ZAPSUPREMO

AI Security Review

Evidence context: threat category none; evidence quality moderate.

ZAPSUPREMO lists itself as "um CRM no Google Chorme" and ships as version 7.4.3.87 with 41 users. Its network-capable code sits in content/assets/js/ and background.js. Three patterns repeat across it: NET-FETCH in background.js:137 and in most of the v_7_4_3_87_0a650f0a-*.js files, NET-JQUERY_AJAX in ...23.js and ...24.js, and NET-SOCKET_IO in ...12.js, ...149.js, ...175.js and ...66.js. A persistent socket plus fetch calls is what a chat-driven CRM looks like. The scanned manifest declares no permissions and no host permissions, which is unusual for that design and is the main gap in the data.

Two obfuscation findings sit in the bundle: 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. A base64 block in a hashed filename and an indirect-call wrapper are worth naming. Both are also ordinary build output: webpack and esbuild emit hashed names, and inline base64 is how bundlers embed fonts, icons and small data. Neither finding is paired with a malware signature. Every malware and malware-signature bucket came back at zero, as did secrets, dependencies and tool-poisoning.

The endpoint list shows why count-based triage fails here. Entries like border.radius.circle, btn.glass, color.border.accent.orange and calls.function.arguments.delta are CSS values and minified property chains the extractor read as hostnames. above.sm, ae.jobs, af.pf and be.audio are two and three letter fragments from the same source. The endpoints that are real are app.wascript.com, which matches the WhatsApp-facing behaviour a Brazilian CRM needs, and api.statsigcdn.com, a feature-flag CDN. The 422 IoC findings and the 754 code-smell findings track the volume of minified JavaScript in content/assets/js/, nothing else.

A skeptic can point at the versioned filenames, the hashed file under funnil/, and a publisher whose only identity is [email protected], a storefront address rather than a company. An extension holding a live socket with no declared permissions could be shipping session data somewhere the manifest does not admit. That concern is real. Runtime capture of the socket.io connection and of the fetch targets in background.js would settle it, and static scanning has taken this as far as it goes. What the static evidence lacks is everything that separates a bundled app from malware: no credential access finding, no cookie handling, no dynamic URL template construction, no obfuscated payload co-located with a malware signature.

Sitting alongside the noise are two facts that keep this from being a clean bill of health. The obfuscation flags are both rated medium, and the manifest permissions did not come through in the scan. Neither is evidence of harm, and neither outweighs a completely empty malware bucket across a 1,201-finding analysis.

Key Reasons

• Zero malware and zero malware-signature findings across 1,201 total findings
• IoC list contains extractor artifacts (border.radius.circle, btn.glass, calls.function.arguments.delta, above.sm) rather than real hosts
• Network findings are generic library detections (NET-JQUERY_AJAX, NET-SOCKET_IO, NET-FETCH) in minified content/assets/js/ bundles
• Both obfuscation flags (OBFUSCATION-LARGE_BASE64 in funnil/, OBFUSCATION-FUNCTION_INDIRECT) match standard bundler output with no co-located malware signature
• No credential, cookie, secret or dynamic URL template findings despite a CRM description

False Positive Considerations

• IoCs extracted from minified CSS tokens and property chains (border.radius.circle, calls.function.arguments.delta, btn.glass)
• 754 low-severity code-smell findings generated by bundled JavaScript in content/assets/js/
• Obfuscation flags on normal build output: hashed filenames and inline base64 in funnil/
• Generic network detections on standard jQuery ajax, fetch and socket.io calls

Reviewed 2026-10-11; recommended action: monitor; model confidence 70%.