BootComp

AI Security Review

Evidence context: threat category none; evidence quality moderate.

BootComp is a CRM layer for WhatsApp Web. Content scripts inject into the chat page to organise conversations, and the background worker talks to a backend over socket.io and fetch. What drove this review is finding volume, not finding behaviour.

Each of the 22 network findings is a generic client pattern. NET-JQUERY_AJAX sits in content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d28923.js, NET-SOCKET_IO in the ...12.js and ...149.js bundles, and NET-FETCH in background.js:137 along with a dozen content-script files. A chat CRM needs a live socket and REST calls to its own service, and no call site here names a destination.

The destination list shows the noise plainly. above.sm, below.sm, border.radius.circle, calls.function.arguments.delta and color.border.accent.orange are property chains and design tokens that the domain extractor read as hostnames. The real services in that list are api.statsigcdn.com, a feature-flag and analytics CDN, and app.wascript.com, whose naming matches a WhatsApp automation product.

Nothing in the bundle matched a malware signature. There are no credential-theft indicators, no hijack domains, and no proxyware or exit-node functionality.

The two obfuscation hits are OBFUSCATION-LARGE_BASE64 in funnil/assets/js/9cf42769-2ad5-4714-8614-9a2b4c455349.CgnacD38.js:1 and OBFUSCATION-FUNCTION_INDIRECT in content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d289260.js:1. Encoded blobs of that kind sit in most bundled front ends holding fonts, icons or a compiled library, and indirect calls are what a minifier emits. The 754 code-smell entries are the standard pattern matches that fire on any non-trivial JavaScript. The publisher is an email address, [email protected], rather than a company, so there is no brand to check against either.

The strongest argument against this reading is the Atlassian cluster in the endpoint list. api.atlassian.com, api.dev.atlassian.com, api.stg.atlassian.com, api.atlassian-us-gov-mod.com and atlassian-isolated.net span a whole corporate tenancy matrix, which is strange company for a WhatsApp CRM. If BootComp opened sessions to those hosts it would be reaching into unrelated corporate workspaces. No network finding sits at a call site that names any of them, and the same list carries junk like artboard.phone and be.audio, so the extractor harvested strings from a lookup table or a bundled blocklist rather than from code that runs. Runtime capture of the socket.io handshake and the fetch in background.js would settle it.

The second item worth resolving is the base64 payload under funnil/. That file carries a hashed name and nobody has decoded the blob yet, so its contents remain unknown.

The manifest data collected here records no permissions and no host permissions, so the scope of content-script injection is not confirmed by this review. Nothing in the evidence describes reading browser history, substituting search, or shipping cookies off-device.

Key Reasons

• Zero malware-signature, malware, secret, dependency and tool-poisoning findings across the whole scan
• All 22 network findings are generic client patterns (NET-JQUERY_AJAX, NET-SOCKET_IO, NET-FETCH) with no named destination at the call site
• The 409 IoC entries are dominated by parser artifacts such as border.radius.circle and calls.function.arguments.delta
• The 754 code-smell entries come from bundled content scripts and carry no behavioural signal on their own
• No browser-hijack, credential-theft, proxyware or typosquatting indicators in the manifest, endpoints or file names

False Positive Considerations

• IoC extractor harvested property chains and design tokens (border.radius.circle, calls.function.arguments.delta, artboard.phone) as domains
• 754 code-smell matches generated by bundled and minified content-script JavaScript
• Obfuscation hits match standard bundler output: an embedded base64 asset and indirect function calls
• Finding volume inflated by dist-style bundles in content/assets/js and funnil/assets/js

Reviewed 2026-10-10; recommended action: no action; model confidence 72%.