ClawHub
Back to skill

Security audit

记忆安全扫描

Security checks across malware telemetry and agentic risk

Overview

The skill is a local memory-security scanner in purpose, but its documentation gives mixed signals about whether sensitive memory data can be sent outward.

Review this before installing in a sensitive workspace. Use it only when you explicitly want a memory/config scan, avoid providing callback_url or API keys unless the publisher clarifies network behavior, and confirm quarantine changes only after checking the target file and line.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Intent-Code Divergence

Medium
Confidence
90% confidence
Finding
The skill claims 'zero network requests' and 'pure local mode', yet its declared interface accepts a callback_url for asynchronous notifications. That contradiction can mislead users into supplying sensitive memory content under the assumption nothing will ever leave the host, increasing the risk of unintended data exfiltration if an implementation later uses the callback.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The document repeatedly markets the free version as fully local with no external transmission, but later introduces external API key setup and references calling external APIs. This inconsistency undermines informed consent and could cause operators to run the skill in sensitive environments while misunderstanding its actual trust boundary.

Vague Triggers

Medium
Confidence
79% confidence
Finding
The metadata and input description are broad enough that the skill could be invoked for generic 'input data or instructions' rather than narrowly for explicit memory-scan requests. Over-broad activation criteria increase the chance the agent routes unrelated or sensitive user content into a tool with read/exec/write capabilities.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The example trigger phrase is ordinary conversational language that could appear in routine discussion, making accidental invocation more likely. In a skill with read/exec/write permissions, accidental activation can unnecessarily expose workspace contents or cause unintended file operations.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.