ClawHub
Back to skill

Security audit

CMIC Skill Scanner

Security checks across malware telemetry and agentic risk

Overview

This documentation-only skill is a security scanner wrapper with disclosed local scanning behavior and optional network/LLM features that users must explicitly enable.

Before using it, verify any downloaded scanner binary against the published checksum or build from source. Use `--engine native` if you do not want a locally resolved external scanner to run, and enable `--upload-url` or `--use-llm` only with endpoints you trust because LLM review can send selected text from the target package.

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 (2)

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The documentation makes a categorical claim that no skill source code is ever transmitted, but elsewhere states that native LLM review may send up to 24 text files from the target package to an endpoint. This is a material contradiction that can mislead users about data exfiltration risk and cause them to enable networked features under false assumptions.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The default prompt 'Audit this local skill target, output findings and risk level.' is broad and can trigger auditing behavior without clear constraints on scope, confirmation, or allowed inputs. In an agent setting, vague default prompts can cause the model to act on unintended local targets or perform analysis on attacker-supplied paths/content, increasing the chance of misuse or unsafe autonomous behavior.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.