Security audit

Video Understanding

Security checks across malware telemetry and agentic risk

Overview

This skill transparently uses Google Gemini and video download tools to analyze user-provided videos, with privacy-sensitive behavior that is disclosed and mostly user-controlled.

Install only if you are comfortable sending video content or video URLs to Google Gemini. Avoid confidential, regulated, private, or unlisted videos unless that processing is acceptable for your account and organization; use the documented purge option after sensitive follow-up sessions.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 95% confidence
Finding: The skill invokes shell commands, reads environment variables, and can read/write local files, yet it does not declare corresponding permissions. This creates a transparency and policy-enforcement gap: users or orchestrators may not realize the skill can download remote content, access secrets like GEMINI_API_KEY, and persist cache/output files locally.

Vague Triggers

High

Confidence: 91% confidence
Finding: The trigger rule is extremely broad, causing this skill to activate for nearly any mention of a video, including casual requests. In context, that is dangerous because the skill may download third-party URLs or upload local/user-provided video content to Google Gemini without a clear, narrowly scoped handoff or fresh consent.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The skill does not prominently warn users that non-YouTube URLs and local video files may be uploaded to Google Gemini and that remote handles/cache entries may persist temporarily. This can expose sensitive video contents, embedded text, voices, or metadata to a third party without informed consent, especially during repeated follow-up analysis.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The script uploads local or downloaded video files to Gemini for analysis without any explicit consent flow or warning that media will be transmitted to a third-party AI service and may be retained temporarily. In a video-understanding skill, users may provide sensitive recordings, making undisclosed remote processing a meaningful privacy and compliance risk.

Missing User Warnings

Low

Confidence: 84% confidence
Finding: Direct YouTube analysis forwards the video URL to Gemini without warning the user that a third-party service will access or process that resource. The privacy impact is lower than uploading a local file, but it still creates undisclosed external data sharing and may reveal user interests or private/unlisted links.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill persists prior model outputs and session context in a local cache directory without any explicit notice, retention controls surfaced to the user, or sensitivity checks. Because outputs may contain transcripts, descriptions, and answers derived from potentially private videos, silent persistence increases data exposure on the host system.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.