed.

3. Map findings to fixes

For each 🚨 finding, look it up in references/threat-patterns.md — every common pattern in that file has a detection query, an impact statement, and a fix.

The most common critical finding is "UPDATE policy lacks WITH CHECK on role/email — privilege escalation risk". That has a turn-key SQL template:

bash

# Edit the table name in the file first if it isn't `public.profiles`
node <project>/deploy/run-migration.cjs \
  <skills-dir>/supabase-security-audit/scripts/hotfix-role-lock.sql

Then re-run audit.js to confirm the finding flipped to ✅.

4. Document the audit

Append a one-paragraph summary to your project's memory or change-log file under a dated heading, including which findings were fixed and any deferred items.

What This Skill Does NOT Cover

• DDoS / rate-limit checks. Supabase + Vercel handle the basics; for finer control add Cloudflare or Supabase's Pro plan rate limiting.
• Application-level CSRF. Supabase auth uses Bearer tokens, not cookies, so CSRF is not a concern for the auth flow itself — but custom session cookies (e.g. an internal dashboard's mc_session) must be audited separately.
• Penetration testing of business logic (e.g. "can a customer call /api/send-portal-link for another customer's id?"). Spot-check those manually by tracing each api/*.js endpoint's auth check.

References

• references/threat-patterns.md — exploitable patterns with detection queries and fixes.
• scripts/audit.js — the audit runner.
• scripts/hotfix-role-lock.sql — turnkey fix for the privilege-escalation pattern.