ClawHub
Back to skill

Security audit

LLM Evaluation Toolkit（LLM评测工具链）

Security checks for vulnerabilities and agentic risk

Overview

This is a local LLM evaluation toolkit whose file access and command use match its stated purpose, though its security audit document has copy-paste inaccuracies.

This skill is reasonable to install if you want a local, offline evaluation helper. Treat the included security audit as imperfect documentation, and review file paths before running the Python commands because they read local evaluation data and can create output files.

SkillSpector

Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (3)

Lp3

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding
The skill advertises executable local tooling via `python tools/eval_toolkit.py ...` and describes capabilities that imply reading and writing local files, invoking Python from the shell, and potentially integrating with external APIs in some modes, yet no permissions are declared. This creates a transparency and policy-enforcement gap: users or the platform may treat the skill as low-privilege documentation while it actually drives code execution and filesystem access.

Intent-Code Divergence

High
Confidence
98% confidence
Finding
The audit report claims to assess the 'LLM evaluation toolkit', but its detailed evidence references a different package ('ai_governance_toolkit') and unrelated verification scripts. This undermines the integrity of the security review and can falsely assure users that the actual skill was audited, allowing unsafe behavior in the real package to go unnoticed.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The report states that the tool only prints to stdout and does not read or write external files, but the skill description explicitly includes dataset management, report generation, and release gating, which normally require file I/O. This contradiction indicates the audit is inaccurate or incomplete, weakening trust in its safety claims and potentially concealing file-system behaviors that were never reviewed.

Static analysis

No suspicious patterns detected.