ClawHub

Sa Copilot

Security checks across malware telemetry and agentic risk

Overview

This appears to be a purpose-aligned architecture/workflow assistance skill with clean malware telemetry, but its activation language is broader than ideal.

Before installing, review when the skill activates. Avoid sharing confidential architecture, customer data, internal documents, credentials, or sensitive project paths unless you trust the remote MCP service and its data handling.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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)

Vague Triggers

Medium
Confidence
86% confidence
Finding
The onboarding trigger is defined using very broad terms such as generic greetings and 'similar' messages, which can cause the skill to activate its guided workflow in conversations that did not clearly request it. In an agent setting with remote MCP connectivity, ambiguous activation increases the chance of unintended tool initialization, user confusion, and premature disclosure or collection of document paths and workflow state.

Vague Triggers

Medium
Confidence
90% confidence
Finding
The tool trigger keywords are broad and overlap with ordinary architecture discussion terms like 'architecture design' and 'design review', so normal conversation may be interpreted as a command to invoke MCP tools. Because these tools are remote and phase-driven, accidental invocation can lead to unintended external requests, workflow transitions, and exposure of project paths or sensitive architectural context.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal