ZAPARETO

AI Security Review

Evidence context: threat category none; evidence quality moderate.

ZAPARETO describes itself as a CRM for Chrome and ships as version 7.4.3.86. The code is a bundled web app, with most of the weight in content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2*.js and a small background.js at the root.

The 22 network findings are library call sites. They mark where bundled code dials out. NET-SOCKET_IO fires at content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb212.js:15 and content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2148.js:64, NET-JQUERY_AJAX at content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb223.js:13, and NET-FETCH at content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2136.js:1 and background.js:137. Socket.io plus jQuery in a CRM client is what a live chat agent is built from, and the file naming pattern, a version string followed by a UUID, is what a release pipeline writes.

The IoC list shows the same signature of noise. Reading the entries turns up border.radius.circle, color.border.accent.orange, btn.glass, calls.function.arguments.delta, above.sm and below.sm. Those are CSS token names and JavaScript property chains, not hosts. Of the endpoints that could resolve, api.dev.atlassian.com and api.statsigcdn.com belong to Atlassian and Statsig, both ordinary SaaS dependencies. app.wascript.com is the one endpoint that might be this product's own service, which matches the description of a CRM talking to a backend.

Two obfuscation findings need naming. OBFUSCATION-LARGE_BASE64 in funnil/assets/js/ed6113f2-bc09-4421-a453-bbbd66fe9488.DCssQ7AT.js:1 marks an inlined base64 blob inside a content-hashed asset file, which is how builds embed fonts and images. OBFUSCATION-FUNCTION_INDIRECT in content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2259.js:1 matches the indirect call pattern minifiers emit. Neither is a packed loader, and nothing in the report decodes strings at runtime or reaches eval.

No malware signature matched anywhere. Nothing reads cookies, session tokens or login pages. The 754 code-smell items are floor noise: they fire on fetch, crypto and environment access, which every bundled dependency contains.

The strongest counterargument is the manifest. ZAPARETO declares no permissions and no host permissions, which sits oddly with a CRM that runs content scripts and calls fetch from background.js:137. A skeptic would say the permission data is missing, so nobody can see what the extension reaches. That reading is fair, and it is also why the code's behavior carries the weight here: the findings show library calls, asset embedding and minifier patterns, and none of the things a malicious wrapper would carry, such as cookie reads or a hardcoded collection endpoint.

A CRM extension that talks to its own backend over socket.io, bundles jQuery and inlines a base64 asset is doing what its description claims. The volume comes from a compressed bundle tripping extractors that read property chains and library names as indicators.

Key Reasons

• Zero malware signatures and zero secret or credential findings across 1185 items; every medium-severity hit is a library call site (NET-SOCKET_IO, NET-JQUERY_AJAX, NET-FETCH) inside bundled content/assets/js release files.
• The 406 IoCs are extractor garbage: CSS token chains and property paths such as border.radius.circle, btn.glass, above.sm and calls.function.arguments.delta rather than hostnames.
• Both obfuscation hits are build artifacts: an inlined base64 asset in funnil/assets/js/ed6113f2-bc09-4421-a453-bbbd66fe9488.DCssQ7AT.js and a minifier indirect-call pattern in content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2259.js.
• The only resolvable endpoints are known SaaS (api.dev.atlassian.com, api.statsigcdn.com) plus the product's own app.wascript.com.
• The 754 code-smell findings are low-severity pattern matches that fire on any bundled dependency.

False Positive Considerations

• Minified bundle triggering 754 low-severity YARA code-smell rules
• XIOC extractor reading CSS class names and object property chains as domains
• Obfuscation heuristic firing on inlined base64 assets and minifier call patterns
• Network findings originating from bundled jQuery, socket.io and fetch libraries instead of extension logic

Reviewed 2026-10-10; recommended action: monitor; model confidence 72%.