Security audit

Barefoot Doctor (EN)

Security checks across malware telemetry and agentic risk

Overview

This markdown-only medical skill is not malware, but it needs review because it gives high-stakes diagnosis, treatment, emergency, pediatric, herbal, and acupuncture guidance with unsafe scoping gaps.

Install only if you treat this as reference material, not medical care. Do not rely on the stated accuracy claims, do not use 120 unless it is your local emergency number, avoid entering identifying health details, and verify medication, herb, acupuncture, pregnancy, pediatric, chronic disease, and emergency advice with qualified local professionals.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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 (8)

Missing User Warnings

High

Confidence: 96% confidence
Finding: These templates instruct the model to provide diagnosis, treatment, medication, pediatric dosing, herbal regimens, acupuncture, and first-aid guidance, but they do not consistently place a strong user-facing warning up front that the AI is not a substitute for a licensed clinician or emergency care. In a medical skill, omission of that warning can cause dangerous overreliance, delayed care, and unsafe self-treatment, especially because the prompts are designed for copy-paste reuse and appear authoritative.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The emergency first-aid template tells the AI to give step-by-step emergency guidance but does not begin with a clear instruction to contact local emergency services immediately for serious emergencies. That structure can encourage users to wait for or follow AI-generated procedures instead of obtaining real-time professional help, which is especially dangerous in choking, poisoning, drowning, electric shock, unconsciousness, and similar time-critical events.

Natural-Language Policy Violations

Medium

Confidence: 91% confidence
Finding: Hard-coding '120' assumes a specific geography and may provide the wrong emergency number for users in other regions. In an emergency context, locale-mismatched instructions can waste critical time, confuse users, or reduce trust in the rest of the guidance.

Natural-Language Policy Violations

Medium

Confidence: 90% confidence
Finding: The checklist hard-codes "Call 120" as the emergency instruction without establishing the user's country or region. In a medical triage context, a locale-specific emergency number can misdirect users in urgent situations, causing dangerous delays in contacting emergency services when they are outside the intended jurisdiction.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The trigger scope is very broad, covering diagnosis, treatment, TCM, herbal medicine, acupuncture, and emergency first aid from loosely specified keywords. In a health skill, ambiguous activation can cause the agent to engage in high-risk medical guidance in inappropriate contexts, increasing the chance of unsafe advice, delayed professional care, or accidental routing of emergency situations into a generic assistant flow.

Missing User Warnings

High

Confidence: 95% confidence
Finding: This file contains actionable medical triage, dosing, herbal, acupuncture, and first-aid guidance but does not prominently frame the material as educational support rather than professional medical advice. In a health skill, omission of clear safety boundaries can cause users to rely on potentially incomplete or context-insensitive instructions, increasing risk of delayed care, contraindicated treatment, or harmful self-management.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: The diagnostic flow hard-codes '120 + stop' as the emergency action, which is region-specific and may be wrong for many users. In an emergency context, giving the wrong emergency number can delay lifesaving care, making the issue materially safety-relevant even if not malicious.

Natural-Language Policy Violations

Medium

Confidence: 94% confidence
Finding: The triage code and emergency protocol sections repeat 'CALL_120' and 'Call 120,' reinforcing a single locale-specific emergency path without verifying user location. Because this skill is explicitly medical and emergency-oriented, the context makes the mistake more dangerous: users may follow the wrong contact instruction during cardiac arrest, poisoning, drowning, or other time-critical events.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.