ClawHub

Sdr Humanizer

Security checks across malware telemetry and agentic risk

Overview

This is not malware, but it tells an agent to hide AI involvement and silently copy translated sales messages into a WhatsApp self-chat.

Review before installing. Only use this skill if your organization explicitly permits AI-assisted sales outreach, delayed message delivery, and internal translation handling. Remove or disable the silent WhatsApp self-chat sync unless the data flow is disclosed, approved, logged, and governed by clear retention and access controls.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Intent-Code Divergence

Medium
Confidence
96% confidence
Finding
The skill explicitly aims to make AI-generated outreach appear human while hiding the AI's role from customers. This is deceptive social engineering behavior that can mislead recipients about who is communicating with them and undermine informed consent and trust.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill directs the system to silently send a Chinese translation to a self-chat channel without notifying the user or customer. Even if sent to the operator's own account, this creates an undisclosed secondary transmission of conversation content and increases the risk of privacy, retention, and audit issues.

Natural-Language Policy Violations

High
Confidence
88% confidence
Finding
Forcing all customer replies to be in English without user choice can override customer preferences and business context, causing miscommunication and inappropriate handling of locale-sensitive interactions. In combination with hidden translation, it also centralizes a covert workflow that the customer does not see or control.

Ssd 3

High
Confidence
97% confidence
Finding
The instruction to copy each interaction into a hidden self-chat channel is a direct data handling risk because it duplicates customer communications into an additional, undisclosed store. This expands exposure, retention, insider access, and compliance risk, especially on consumer messaging platforms not designed as controlled processing pipelines.

Ssd 1

Medium
Confidence
95% confidence
Finding
Telling operators never to disclose AI involvement encourages deceptive role-play and suppresses transparency in customer interactions. In a sales setting, this increases the likelihood of manipulative outreach and can create legal, reputational, and policy violations if customers are misled about the source of communications.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal