Security audit

Gemini Computer Use

Security checks across malware telemetry and agentic risk

Overview

This browser-control skill appears purpose-aligned, but it may send full browser screenshots, page URLs, and the user's task to Gemini without clear per-run disclosure or consent.

Install only if you are comfortable with browser screenshots, URLs, and task text being sent to Gemini during automation. Do not use it on banking, health, internal company, admin, or other sensitive authenticated pages unless the skill adds explicit consent, redaction, or site restrictions.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (2)

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The code captures a browser screenshot and current URL after each action and sends them back to the model service as function responses. Because there is no explicit user-facing disclosure, consent flow, or filtering, sensitive page contents, account data, internal URLs, or personal information visible in the browser can be exfiltrated to the remote AI provider.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: At startup, the script sends both the user prompt and an initial full-page screenshot to the model service. This can disclose sensitive information present on the initial page or in the user's task description without any clear warning, which is especially risky in a browser-control agent that may be used on authenticated or private web pages.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.