ClawHub
Back to skill

Security audit

PKULaw

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed PKULaw legal-search connector, with a generic tool-call path that users should keep limited to read-only schemas.

Before installing, be comfortable using an OOMOL-connected PKULaw account and the oo CLI. For call_tool, inspect the live schema and only run it when the exposed PKULaw tool is clearly read-only; ask for explicit confirmation if any schema suggests changing, posting, deleting, or account-affecting behavior.

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 (2)

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The skill metadata and description frame this integration as a search/read-only PKULaw capability, but the exposed `call_tool` action allows generic invocation of whatever tools the connector currently exposes. That creates a capability mismatch: downstream agents or users may treat the skill as read-only and invoke it in contexts where broader operations were not intended, increasing the risk of unexpected state changes or abuse if the connector later exposes non-read actions through the generic interface.

Intent-Code Divergence

Medium
Confidence
97% confidence
Finding
The safety section says untagged actions are safe reads, but `call_tool` is untagged even though it is a generic dispatch mechanism whose behavior depends on live connector capabilities. This can mislead an agent into treating `call_tool` as inherently safe and executing arbitrary backend operations without confirmation, especially if the connector adds write/destructive tools later or exposes side-effecting functionality through the generic endpoint.

Static analysis

No suspicious patterns detected.