video-overlay-cleanup-agent

• Outcome: Pass
• Latest audit: May 29, 2026 · 8:38 PM UTC
• Version: 1.0.1

Overview

This is a coherent video-cleanup skill with local FFmpeg workflows and an explicit Gemini mode that uploads selected frames and masks when used.

Install if you are comfortable with local video file processing and, when using Gemini mode, uploading selected video frames and masks to Google Gemini. Prefer the local removelogo path for sensitive recordings, use trusted custom editor commands only, and choose output/work directories carefully because intermediate files may be overwritten or cleaned up.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

usage_output=str(usage_path) if usage_path else "",
        )
        try:
            subprocess.run(shlex.split(command), check=True)
        except subprocess.CalledProcessError as exc:
            raise SystemExit(
                f"Frame editor failed on frame {frame.name} with exit code {exc.returncode}: {command}"

• Confidence: 93% confidence
• Finding: The script executes a user-supplied command template via subprocess after formatting it with file paths and other values. Although shell metacharacter injection is reduced by avoiding shell=True, this still permits arbitrary program execution because --editor-cmd is fully attacker-controlled and is run once per frame, which can lead to destructive local command execution or large-scale abuse if untrusted input can reach this parameter.

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The script sends local image frames and masks to the external Gemini API for processing, which creates a real data-exfiltration/privacy boundary not evident from a skill framed mainly as local video-overlay cleanup. If users process sensitive recordings, screen contents may leave the local environment and be exposed to third-party retention, logging, or policy risks.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The skill instructs users to send video frames to Gemini for masked repair but does not clearly warn that frames may be transmitted to a third-party provider and may contain sensitive visual data. Users cleaning screen recordings are especially likely to expose notifications, status bars, messages, account details, or other personal/business information unintentionally.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The documentation explicitly states that original frames and masks are sent to Gemini, but it does not warn users that this may transmit potentially sensitive visual data to a third-party service. In a video-cleanup skill, frames can contain personal information, confidential screens, notifications, or watermarks, so omission of a privacy/data-transmission warning can lead to unintended disclosure.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The script invokes ffmpeg with the `-y` flag, which forces overwriting output files without prompting. In this skill, the output directory and filename pattern are user-controlled, so running the script against an existing frame directory can silently destroy prior extracted frames or unrelated matching files, causing data loss and making recovery difficult.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The code uploads image data to an external service without any explicit user-facing disclosure or consent mechanism in the script. For a tool likely to process screen recordings, this is dangerous because captured frames can contain passwords, chats, account data, or other sensitive visual information that users may assume remains local.

VirusTotal

65/65 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.