Gemini Computer Use

• Outcome: Pass
• Latest audit: May 27, 2026, 8:59 PM
• Version:
  1.0.0

Overview

This skill does what it claims: it runs a Gemini-powered browser automation loop, but users should know browser screenshots and URLs are sent to Gemini.

Install only if you are comfortable with a third-party Gemini model seeing the browser page, URL, and your prompt. Use a sandboxed browser profile, avoid logged-in or sensitive sites unless intentional, keep turn limits low, and use --exclude to block actions you do not want automated.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (2)

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The code captures a live browser screenshot and current URL after each action and sends both to the remote Gemini model as function responses. This can expose sensitive on-screen data, session state, internal URLs, or user-entered content without any explicit notice, consent flow, or redaction, which is especially risky in a browser-control agent handling arbitrary websites.

Missing User Warnings

Medium

• Confidence: 98% confidence
• Finding: The initial request sends both the user's prompt and a full browser screenshot to the external model immediately after page load, with no warning that page contents will leave the local environment. In a computer-use skill, screenshots may contain credentials, personal data, confidential documents, or authenticated application state, making silent transmission a real privacy and data-leak risk.

VirusTotal

65/65 vendors flagged this skill as clean.