larkspurSecurity review · AI-built appsBook a call

For apps built with Lovable, Bolt, Replit and Supabase

Your AI-built app, made safe to sell.

These tools let you ship in a weekend. The part they skip is the part that matters once real users arrive: who can read whose data, where your keys live, and what happens when something breaks. That part is what I fix.

Leak Check · Finding 01Critical

Signed in as user A, we could read user B’s policy documents.

Row security
offon
Read rule
auth.role() = ’authenticated’auth.uid() = owner_id
Test
A reads B · alloweddenied
OpenFixedProven by a test that runs on every deploy
Illustration of a typical finding
01The problem

The tools are good at shipping. They are not good at saying no.

1701
of 1,645

apps showcased on Lovable let anyone read their database without logging in.

1,0462
of 1,072

Supabase-backed apps built with Lovable, v0, Bolt and Replit had at least one vulnerability. 173 had a critical one.

11%3
of 20,000

indie launches exposed Supabase credentials in their frontend code, many of them the key that skips every rule.

02Free self-check

Five minutes, tonight, no tools.

Mark each one as you go. If something turns up, fix it, or send me what you found and I will tell you what it means.

The full 30-point launch checklist, free

Using Supabase? Paste your schema into the free RLS policy checker and it reads your rules back to you, table by table, with the fix for each one. It runs in your browser.

  1. 1

    Look for “Unrestricted” tables.

    In your Supabase dashboard, open the Table Editor. Any table labeled Unrestricted has Row Level Security off and can be read by anyone with your public key.

  2. 2

    Read your rules out loud.

    A rule that says “authenticated users can read” lets every signed-in user read every row. It should say only the owner can, usually auth.uid() = user_id.

  3. 3

    Try it with two accounts.

    Sign up twice. As the first account, try to open, edit or delete something that belongs to the second. If it works for you, it works for anyone.

  4. 4

    Search your frontend for keys.

    Open your live site, view the source and the JavaScript files, and search for service_role, sk_live and sk-. None of them belong in a browser.

0 of 4 checked.

03Scope

What gets checked.

  1. 3.1

    Who can see what

    Row Level Security on every table and storage bucket, written as “owns this row”, then proven with a cross-account test.

  2. 3.2

    Where your keys live

    Secrets out of the browser and the repo, privileged work moved to the server, exposed keys rotated.

  3. 3.3

    Sign-in and accounts

    Email confirmation, password rules, sign-up abuse, reset links, and admin pages checked on the server, not just hidden.

  4. 3.4

    Money and AI spend

    Rate limits on anything that calls a model, Stripe amounts set on the server, webhooks that verify their signature.

  5. 3.5

    When something breaks

    Alerts that reach a person, uptime checks, backups with a restore actually tested, and tests that run before every deploy.

  6. 3.6

    What buyers will ask for

    A data map, an access list, audit logging and incident notes: the answers to the security questionnaire your first big customer will send.

04Fee schedule

Fixed prices. You know the number first.

ServiceWhat you getTimeFee
Leak CheckCross-account tests on your data and files. Keys, sign-in settings and open endpoints. A plain-English report ranked by risk.48 hours$490Credited in full to a Sprint booked within 30 days
Production-Ready SprintEverything in the Leak Check, fixed. Rules with tests that prove them. Monitoring, backups and a deploy pipeline. An evidence pack for security questionnaires. A runbook and a handoff call.About 14 days$4,500Fixed price, agreed before work starts
Technical leadReview AI-written changes before they ship. Monthly re-tests of rules and keys. Answers for your customers’ security reviews. Architecture calls when you need them.Monthly$2,500Up to 8 hours. $4,000 for up to 16. Cancel any month

GuaranteeIf a critical issue in scope turns up within 30 days of a Sprint, I fix it at no charge.

05Who does the work

One person. The one you talk to.

I am a technical lead by day, and I build at AI speed the rest of the time. Then I go back and harden what I built. Fast and safe is not a trade-off. It is two passes.

I have shipped on Supabase, and I work with CMMC, the NIST 800-171 framework that covers the same ground as SOC 2 and ISO 27001: access control, logging, incident response and vendor risk. That is where the evidence pack comes from.

No account managers. Based in Denver, working with founders anywhere.

Adam ZaouagueTechnical lead · Larkspur Digital

06Questions
Do you test my app without asking?
Never. I test only after you say yes in writing, on test accounts you create for me, on a staging copy when you have one. Nothing of yours is kept after the report.
Will you rewrite my app?
No. The point is to keep what works. Most AI-built apps need targeted fixes, not a rebuild. If something does need rebuilding, you will hear that before any work starts.
Will this make me SOC 2 compliant?
No one honest can promise that from a two-week sprint. What you get is the groundwork: the controls fixed and the evidence written down, so a real audit starts far ahead.
What if you miss something?
If a critical issue in scope turns up within 30 days of a Sprint, I fix it at no charge.

Thirty minutes. A straight answer.

Tell me what you built and who uses it. I will tell you where I would look first, and whether you need me at all.