Back to skill

Security audit

Blood Pressure Therapy

Security checks for vulnerabilities and agentic risk

Overview

This health skill is coherent in purpose, but it needs Review because it persistently stores sensitive health data without clear consent or deletion controls and makes strong medical claims for an audio intervention.

Review carefully before installing. Do not enter sensitive health details unless you are comfortable with them being saved to both a local file and long-term agent memory. Treat the audio link as unverified wellness content, not a substitute for medication, clinician advice, urgent care, or validated blood-pressure monitoring.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:62
Finding

Persistent Storage of Sensitive Health Data Without Adequate Privacy Controls

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:62-82; api/profile_inquired.md:27-31
Vulnerability Type: Unprotected persistent storage of sensitive health information
Risk Level: Medium

Vulnerable Instructions

The following is a faithful English rendering of the complete relevant segment from SKILL.md:62-82:

markdown
First check whether the local `hc_profile.md` file exists:

- If it exists: Read the existing profile data and confirm whether the information needs to be updated.
- If it does not exist: Collect the user's basic health data according to the rules in
  `api/profile_inquired.md`. After collection, save it to `hc_profile.md` in JSON format.

Information to collect (10 items):

| Name | Key |
|---|---|
| Genetic and family history | geneticHistory |
| Age | age |
| Sex | gender |
| High-salt diet | isHighSalt |
| Overweight or obesity | isFat |
| Sedentary lifestyle or insufficient exercise | isLackExercise |
| Staying up late or severe sleep deprivation | isLackSleep |
| High stress, anxiety, or irritability | hasStress |
| Smoking or excessive alcohol consumption | isSmokerDrink |
| High-fat and high-sugar diet | isHighOilSugar |

Note: All collected data must be stored in long-term memory using the
`update_memory` tool so that it can be retrieved in subsequent conversations.

The storage instruction in api/profile_inquired.md:27-31 additionally states:

markdown
Ensure that the JSON is displayed in pretty format.

Store the JSON data generated above by overwriting:
`./hc_profile.md`

Technical Analysis

The Skill collects health and behavioral information including age, sex, family medical history, obesity, sleep patterns, stress, smoking, alcohol use, diet, and blood-pressure context. This is sensitive personal health information.

The instructions require the data to be persisted through two separate mechanisms:

  1. A predictable ...[truncated 2473 chars]
Remediation
View remediation

Remediation Suggestions

  1. Default to session-only processing and do not persist health information unless persistence is necessary for a user-requested feature.
  2. Obtain explicit, informed, opt-in consent before storage. Request separate consent for local storage and long-term Agent memory.
  3. Remove the unconditional update_memory requirement and avoid storing duplicate copies.
  4. Collect only the minimum fields needed for the immediate assessment.
  5. Replace the predictable shared Markdown path with a user-scoped, access-controlled data store.
  6. Encrypt sensitive records at rest and restrict access to the current authenticated user.
  7. Define and enforce a short retention period.
  8. Provide commands for users to inspect, correct, export, and permanently delete their records.
  9. Bind every profile to a verified user or session identity and reject retrieval when identity isolation cannot be guaranteed.
  10. Validate the schema and prevent stale or cross-user data from being treated as the current user's medical profile.
  11. Clearly disclose what information is collected, why it is needed, where it is stored, and who can access it.

other

Warning
Location
references/principle.md:37
Finding

Unsupported Medical Efficacy and Safety Claims May Cause Unsafe Reliance

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:4, SKILL.md:50, SKILL.md:113-118, references/principle.md:37-40
Vulnerability Type: Unsupported medical treatment, efficacy, and safety claims
Risk Level: Medium

Vulnerable Instructions

A faithful English rendering of SKILL.md:113-118 is:

markdown
- Hypertension healing audio:
  https://myxt.com/link/738cba02-d41a-453a-99db-9be5545c1cd7
- Explain to the user:
  - The audio lasts approximately 25 minutes. It is recommended that it be
    listened to with headphones in a quiet environment.
  - The audio uses specific frequencies to stimulate brain nerves, assist in
    activating the vagus nerve, and regulate hormone secretion.
  - It is recommended that blood pressure be measured again after use to
    observe changes.

The Skill also characterizes the product at SKILL.md:4 and SKILL.md:50 as:

markdown
A targeted non-drug audio adjunctive therapy system that takes effect rapidly
within 25 minutes.

Targeted non-drug audio treatment, effective in 25 minutes.

A faithful English rendering of the complete claims in references/principle.md:37-40 is:

markdown
1. Safety: Avoids drug side effects and metabolic burden.
2. Rapidity: Blood-pressure changes can be measured within 15 minutes.
3. Personalization: Precise AI-based matching and intervention.
4. Non-invasiveness: Only listening to music is required, without changing
   lifestyle habits.

Technical Analysis

The repository does not provide clinical studies, citations, validation results, contraindication analysis, regulatory status, uncertainty estimates, or applicability limitations supporting the claimed intervention.

In particular, the included AHA/ACC reference only provides blood-pressure classification thresholds. It does not substantiate that the linked audio:

  • Lowers blood pressure within 15 or 25 minutes.
  • Activates the va ...[truncated 2795 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the quantified 15-minute and 25-minute efficacy claims unless supported by applicable, peer-reviewed clinical evidence.
  2. Remove claims that the audio activates the vagus nerve, regulates hormones, or modifies specific metabolic pathways unless those exact claims are adequately substantiated.
  3. Do not describe the audio as a treatment, targeted therapy, or personalized clinical intervention without appropriate evidence and authorization.
  4. Present the audio, at most, as optional relaxation or wellness content that is not proven to treat hypertension.
  5. State prominently that it must not replace prescribed medication, clinician review, validated blood-pressure monitoring, or evidence-based lifestyle measures.
  6. Remove the assertion that users need not change lifestyle habits.
  7. Avoid broad safety claims. Document known limitations, hearing-related considerations, and situations in which headphone use may be inappropriate.
  8. Provide standardized blood-pressure measurement instructions, including rest, posture, cuff fit, repeated readings, and device validation.
  9. Add symptom-based emergency guidance and clarify that very high readings require prompt reassessment and appropriate medical care.
  10. Cite the complete source, population, intervention, comparison, outcome, and limitations for every retained clinical claim.
  11. Clearly identify the operator and privacy policy of the third-party link before directing users to it.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (12)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill instructs collection and long-term storage of sensitive health information but does not require an explicit privacy notice, consent flow, retention period disclosure, or deletion option. In a health-management context, this makes the issue more dangerous because users may disclose regulated or highly sensitive data without understanding how it will be stored or reused.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The description, workflow, warnings, and recommended phrasing are presented entirely in Chinese, and the skill does not state that language output is optional or user-selectable. This can violate language/locale policy when a skill implicitly forces a specific language without user opt-in or a documented region-specific justification.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger list is broad and generic for common blood-pressure-related phrases, increasing the chance the skill activates in situations where the user did not intend to enter a health-data collection workflow. In a medical context, unintended invocation can lead to unnecessary collection of sensitive health information or unreviewed health suggestions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly requires long-term memory storage of detailed health-profile data, including sensitive medical and lifestyle information, for future reuse. This exceeds what is minimally necessary to recommend an audio session and creates privacy, retention, and secondary-use risk if the data is later exposed, reused out of context, or accessed by other components.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The instruction to store all collected health data in long-term memory for later reuse creates an unnecessary persistence channel for sensitive medical information. Because the skill handles blood pressure, family history, age, lifestyle, and stress indicators, compromise or unintended reuse of this memory could materially affect user privacy and trust.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

All natural-language instructions are written as mandatory Chinese interaction steps, including asking questions and generating output, with no indication that the user may choose another language. This creates a language/locale constraint without opt-in or documented justification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill instructs the agent to collect sensitive health-related profile data and overwrite it into a local file (./hc_profile.md). Persisting medical-risk information to disk is security-relevant behavior, and because this storage behavior is not reflected in the stated functionality/manifest, it creates a transparency and consent gap around handling of sensitive personal data. In a health context, undisclosed local storage materially increases privacy risk if the file is later accessed, reused, or exfiltrated by other components.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The markdown explicitly instructs overwriting ./hc_profile.md, which is a state-changing file operation, but provides no user warning or confirmation step. Because the file contains health-profile information, silent overwrite can destroy prior records, corrupt expected state, or replace legitimate data without the user's awareness, increasing both integrity and privacy risks.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document explicitly states that the system uses sensitive health-related attributes such as age, sex, daily blood pressure, medication status, and inferred cause categories to generate personalized interventions, but it provides no privacy notice, consent model, retention limits, or handling safeguards. In a health context, this omission is risky because users may disclose regulated or highly sensitive medical data without understanding how it will be stored, processed, shared, or protected.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file presents the skill content entirely in Chinese, including the title and table labels, without indicating any language choice or opt-in. Under the policy, forcing a specific language without user selection is a natural-language policy concern unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The entire skill content is presented only in Chinese, with no indication that the skill is region-specific or that users can opt into another language. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The entire file is presented exclusively in Chinese, and there is no statement that the skill is China-specific or that users may choose another language. This can violate language/locale policy when a skill implicitly forces a specific language without opt-in or justification.

Content

No source excerpt is available for this finding.