Security audit

douyin-video-parser

Security checks across malware telemetry and agentic risk

Overview

The skill appears to perform the stated Douyin transcription workflow, but its offline wording and broad triggers should be clarified.

Install only if you are comfortable with the agent contacting Douyin/iesdouyin for the video, possibly HuggingFace for the Whisper model on first use, and writing transcript files locally. Use it for explicit Douyin links and consider tightening the triggers and removing the stale __pycache__ file before publishing broadly.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (5)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 94% confidence
Finding: The skill instructs the agent to use shell execution, network access, and local file read/write, but it declares no permissions or equivalent user-facing capability disclosure. This creates a trust and containment problem: an agent may invoke a skill that downloads remote content and writes files locally without clear authorization boundaries or informed consent.

Description-Behavior Mismatch

Medium

Confidence: 89% confidence
Finding: The skill markets itself as '完全免费、离线' and not dependent on external services, while the documented flow clearly performs outbound HTTP requests to Douyin and may download models from HuggingFace. Misrepresenting network behavior is security-relevant because users and calling agents may approve it under false assumptions about privacy, egress, and data handling.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: Claiming the workflow is '完全本地' and '之后不再联网' understates the actual remote dependencies in the documented steps. In agent environments, inaccurate locality claims can bypass user scrutiny and organizational controls around network access, external content ingestion, and supply-chain downloads.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The documentation does not clearly warn that the skill downloads remote video content to local storage and processes it automatically. This matters because automatic remote fetch plus local persistence can expose users to unexpected storage, privacy, copyright, or policy risks, especially in agentic contexts where a model may act with limited user review.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The trigger list includes very generic phrases such as '拆视频', '分析视频', '听视频', and '读视频', which can match many unrelated user requests and cause the skill to be invoked outside its intended Douyin-only scope. Because this skill downloads remote media and performs local transcription/analysis, over-broad routing increases the chance of unintended network access, unnecessary processing, and confusion or misuse by the agent.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.