ai-subtitle-generator

• Outcome: Review
• Latest audit: May 29, 2026 at 9:35 PM UTC
• Version:
  1.0.0

Overview

The skill is not clearly malicious, but its subtitle workflow asks users to enable a broad paid SkillBoss API setup that goes well beyond subtitle generation.

Review the remote SkillBoss setup before running it, use spending limits or a restricted key if available, and only submit audio or video you are comfortable sending to SkillBoss and downstream model providers. For subtitle-only work, prefer a workflow limited to transcription/subtitle models rather than enabling the full SkillBoss API surface.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• 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
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Description-Behavior Mismatch

Medium

• Confidence: 96% confidence
• Finding: The skill is presented as a narrow subtitle-generation tool, but its setup instructs the agent to install and auto-configure a much broader platform with hundreds of unrelated APIs. This violates least privilege and increases the chance that user data, credentials, or future agent actions are routed through an overbroad third-party integration far beyond the stated task.

Description-Behavior Mismatch

Medium

• Confidence: 92% confidence
• Finding: The recommended models and agent instructions prominently suggest general chat LLMs for a subtitle-generation skill, which is inconsistent with the stated purpose and can cause agents to send audio/transcription tasks or user content to inappropriate models or workflows. This mismatch can lead to incorrect handling of sensitive media and unnecessary data exposure to services not suited for transcription.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The one-command setup auto-configures an external service with very broad capabilities but does not warn about the scope of access, what data may be sent, or what services become available afterward. In an agent context, this can silently expand the system's authority and increase the likelihood of unintended third-party data sharing.

VirusTotal

65/65 vendors flagged this skill as clean.