What a security scan covers
The same checks on every plan, free or paid. 107 of the rules were written for mistakes that work fine until someone looks. Pick your stack to see what they catch in your app, and what they miss.
- Secrets
- 214 credential formats from 122 providers, plus private keys and JSON Web Tokens. Git history too.
- Dependencies
- Known vulnerabilities in 13 ecosystems. Development packages too, in apps built with Vite, SvelteKit, Astro, Nuxt or Remix.
- Code
- 107 rules for mistakes that work fine until someone looks, plus 344 general rules at Low.
- Configuration
- Dockerfile, Kubernetes, Terraform, 4 whole-app web checks, and 27 mobile checks.
New: npm packages your AI may have made up
We now flag known-malicious npm packages, and direct npm dependencies that don't exist, are brand-new and unused, or barely used. An AI assistant can invent a package name, and anyone can register that name with harmful code. We also check the Node version your project pins, if it pins one, for known vulnerabilities and end of life.
- npm packages
- Known malicious package Critical
- Package that doesn't exist on npm High
- New package almost nobody uses Moderate
- Package almost nobody uses Low
Every package your npm, Yarn, pnpm or bun.lock lockfile installs from the npm registry, for the malicious list. The direct dependencies in package.json, looked up on the npm registry, for the other three.
What these don't check
- Names that imitate a popular package, and install scripts.
- A real package that isn't the one you meant.
- A Node version your host picks for you, as on Vercel, Netlify or Lovable. The report says when no pin was found.
- Packages from PyPI, crates.io, Go and RubyGems. Only npm is checked.
- Whether a package installed by another package exists. Only the direct dependencies in package.json are looked up, and never ones from a private registry.
- A malicious package OSV hasn't listed yet. Our copy of its list is refreshed weekly.
bun.lockb. The report says when its packages weren't checked against the malicious list.
Also new: login limits, web hardening and more payment providers
We now check the development packages of apps built with Vite, SvelteKit, Astro, Nuxt or Remix, which can ship them to the browser. We also flag login, password reset and sign-up routes with no rate limit we can see, webhooks from Lemon Squeezy, Paddle, Clerk and Resend that anyone can fake, secrets written to your logs, error details and password hashes sent to the browser, loose security headers, and Vercel cron jobs anyone can start. 10 new checks and 4 whole-app checks, in JavaScript and TypeScript.
- Login and sign-up limits
- Login or password reset with no limit on tries Moderate
- Sign-up with no limit on new accounts Low
Next.js, Express, Hono, Fastify, Koa and Nest routes, in a repository with no rate limit or captcha anywhere we can see.
- Webhooks anyone can fake
- Lemon Squeezy webhook anyone can fake Critical
- Paddle webhook anyone can fake Critical
- Clerk webhook anyone can fake High
- Resend webhook anyone can fake Moderate
Next.js route handlers (App and Pages Router), Express, Hono and Supabase Edge Functions. Stripe as before.
- Secrets and data leaks
- Password hash sent to the browser High
- Secret written to the logs Moderate
- Error details sent to the browser Moderate
JavaScript and TypeScript, with Prisma, Mongoose and Drizzle for user rows.
- Security headers
- Script rules that let any script run Moderate
- Helmet protections turned off Moderate
- Referrer or permissions header too open Low
- App sets no security headers Low
next.config, middleware, vercel.json, Netlify and _headers files, HTML, and Express with Helmet.
- Scheduled jobs
- Vercel cron job anyone can start Moderate
Routes listed under crons in vercel.json.
What these don't check
- Each check reads one file at a time. A signature check, secret check or field filter in a helper in another file is not seen.
- A webhook check that is skipped when the secret is not set counts as a check.
- Paddle Classic webhooks.
- Headers set in your host's dashboard or a CDN. "App sets no security headers" may be a false alarm, and the finding tells you how to check your live site.
- A secret copied into a variable with a neutral name before it is logged.
- A rate limit set at your host, such as Cloudflare or Vercel Firewall, and login code in Python, SvelteKit, Remix or Supabase Edge Functions. Any limiter or server-side captcha check anywhere in the repository quiets both login checks.
Added earlier in October
Request input that reaches a shell, a database query or a URL, passwords and login tokens, Supabase Storage and Edge Functions, GitHub Actions workflows, AI output, and Stripe payments.
- Request input
- Request input run as a shell command Critical
- Request data loaded with an unsafe deserializer Critical
- Request input used as a MongoDB query High
- Your server fetching a URL the request picks High
JavaScript and TypeScript: route params, query and body, followed within one function.
- Passwords and login tokens
- Passwords stored or checked as plain text High
- Passwords hashed with MD5 or SHA Moderate
- Login token read without checking its signature Moderate
JavaScript and TypeScript.
- Supabase
- Edge Function with full database access and no caller check Critical
- Stored files anyone can change or delete Critical
- Storage anyone can upload to High
- Stored files any signed-in user can change or delete High
Storage policies in .sql migrations, and Edge Functions under supabase/functions.
- GitHub Actions
- A workflow that runs text from an issue or comment Critical
- A workflow that runs a stranger's pull request with your secrets Critical
- A workflow that runs a pull request's title, branch or commit text High
Workflow and action files under .github.
- AI integration
- AI output run as code or a command Critical
- AI output run as SQL Critical
- LangChain allowed to run code on your server Critical
- AI output shown as HTML High
- An AI API key in the browser High
- LangChain allowed to call any address High
- AI choosing the URL or file your server uses Moderate
OpenAI, Anthropic, Vercel AI SDK, Google GenAI, LiteLLM and LangChain, plus tool and MCP-server handlers.
- Stripe payments
- A webhook anyone can fake Critical
- A failed signature check that is ignored Critical
- The signature check skipped when the secret is missing High
- The buyer setting their own price High
Express, Next.js route handlers, Fastify, Flask, FastAPI and Django.
- Login and API security
- Login tokens signed with a secret written in the code High
- Login token signature check turned off High
- Flask session key written in the code High
- Whole project folder served to the web High
- Any website able to use your users' logins Moderate
JavaScript, TypeScript and Python.
What these don't check
- Request input and AI output are followed within one function or file. Data that passes through your own helper in another file is not followed.
- A MongoDB sanitiser set up in another file. Routes in their own files may be flagged even though the app is protected.
- Whether sign-up is open. An admin-only Storage bucket in an app with sign-ups turned off may still be flagged.
- A pull request checked out by a script step in a workflow.
- AI providers other than the ones above, or raw HTTP calls to an AI API.
- A Stripe signature check in shared middleware or a guard in another file. That handler may be flagged even though it is protected.
- Whether an AI endpoint has a login or a rate limit.
What it covers for your app
Pick what you built with.
- Secrets
Covered: Stripe, OpenAI, Anthropic, AWS and GitHub keys, in your files and git history
Covered: A Supabase service-role key, found as a JSON Web Token
Not covered: A Postgres DATABASE_URL connection string
- Dependencies
Covered: package-lock.json, yarn.lock, pnpm-lock.yaml and bun.lock
Not covered: bun.lockb, the binary lockfile (commit the text bun.lock)
Not covered: Development packages, skipped on purpose
- Code
Covered: Supabase row-level security: open policies, RLS switched off in a migration, policies trusting user_metadata
Covered: Admin checks only in "use client" components, swapped IDs, server actions with no auth check
Covered: Server secrets in NEXT_PUBLIC_ variables
Covered: AI output run as code or SQL, or shown as HTML, and AI keys shipped to the browser
Covered: Stripe webhooks anyone can fake, and checkouts where the buyer sets the price
Covered: Lemon Squeezy, Paddle, Clerk and Resend webhooks anyone can fake
Covered: Storage policies open to anyone, and Edge Functions with full access and no caller check
Covered: Secrets in your logs, and error details or password hashes sent to the browser
Partly: Security headers in next.config, middleware and vercel.json, not ones set in the Vercel dashboard
Partly: Your own login and sign-up routes with no rate limit, not limits set in Vercel Firewall or Supabase Auth
- Configuration
Covered: Your Dockerfile, if you have one
Covered: Vercel cron jobs listed in vercel.json that never check CRON_SECRET
Not covered: Vercel project settings
Try
Where it stops
Secrets
- Database connection strings: MongoDB, Postgres, Redis, AMQP.
- Providers without their own rule, such as Postmark, PayPal, Vercel, Segment and Docker Hub. Only a key-named, random-looking value is flagged.
Dependencies
- Development packages, except in apps built with Vite, SvelteKit, Astro, Nuxt or Remix.
- A dependency file that can't be read. The report says so, and never shows it as clean.
bun.lockb, andgo.sumwithoutgo.mod.- Unpinned ranges in
requirements.txt, and Maven versions set by a parent or BOM. - CocoaPods beyond the 8 pods on our own list.
Code and configuration
- Ruby, Rust, Elixir, PHP and Objective-C code.
- Docker Compose files. Helm, CloudFormation and ARM templates are not claimed.
Everything else
- Anything behind sign-in on your running app. A site scan reads only what your site serves to anyone, and a repository scan never runs anything.
- On an individual plan, a repository over 500 MB or 20,000 files: no git history for secrets, and code rules read the first 20,000 files. The report says so.
A clean report means these checks found nothing blocking on that commit. It is a good sign and never a guarantee.
Guides
How to check these yourself, and what to change.
Try it on your own code.
Free to scan. No card. Compare plans