Methodology
The automated audit is deterministic code: HTTP requests, an HTML parser, headless Chromium, file scans, and the OSV.dev vulnerability database. There is no language model in the loop and nothing is estimated. The same input produces the same result, apart from things that really change between runs, like response time.
Scoring
Each check has a weight by severity: critical 10, high 6, medium 3, low 1. A pass earns the full weight, a warning half, a failure nothing. The score is earned ÷ possible × 100 over checks that ran. If any critical check fails, the score is capped at 40.
Safety
Every outbound request is checked against private, loopback, link-local and metadata ranges, and the connection is pinned to the validated IP. Chromium’s subrequests go through the same filter. Uploaded sites are rendered from memory and previewed under a sandboxed CSP, so their scripts can’t reach your session. Real .env files are never stored.
What a pass doesn’t mean
Automated checks cover what can be measured from outside. They can’t find logic bugs, broken auth, or a confusing product. A high score means fewer known problems, not a secure or good product. That’s what expert reviews are for.
Not run ≠ passed
If Chromium is unavailable, OSV.dev is unreachable, or a check doesn’t apply, it is marked “not run” and left out of the score rather than counted as a pass.
Security · 17
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
Served over HTTPS sec-https | The final URL after redirects uses https. (HTTP request) | critical | URL |
Valid TLS certificate sec-tls | Node's TLS stack verifies the chain; warns if expiry is under 14 days. (HTTP request) | critical | URL |
HTTP redirects to HTTPS sec-http-redirect | Requests http://host/ without following redirects and checks for a 3xx to https. (HTTP request) | medium | URL |
Strict-Transport-Security sec-hsts | Header present with max-age ≥ 180 days. (HTTP request) | medium | URL |
Content-Security-Policy sec-csp | Parses script-src (or default-src); warns on 'unsafe-inline' without a nonce or hash. (HTTP request) | medium | URL |
Clickjacking protection sec-framing | X-Frame-Options DENY/SAMEORIGIN or a CSP frame-ancestors directive. (HTTP request) | medium | URL |
X-Content-Type-Options sec-nosniff | Header equals nosniff. (HTTP request) | low | URL |
Referrer-Policy sec-referrer | Header set and not unsafe-url / no-referrer-when-downgrade. (HTTP request) | low | URL |
No version disclosure sec-version-leak | X-Powered-By absent and Server header contains no version number. (HTTP request) | low | URL |
Cookie attributes sec-cookies | Every Set-Cookie on the first response has Secure and SameSite; session-like names also need HttpOnly. (HTTP request) | medium | URL |
/.env not public sec-exposed-env | GET /.env must not return KEY=value lines. (HTTP request) | critical | URL |
/.git not public sec-exposed-git | GET /.git/HEAD and /.git/config must not return git metadata. (HTTP request) | critical | URL |
No secrets in client code sec-client-secrets | Pattern scan of the HTML and up to 12 same-origin scripts. Supabase JWTs are decoded and flagged only when role is service_role. (HTTP request) | critical | URL |
Production source maps sec-source-maps | Requests the first bundle's .map (or SourceMap header target). (HTTP request) | low | URL |
No http:// subresources sec-mixed-content | Scripts, styles, images, iframes, media and form actions referenced over plain http. (HTML parse) | medium | URL, Built site |
No real .env files sec-committed-env | .env, .env.local, .env.production… with assigned values. Templates (.env.example) are allowed. (File scan) | critical | Files |
No hardcoded secrets sec-file-secrets | Same pattern set as the client scan, over every text file under 2 MB. (File scan) | critical | Files |
Privacy · 1
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
No tracking before consent priv-trackers | Records every host Chromium contacts while the page loads, with no clicks, and every cookie set. Third-party requests to 40+ known advertising, cross-site analytics and session-replay domains, or any third-party cookie, warn. Cookieless analytics (Plausible, Fathom, Vercel, Cloudflare) and strictly necessary bot-protection or load-balancer cookies (Cloudflare __cf_bm, Akamai, AWS ALB, Imperva) are not counted. (Headless Chromium) | medium | URL, Built site |
Performance · 9
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
Server response time perf-ttfb | Time to first byte from the audit server: < 800 ms pass, < 1.8 s warn. (HTTP request) | medium | URL |
Short redirect chain perf-redirects | ≤ 1 redirect passes, 2 warns. (HTTP request) | low | URL |
HTML compressed perf-compression | content-encoding is br, gzip, zstd or deflate for HTML over 1 KB. (HTTP request) | medium | URL |
Page weight perf-weight | Bytes transferred by Chromium during load: < 2 MB pass, < 4 MB warn. (Headless Chromium) | medium | URL, Built site |
Largest Contentful Paint (mobile) perf-lcp | A cold-cache load as Lighthouse's emulated phone (Moto G Power, 412px, Android Chrome) under its mobile profile (Slow 4G: 150 ms RTT, 1.6 Mbps down), applied with DevTools throttling. The CPU is slowed to a phone-class speed calibrated with Lighthouse's BenchmarkIndex (target 1000, i.e. Lighthouse's default 4× on a current laptop), so results don't depend on the audit server's hardware. ≤ 2.5 s pass, ≤ 4 s warn, otherwise fail; nothing painted within 45 s fails. (Headless Chromium) | high | URL |
First Contentful Paint (mobile) perf-fcp | Same throttled phone load: ≤ 1.8 s pass, ≤ 3 s warn (Lighthouse's mobile boundaries). (Headless Chromium) | medium | URL |
Total Blocking Time (mobile) perf-tbt | Same throttled phone load, Lighthouse's definition (when the audit server's CPU is slower than the phone target, long-task durations are first scaled to the target speed, as Lighthouse's simulated throttling does): each long task's time beyond 50 ms between first paint and Time to Interactive (the start of the first 5 s window with no long tasks and at most two requests in flight). ≤ 200 ms pass, ≤ 600 ms warn. Shown but not scored when the audit server has fewer than 4 CPU cores, where the browser's own threads inflate it. (Headless Chromium) | medium | URL |
Cumulative Layout Shift perf-cls | Sum of layout-shift entries without recent input: < 0.1 pass, < 0.25 warn. (Headless Chromium) | medium | URL, Built site |
No oversized assets perf-heavy-assets | Images > 400 KB, scripts > 1 MB, video > 8 MB. (File scan) | medium | Built site |
Accessibility · 15
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
Text contrast (WCAG 1.4.3) a11y-contrast | For up to 800 rendered text elements, composites background layers and resolves any CSS color via canvas, then compares to 4.5:1 (3:1 large text). (Headless Chromium) | medium | URL, Built site |
No sideways scroll at 390px resp-overflow | document scrollWidth vs. innerWidth in a 390×844 mobile context; lists offending elements. (Headless Chromium) | medium | URL, Built site |
Touch targets ≥ 24px (WCAG 2.5.8) resp-tap-targets | Bounding boxes of links, buttons and inputs on the mobile layout; inline text links are exempt. (Headless Chromium) | low | URL, Built site |
WCAG 2.2 A/AA scan (axe-core) a11y-axe | Runs Deque's open-source axe-core engine (the one behind Lighthouse and most accessibility tools) on the rendered desktop page with the WCAG 2.0–2.2 A/AA rule tags. Serious or critical violations fail; moderate or minor ones warn. Rules already scored by other checks are excluded. (Headless Chromium) | high | URL, Built site |
Other pages: headings and alt text a11y-pages | On each crawled page's server HTML: images with no alt attribute fail, a missing <h1> warns. JavaScript-rendered app shells are not judged. (HTML parse) | medium | URL |
Keyboard focus is visible (WCAG 2.4.7) a11y-focus-visible | Judged on pixels: after a Tab press puts the page in keyboard mode, the area around each of the first 12 visible links and controls is captured unfocused and focused (transitions finished, text caret hidden). Identical pixels mean no visible focus indicator. Any warns; over 30% fails. (Headless Chromium) | medium | URL, Built site |
Document language a11y-lang | <html lang> present. (HTML parse) | medium | URL, Built site |
Viewport and zoom a11y-viewport | width=device-width set and neither user-scalable=no nor maximum-scale ≤ 1. (HTML parse) | medium | URL, Built site |
Images have alt a11y-img-alt | Visible <img> without an alt attribute (alt="" counts as decorative). (HTML parse) | medium | URL, Built site |
Form fields labelled a11y-form-labels | label[for], wrapping <label>, aria-label, aria-labelledby or title. Placeholder alone fails. (HTML parse) | medium | URL, Built site |
Buttons have names a11y-button-names | Text, aria-label, aria-labelledby, img alt, svg <title> or title. (HTML parse) | medium | URL, Built site |
Links have names a11y-link-names | Same accessible-name rules as buttons. (HTML parse) | medium | URL, Built site |
One main heading a11y-h1 | Exactly one visible <h1>. (HTML parse) | low | URL, Built site |
Heading order a11y-heading-order | No jumps of more than one level (h2 → h4). (HTML parse) | low | URL, Built site |
Unique IDs a11y-duplicate-ids | No id value used twice. (HTML parse) | low | URL, Built site |
Design & UX · 7
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
No template or placeholder content ux-placeholder | Rendered page: template default titles (Create Next App, Vite + React, React App, Document…) and descriptions, starter text ("Get started by editing", lorem ipsum), the starter logo row (next.svg + vercel.svg, vite.svg, react.svg), and broken render output: a visible text node that is exactly undefined, NaN or null, $NaN, [object Object], or unrendered {{ }}. Code blocks are ignored. (Headless Chromium) | medium | URL, Built site |
Images load and keep proportions ux-images | Visible images in the first two screens that finished loading with no pixels fail; images drawn more than 10% off their natural aspect ratio with object-fit: fill warn. SVGs are skipped. (Headless Chromium) | medium | URL, Built site |
No overlapping text ux-overlap | On the first 1.5 screens at 1280px: lines of text (each line's own rectangle, so floats and padding don't count) from two unrelated elements overlapping by over 30% of the smaller line, where the overlap is the topmost thing drawn and neither is clipped by an ancestor. Fixed-position, rotated and identical stacked text are ignored. (Headless Chromium) | medium | URL, Built site |
No cut-off text ux-clipped | Leaf text elements whose content is larger than their box under overflow: hidden or clip, with no text-overflow: ellipsis or line-clamp. (Headless Chromium) | low | URL, Built site |
Comfortable line length ux-line-length | Paragraphs, list items and cells over 240 characters, estimated at over 110 characters per line at 1280px (width ÷ half the font size). (Headless Chromium) | low | URL, Built site |
Consistent typography ux-fonts | Typefaces weighted by how much visible text each sets; more than three with at least 2% each warns. Icon fonts are ignored. (Headless Chromium) | low | URL, Built site |
Readable text on phones ux-text-size | Like Lighthouse's legible-font-size audit: the share of visible text characters on the 390px layout set below 12px. Over 10% warns; over 40% fails (Lighthouse's threshold). Text hidden from assistive tech is skipped. (Headless Chromium) | medium | URL, Built site |
Search & sharing · 9
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
Page title seo-title | <title> present; 10–70 characters. (HTML parse) | medium | URL, Built site |
Meta description seo-description | Present; 50–170 characters. (HTML parse) | low | URL, Built site |
Social preview tags seo-social | og:title and og:image. (HTML parse) | low | URL, Built site |
Canonical URL seo-canonical | <link rel=canonical>. (HTML parse) | low | URL, Built site |
Favicon seo-favicon | <link rel=icon> or apple-touch-icon. (HTML parse) | low | URL, Built site |
Excluded from search seo-noindex | Only reported when meta robots contains noindex. (HTML parse) | high | URL, Built site |
Unique titles and descriptions seo-pages | Across the audited page and every crawled page: a missing or duplicated <title> fails, a missing meta description warns. JavaScript-rendered app shells are not judged. (HTML parse) | medium | URL |
robots.txt seo-robots | /robots.txt answers 200 with non-HTML content. (HTTP request) | low | URL |
XML sitemap seo-sitemap | /sitemap.xml is a urlset/sitemapindex, or robots.txt declares one. (HTTP request) | low | URL |
Reliability · 6
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
Page responds 2xx rel-status | Final HTTP status after redirects. (HTTP request) | critical | URL |
No JavaScript errors rel-console-errors | console.error calls and uncaught exceptions during load: 0 pass, ≤ 2 warn. (Headless Chromium) | medium | URL, Built site |
All resources load rel-failed-requests | Network failures and 4xx/5xx responses; third-party-only failures warn. Requests the browser cancels itself (ERR_ABORTED, e.g. dropped prefetches) are not failures. (Headless Chromium) | medium | URL, Built site |
No broken internal links rel-broken-links | Crawls up to 20 same-origin pages, starting from the links on the rendered page and sitemap.xml, honouring robots.txt. Any visited page answering 404, 410 or 5xx (or failing DNS/TLS) fails; access control (401/403), other 4xx, timeouts and resets are not counted. Two requests at a time; the crawl stops as soon as the site answers 419, 429 or 503. (HTTP request) | medium | URL |
Builds from a clean install rel-build | Projects with a build script and no build output are installed and built in a throwaway Vercel Sandbox microVM with no secrets: registries-only network while installing, no outbound network while running. A failed install fails; a failed build warns, since it may only need env vars. (Vercel Sandbox build) | medium | Files |
No broken local references rel-broken-refs | Every local src/href with a file extension resolves to a file in the upload. (File scan) | medium | Built site |
Code & dependencies · 4
| Check | How it’s measured | Severity | Runs on |
|---|---|---|---|
Known vulnerable dependencies code-dependency-vulns | Exact versions from package-lock, pnpm-lock, yarn.lock, requirements.txt (== pins), poetry.lock or uv.lock, queried against OSV.dev. High/critical fails. (OSV.dev database) | high | Files |
Lockfile committed code-lockfile | package.json without any lockfile. (File scan) | low | Files |
.gitignore excludes .env code-gitignore-env | Root .gitignore has a .env pattern (source projects only). (File scan) | medium | Files |
README present code-readme | GitHub imports only. (File scan) | low | Files |