Skip to content

Security 

Walkthru drives a real browser on sites you care about. Here is exactly what it can touch, what it cannot, and how to reach us if something looks wrong.

Last updated

How we think about it

  • Your browser, your session. Tests run in the tab you are already signed in to. You never give us a password for your site.
  • Least data. The extension sends a masked text outline of the page, not the page itself.
  • Rules in code. Safe mode, site boundaries and step limits are enforced by deterministic code on both the extension and the server, not by the AI.
  • Passive checks only. We never send attack payloads to your site.

Extension permissions

Every permission the extension asks for, and why:

sidePanel
Shows the test controls and live steps beside the page.
activeTab and tabs
Reads the address of the tab you start a test in, and follows it as the test user navigates.
scripting
Adds the page reader and Scout to that tab when a test starts. Nothing runs on other tabs.
Screenshot access (asked once)
Chrome only allows screenshots of a tab with broad site access. It is requested when you first start a test, and used only for the tab under test.

A run stops if the test user leaves the site you started on, after 12 steps or after 4 minutes.

Where test users may act

The server decides, before a run starts and again at every step. The AI cannot change it, and neither can a modified copy of the extension, because every step goes through our API.

  • Visitor mode (any site you have not verified): read, scroll, click links, menus and tabs, and use the site's search box. No other form fields, no signing in, and no liking, following, posting, messaging, buying or adding to a cart. A page that shows a signed-in account ends the run.
  • Owner mode (a domain you verified with a meta tag, a file or a DNS record, checked again at every run): what a real visitor does, with safe mode still blocking payments, deletions and cancellations.
  • Refused: banking and payments, trading and crypto, government and tax, healthcare portals, webmail, social networks, dating, adult, gambling, and password managers or sign-in providers, unless you verified that exact domain.

Test users never solve or bypass CAPTCHAs or bot protection; you solve them yourself, or the run stops and the report says so plainly. Test runs against one unverified site are limited per hour, and every run is logged with its actions (never typed values) for 90 days.

What our scanner does

The security check is a hygiene review, not a penetration test. It reads response headers, https redirects, the certificate from one normal TLS handshake (plus one that offers only TLS 1.0 and 1.1), your CAA record over DNS and one ordinary request with an Origin header to see how CORS answers. On verified domains only, it also reads a list of well-known public paths (environment files, backups, database dumps, server status pages), your JavaScript bundles and their source maps, and the DNS of subdomains your pages link to. Every request is an ordinary read. It follows at most 10 same-site links (50 on paid plans) while honouring robots.txt.

Checks for exposed files, keys leaked into JavaScript, public source maps and subdomain takeover only run on domains whose owner has verified them. The scanner refuses private and local network addresses, and re-checks every redirect before following it.

Protecting your data

  • Production web and API traffic uses HTTPS. Private API calls require your signed-in session; public scans and shared reports have separate access rules.
  • Database rows use row-level security. Private runs are owner-only; public links and team workspace sharing grant separate, deliberate access. Removing a workspace member removes their workspace access.
  • Website sign-out requests extension disconnect in this browser and warns if it cannot be confirmed. Account changes stop further actions; an action or request already sent may finish. End an unfinished server run from its original account's dashboard.
  • Private API session checks are cached for up to five minutes, never past token expiry. Supabase access tokens may remain valid until expiry after sign-out; logout does not revoke public links or workspace access.
  • Screenshots are stored privately and shown through links that expire after an hour. They are deleted after 30 days.
  • Deleting your account removes stored files first, then run history, then the account, so nothing is left behind unreachable.
  • API keys for the MCP server and deploy-hook URLs are shown once and stored only as SHA-256 fingerprints. Each key can reach only its owner's runs, every call re-checks the plan, and a key can be revoked at once from the MCP page. Deploy hooks run at most one check every 10 minutes.

Report a vulnerability

If you find a security problem in Walkthru, email [email protected] with the steps to reproduce it and the impact you expect. We will confirm we received it within three business days and keep you updated until it is fixed.

Please test only against your own account, avoid touching other people's data, and give us a reasonable time to fix the issue before sharing it. We will not take action against good-faith research that follows these rules.

Report abuse or opt your site out

If Walkthru was used against your site, or you want test users and our scanner kept off it, email [email protected] with the domain. We act within 24 hours. How our scanner identifies itself is on the Walkthru bot page.