WP Front Door

How the check works

WP Front Door looks at a WordPress site the way a stranger would: from outside, logged out, with ordinary requests. This page covers exactly what it asks for, what it refuses to do and what it cannot see.

The six requests

Every request is a GET, sent with the user agent WPFrontDoor/1.0 (+https://hamzaahmadaslam.com/work/wp-front-door) so a site owner can see who asked. It never logs in, never tries a password and never submits a form. The second request for the page uses Googlebot's user agent with the checker's name still on the end, Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html) WPFrontDoor/1.0 (+https://hamzaahmadaslam.com/work/wp-front-door), because hacked sites show their spam to anything that says it is Googlebot. It cannot pass Google's own address check and does not try to.

The address you gave
the page itself: its HTML, response headers, redirects and time to first byte
The same address, as Googlebot
the copy a search engine gets, compared with the visitor's to catch spam shown only to Google
/wp-json/
whether the REST API answers, which helps confirm WordPress
/wp-json/wp/v2/users?per_page=1
whether account names are listed to visitors
/xmlrpc.php
whether the XML-RPC endpoint answers
/readme.html
whether the stock readme names the version

Each request gets 8 seconds and reads at most 1.5 MB. Redirects are followed one hop at a time, five at most. The checker runs on Vercel, currently in its Washington, D.C. region, which is where the first-byte time is measured from.

Addresses it will not check

Only http and https addresses on the standard ports, with no user name or password in them and no .local host names.

Before anything is fetched, the host name is looked up and refused if it points at a private, loopback or link-local address. The same check runs again inside the connection's own lookup, so a name that changes its answer between the two cannot slip through. Each redirect is checked the same way, which stops a public site bouncing the checker into a private network.

One address can run ten checks a minute, and one site can be checked twenty times in ten minutes. Both limits count per server instance.

What it cannot see

  • Anything behind a login: the dashboard, members' pages, a store's account pages.
  • Plugin versions and known vulnerabilities. It sees plugin folder names in the page's file paths, and plugins that load no files on the page stay invisible.
  • Other pages. It reads the one address you give, so a slow product page or a heavy blog post needs its own check.
  • What visitors experience in the browser. There is no rendering, no image weight and no Core Web Vitals, only the HTML and the first byte.
  • Speed from where your visitors are. Every request comes from one place, so distance to your host is part of the first-byte time.
  • A warm cache. It asks once, so a page that was just purged can miss. Run the check a second time before trusting a cache miss.
  • Whether a site is clean. The signs of a hack come from one page read from outside. Malware in files or the database, spam on other pages, and cloaking that checks for Google's real addresses or acts only on visitors arriving from a search all stay out of sight.

If a firewall or bot challenge answers instead of the site, the report says so, because every other finding then describes the challenge page.

How the scores add up

Security, caching and page weight each start at 100. A problem takes 25 points off its area and a warning takes 10. Notes cost nothing. The overall score weights security at 40% and the other two at 30% each. Signs of a hack have no score: a pattern is worth a look, not a grade.

The fix-first list takes the problems and warnings that have a fix, problems first, and keeps five. A Safe Browsing listing goes to the top. The full list of checks, with the thresholds each one uses, is spread across the fix guides.

What happens to the page it reads

No report is written anywhere. The report is built from the responses in memory and discarded once the page is sent. Reloading a report runs a fresh check.

One optional step sends text out. When the site owner has switched on Jev, the page's title, description and the start of its visible text go to it once, to read whether the site is a store, a publisher, a business or a blog. The fix-first list is then ordered for that kind of site. Jev is asked only when the check is started from the form on this site, a few times an hour per address. Its answer (the kind of site and how sure it is) is kept for a day under a hash of what was sent, so the same page is not read twice in a day, and a shared report link or the JSON API can use that answer but never asks Jev itself. A page containing text written to steer AI readers is never sent, and the report says so.

When the site owner has set a Google Safe Browsing key, one more request goes out, to Google rather than to the site: the page's address, the site's root and the addresses of scripts the page loads from other sites, to ask whether Google lists any of them as malware or phishing. Nothing about the answer is kept. Without the key, the report links to Google's Transparency Report for the site instead.

Who built it

Hamza Ahmad Aslam, a WordPress and WooCommerce developer. The WP Front Door case study covers why it exists and how it was built. For scripts and dashboards there is a JSON API with the same report.