ClawHub
Back to skill

Security audit

Chatmosp Kmc Simulator

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed local KMC simulation skill that runs external Wine/Python tools, with documentation confusion but no hidden data access or malicious behavior found.

Before installing, confirm you trust the external mosp-for-chatMOSP code and Windows engine it will run through Wine, and only run the Wine installation commands if you are comfortable changing system packages. The publisher should correct the SKILL.md/SKILL_cn.md language-routing labels to reduce confusion.

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The language-routing block tells the agent to read `SKILL_cn.md` when the user writes in English, but this file is itself labeled `SKILL_cn.md` while later text says Chinese users should continue using this file as `SKILL.md`. This creates ambiguous authority and can cause the agent to follow the wrong instruction source, which is a prompt-integrity issue because control flow depends on attacker-editable content rather than stable metadata.

Natural-Language Policy Violations

Medium
Confidence
96% confidence
Finding
These routing instructions can force the agent to switch to a different file based solely on the user's language, despite contradictory labels and no trusted external arbitration. An attacker or careless author could use this pattern to hide materially different instructions in the alternate file, causing the model to bypass expected review context or apply the wrong operational rules.

Static analysis

No suspicious patterns detected.