How CRXaminer works

CRXaminer shows what a Chrome extension is able to do before you install it. It reads the extension's package and its Chrome Web Store listing. It never runs the extension.

What a scan does

  1. Reads the store listing: name, developer, users, ratings, badges and last update.
  2. Downloads the package from Google and reads its manifest.json, the file where an extension declares what it needs. It also lists the URLs that appear in the extension's files.
  3. Applies the rules to the manifest to rate its access.
  4. Runs an AI check that compares what the extension asks for with what its listing says it does.

Reading the result

  • Access is what the extension could do if it turned bad. Fixed rules decide it, so the same manifest always gets the same tier.
  • Trust is what the listing shows about who publishes it and how it is received.
  • The overall level combines the two, with a one-line headline and advice.
  • A low level does not prove an extension is safe. CRXaminer does not review the extension's code.
  • Scans are saved, so everyone sees the same result. A scan older than 24 hours can be run again from its page.

Access: what it could do

The rules read only the manifest. They rate what the extension is allowed to do, assuming the worst, not what its code actually does. Each finding has a tier, and the extension's access is the highest tier found. Nothing is added up.

Full control

It could take over accounts, traffic or other parts of the browser or computer.

  • All sites plus cookies. It can read the cookies that keep you signed in on every website, so it could take over your accounts.
  • All sites plus blocking web requests. It can watch, block and change the traffic of every website while it happens.
  • Scripts inside every website's own code. Its scripts run in the page's own JavaScript on every website, where they can change what the page's code does.
  • Updates from outside the Chrome Web Store. Its updates would come from a server other than the Chrome Web Store, so they skip the store's review.
  • The debugger permission. It can take full control of the pages it attaches to, including sites where you are signed in.
  • The management permission. It can switch off or remove your other extensions, including security tools.
  • The nativeMessaging permission. It can hand work to programs on your computer, which Chrome's protections do not limit.
  • The proxy permission. It can send all your browser traffic through a server it chooses.
Broad reach

It could see or change a great deal of what you do in the browser.

  • Access to all websites. It can read and change every website you visit, including sites where you are signed in.
  • Scripts on all websites. Its scripts run automatically in every website you open.
  • Text can be run as code. Its policy allows 'unsafe-eval' in its own pages, which makes it easier to run code that was never reviewed.
  • Changes your search engine. Searches you type in the address bar go to a site it names, which sees every search.
  • 9 permissions

    browsingData, clipboardRead, contentSettings, desktopCapture, history, privacy, tabCapture, webAuthenticationProxy, webNavigation

Scoped

It could reach particular data or sites, not everything.

  • Access to sensitive sites. It asks for access to named sites such as banks, payment services, email or code hosting.
  • Changes your homepage or startup pages. It sets the page Chrome opens at startup or for the home button.
  • Replaces Chrome pages. It shows its own page instead of Chrome's new tab, history or bookmarks page.
  • Scripts inside named sites' own code. Its scripts run in the page's own JavaScript on the sites it names, where they can change what the page's code does.
  • Messages from many web pages. Pages on whole groups of subdomains, or over plain http, can send messages to the extension.
  • Every file open to all websites. Any website can load every file in the extension.
  • Google account data. It asks for access to data in your Google account, such as mail, files or calendar.
  • 22 permissions

    bookmarks, clipboardWrite, cookies, declarativeNetRequest, declarativeNetRequestWithHostAccess, downloads, downloads.open, geolocation, identity, identity.email, input, pageCapture, readingList, scripting, sessions, tabs, topSites, userScripts, vpnProvider, webRequest, webRequestAuthProvider, webRequestBlocking

Routine

It only asks for things most extensions need.

  • Access to local files. It can read files on your computer, but only if you turn on file access for it.
  • WebAssembly allowed. Its pages may run WebAssembly; this is the standard way to do so in Manifest V3.
  • Manifest V2. It uses the older Manifest V2 format, which has fewer restrictions and is being phased out.
  • Messages from named websites. The websites it lists can send messages to the extension; a flaw in how it handles them could be used by those sites.
  • Files open to all websites. Any website can load some of its files, which lets sites detect that you have it.
  • Google sign-in. It can sign you in with your Google account and read your name and email address.
  • 45 routine permissions

    accessibilityFeatures.modify, accessibilityFeatures.read, activeTab, alarms, audio, background, certificateProvider, contextMenus, declarativeContent, declarativeNetRequestFeedback, dns, documentScan, downloads.shelf, downloads.ui, enterprise.deviceAttributes, enterprise.hardwarePlatform, enterprise.networkingAttributes, enterprise.platformKeys, favicon, fileBrowserHandler, fileSystemProvider, fontSettings, gcm, idle, loginState, notifications, offscreen, platformKeys, power, printerProvider, printing, printingMetrics, processes, search, sidePanel, storage, system.cpu, system.display, system.memory, system.storage, tabGroups, tts, ttsEngine, unlimitedStorage, wallpaper

Minimal

The rules found nothing to flag.

Optional permissions are rated one tier lower, because Chrome asks you before granting them. On a scan page, the Manifest tab shows the line behind each finding.

Trust: who stands behind it

Trust comes from the store listing. The scan page lists the signals it saw.

Established

A recognised publisher or a large, long-standing user base.

The store's Established publisher badge and 100,000 users or more, or 1,000,000 users and 1,000 ratings or more.

Some track record

Enough users and ratings, or a store badge, to show it is in real use.

The Established publisher or Featured badge, or 10,000 users and 50 ratings or more.

Unproven

Few users or ratings, or no listing data, so there is little to go on.

Less than that, or no listing data.

Warning signs

Something about the listing or its stated purpose does not add up.

The AI check found serious warning signs, or raised concerns about an extension that was already unproven.

Few users, few ratings, a low rating or no update for 2 years are shown as cautions but do not lower trust on their own.

How they combine

More access or less trust never lowers the overall level.

Overall level by access tier (rows) and trust level (columns)
Access ↓ · Trust → Established Some track record Unproven Warning signs
Full control Medium High Critical Critical
Broad reach Low Medium High Critical
Scoped Low Low Medium High
Routine Minimal Minimal Low Medium
Minimal Minimal Minimal Minimal Low

The AI check can lower trust, never raise it, and it never changes access.

For the full story, read the CRXaminer write-up on Astarte.