Security audit

Video No Subtitle Transcribe

Security checks for vulnerabilities and agentic risk

Overview

The skill appears purpose-aligned for video transcription, with disclosed network use and an optional third-party fallback that users should treat carefully.

Install only if you are comfortable with yt-dlp contacting the source video site, the script downloading a Whisper model from ModelScope, and the optional browser fallback sharing a YouTube URL with youtube.iiilab.com after local download attempts fail. Avoid the third-party fallback for private, unlisted, sensitive, or client-owned videos unless you explicitly accept that disclosure.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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 (4)

Intent-Code Divergence

Medium

Confidence: 94% confidence
Finding: The skill claims it is 'completely free, runs locally, no privacy leakage' while its documented fallback explicitly sends a YouTube URL to a third-party site. That contradiction can mislead users or downstream agents into disclosing private, unlisted, sensitive, or client-owned video references without fully appreciating the data-sharing risk.

Missing User Warnings

Low

Confidence: 87% confidence
Finding: The changelog documents a fallback that sends users to a third-party download site (youtube.iiilab.com) after normal retrieval fails, but it provides no warning about data exposure, trust boundaries, or the site's handling of requested URLs and downloaded media. In a transcription skill, this can cause users to unknowingly disclose viewing targets or media content to an external service, increasing privacy, integrity, and supply-chain risk.

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The description includes broad trigger phrases such as 'no subtitles', 'transcribe video', and 'extract speech', which can cause the skill to activate in contexts beyond the user's intent. Over-broad invocation increases the chance an agent will download or process media, potentially contacting external services or handling sensitive content, when a narrower or safer tool should have been used.

Natural-Language Policy Violations

Medium

Confidence: 81% confidence
Finding: The skill hard-codes a China-specific mirror recommendation and states regional assumptions like 'YouTube requires a proxy (mainland China)' without user opt-in or region scoping. This can steer agents to use a specific external source or proxy behavior that may be inappropriate for the user's jurisdiction, trust model, or network environment.

Static analysis

No suspicious patterns detected.