Back to skill

Security audit

将 Markdown 技术文档自动转换成带配音旁白的专业视频

Security checks across malware telemetry and agentic risk

Overview

This looks like a real document-to-video helper, but it needs Review because it includes instructions for copying and modifying installed agent skills across environments plus some unsafe cleanup and overwrite commands.

Install only if you are comfortable with a skill that runs local build commands, sends narration text to an external TTS provider, and contains maintenance instructions that can modify installed skill copies. Avoid using it on confidential documents unless external TTS is approved, and do not run the cleanup or sync commands without checking the current directory and destination paths first.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Lp3

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding
The skill clearly instructs users to read and write local files and execute shell commands (`pip3`, `ffmpeg`, `npm`, `rm -rf`), yet no permissions are declared. This creates a transparency and consent problem: an agent platform may expose file and shell capabilities without users understanding the scope of actions the skill expects to perform.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The document instructs the agent to interact with and rely on modifications inside the skill's own installation directory under ~/.hermes/skills/doc-to-video/, including behavior tied to a self-update mechanism. For a markdown-to-video skill, self-referential mutation of installed skill files is outside normal rendering scope and creates a path for persistence, integrity loss, or unauthorized skill tampering.

Context-Inappropriate Capability

Medium
Confidence
94% confidence
Finding
The guide includes synchronization of skill files into another workspace (OpenClaw), which expands the skill's reach beyond generating videos and encourages cross-environment propagation of skill content. This is dangerous because it can replicate modified or compromised skill artifacts into other agent runtimes or workspaces without a clear business need.

Missing User Warnings

Medium
Confidence
86% confidence
Finding
The script sends narration text to the external `edge_tts` service, which means any content placed in `SCENES` is transmitted off-host. For a documentation-to-video skill, this is contextually relevant because users may paste proprietary docs, internal procedures, or sensitive text into narration without realizing it leaves their environment.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The script uses `ffmpeg -y` to overwrite `/tmp/noaudio.mp4` and the final output without any confirmation or safety checks. In normal use this can cause accidental data loss by replacing existing files, especially if the script is re-run or filenames are reused; the doc-to-video context makes this more operationally risky because repeated rendering is expected, but it does not indicate malicious behavior.

Tool Parameter Abuse

High
Category
Tool Misuse
Content
cd ~/vscode
cp -R my-first-video-project my-second-video-project
cd my-second-video-project
rm -rf audio/* out/*.mp4 out/*.jpg node_modules package-lock.json build
npm install --no-audit --no-fund   # 3s,缓存命中
```
Confidence
95% confidence
Finding
rm -rf audio/* out/*.mp4 out/

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.