ClawHub
Back to skill

Security audit

Word文档工具

Security checks for vulnerabilities and agentic risk

Overview

The skill is a Word automation helper, but it asks for command and file-write authority while giving weak and partly misleading safety boundaries.

Install only if you are comfortable letting the agent operate on local Word documents and run local commands for Word automation. Prefer using it on copies of documents, review any command or overwrite action before approval, and avoid sensitive files unless the agent environment provides containment you trust.

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
94% confidence
Finding
The skill explicitly advertises command execution via the declared tools (`exec`) while also assuring users that commands run in a 'safe sandbox'. That assurance is misleading because the manifest itself does not enforce sandboxing, and users or agents may be induced to run OS-level commands with reduced caution. In a document-processing skill, this mismatch can hide real host-impacting behavior such as file modification, process control, or data access.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The skill promotes document manipulation, export, and command execution capabilities without a prominent warning that it may modify files or invoke OS commands. This can lead users or agents to supply sensitive paths or approve actions without understanding that local documents may be changed, exported, or processed by external command execution pathways. In an automation skill, lack of explicit consent boundaries increases the chance of unsafe or unintended destructive actions.

Static analysis

No suspicious patterns detected.