WORKZAP

AI Security Review

Evidence context: threat category none; evidence quality moderate.

WORKZAP is a Chrome CRM client at version 7.4.3.87, published by [email protected] and installed by four users. The manifest snapshot lists no host permissions and no API permissions, while the code itself ships content scripts and a background.js that calls fetch at line 137. What follows is 1,188 findings, and their shape matters far more than their total.

Start with what is missing. No malware signature matched. No malware family matched. No secret was lifted from the bundle. No tool-poisoning indicator fired. In a package this size, that absence is the most useful thing in the set.

The 22 network findings are patterns rather than destinations. NET-FETCH appears in background.js:137 and across a long run of content/assets/js/v_7_4_3_87_0a650f0a-*.js chunks. NET-SOCKET_IO appears in four of the same chunks, and NET-JQUERY_AJAX in two. A CRM that holds a live link to its backend uses sockets and fetch. These hits say the code talks to a server, which is the product's job. Not one of them names an endpoint worth flagging.

The endpoint list is where the extractor comes apart. border.radius.circle, btn.glass, color.border.accent.orange, calls.function.arguments.delta, above.sm and below.sm are not hosts. They are property chains and design-token strings pulled out of minified JavaScript. Of the hostnames that do exist, api.statsigcdn.com serves feature flags, api.dev.atlassian.com and api.stg.atlassian.com are Atlassian's own developer APIs, and app.wascript.com is the vendor's application host. None of these is a collection point.

The 754 code-smell findings are the same problem at scale. Those rules match ordinary JavaScript: eval-adjacent patterns, randomness calls, environment variable reads. Every bundled dependency triggers its own set, and the count multiplies.

That leaves the two obfuscation findings, which are the honest case against calling this clean. OBFUSCATION-LARGE_BASE64 points at funnil/assets/js/9cf42769-2ad5-4714-8614-9a2b4c455349.CgnacD38.js, and OBFUSCATION-FUNCTION_INDIRECT points at content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d289260.js. A large base64 string inside an assets/js folder is normally an inlined font, sprite or icon set, and indirect calls are what minifiers emit. Normally is doing real work in that sentence. Decoding the blob in the funnil chunk would settle it; until then it stays an open thread.

A skeptic would point at the publisher. An email at a domain called extensao.store, no company name, four users, and chunk files named with UUIDs and a version string. That is the profile of a throwaway upload. It is also the profile of a small shop shipping a niche CRM, and a file listing cannot separate the two. What the findings can say is narrower and more useful: nothing in this bundle reads credentials, matches a malware signature, or reports to an endpoint that is not a recognised service or the vendor's own host. Suspicion about who published it is not evidence about what it does.

Key Reasons

• Zero malware-signature, malware, secret and tool-poisoning matches across 1,188 findings
• All 22 network findings are protocol patterns (NET-FETCH, NET-SOCKET_IO, NET-JQUERY_AJAX) rather than suspicious endpoints
• The endpoint list is dominated by extractor artifacts such as border.radius.circle, btn.glass and calls.function.arguments.delta
• Named hosts resolve to legitimate services (api.statsigcdn.com, api.dev.atlassian.com) or the vendor's own app.wascript.com
• 754 of 1,188 findings are low-severity code-smell rules that fire on any bundled JavaScript

False Positive Considerations

• 754 low-severity code-smell findings from bundled/minified JavaScript
• 409 IoC entries including property chains and CSS design tokens misparsed as domains
• Network findings are protocol detections (fetch, socket.io, XHR), not endpoint detections
• High finding count in hashed bundle chunks creates multiplicative per-file matches

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