vibe-code-security-guard
Security exposure scan for apps built with AI coding tools (Lovable, Bolt, v0, Replit, Cursor, Claude Code, Codex). Checks a deployed app's URL for leaked API keys (OpenAI, Anthropic, Gemini and 60+ other providers), exposed source maps, publicly readable Supabase tables, and missing security headers, and returns an A–F grade with fix prompts. Use when the user shares a deployed app URL and asks whether it is safe, wants a pre-launch check, or wants a re-scan after changes. Runs the magnova/vibe-code-security-scanner Actor through the Apify plugin for OpenClaw; each site scanned costs $0.05 on the user's own Apify account.
Vibe Code Security Guard
AI app builders ship fast, and their most common failures are silent: a live API key sitting in the page's JavaScript, a Supabase table anyone can read, source maps that hand over the original code. This skill runs a read-only exposure scan of a deployed app's public surface and reports what a stranger can already see — with a grade and a fix prompt for each problem.
Prerequisites
- The Apify plugin for OpenClaw is installed and allowlisted:
openclaw plugins install @apify/apify-openclaw-plugin APIFY_API_TOKENholds the user's own Apify API token. All runs bill to that account.- You call Actors through the plugin's
apifytool:start(returns a runId), thencollect(returns the results).
If the plugin or token is missing, stop and tell the user exactly what to install or set. Do not probe the site yourself with ad-hoc requests in place of the scan.
Cost, stated plainly: $0.05 per site per scan, charged to the user's Apify account. Tell the user the cost before the first scan of a new site.
Permission rule — read this first
Only scan sites the user owns or has written permission to test. If the user gives you a URL and it is not clearly their own app, ask one question — "Is this your site, or do you have the owner's permission to test it?" — and wait for the answer. The scanner's deeper checks (public-read tests on Supabase tables) run only in owner mode, which requires the user's explicit confirmation of ownership for that site.
Scan procedure
- Confirm the URL(s) and permission (above). One scan can take several URLs in a single run.
- Start the Actor with the
apifytool:- actorId:
magnova~vibe-code-security-scanner - input:
urls: the URL list. AddiOwnThisSite: trueonly when the user has confirmed they own the site(s) — this enables the read-only Supabase table checks. LeavecheckSourceMapsat its default unless the user asks to skip it.
- actorId:
- Collect the results and report per site (format below).
Report format
🛡️ <site> — Grade: <A–F>
Critical / high findings first, in plain language:
- <what was found> — <where: file, table, or header>
Fix: <the fix prompt or instruction the scan returned>
Then medium/low findings the same way. Close with the single
most important next step.
Rules for reporting:
- Never print a full secret. If a finding includes a leaked key or token value, name the provider and where it was found and mask the value (first 4 characters at most). Tell the user to treat that key as compromised and rotate it — deleting it from the site later does not un-leak it.
- Report only what the scan returned. Do not add vulnerabilities it did not find, and do not soften findings it did.
- If a site comes back clean, say what was checked and that nothing was found — a clean scan is a point-in-time exposure check, not a guarantee of safety.
Re-scans and monitoring
- After the user fixes findings, offer a re-scan of the same URL to verify the grade improved.
- For ongoing watching of the user's own apps, offer a weekly
scheduled re-scan. If the user picks (or creates) a key-value
store in their Apify account for scan history and gives you
its ID, pass it as
historyStorewithonlyNewFindings: trueso weekly reports list only what changed. - Keep a short local note of each scan (date, URL, grade) in
state/vibe-code-security-guard.jsonso you can answer "did that site improve?" without re-scanning.
Notes
- This is an automated check of a site's public surface — leaked secrets in shipped code, exposed source maps, public database reads, missing headers. It is not a penetration test and does not review server-side code the site does not expose.
- Owner mode's table checks are read-only: they test whether rows can be read with the site's own public key, and never write, modify, or delete anything.
- Downloads
- 87All time
- Security audit
- Pass
- Last updated
- 3d ago
- Current version
- v1.0.0
- License
- MIT-0
