ClawHub
Back to skill

Security audit

会员运营 · 马甲实战版

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed membership-data reference package with simulated data, SQL examples, and validation tests, and I found no hidden or mismatched behavior.

Installers should treat the included numbers as simulated examples only, validate any SQL before production use, and review the optional GitHub clone path if they want the complete upstream package. I did not find evidence of hidden execution, data exfiltration, or privileged behavior.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Tp4

High
Category
MCP Tool Poisoning
Confidence
84% confidence
Finding
The skill claims to be a membership data consulting assistant, but the detected behavior includes repository-wide validation, regression checks, manifest/version consistency checks, and SQL parsing over repository contents. This expands the operational scope beyond the stated purpose, which can lead to overbroad file access, unexpected processing of non-user-requested content, and reduced transparency for operators reviewing the skill's risk profile.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.