Scanning
How a scan works
The steps from URL to fix report, and the safety rules around them.
A scan has two parts. First, fixed code runs a set of checks against your site and records exactly what it saw. Then an AI model reads those results and writes the report. The AI never touches your site and never adds findings of its own.
1. You submit a URL
Enter a domain or full address. Without a scheme we use https://. You choose a normal or deep scan; deep needs a verified domain.
2. Safety check
We refuse private and internal addresses (for example localhost, 10.0.0.0/8 or cloud metadata addresses). The address is checked again on every redirect and every connection, so a public name that points inside a network is blocked too.
3. The checks run
Each scan runs in its own isolated process on a server used only for scanning. It loads your page, follows up to 5 redirects, then runs every check in a fixed order. The checks are written in code, not chosen by an AI, so the same site gives the same result twice.
- Limits: up to 100 requests for a normal scan and 600 for a deep scan (plus one connection to each of 22 server ports and nuclei templates), with a 4-minute time limit. Certificate logs, OSV.dev, CISA, endoflife.date and wordpress.org are read directly, not through your site.
- If a check cannot finish (a timeout, a blocked request), it is marked not checked. It never counts as a pass.
- Secret values we come across, for example in an exposed .env file, are hidden in the evidence. We keep only the key names.
You can watch each check live on the scan page, or follow the events stream from code.
Software we found
Every report lists the software and versions the scan could see, from response headers, the page and its scripts. Each one gets a verdict with the public source that proves it: no known bugs (OSV.dev has no advisory for that exact version), known bugs or attackers use a bug in it (OSV.dev, then CISA's list of exploited bugs), no longer gets security fixes or newer release out (endoflife.date), version hidden, or not judged when we have no source for that software. A hidden version is never called safe.
4. The report is written
The top of the report is written by code, not by an AI, so it always says the same thing for the same result: the verdict, what to do first, and the eleven steps a hacker takes, each with what we did and what we found in plain words. Below that, in the part for your developer, a language model explains each finding with the evidence and the exact fix for the setup it detected (for example Cloudflare, nginx or a framework config). It is told to use only the findings it was given and never to invent vulnerabilities. If the AI is unavailable, or nothing was found, that part comes from a fixed template.
5. You or your agent fix it
Download the report as Markdown, or copy the ready-made prompt at its end into Claude Code, Codex or Cursor in your project. The prompt asks the agent to fix issues most severe first, explain each change, and ask before any DNS, hosting or secrets change. See Agent skill.
6. Rescan
Rescan after you deploy. Rescans are free. The new report starts with a comparison with your last scan of the same page: fixed, still open, new, and could not recheck. We call a finding fixed only when the same page was scanned again and its check ran. If your last scan was of a different page, its findings are listed as could not verify. Scans without an account are compared only with earlier scans from the same browser.
See one recorded scan, a fix and a rescan, step by step: How a scan works.
Where scans come from
All scans are sent from the WebSecure scan server (IPv4 46.225.187.204) with the User-Agent Mozilla/5.0 (compatible; WebSecureScanner/1.0). Scans never try passwords, never try to log in, and never run denial-of-service tests. Server ports are only checked on verified domains, with one connection per port. If you think we scanned your site without permission, write to security@hoopgram.com.
Live progress
While a scan runs, its page shows a progress bar, the check running right now, how many checks are done and how many things were found so far. When the checks finish, the report is written, usually in under 30 seconds.