ClawHub
Back to skill

Security audit

LLM Quality Evaluation（LLM质量评测）

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent local LLM evaluation guide and toolkit with no network use, credential access, persistence, or automatic high-impact actions found.

Installers should treat this as a local documentation-and-helper skill. The main toolkit does not collect data or go online, but the bundled verification scripts can rewrite files in the package's verify directory if run manually. Also note that the license text and separate knowledge-copyright language may be inconsistent for redistribution, though that is not a security concern.

SkillSpector

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
90% confidence
Finding
The declared purpose is LLM quality evaluation, but the finding indicates additional hidden behavior that inspects package files, performs security/compliance validation, and writes security_results.json and security-radar.svg outputs. That mismatch matters because users may invoke a seemingly harmless evaluation skill while it processes unrelated local files and generates artifacts, expanding the trust boundary without clear disclosure.

Static analysis

No suspicious patterns detected.