Security audit

倪海厦skill·经方中医AI

Security checks across malware telemetry and agentic risk

Overview

This skill is not malware, but it should be reviewed because it can give diagnosis-like and treatment-like medical guidance for serious conditions without consistent safety limits.

Install only if you want a historical/educational TCM reference and can keep it from being used as medical advice. Do not rely on it for diagnosis, prescriptions, herbs, dosing, acupuncture, emergencies, pregnancy, children, serious illness, medication changes, or decisions about standard medical care; consult licensed clinicians instead.

SkillSpector

By NVIDIA

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (46)

Intent-Code Divergence

High

Confidence: 97% confidence
Finding: The document says it cannot replace professional medical advice or provide concrete prescriptions, yet elsewhere it gives extensive disease-specific treatment instructions, formulas, and dosing. That contradiction is dangerous because downstream agents may follow the more actionable sections and present high-risk medical guidance as if it were safe educational content.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The skill contains actionable guidance for acute and severe conditions, including emergency handling and treatment suggestions, while being framed as a knowledge/roleplay skill rather than a regulated clinical tool. In this context, users or agent systems could treat historical TCM content as real-time medical instructions, creating severe risk of delayed care, harmful self-treatment, or injury.

Description-Behavior Mismatch

Medium

Confidence: 96% confidence
Finding: The skill is presented as a TCM diagnostic/knowledge corpus, but this file also includes feng shui and fortune-telling material that is unrelated to the declared medical purpose. In a medical-facing skill, mixing occult or non-medical guidance into the corpus can mislead users about the scope and credibility of the system, increasing the risk of unsafe or deceptive health advice.

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: Including fortune-telling or feng shui capability in a skill whose stated purpose is medical diagnosis/treatment is a scope-integrity failure. If surfaced to users, this can lend false authority to non-evidence-based recommendations and contaminate medical decision support with unrelated, potentially manipulative content.

Description-Behavior Mismatch

Medium

Confidence: 93% confidence
Finding: The page's footer disclaimer says the project is only for TCM learning and research, but the app promotion explicitly advertises '独立诊断' and symptom-guided diagnostic functionality. That mismatch can mislead users into relying on an unregulated tool for medical decision-making, which is especially risky because the skill also presents treatment-oriented recommendations.

Description-Behavior Mismatch

Medium

Confidence: 86% confidence
Finding: The module claims to support early-stage differentiation for common exterior-pattern illness, but the content expands into extensive disease management, emergency-style advice, invasive procedures, and treatment of serious conditions. That mismatch increases the chance that downstream agents or users will overtrust the skill for decisions far beyond its stated scope, creating unsafe medical use without appropriate safeguards.

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: Embedding concrete cancer and major-disease treatment formulas inside a skill framed as classical knowledge interpretation materially raises safety risk. Users or consuming agents may present these recipes as actionable treatment recommendations for life-threatening illness, delaying evidence-based care and exposing users to harmful substances or contraindicated regimens.

Description-Behavior Mismatch

Medium

Confidence: 89% confidence
Finding: The file is presented as a faithful detailed extract of a classical text, but parts of it introduce modern diagnostic claims and strong efficacy assertions as if they were established clinical guidance. That mismatch can mislead downstream users or agents into treating commentary as authoritative medical instruction, increasing the chance of harmful self-treatment or misinformation propagation.

Intent-Code Divergence

Medium

Confidence: 84% confidence
Finding: A document framed as 'core points' or 'lecture excerpt' shifts into assertive treatment-effect claims, which is an intent mismatch. Users may reasonably assume they are reading neutral study notes when they are actually being nudged toward concrete medical actions and beliefs.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: The file includes public-health-substituting claims in a knowledge module whose stated purpose is classical medicine study, including language implying self-cultivation or traditional practices are more fundamental than modern preventive measures. In a skill context, that can steer users away from evidence-based prevention or treatment and creates elevated real-world harm risk.

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: This module embeds astrology, feng shui, and fate-analysis content inside a skill presented primarily as TCM diagnostic and materia medica guidance. That scope expansion is dangerous because users may treat metaphysical outputs as health-relevant advice, and the unrelated content increases the chance of deceptive authority transfer from medical-seeming material to non-medical claims.

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: The file advertises uses including 命理风水 alongside acupuncture and herbal reference content, extending the skill beyond its stated core medical mindset into metaphysical guidance. In a medical-context skill, this is risky because it can mislead users into viewing unsupported spiritual or fate-based claims as part of legitimate diagnosis or treatment reasoning.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The README prominently markets the skill as able to perform syndrome differentiation, formula selection, and symptom interpretation before presenting any meaningful safety warning. In a medical skill, capability claims that imply diagnostic or treatment assistance can encourage unsafe reliance, especially when users may act on recommendations without clinician oversight.

Missing User Warnings

Medium

Confidence: 98% confidence
Finding: The usage examples give concrete symptom-to-diagnosis and symptom-to-treatment guidance such as identifying a condition and naming specific formulas, without an immediate warning that this is not medical advice. Because the project is framed as an AI skill, these examples materially increase the risk that users will self-diagnose or self-treat based on incomplete or unsafe information.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The file lacks strong, user-facing medical safety warnings near the main workflow and activation logic, despite encouraging disease analysis and formula selection. In a medical skill, absence of prominent safety constraints increases the chance that an agent will provide diagnosis-like or treatment-like output without proper triage or referral.

Missing User Warnings

High

Confidence: 99% confidence
Finding: Dangerous emergency and treatment content appears without immediate, conspicuous risk warnings or refusal logic. In a medical-context skill, this sharply elevates harm because users may follow urgent care instructions from the model instead of seeking emergency services or licensed care.

Missing User Warnings

High

Confidence: 96% confidence
Finding: This file presents numerous case summaries involving serious conditions such as heart disease, stroke, kidney failure, diabetes, and cancer-adjacent contexts, alongside named formulas and strong claims about treatment outcomes, but provides no warning that the material is historical/reference content rather than medical advice. In a skill explicitly designed to emulate a named practitioner’s diagnostic and prescribing style, users could reasonably treat these summaries as actionable guidance, creating a significant risk of delayed care, medication substitution, or unsafe self-treatment.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The flagged section contains procedural acupuncture/needling guidance, contraindication tables, and emergency-response style instructions presented as actionable medical content without safety boundaries, training requirements, or urgent referral warnings. In a consumer-facing agent skill, this can enable unqualified users to attempt invasive procedures or delay evidence-based care, creating a meaningful risk of physical harm.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The demo presents concrete diagnosis and treatment guidance ('太阳伤寒，麻黄汤证', '一剂汗出即愈') without an immediate safety warning, clinician referral, contraindication notice, or emergency escalation guidance. In a medical context, concise authoritative treatment advice can cause users to self-diagnose or self-medicate, creating direct health and safety risk.

Missing User Warnings

High

Confidence: 95% confidence
Finding: The installation and app-promo sections market the tool across multiple platforms and describe autonomous diagnostic use without nearby warnings about medical limitations. This increases the chance of broad deployment of treatment-oriented behavior while minimizing user awareness of the risks and the need for professional oversight.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The file presents diagnostic and treatment-selection guidance for fever/cold-like illness without any prominent warning that users should not self-diagnose or self-medicate. In medical context, omission of a safety disclaimer is dangerous because users may treat the material as direct clinical advice rather than historical or educational content.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The document repeatedly includes actionable medication formulas, dosage-like instructions, toxic substances, strong purgatives, moxibustion/fire-based practices, bloodletting, and needling guidance, yet provides no overarching safety framework or contraindication controls. In this skill context, that makes the content especially dangerous because an agent could operationalize high-risk medical instructions for lay users, leading to poisoning, burns, delayed treatment, or other serious harm.

Natural-Language Policy Violations

High

Confidence: 95% confidence
Finding: 该处包含明显的群体性侮辱与贬损语言，针对特定国籍/族群用户进行羞辱。放在一个面向用户的医疗技能中，这会正常化歧视性表达，并可能诱导代理以辱骂、羞辱方式回应用户，造成情绪伤害与不当影响。

Natural-Language Policy Violations

High

Confidence: 96% confidence
Finding: 该段继续使用带有国籍属性的侮辱性称呼，属于面向群体的贬抑内容。作为技能知识库的一部分，这种语气会被模型模仿并扩散到实际对话中，形成明显的安全与合规风险。

Natural-Language Policy Violations

High

Confidence: 97% confidence
Finding: 该处存在重复性、泛化式针对中国人的贬损表述，具有强化偏见和仇视叙事的效果。若被代理继承，会显著增加对用户进行歧视、羞辱或排斥的概率。

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.