It looks finished. Does it actually work?
Claude / Codex / Gemini / Cursor / Copilot / v0 / Bolt / Windsurf / Replit — any AI-built codebase. We find the security holes, the liabilities and the buttons that lead nowhere, then write the fix. Free scan, then $99 one-time audit or $49/mo re-scan monitor.
Built a Lovable app? The Lovable Security Audit covers Lovable-specific patterns: Supabase RLS bypass, removed-auth-header IDOR, and direct Supabase API misconfigurations.
$99 one-time audit · or $49/mo for ongoing re-scans
What the AI-built Code Audit fixes
Security Vulnerabilities
The problem
AI-generated code builds the happy path first and skips the defensive checks.
What we find
Missing ownership checks that let one user read another's data, API keys and credentials left in source, Supabase tables with no row-level security, user input reaching SQL, vulnerable dependencies, and a robots.txt that re-exposes private paths.
Product & Service Liabilities
The problem
AI-built products launch leaning on third-party APIs, with nothing handling the failures.
What we find
Payment webhooks that skip signature checks or answer 200 before verifying, so paid events are lost. Gated data replaced by placeholder text. Unvetted AI providers and prompt-injection surfaces. Hidden instructions inside agent files like CLAUDE.md. Copy that promises a turnaround time or a manual review the code never performs.
Broken Paths & Missing Buttons
The problem
The screens look finished while the wiring behind them was never completed.
What we find
Buttons and fetch calls that point at routes that do not exist, pages that can never render because a redirect shadows them, and responses used without checking for errors, so failures vanish silently.
Privacy policy, terms and cookie banner on your live site are checked by the Website Privacy scan.
Scan my live siteWhat the audit covers
81 deterministic rules plus 5 categories the AI review assesses from what the scan collects. Each is a specific, named check, not a generic checklist.
Checked against your actual code — not a generic checklist.
Free scan covers deterministic checks · full findings need the $99 AI-powered audit
A repository scan reads code and config. It cannot see your backups, who can reach your production credentials, or your branch protection, so it never reports those as safe.
IDOR / broken object-level auth
Full auditCan user A access user B's data?
Exposed secrets and credentials
Full auditAPI keys, DB URLs, wallet keys, and Supabase, Neon, Fly.io, DigitalOcean and Netlify tokens in code
robots.txt that re-exposes private paths
Full auditAI-crawler groups with no Disallow replace your * rules (RFC 9309), so /api and /dashboard become crawlable
Delivery promises the code does not keep
Full auditCopy that promises a turnaround time or a manual review while the handler finishes and publishes instantly
Pages that can never render
Full auditPage files shadowed by a next.config redirect — edits to them never reach visitors
Missing Row Level Security (RLS)
Full auditSupabase tables open to any authenticated user
SQL injection paths
Full auditUser input reaching raw SQL or ORM interpolation
AI dependency risks
Full auditUnvetted model providers, prompt injection surfaces
Vulnerable dependencies
Full auditKnown CVEs in your lockfile
Missing rate limiting
Free scanPublic endpoints with no throttle — open to abuse and credential stuffing
Unchecked HTTP responses
Free scanfetch() calls that invoke .json() without a .ok guard — silently swallows errors
Webhook silent-success
Free scanHandlers that return 200 before verifying payload — payment events discarded forever
Broken button → route contracts
Free scanUI buttons calling fetch('/api/…') paths with no matching route handler
Hidden text in agent instruction files
Free scanInvisible characters, hidden comments and unpinned launchers in CLAUDE.md, AGENTS.md, .cursorrules and .mcp.json (needs GitHub sign-in)
Secrets shipped to the browser
Free scanNEXT_PUBLIC_ / VITE_ variables named like secrets, and Supabase service-role keys in client code or hard-coded (needs GitHub sign-in)
Prices and roles the client controls
Free scanAmount, plan, role or credits read from the request body and handed to Stripe or written to your database (needs GitHub sign-in)
Destructive scripts with no guard
Free scanSeed and reset scripts, and install or deploy commands, that can wipe a production database (needs GitHub sign-in)
Webhooks that process the same event twice
Free scanSignature-verified handlers that never record or check the event id, so a provider retry runs the payment logic again (needs GitHub sign-in)
Unsanitised HTML
Free scandangerouslySetInnerHTML, innerHTML and v-html fed non-literal values without a sanitizer (needs GitHub sign-in)
Delete buttons with no confirm
Free scanDelete, revoke or cancel buttons and form actions with no confirmation step or undo found in the same file (a heuristic: needs GitHub sign-in)
Agent permissions and CI / install risks
Free scanCommitted bypass-permission agent settings, pull_request_target running PR code, expression injection in workflows, download-and-run install scripts (needs GitHub sign-in)
Tests that never test a denial
Free scanAn app with login whose sampled tests never assert a 401, 403 or cross-tenant refusal (needs GitHub sign-in)
Also checked in every file of the repository
These read the code itself and are reported as "Needs review": they point at a pattern, a person confirms it.
- A webhook handler that ignores a failed release of its processed-event record, so a retried event can be dropped as a duplicate.
- A cron, internal or webhook route that reads its secret from the URL query string, where logs keep it.
- A Google or GitHub sign-in that trusts the profile email without checking that the provider verified it.
- Placeholder text such as "[To be specified]" in a page or email that customers can see.
- A session cookie whose Secure flag follows the request instead of being forced on in production.
- A request token or secret compared with === instead of a timing-safe comparison.
- Credits or a balance that are read and then written back, so two simultaneous requests can spend the same credit.
- A rate limiter keyed on the raw client address, which an IPv6 client can rotate to dodge the limit.
- Session replay that records page text, so signed-in screens with keys or customer data reach the analytics vendor.
- A client component whose first render uses the clock or a random number, so the page re-renders on load and wipes what the visitor typed.
- A Claude API call that sends a parameter its model rejects (forced tool choice, a thinking budget), so every call fails.
AI remediation patches — copy and apply
Every finding comes with a ready-to-paste code fix. One click to copy, or one click to open a real pull request on your own repo.
- Per-finding patches — not generic advice
- One-click copy to clipboard
- Create Pull Request — opens a real PR on your own repo
const session = await getServerSession();
if (!session?.user?.id ||
session.user.id !== params.id) {
return NextResponse.json(
{ error: 'Forbidden' }, { status: 403 }
);
}How it works
Scan your GitHub repo (free)
Paste the URL of your Claude / Codex / Gemini / Cursor / Copilot / v0 / Bolt / Windsurf / Replit app. Public or private (private requires GitHub OAuth).
AI analysis runs (under 10 min)
We check security, liability and broken-path patterns against your actual code — IDOR, credentials, RLS, injection paths, rate limiting, unchecked responses, webhook handling, button→route gaps, and more.
Unlock the audit report ($99) or subscribe to re-scans ($49/mo)
Unlock the full report + AI remediation patches with a one-time $99 audit, or subscribe to the $49/mo monitor for ongoing re-scans as your code changes.
Apply the patches
Copy patches directly, or click Create Pull Request to open a real PR on your own repo. Every finding has a ready-to-paste fix, not just a description of the problem.
Find it before your customers do
Your customers trust you with their data. A full repo audit costs $99 one-time, with remediation patches included. Shipping fast? Subscribe to the $49/mo monitor so new issues don't slip through after every deploy.
Scan free first at regulatorysignals.com
Where to go next
EU AI Act risk checks, reports, audit pack, and policies — one program.
Free to startStart Free Check11 MCP-specific security rules. Signed cert. Public registry listing.
$249 one-timeGet AuditedFree website compliance scan, paid report, and daily regulatory feed.
Free to startStart Free Scan