web-access
Audited by Socket on Oct 20, 2026
3 alerts found:
AnomalySecurityx2SUSPICIOUS: the skill's purpose matches browser automation and authenticated web access, but its trust footprint is broad. The main risks are the unverifiable local proxy/script chain, arbitrary DOM/eval control over the user's real Chrome session, optional third-party content routing via Jina, and autonomous actions like uploads/clicks in logged-in sites. No clear evidence of credential theft or malicious exfiltration is present, but this is a high-impact capability set that deserves medium-to-high caution.
This package/module is a powerful local Chrome DevTools Protocol proxy that exposes endpoints to (1) execute arbitrary JavaScript in the browser via Runtime.evaluate (/eval) and (2) perform file writes on the host via /screenshot using a user-controlled path. It also allows user-driven navigation, clicking, and file input injection into the browser. While there is no obvious obfuscated payload or external C2 in the shown fragment, the capability set is consistent with a remote-control/sabotage tool if the HTTP interface is reachable by an attacker. Additionally, the provided fragment appears truncated in enablePortGuard, limiting complete assessment. If used, it should require strong authentication/authorization and enforce strict allowlists/safe paths; otherwise, it presents a significant security risk.
This fragment describes a local HTTP-to-CDP automation proxy with multiple high-impact primitives: arbitrary in-page JavaScript execution (/eval), navigation to caller-supplied URLs, simulated user interactions (/clickAt), and explicit bridging of local filesystem content into the browser via /setFiles, plus a caller-influenced filesystem write via /screenshot(file=...). If the service is reachable by untrusted parties or lacks strict authentication/authorization and strong validation (localhost binding, URL allowlisting, and path confinement for file inputs/screenshots), it could be repurposed for data theft/exfiltration and other malicious browser automation outcomes. No direct evidence of packaged malware is shown—this is a capability-based risk assessment driven by the documented API surface.