YouTube Chinese Subtitle Burn-in

• Outcome: Pass
• Latest audit: May 28, 2026, 6:50 AM
• Version:
  0.1.0

Overview

This is a disclosed YouTube subtitle and video-processing skill that uses expected media tools and project files, with no evidence of hidden data access or destructive behavior.

Install this only if you want an agent to download/process YouTube media, run ffmpeg/yt-dlp, and create local subtitle/video outputs. Check that you have rights to transform the source video, specify Chinese-only versus bilingual output up front, and review any proposed reusable feedback updates before repackaging the skill.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (4)

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

print(f"Window: start={start:.3f}s duration={duration:.3f}s subtitle_cues={len(selected)}")
            print(f"Render mode: {mode}")
            return 0
        result = subprocess.run(
            [
                "ffmpeg",
                "-y",

• Confidence: 87% confidence
• Finding: result = subprocess.run( [ "ffmpeg", "-y", "-ss", f"{start:.3f}", "-t", f"{duration:

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 95% confidence
• Finding: The skill directs the agent to read and write many project files and invoke shell tools like ffmpeg, yt-dlp, and Python scripts, but it does not declare corresponding permissions. That creates a capability/permission mismatch where operators or policy layers may underestimate what the skill can do, reducing transparency and weakening review and containment.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The invocation description is broad enough to match many generic requests about subtitles, translation, thumbnails, descriptions, and feedback handling, not just the narrowly intended YouTube burn-in workflow. Over-broad routing can activate a shell-capable, file-writing skill in contexts where a simpler, lower-privilege skill would suffice, increasing the chance of unnecessary downloads, file modification, or processing of untrusted inputs.

Natural-Language Policy Violations

Medium

• Confidence: 95% confidence
• Finding: The workflow explicitly defaults to producing Simplified Chinese hard-subtitled output for any user who provides only a URL, without first obtaining clear consent for translation, subtitle language, or media modification. This can cause unauthorized transformation of third-party content, user surprise, and accidental disclosure or publication of altered outputs that the user did not intend to create.

VirusTotal

VirusTotal findings are pending for this skill version.