SellerHub360

AI Security Review

Evidence context: threat category none; evidence quality moderate.

SellerHub360 is a WhatsApp Web CRM: a content script that layers a sales dashboard over the chat interface. The manifest as extracted shows empty permission and host permission arrays, so the behavioural evidence sits in background.js and the bundled scripts under content/assets/js/.

The network findings record library behaviour. NET-FETCH-background.js-137 and NET-FETCH on content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2136.js:1 mark fetch() calls. NET-SOCKET_IO on v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb212.js:15 and ...bb2148.js:64 mark socket.io client code. NET-JQUERY_AJAX on ...bb224.js:181 marks jQuery's ajax transport. A CRM that live-syncs WhatsApp conversations needs a WebSocket plus an HTTP client, and socket.io is the standard choice. None of these findings identify a destination host.

The endpoint list is where the extractor falls apart. 'border.radius.circle', 'btn.glass', 'calls.function.arguments.delta' and 'avatar.online' are CSS keywords, API field paths and property chains that the string parser promoted to domains. '4g.ve', 'above.sm', 'ae.online' and 'af.pf' are two-letter fragments of the same kind. The entries that resolve to something real are app.wascript.com, the vendor's own service, api.statsigcdn.com, a feature-flag CDN, and a set of api.*.atlassian.com hosts that sit inside bundled client libraries.

The two obfuscation alerts are minification artifacts. OBFUSCATION-FUNCTION_INDIRECT on content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2259.js:1 flags indirect function invocation, which is how every webpack and esbuild bundle calls modules through its runtime. OBFUSCATION-LARGE_BASE64 on funnil/assets/js/ed6113f2-bc09-4421-a453-bbbd66fe9488.DCssQ7AT.js:1 flags an embedded base64 blob, the usual shape of a bundled font, icon sprite or inlined image. Neither is the deliberate string scrambling that hides intent.

What is absent carries weight. No malware signature matched, no secret was recovered, no tool-poisoning indicator fired, and no dependency finding is present. The remaining volume is code-smell, which fires on ordinary JavaScript patterns, plus IoC strings of the kind already shown to be garbage.

A skeptic could argue the opposite case. The publisher is an email address rather than a named company, the file names are content-hashed and opaque (v_7_4_3_86_085ff17f-...), and a base64 blob plus indirect calls is also what a hidden payload looks like. That reading fails on specifics: hidden payloads normally arrive with at least one credential-targeting indicator, one non-generic external host, or a malware family match, and this extension has none. The hashed names share one consistent version prefix across the whole content/ tree, which is what a build pipeline produces. What would change the picture is a socket.io endpoint resolving outside wascript.com, or a manifest entry granting access to sites beyond web.whatsapp.com.

Key Reasons

• No malware signature, secret, dependency or tool-poisoning finding anywhere in the extension.
• Network findings are framework detections (socket.io, fetch, jQuery ajax) in bundled content scripts, with no destination host attached.
• Endpoint list is dominated by parser garbage such as 'border.radius.circle' and 'calls.function.arguments.delta'; the real hosts are app.wascript.com, api.statsigcdn.com and Atlassian API strings inside bundled libraries.
• Both obfuscation alerts match standard webpack/esbuild output: an embedded base64 asset and indirect module calls.
• Behaviour matches the stated product; a WhatsApp Web CRM needs a live socket and an HTTP client.

False Positive Considerations

• IoC extractor promoted CSS values and JavaScript property chains to domains
• Hundreds of low-severity code-smell hits on minified JavaScript
• Content-hashed bundle files under content/assets/js/ multiply framework detections
• Obfuscation rules firing on webpack output rather than deliberate string hiding

Reviewed 2026-10-13; recommended action: suppress false positive; model confidence 75%.