ClawHub
Back to skill

Security audit

Microsoft MarkItDown

Security checks across malware telemetry and agentic risk

Overview

This is a coherent MarkItDown helper that converts user-supplied documents or URLs, with no evidence of hidden behavior or malicious data handling.

Install if you are comfortable using Microsoft's MarkItDown and its Python dependencies. Prefer pipx or a virtual environment, review upstream network behavior before converting URLs or YouTube links, and avoid sending confidential documents, images, or audio through workflows unless you understand where processing occurs.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (4)

Lp3

Medium
Category
MCP Least Privilege
Confidence
75% confidence
Finding
The skill instructs users to install and run shell commands (`pipx`, `pip`, `brew`, `apt`, virtualenv activation) but does not declare any permissions or execution expectations. In an agent ecosystem, undocumented shell capability increases risk because the agent may perform package installation or system modifications without clear consent boundaries.

Description-Behavior Mismatch

Medium
Confidence
89% confidence
Finding
The wrapper explicitly accepts an input_path documented as a file path or URL and supports output formats beyond Markdown, which expands capability beyond the declared skill purpose. In an agent context, allowing arbitrary URLs can enable server-side request forgery or unexpected network access through the underlying converter, especially if the skill is invoked on attacker-controlled input.

Vague Triggers

Medium
Confidence
88% confidence
Finding
The trigger phrase `convert document` is very broad and likely to match many ordinary user requests, causing unintended invocation of this skill. Overly generic routing can expose user files to processing paths they did not intend, especially given the skill handles local files, OCR, audio, and remote URLs such as YouTube.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill advertises OCR, audio transcription, and YouTube extraction without warning that these operations may process sensitive local documents, images, speech, metadata, or remote content. Users may unknowingly submit confidential material, and URL-based processing can introduce additional privacy and provenance concerns.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.