Exposed secrets
OpenAI, Stripe, AWS, GitHub, Supabase keys committed to your repo or shipped to the browser. Values always redacted in your report.
You shipped fast with AI. Now find out if you also shipped your API keys, open database, or an unprotected admin route — before a stranger does. Plain-English report + copy-paste fixes for your coding agent.
Built by Breaking Circuits' security research practice — the team that dissects real malware and supply-chain attacks and publishes the detections in our Intel & detections site. Tested on 8 targets — 6 real apps plus vulnerable & clean fixtures. Well-built apps come back GREEN, planted disasters come back RED. We fixed our false positives in public before selling a single scan.
Paste a public repo or live URL. We run the first checks instantly — full scan covers every category below + fix prompts.
Escape scanned 5,600 vibe-coded apps and found 2,000+ vulnerabilities and 400+ exposed secrets. These are the patterns behind almost all of them — and exactly what Vibe Scan hunts for.
OpenAI, Stripe, AWS, GitHub, Supabase keys committed to your repo or shipped to the browser. Values always redacted in your report.
Supabase / Firebase rules that let anyone read your users table, storage buckets left public, or a service-role key on the client.
Admin routes with no auth, debug mode left on, endpoints anyone can call without logging in.
The classic vibe-code trap: a “public” env var that isn't public at all. Anything named SECRET / PRIVATE on the client flags hard.
No rate limiting on AI / email / paid API routes — one viral post or bot can drain your credits overnight.
HTTPS, security headers, committed .env files, and .gitignore gaps that turn a future mistake into a leak.
We don't ask you to become a security expert or hire one. We translate findings into the language your builder already speaks: prompts.
Public GitHub repo, private repo via read-only access, zip, or live URL. Takes 20 seconds. You confirm it's yours to scan.
Intake → Scanner → Verifier → Report. Each agent has one job and checks the last one's work. That's why we killed our own false positives before launch.
Plain-English report: what it is, why it matters, where it is (file + line), and a copy-paste prompt like “Move STRIPE_SECRET_KEY server-side and rotate it…” to give your coding agent.
No 40-page PDF you'll never read. A short, ranked list you can clear in an afternoon — each finding with its fix prompt ready to copy.
In plain English: Anyone who visits your site can view-source and steal this key, then make charges / refunds as you. The value is redacted below — rotate it, don't just move it.
src/lib/pay.ts:4 — NEXT_PUBLIC_STRIPE_SECRET_KEY=sk_live_••••8xK2
The service-role key bypasses Row Level Security entirely. On the client, it's effectively “anyone can read/write the whole database.” Your anon key is fine in public — this one isn't.
Supabase anon keys and Stripe pk_live_ publishable keys are public by design. Early versions of scanners (including ours, v0.1) flagged them and scared people for no reason. We decode JWTs and check names so we only flag real secrets.
We tested on purpose-built vulnerable apps, clean apps, and 6 real production templates — including Vercel's own Next.js + Supabase + Stripe template. When we got it wrong, we published the false positives and fixed them:
Vibe Scan comes out of Breaking Circuits' threat research. When our watch dissects a live campaign, we turn it into detections — Sigma rules, YARA rules, IOC sets, and hunt notes — and publish them free. The full collection now lives on its own site.
Attackers rewrote AOMEI's update configs for 5–10 minutes at a time, then reverted them — vendor checks come back clean while endpoints stay compromised. The pack hunts what persists: the scheduled task, the sideloaded DLL, the C2 staging.
Stage 2 hides a credential stealer in a fake font file, executed by a VS Code folderOpen task and confirmed across 79 cloned repos. The pack weights behavior over hashes, so it outlasts the rotating samples.
A human audit for this starts at $2,500 and takes weeks. Vibe Scan is automated, focused, and priced for indie builders who just need to know if it's safe to share the link.
One-time. For “I'm about to share / launch this.”
Owner-authorized only • No code changes made to your app • Not a pentest
For apps you keep shipping. We re-scan so a new key doesn't slip in next week.
Starts with a full $39 scan included in month one.
$2,500+ • 2–4 weeks • 40-page PDF • Built for funded startups with compliance needs.
$0 today • One leaked Stripe / OpenAI key can mean $$$$ in abuse, a drained account, and user data exposed.
Minutes • Ranked, plain-English • Fix prompts your agent can execute today. The launch-day check.
The scan finds problems in the app you already built. The kit stops your coding agent from introducing them in the next one: security-hardened rules files for AGENTS.md, Claude Code, and Cursor, plus the pre-launch checklist, the secrets guide, and the fix-prompt library. Plain text and Markdown — delivered to your checkout email as a small ZIP, seconds after checkout.
Kit purchases join the same 14-day refund policy as scans. Honest limits: the kit reduces the mistakes a scan finds — it is not a guarantee, and it is not a penetration test.
Drop the files into your project once; every future build by your agent follows them. Standard price is $39 once the launch ends.
Get the kit — $29 → Scan + Kit bundle — $59 →The bundle is $9 less than buying the scan ($39) and kit ($29) separately. At bundle checkout you'll enter your app/repo URL and confirm it's yours to scan, like a scan. The kit alone needs no intake — the ZIP goes to your checkout email.
The complete AGENTS.md from the kit — the secrets rules and the “before you say done” self-check included — free, no checkout. The full kit adds the Claude Code and Cursor variants, the human pre-launch checklist, the secrets guide, and the fix-prompt library.
A human pass over one app — repo plus live URL — delivered personally by BC, founder of Breaking Circuits, with a written findings note. Open to anyone who needs one, scheduled with the buyer after checkout — no slot limit.
Vibe Scan reaches vibe coders at the moment they're fixing their app's security. Three founding sponsor slots are open for launch: this page, the Intel & detections site, the free-scan results, and every paid report. Sponsors never influence scan verdicts, placements are always labeled, and sponsors never receive customer scan data.
Labeled placement on this page, seen by builders checking whether their app is safe to share.
Labeled placement on the free-scan results, at the moment a builder sees their first findings.
Labeled placement in every paid report, next to the fixes builders act on.
Vibe Scan comes out of Breaking Circuits' threat research. When our watch dissects a live campaign, we turn it into detections — Sigma rules, YARA rules, IOC sets, and hunt notes — and publish them free. Same-day flags, written for the people who have to hunt this stuff.
Attackers rewrote AOMEI's update configs for 5–10 minutes at a time, then reverted them — vendor checks come back clean while endpoints stay compromised. The pack hunts what persists: the \RadminComponentRuntime scheduled task, Radmin running with --runtimechk, SHINYROAD DLL sideloading, and C2 staging.
Stage 2 hides a credential stealer in a fake font file: a .vscode/tasks.json folderOpen task executes Node.js out of fa-solid-400.woff2, exfiltrates to a fixed server, and plants a cron job pulling new commands every 5 minutes. Confirmed across 79 cloned repos — and the folderOpen pattern now repeats across campaigns, so the pack weights behavior over hashes.
The imgdecode npm package loads a compiled Windows DLL that downloads AES-encrypted shellcode from GitHub and injects it into dllhost.exe — no install script, so install-script blockers never fire. Rules key on the three DLL builds' shared PDB path, the console check that dodges sandboxes and CI, and the EdgeUpdate persistence artifacts.
An infostealer going after the AI toolchain itself: credentials and session files for Claude Code, Codex, Cursor, Copilot, JetBrains AI Assistant and more — plus AWS keys, browser data, and wallets — posted to a /gate endpoint. The behavioral Sigma rule watches reads of AI-assistant credential stores, which outlasts the rotating sample hashes.
Access-refresh phishing via the lookalike domain my-github[.]com: a fake GitHub security alert leads to a reverse proxy that harvests password + 2FA in real time, then silently authorizes an OAuth app with repo, workflow, and read:org scopes — the workflow scope is what lets the operator push the next batch of malicious workflows after victims clean up. Strongest hunt: OAuth client ID Ov23ctrqCqZ79UaYKwyq in authorized apps.
A fake Web3 accelerator whose npm packages were malicious from publication — including a backdoored AI coding agent that rerouted the victim's LLM traffic through attacker infrastructure and persisted an attacker-controlled MCP server in the agent's config. Strongest hunt: ~/.apex/agent/mcp.json pointing at arena.apexaccs.org.
Trojanized OVHcloud CLI releases v0.11.0–v0.15.0 stole cloud, SSH, and AI-assistant secrets during a ~94-minute window, after a developer's hijacked GitHub account was used to replace already-published release assets. Sigma only — no YARA in this pack: no sample hashes have been published for this one. Strongest hunt: an envchecker-* staging temp directory.
PyPI supply-chain compromise: all seven 1.107.0 wheels — plus container images — carried Python startup hooks (cook.pth / cook.py / sitecustomize.py) stealing cloud and AI dev-tool credentials, with an hourly unsigned remote-Python loader on top. Strongest hunt: a cook.pth file containing just “import cook” at the top level of site-packages.
Two of the rules, as published here. All eight packs are published in full — complete Sigma and YARA rule sets, IOC sets, and response notes — in the public Breaking Circuits detections repo on GitHub, free to use. Corrections and contributions are welcome via the repo.
Vibe Scan is one piece of a bigger operation. The research, the automation, and the takedown work feed each other.
The parent: security research and threat intel. The watch that produced the packs above — and the standard this scanner is held to.
Applied automation under Breaking Circuits — home of Vibe Scan, the safety scan for vibe-coded apps: bcautomations.online ↗
Breaking Circuits' scam takedown work: tracking and taking down scam infrastructure. A grant push for the takedown site is underway — Vibe Scan is part of the same Breaking Circuits infrastructure.
Vibe Scan is a service of BC Automations, brought to you by Breaking Circuits.
Vibe Scan is an automated safety scan of a repository and/or live app URL that you provide. You give us a public GitHub repository, a live app URL, or both. We run automated checks and deliver a written report.
That report includes a RED / YELLOW / GREEN verdict, a ranked list of findings (with file and line where we found them, when applicable), and a suggested fix prompt for each finding that you can paste into your coding agent.
The price is $39 one-time per scan. We may also offer an optional monitoring subscription, Vibe Scan Watch, at $12 per month.
Vibe Scan is not a penetration test. It is not a security audit by a human. It is not a guarantee that an app is safe or free of vulnerabilities.
Our findings are pattern-based. That means we look for known patterns — exposed secrets, open data rules, missing protection, and similar issues. Pattern-based checks can miss issues, including issues that are hidden, split across files, or outside the checklist we run.
A GREEN result means “no issues found by these checks” — not “this app is secure.”
You must own the app or repository you submit, or have the owner's authorization to have it scanned. You confirm this at checkout by typing YES.
We scan only what you submit. Checks on live URLs are non-intrusive: we do not attempt exploitation, we do not attempt to log in, and we do not run load testing.
We never test whether a leaked credential is still valid. If we find a possible secret, we show it redacted and tell you to rotate it — we do not use it.
Reports are delivered to the email address you use at checkout.
If we cannot access the repository or URL you submitted, we will tell you. What happens next — including a refund if it can't be resolved — is covered in our Refund Policy.
The Agent Rules & Safety Kit is a digital download — a ZIP of plain-text and Markdown files — delivered to your checkout email after purchase. The Scan + Kit bundle combines one Full Safety Scan with the kit, under this same license.
License summary: one buyer may use the kit's files in their own projects, personal and commercial. You may not redistribute, resell, sublicense, or publicly repost the kit or its files, in whole or in part. The kit reduces common mistakes; like the scan, it is not a guarantee of security and not a penetration test.
Kit purchases are covered by our Refund Policy (14 days).
To the maximum extent permitted by law, our liability is limited to the amount you paid for the scan.
We are not liable for breaches or losses involving your app, including breaches or losses involving issues our scan did not detect.
We may update these terms. The version posted at the time of your purchase applies to that purchase.
If your repository or URL is inaccessible and it can't be resolved with you, we refund 100% automatically — no request needed.
If a report was delivered and you believe a finding is wrong, contact us within 14 days of delivery.
We will re-check that finding by hand. If we got it wrong, you choose: we correct the report, or we refund you in full.
If you simply change your mind after delivery of a completed report, refunds are at our discretion. We are saying that plainly here rather than hiding it — the two cases above are guaranteed, this one is not.
Please contact us first before filing a chargeback. We resolve refund requests quickly — within 5 business days — and contacting us first is almost always faster for you.
If you purchased Vibe Scan Watch: you can cancel anytime. Cancellation stops future billing. We do not provide partial-month refunds.
We collect the email address and the repository and/or URL you provide at checkout. That's what we need to run the scan and send you the report.
Payment is processed by Stripe. We never see or store your full card number — Stripe handles payment details on its side.
To run the checks, your repository is copied temporarily. We work from that temporary copy.
Any secrets we find are shown redacted in your report — never in full. The report is designed so it doesn't become a second leak.
The working copy is deleted within 7 days of report delivery.
We keep a copy of your delivered report and basic order records for support, re-checks (see the Refund Policy), and accounting.
Reports are kept for 12 months, then deleted.
Those are plain statements, not qualifications. If we ever needed to change one, we would update this policy first.
We don't run third-party advertising trackers on this page. We don't use your visit to build an advertising profile.
For privacy questions, or to request deletion of your report and related records, contact us via bcautomations.online. We will handle deletion requests for the records we control — basic order records we must keep for accounting may need to be retained as required by law.