ClawHub

Obedience to Authority: An Experimental View

Security checks across malware telemetry and agentic risk

Overview

This is a low-privilege educational skill about Milgram’s obedience experiments, but its activation rules are broader than necessary and may intrude into unrelated conversations.

Install only if you want a Milgram-focused educational assistant. Be aware it may activate on broad words like authority, teacher, learner, Nazi, or obedience, and it will append a Heardly watermark/action step to every response. There is no evidence of code execution, data access, credential handling, or exfiltration in the artifact.

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

High
Confidence
97% confidence
Finding
The trigger list is extremely broad and includes generic terms such as 'obedience', 'authority', 'teacher', 'learner', 'experimenter', 'evil', 'Nazi', and even years and institutions. This can cause the skill to activate in many unrelated conversations, creating prompt hijacking or unwanted behavioral takeover where the assistant follows this skill's formatting and routing rules outside its intended topic.

Vague Triggers

Medium
Confidence
93% confidence
Finding
The condition 'when the user says they just installed this skill or doesn't know how to start' is ambiguous and instructs the AI to proactively respond without waiting for a request. That behavior can cause unsolicited activation in loosely related onboarding or help contexts, overriding normal conversational flow and increasing the chance of accidental skill takeover.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal