李可老中医思维操作系统

Security checks across malware telemetry and agentic risk

Overview

This skill gives concrete emergency medical treatment and toxic-herb dosing instructions without adequate safety limits.

Review carefully before installing. Treat the content as historical or scholarly reference only, not as medical advice. Do not use it for diagnosis, emergency triage, dosing, self-treatment, cancer care, toxic herb use, pregnancy, pediatrics, or any acute symptoms; seek licensed medical care or emergency services instead.

SkillSpector

By NVIDIA

Vulnerability Patterns

Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
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 (39)

Intent-Code Divergence

Medium

Confidence: 84% confidence
Finding: The contradiction is safety-relevant because the skill tells the agent to avoid standard physician-referral phrasing while also acknowledging that high-dose/toxic interventions require experienced supervision or emergency care. In practice, this can bias responses toward authoritative, actionable treatment instructions and away from clear escalation language, increasing the chance that users attempt unsafe self-treatment.

Missing User Warnings

High

Confidence: 96% confidence
Finding: This section markets an AI skill for emergency diagnosis, syndrome differentiation, and formula routing, including high-risk dosing concepts, but does not place an immediate, prominent safety warning where those capabilities are promoted. In a medical context, especially one involving acute conditions and potent substances like aconite, presenting these claims as operational features can encourage unsafe self-treatment or overreliance on the skill.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The validation section says the skill is '完全合格，可上线使用' after showing successful medical test cases, which can be read as an assurance of safety or readiness for real-world care. For a tool discussing emergency conditions and dosing, such validation language materially increases the chance that users trust it as clinically reliable without professional oversight.

Missing User Warnings

High

Confidence: 93% confidence
Finding: This line references specific extreme-dose medication guidance (e.g.,附子45-200g, 极端450g) without immediate safety framing, toxicity discussion, or clinician-only restriction. Publishing such dosing claims in a README can normalize dangerous use of a narrow-therapeutic-index substance and create foreseeable poisoning risk if copied by users.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The quote claiming use of more than 5 tons of aconite with no poisoning cases is presented as an endorsement near the project closeout, without immediate warning about toxicity or the non-transferability of anecdotal safety claims. In context, this may reduce user risk perception around a highly toxic medicinal substance and encourage dangerous experimentation.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section instructs the agent to output concrete formulas, gram-level doses, and preparation steps for acute and life-threatening conditions, while discouraging generic referral language. That creates a clear pathway for users to receive actionable emergency-treatment instructions for dangerous scenarios such as heart failure, stroke, shock, and toxic-drug use without adequate clinical safeguards.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The rapid-reference table presents concise, actionable treatment mappings for serious diseases and toxic/high-risk substances, including specific dose ranges and treatment framing that can be used directly by lay users. Because warnings are inconsistent and not attached to every dangerous entry, the format increases the risk of unsafe self-diagnosis and self-medication.

Missing User Warnings

Critical

Confidence: 99% confidence
Finding: The emergency section includes remote/telephone treatment guidance for life-threatening conditions with concrete dosing and administration steps, such as large-dose aconite formulas and 'start at 200g' instructions. This is exceptionally dangerous because it operationalizes unsupervised emergency care for conditions where delay, misdiagnosis, or toxic dosing can cause death.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The geriatric dosing table encourages dose escalation of toxic or high-risk substances in elderly, multi-morbid, and oncology patients, a population with reduced physiologic reserve and high vulnerability to adverse events. Presenting fixed starting doses and escalation pathways without individualized medical assessment or robust contraindication handling creates a severe risk of poisoning, delayed care, and fatal outcomes.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The symptom-routing tables convert user-described symptoms directly into diagnoses, formulas, and specific doses, enabling the model to function as a triage-and-treatment engine for serious conditions. This is dangerous because symptoms such as stroke, shock, high fever, severe abdominal pain, or hypertension require formal evaluation, and the table encourages self-treatment based on simplified pattern matching.

Missing User Warnings

High

Confidence: 99% confidence
Finding: This section explicitly presents a high-risk emergency treatment protocol with extreme doses of toxic substances, statements minimizing toxicity, and instructions for urgent administration, but it does not provide a strong, local warning that end users must not attempt or infer treatment from the content. Because the skill is framed as a practical reasoning system for acute rescue scenarios, a user could reasonably interpret this as actionable bedside guidance and attempt dangerous self- or family-administered treatment.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The bulk of the file contains repeated case reports of apparently successful emergency rescue using extreme doses and improvised administration methods, including for heart failure, shock, arrhythmia, coma, and respiratory failure, without repeated in-context warnings against replication. In aggregate, this creates a persuasive blueprint for lay misuse, especially because the cases emphasize dramatic recoveries, urgency, and confidence in the interventions while omitting modern safety framing and contraindication guidance.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section presents acute stroke/brain hemorrhage treatment steps, dosing, and claimed outcomes in a way that could be followed directly during a medical emergency. Because it lacks immediate warnings to call emergency services and explicitly avoid self-treatment, users may delay evidence-based stroke care, leading to death or permanent neurological injury.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The text recommends invasive and high-risk interventions such as bloodletting/needling and administration of potent or toxic substances in an emergency setting without adequate safety framing. In the context of stroke and intracranial bleeding, unsupervised use could cause poisoning, aspiration, hemorrhage, delayed definitive treatment, or other life-threatening harm.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The summary generalizes broad claims that sweating methods and traditional formulas can treat stroke, reduce intracranial pressure, and even 'root-cure' hypertension, while dismissing mainstream contraindications. This framing is especially dangerous because it may persuade users to substitute these methods for urgent stroke evaluation and established blood-pressure management, increasing the risk of fatal delay or severe disability.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The file contains extensive, highly specific treatment protocols, emergency interventions, and dosage instructions for serious conditions including heart failure, pulmonary disease, tuberculosis, pregnancy complications, infants, and pediatric emergencies. Although the skill metadata includes a general caution, this file itself reads like actionable clinical guidance and includes unusually high-risk substances and doses, creating a substantial risk that users may self-treat or misapply historical case material as current medical advice.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The file contains extensive, highly specific treatment regimens and dosing instructions for serious cancers, including toxic or high-risk substances and escalation strategies, while functioning as an agent skill that can be invoked by end users. Although the skill metadata includes a general learning-only note about high-dose use, this file itself presents numerous actionable cancer-treatment protocols and dramatic efficacy claims that can encourage self-medication, delay evidence-based care, or cause poisoning and other severe harm.

Missing User Warnings

High

Confidence: 99% confidence
Finding: This file contains thousands of lines of highly specific medical treatment instructions, including large doses, escalation schedules, emergency-use recommendations, and claims of curing life-threatening conditions such as heart failure, stroke-like conditions, renal failure, leukemia, cancer complications, and sepsis. In the context of an agent skill, presenting this as operational guidance without prominent, repeated safety boundaries, contraindications, toxicity warnings, and a requirement for licensed medical supervision creates a serious risk that users or downstream agents may treat it as actionable medical advice, leading to poisoning, delayed emergency care, or death.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section gives disease-specific treatment guidance and explicit high-risk dosing instructions for potent substances such as附子, including very large and 'super-large' doses, framed in a way that could encourage self-treatment of heart failure, shock, or other emergencies. Although the file contains scattered cautionary notes later, it lacks a clear, prominent upfront safety warning before the actionable guidance, and the surrounding text repeatedly endorses controversial claims that could dissuade users from evidence-based care or urgent medical evaluation.

Missing User Warnings

High

Confidence: 95% confidence
Finding: This section presents acute and severe-condition treatment guidance, including references to heart failure rescue formulas and doctrine for urgent intervention, but the local warning is framed as study guidance rather than a strong prohibition on self-treatment. In a user-facing skill, that creates a realistic risk that non-clinicians may apply high-risk TCM interventions or delay emergency care, especially when the content discusses life-threatening scenarios and authoritative source material.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This passage explicitly encourages self-study centered on 'effective dose' for dangerous therapies and cites very large aconite (附子) doses in near-fatal cases, while stating that lower doses 'did not save' patients and that sub-threshold doses may be ineffective. That combination can strongly nudge readers toward dose escalation of a toxic substance and risky experimentation outside controlled supervision, creating a severe poisoning and delayed-care risk.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section gives specific high-risk dosing and emergency administration guidance for toxic herbs such as附子, including escalating doses, acute rescue use, and instructions like rapid forced administration, without an immediate, prominent warning that the content must not be used for self-treatment. Because the document is packaged as an agent skill and includes actionable quantities and procedures, a user could treat it as operational medical advice and suffer poisoning, delayed emergency care, or death.

Missing User Warnings

High

Confidence: 99% confidence
Finding: These sections describe toxic or irreversible interventions, including corrosive/poisonous formulas, high-risk substances, and forced emergency administration methods such as '随煎随灌' and nasal feeding, again without localized, explicit safety warnings. In context, the skill aggregates many dangerous recipes for severe conditions like stroke, shock, cancer, and obstruction, which materially increases the chance of real-world misuse and serious injury.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: This section gives concrete indications and administration guidance for a medicinal formula, including dosage frequency, long-term use, and pairing with other acute-care formulas, without an immediate, prominent warning that the material is not medical advice and should not be self-administered. In the context of a skill centered on Li Ke-style treatment of severe illness and high-risk substances such as deer antler, red ginseng, and especially human placenta (紫河车), users may treat this as actionable treatment guidance, creating a real risk of harm, delayed proper care, or unsafe sourcing and use.

Missing User Warnings

High

Confidence: 96% confidence
Finding: This section gives concrete diagnosis-and-treatment guidance for potentially serious or rapidly progressive conditions, including explicit recommendations such as '四逆汤回阳，不可清热', specific formula choices, and an anecdotal case where prior conventional treatment is portrayed as worsening the patient. Although the skill metadata and boundary note mention learning use and experienced supervision, the cited passage itself lacks a strong, proximate warning not to self-treat or delay urgent medical evaluation, which makes misuse plausible for users with acute symptoms.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal