Security audit

Meta Video Ad Analyzer

Security checks across malware telemetry and agentic risk

Overview

This skill appears to do the advertised video ad analysis, but users should treat media sent through it as cloud-processed and manage generated thumbnails carefully.

Install only if you are comfortable processing ad videos or images with Google cloud AI services. Use a dedicated least-privilege Google service account, avoid confidential or regulated media unless your policies allow it, and ensure generated thumbnails under static/thumbnails are not publicly exposed or retained longer than needed.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 92% confidence
Finding: The skill advertises capabilities that imply reading environment variables, accessing local files, and invoking shell-dependent tooling like ffmpeg, but it does not declare corresponding permissions or trust boundaries. This creates a transparency and governance gap: users or platforms may authorize the skill without understanding that it can access credential paths, process arbitrary local media files, and execute external binaries.

Description-Behavior Mismatch

Medium

Confidence: 84% confidence
Finding: The code persists thumbnails derived from user-supplied media into a web-served static directory, creating data retention and potential unauthorized disclosure risk. In an agent skill context, silently writing analyzed content to disk can expose sensitive creative assets or user-provided imagery beyond the original processing intent.

Description-Behavior Mismatch

Medium

Confidence: 84% confidence
Finding: Image thumbnail generation saves user-provided image derivatives to disk under a static path, which extends processing beyond transient analysis and may leak sensitive input content. Because the skill advertises analysis functionality, this hidden persistence increases privacy and exposure risk in normal use.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill sends video frames, images, and audio/transcripts to external Google AI services such as Gemini Vision and Google Cloud Speech, but the description does not clearly warn users that potentially sensitive media content leaves the local environment. This can lead to unintentional disclosure of confidential, regulated, or personal data contained in ads, uploaded videos, overlays, or speech.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill sends video frames, images, OCR-derived text, and potentially whole videos to external AI services without any point-of-use consent or warning in this code path. This is dangerous because uploaded ad creatives may contain confidential, regulated, or user-sensitive content, and transmission to third-party processors changes the trust boundary.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.