Version 6.6 is BootComp, a commercial white-label WhatsApp Web CRM on the platform: embeds the vendor brand table Wl_id with 40 labels where id 8 = "BootComp", matching sigeID 8 and chromeStoreID bhkibpbnfknollmiebgefgadfkjhnaph in I enumerated every network sink across all 10 JS files (fetch, XMLHttpRequest, jQuery.ajax, axios, sendBeacon, WebSocket): the only vendor endpoints are [domain withheld], (email+token+labelID),, and, the last posting only the operator's OWN profile fetched via z.Conn("getMyUserId") when the user installs a CRM template. I also checked the WhatsApp-message channel as a non-HTTP exfil path: there is no hardcoded JID or phone literal anywhere, every @[domain withheld] id is built from the chat the user is acting on, the mass-send CampanhaClass iterates the user's own selected list, and the webhook POST in negacao. targets the user's own configured o.url with user-toggled fields. AI keys go only to [domain withheld] and [domain withheld] from a user-typed input; the "post-user" runtime message carrying {login,senha,token} is handled in-page at negacao char offset 104532 and only sets plugin_page state, never leaving the browser. The difference against the published manifest is version drift, not a hidden payload: I unzipped the stored 6.7.0 package ([domain withheld]), whose manifest is name "Extract Leads", permissions [storage, alarms, tabs], WAR matches <all_urls> and a [domain withheld] content script, exactly matching the published summary, so the listing was rebranded from BootComp to a Bing-Maps lead scraper after 6.6. One disclosure gap meets rule 7(e)(ii) and I record it here rather than suppress it: the registar-cliente POST sends personal info (own number, name, email, business profile) to [domain withheld], which is not the publisher domain on file ([domain withheld]), while the CWS disclosure is "none declared". It is the operator's own data, sent once on an explicit template install, to the white-label platform vendor, so it is below the medium severity floor for findings under a benign verdict. Contradicting signals (publisher malicious, the 96% permhash cluster, the internal note) all trace to the same 2026-09-13 malext feed import, whose own basis text says it is "not an independent code review"; the permhash is the generic [unlimitedStorage, storage, alarms, tabs] set shared by 21,929 versions and carries no information. Provenance: the provided bundle held 12 of the package's 25 files, so the missing Vite chunks (negacao, main, perfilContato, useTradutor) were taken from the stored 6.6 package [domain withheld] on local storage; the megabyte libraries were sink-swept by regex over their full text, not read line by line.