Every shipped script is small, unminified and does exactly what the listing describes: a new-tab page with 10 local wallpapers ( loads), a clock (:165), shortcut tiles, and search via chrome.search.query (:93) using the user's default engine. No content scripts, no APIs, no data ever leaves the browser — the only outbound requests are two GETs to the publisher's own [domain withheld] (:10,:41) carrying no payload, plus Google Fonts and Google's favicon service. The one real capability worth noting is that both endpoints return server-controlled content that is rendered into the new-tab page (:54 `popupContainer.innerHTML = html`, and the shortcut tiles built from remote JSON at:238-248), which the publisher could later use to push ads or phishing tiles; it is not remote *code* — the manifest declares no CSP, so MV3's default `script-src 'self'` blocks both `<script>` in innerHTML and jQuery's script-eval on.append(), and inline event attributes are blocked too. I tried to retrieve both payloads to check for redirects and both returned a Cloudflare managed-challenge 403, so the channel is unexercised rather than cleared. This contradicts the ML score and the permhash cluster (the cluster is generic new-tab permissions, `search` + one first-party host, so it is a template signal rather than a behavioural one) and agrees with chrome-stats and the publisher's 1% sibling-malicious rate; the malext-feed note is an unreviewed bulk import, not a code review. Confidence is held at medium rather than high because is referenced by:17 but absent from the extract, so ~87 KB of the reported 105 KB of JS was not read, and one blocked fetch cannot rule out geo- or time-varied payloads.