musk-eeg

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:40 PM UTC
• Version:
  1.0.5

Overview

This is a disclosed local EEG/neuroscience lookup skill, but its celebrity-style medical-adjacent answers should be treated as educational, not clinical advice.

Install only if you are comfortable running a local Python script and supplying a trusted database ZIP. Verify the database source before first use, because the script will unpack it into the skill data folder. Treat answers as stylized educational summaries with citations to check, not medical advice, diagnosis, or treatment guidance.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (7)

Description-Behavior Mismatch

Medium

• Confidence: 90% confidence
• Finding: The skill promises sourced, retrieval-grounded answers for every claim, but its own examples and persona guidance include uncited assertions and analogical expansions that can exceed the retrieved Wikipedia content. In a neuroscience and medical-adjacent domain, this mismatch can mislead users into treating unsourced or speculative statements as verified facts.

Intent-Code Divergence

Medium

• Confidence: 94% confidence
• Finding: The documentation says the model must only 'concatenate + translate' Wikipedia content, but the prescribed frameworks encourage extrapolation, probability estimates, and value-laden conclusions. That creates a prompt-level integrity failure where the agent may present speculative neuroscience or medical-adjacent claims as grounded retrieval output.

Description-Behavior Mismatch

Medium

• Confidence: 88% confidence
• Finding: The script automatically extracts a ZIP archive into the local data directory at runtime, which is a filesystem-modifying side effect not implied by a simple read-only retrieval tool. If the ZIP file is replaced or tampered with, extraction can overwrite files or introduce unsafe content, and the code does not validate archive members or enforce safe extraction paths.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The README presents the skill as able to answer questions about EEG, seizures, sleep disorders, depression, anxiety, and other medical topics in an authoritative persona, but only later includes a limited disclaimer that Musk is not a medical expert. That creates a real safety risk: users may over-trust stylized, simplified outputs for health decisions, especially in seizure, sleep apnea, or neuropsychiatric contexts where bad advice can delay proper care.

Natural-Language Policy Violations

Medium

• Confidence: 87% confidence
• Finding: Forcing responses in Elon Musk's voice without opt-in increases impersonation and authority-risk, especially in a neuroscience and health-adjacent context where users may over-trust the style as expertise. It also removes user control over tone and disclosure, making misleading roleplay more likely.

Natural-Language Policy Violations

Medium

• Confidence: 90% confidence
• Finding: Mandating first-person Elon Musk roleplay strengthens the impersonation risk and can blur the line between sourced facts, opinion, and celebrity voice. In this subject area, that can amplify misinformation or unsafe overconfidence when discussing seizures, depression, consciousness, or medical interpretation.

Vague Triggers

Medium

• Confidence: 92% confidence
• Finding: The trigger scope is extremely broad, spanning EEG, neuroscience, cognition, mental health, consciousness, sleep disorders, and general brain terms. Over-broad activation raises the chance that this persona-driven, retrieval-constrained skill will intercept queries outside its safe competence, producing misleading or inappropriate responses in sensitive medical contexts.

VirusTotal

67/67 vendors flagged this skill as clean.