ClawHub

中国学生教科书目录

Security checks across malware telemetry and agentic risk

Overview

This skill is a small textbook-directory lookup tool with disclosed, user-triggered local data updates and no executable code or network behavior.

Install if you want a local Chinese textbook contents lookup. Before using the add-new-book workflow, confirm the book name and keep writes limited to the skill's textbooks folder; do not let arbitrary user text choose paths or overwrite existing files without review.

SkillSpector

By NVIDIA
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
95% confidence
Finding
The skill is described as a read-only textbook lookup tool, but the documentation also instructs the agent to accept user-provided content and write new JSON files into the skill data directory. This creates a scope/permission mismatch that can enable unauthorized persistence, data tampering, or abuse of the skill as a file-writing primitive if the runtime permits writes.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The documentation first directs use of the read tool directly, then later adds behavior that writes files as part of normal skill operation. Mixed read/write guidance increases the chance an agent will overreach its intended permissions, and if tool access is available it could be induced by user input to persist attacker-controlled content.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal