Dev.to WebDev πŸ›  Dev πŸ‘ 0 πŸ“– 4 min read

What You Can (and Cannot) See About Web App Security From the Outside: Lessons From Auditing 413 Public Launches

Disclosure: I am the creator of AppSeal, a passive external security scanner. Below is a technical breakdown of what an unauthenticated black-box audit actually reveals, aggregated patterns from 413 public web launches,

What You Can (and Cannot) See About Web App Security From the Outside: Lessons From Auditing 413 Public Launches

Disclosure: I am the creator of AppSeal, a passive external security scanner. Below is a technical breakdown of what an unauthenticated black-box audit actually reveals, aggregated patterns from 413 public web launches, and where external scanning hits a hard wall.

Shipping a web app today takes hours instead of weeks. Between AI code editors, managed databases, and zero-config edge hosts, you can go from an empty repo to a live domain over a single weekend.

The trade-off is that deployment hardening is usually the last thing on the checklist. Default framework configs rarely ship with strict security headers, and it is surprisingly easy to leave source maps, debug routes, or public API keys exposed in a client bundle.

To see which configuration gaps show up most often in the wild, I took a snapshot on November 14 across 413 public web projects shared by developers on Show HN and GitHub.

Defining the Scope (and Keeping the Data Honest)
Before looking at the numbers, three important caveats:

Non-representative sample: These 413 domains are public indie launches and open-source demos, not a random sample of all production software on the internet.

Hardening gaps vs. active exploits: A missing Cross-Origin-Opener-Policy header is not a breach. It is a defense-in-depth omission that lowers browser isolation.

No domain shaming: All figures below are strictly aggregated.

What External Black-Box Checks Actually Inspect
Without source code access or credentials, an external scanner operates strictly at the HTTP, DNS, and rendered DOM layers. Running ~140 lightweight probes plus a headless SPA render surfaces five distinct categories of issues:

Transport & Browser Isolation Headers: CSP, HSTS, X-Frame-Options, COOP, Permissions-Policy, Referrer-Policy, overly permissive CORS (*), and cookie security attributes.

Client Bundle & Artifact Exposure: Publicly reachable .env files, .git/HEAD, .js.map source maps, and regex/entropy matches for third-party API keys embedded in frontend JavaScript.

Debug & Admin Surface: Unauthenticated Swagger/OpenAPI specs, GraphQL introspection, framework debug pages, and .bak files.

Runtime & Content Hygiene: Uncaught stack traces rendered into the DOM, mixed HTTP/HTTPS assets, missing Subresource Integrity (SRI) on CDN scripts, and broken social preview metadata.

AI & Email Posture: llms.txt, AI crawler directives in robots.txt, and SPF/DMARC DNS records.

Key Takeaways From the 413-Project Snapshot
Only 46 out of 413 projects (~11%) scored an A or A+ (90+ out of 100), while the majority clustered between 60 and 85 points.

Here is how often common headers and metadata tags were missing across the dataset:

Cross-Origin-Opener-Policy (COOP): 90% missing

Permissions-Policy: 81% missing

Content-Security-Policy (CSP): 78% missing

X-Frame-Options: 74% missing

Referrer-Policy: 51% missing

X-Content-Type-Options: 47% missing

Strict-Transport-Security (HSTS): 33% missing

Open Graph tags (og:image, og:title, og:description): 29% to 31% missing

Hosting defaults played a noticeable role in baseline scores:

Projects hosted on Lovable (66 domains) averaged 83/100 because the platform injects a solid baseline of security headers out of the box, though none reached Grade A due to stricter isolation headers like COOP.

Custom domains (277 domains) averaged 72/100 with the widest variance - spanning from A+ setups behind tuned Cloudflare rules down to unconfigured reverse proxies.

GitHub Pages (39 domains, avg 66/100) and default Vercel deployments without a custom vercel.json (30 domains, avg 64/100) trailed behind because static hosts either restrict response header control or ship minimal defaults unless explicitly configured.

Fixing the Baseline in Next.js / Vercel
If you deploy on Next.js or Vercel, adding a headers() block in next.config.js closes most of the passive header warnings in one commit:

JavaScript
const securityHeaders = [
{ key: 'X-Frame-Options', value: 'DENY' },
{ key: 'X-Content-Type-Options', value: 'nosniff' },
{ key: 'Referrer-Policy', value: 'strict-origin-when-cross-origin' },
{ key: 'Cross-Origin-Opener-Policy', value: 'same-origin' },
{ key: 'Permissions-Policy', value: 'camera=(), microphone=(), geolocation=()' },
];

export default {
async headers() {
return [
{
source: '/:path*',
headers: securityHeaders,
},
];
},
};
(Note: Add Content-Security-Policy separately once you have mapped your exact script, font, and API connect-src origins so you do not break third-party analytics or checkout widgets.)

Where External Scanning Hits a Wall
It is dangerous to treat an A+ external grade as a clean bill of health. Passive URL checks have blind spots you must cover with other tools:

Row-Level Security & BOLA/IDOR: An unauthenticated scanner cannot log in as User A and try to fetch User B's rows from your Supabase or REST API.

Git History & Dependencies: Keys committed and later deleted from a public repo, or vulnerable npm packages, require secret scanners (gitleaks) and dependency audits (npm audit, Semgrep) inside your CI pipeline.


Unlinked Server Routes: API endpoints not referenced in your frontend bundle or standard discovery paths remain invisible from the outside.

Think of external scanning as a pre-flight smoke test before sharing your launch link. If you want to run a quick check on your own domain, AppSeal gives you one full scan per day with zero signup and outputs an AI-ready fix-prompt.txt for any gaps it finds.

πŸ“° Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes β€” full credit and traffic to the original publisher.