Security audit

文件RAG

Security checks across malware telemetry and agentic risk

Overview

This skill is a local file/RAG helper, but it asks for read, write, and command execution authority while giving broad instructions and unsupported security assurances.

Review this skill before installing and use it only for files and commands you would be comfortable giving an agent direct access to. Do not rely on its claimed encryption, permission controls, sandboxing, command allowlists, sharing, or collaboration features unless those are provided elsewhere and independently verified. Avoid using it on secrets, credentials, confidential folders, or destructive file operations without explicit user review.

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 Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The skill documentation claims security-relevant capabilities such as data encryption and permission management, but the file provides only descriptive markdown and no implementation details or enforceable controls. This can mislead users into trusting the skill with sensitive files under a false assumption of protection, increasing the chance of data exposure or misuse.

Intent-Code Divergence

Medium

Confidence: 97% confidence
Finding: The FAQ explicitly states that the skill ensures data security through encryption and permission controls, but no such mechanisms are defined anywhere in the skill. False assurances about security features are dangerous because operators may process confidential data believing protections exist when they do not.

Intent-Code Divergence

Low

Confidence: 90% confidence
Finding: The FAQ advertises file sharing and collaborative editing support without any interface, workflow, or implementation evidence in the skill. While less severe than false cryptographic claims, this can still cause unsafe operational assumptions, including unintended data sharing through external means or reliance on nonexistent access controls.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The invocation guidance is overly broad, encouraging use for generic file processing, conversion, extraction, and automation scenarios without meaningful scoping or safety boundaries. In a skill that exposes read, write, and exec capabilities, vague triggers can cause the agent to invoke the skill in high-risk contexts involving sensitive files or impactful system operations.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill describes file operations, command execution, and API integration, but it does not provide an upfront warning that these actions may modify files, run system commands, or expose data externally. Given the declared toolset includes read, write, and exec, the lack of prominent risk disclosure makes accidental destructive actions and sensitive data handling mistakes significantly more likely.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.