ClawHub
Back to skill

Security audit

guide-generator

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a disclosed guide-writing workflow, but it asks agents to use local PowerShell curl and has broad triggers plus file-saving behavior that warrant review before install.

Install only if you are comfortable with a guide skill that may perform broad web research and may use local PowerShell curl for WeChat links. Prefer changing it so shell fetching is removed or tightly allowlisted, and confirm any file save path before writing.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (4)

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The skill explicitly instructs use of local PowerShell `curl.exe` to retrieve WeChat article links, which expands the skill from web research into local command execution. Even if the intended use is benign, allowing shell-capable behavior creates a command-injection and environment-abuse surface if URLs or arguments are influenced by untrusted content, and it exceeds the minimum privileges needed for guide generation.

Description-Behavior Mismatch

Medium
Confidence
88% confidence
Finding
The skill directs the agent to save output by default to a fixed local filesystem path, introducing write access to the host beyond simply generating a response. This can cause unintended persistence, privacy issues, or file clobbering if invoked unexpectedly, especially because the save behavior is coupled to a broad consumer-facing skill.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The Chinese trigger phrases are broad everyday requests like making a guide, planning an itinerary, or asking whether something is worth visiting. Overly generic activation criteria increase the chance of unintended skill invocation, which becomes more dangerous here because the workflow includes web access and optional filesystem writes.

Vague Triggers

Medium
Confidence
82% confidence
Finding
The English trigger phrases such as 'make a guide for' or 'food recommendations in [city]' are common conversational requests and are not distinctive enough to safely route to this specific skill. Because the skill performs parallel research and may persist files, accidental invocation can cause unnecessary external access and side effects from ordinary conversation.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.