ClawHub
Back to skill

Security audit

Auto Insight Engine

Security checks across malware telemetry and agentic risk

Overview

This skill is not clearly malicious, but it delegates execution to unbundled local code while describing broad network collection, third-party analysis, and knowledge-base ingestion with weak scoping and disclosure.

Review before installing. Only use this in an environment where you trust the separate local auto_insight_engine module it imports, and confirm what data will be sent to external sources or NVIDIA NIM and what will be written into any knowledge base. The unrelated Bilibili-learned sections should be removed or split into separate skills before broad deployment.

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 (4)

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The skill manifest is internally inconsistent: it advertises an automated multi-source insight collection engine, but large portions of the file contain unrelated, auto-ingested Bilibili-derived 'learned skills' about medical/AR software, subtitles, and other topics. This kind of scope drift can cause the agent to activate on irrelevant prompts, mislead users about what will run, and hide risky capabilities or prompt-routing behavior behind noisy merged content.

Vague Triggers

Medium
Confidence
88% confidence
Finding
The trigger scenario states that the skill should activate when a user 'needs ... related help,' which is broad enough to match many ordinary requests without clear user intent to invoke this specific skill. In a network-enabled skill that aggregates external data, ambiguous triggering increases the risk of unintended execution, unnecessary external calls, and surprising data handling.

Vague Triggers

Medium
Confidence
88% confidence
Finding
This trigger uses similarly ambiguous 'related help' language for a merged sub-skill, making activation conditions overly permissive and poorly bounded. Because the file already mixes unrelated capabilities, such generic routing language further increases the chance that the wrong skill content is selected or invoked.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill description openly describes collection from six external sources, deep analysis via NVIDIA NIM, and KB ingestion, but it provides no user-facing warning about network access, third-party processing, or downstream storage. That omission is dangerous because users may unknowingly trigger broad external data collection and analysis workflows with privacy, cost, and compliance implications.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.