Security audit

倪海厦数字人对谈

Security checks across malware telemetry and agentic risk

Overview

This is not malware, but it is a health-advice persona skill that can give forceful treatment guidance and discourage standard medical care without safety limits.

Review carefully before installing. Treat this as historical/persona roleplay material only, not medical guidance. Do not rely on it for diagnosis, prescriptions, herb dosing, acupuncture or bleeding procedures, emergencies, pregnancy, children’s care, cancer, diabetes, antibiotics, insulin, vaccines, imaging decisions, or decisions to stop or refuse prescribed treatment; consult licensed medical 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 (11)

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The trigger list includes generic phrases such as “中医问题” and broad identity-invocation wording that can match ordinary user conversation, causing the skill to activate when the user did not clearly consent to this specific persona and medical framing. In this skill’s context, accidental invocation is more dangerous because the skill is configured to provide authoritative-sounding health guidance and explicitly discourages mainstream treatment approaches.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The skill advertises authoritative medical-style advice while promoting strong anti-mainstream-treatment positions, but provides no safety warning, limitation notice, or recommendation to seek licensed care for serious symptoms. This is dangerous because users may treat the output as actionable healthcare guidance and delay or reject evidence-based treatment, especially given the skill’s instructions to criticize Western medicine and avoid endorsing it.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section provides concrete medical diagnosis and treatment guidance, including condition-specific prescriptions, dosing ratios, and recommendations for serious illnesses such as gastric ulcer/cancer, pancreatic cancer, postoperative adhesions, and respiratory symptoms, without any safety disclaimer or instruction to seek qualified medical care. In the context of a conversational skill designed to answer users in an authoritative master's voice, users may treat this as actionable medical advice and delay appropriate diagnosis or treatment, creating a substantial risk of physical harm.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section provides concrete treatment guidance, named formulas, indications, and disease-management rules for serious conditions without any safety disclaimer, clinician involvement, or instruction not to self-medicate. In the context of a conversational skill designed to answer health questions in an authoritative voice, users could act on this content directly and delay appropriate diagnosis or treatment.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The document makes categorical, alarming claims against chemotherapy and antibiotics without evidence, nuance, or warnings not to stop prescribed treatment without medical supervision. Because the skill is explicitly framed as an authoritative medical persona, this can persuade vulnerable users to reject or discontinue life-saving conventional care, creating severe risk of preventable harm or death.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: These simplified symptom heuristics and body-sign rules encourage lay self-diagnosis despite being medically unreliable and lacking warnings about emergencies, overlap between conditions, or the need for professional assessment. In a health advice skill that imitates a famous authority, such rules are especially likely to be trusted and can lead users to miss urgent conditions or pursue inappropriate remedies.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The file provides concrete treatment-style instructions, including dosing, contraindications, and disease-specific recommendations, but does not warn users to seek licensed medical care or avoid self-treatment. In the context of a conversational skill explicitly designed to answer medical and health questions in an authoritative persona, this materially increases the risk that users follow unsafe advice, delay evidence-based care, or misuse remedies for serious conditions such as cancer, heart disease, or diabetes.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This note presents detailed medicinal claims, contraindications, and quasi-prescriptive guidance for potent substances such as 牛膝, 鹿茸, 牛黄, 石钟乳 and others, but it does not include a clear warning against self-diagnosis, self-medication, dosage experimentation, or substituting the content for licensed medical care. In the context of a skill explicitly designed to answer health questions in an authoritative persona, users may treat these statements as actionable treatment advice, increasing the risk of unsafe self-treatment, toxicity, pregnancy harm, drug interactions, or delay of appropriate care.

Missing User Warnings

High

Confidence: 97% confidence
Finding: This section gives concrete acute-care and treatment instructions, including stroke rescue principles, medication/formula selection, dosing-style details, and instructions such as continuing medication until sweating occurs, without any safety disclaimer or direction to seek licensed medical care. In the context of a conversational skill designed to answer users in an authoritative medical voice, this can plausibly prompt users to self-diagnose, delay emergency treatment, or attempt unsafe herbal interventions for serious conditions.

Missing User Warnings

High

Confidence: 98% confidence
Finding: These notes include actionable treatment instructions and emergency interventions such as moxibustion, bleeding, needling techniques, and acute-response measures without any safety framing, contraindications, or instruction to seek licensed medical care. In the context of a conversational skill that may present the material authoritatively as advice, users could attempt unsafe self-treatment, delay emergency care, or injure themselves.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The document makes categorical diagnostic claims such as determining serious diseases from sleep timing and asserting that X-rays are unnecessary, with no warning against replacing standard evaluation. In this skill's context—designed to answer health questions in an authoritative 'master' voice and criticize conventional medicine—such claims are more dangerous because they can encourage users to forgo timely diagnosis for cancer, gallstones, liver disease, or other serious conditions.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.