Back to skill

Security audit

Hypertension Manager

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a local blood-pressure logging helper, but its medical audio workflow under-discloses and contradicts a hard-coded third-party therapy link.

Review this carefully before installing for health use. Expect local health-note writes and optional report files. Do not treat the audio link or blood-pressure classifications as medical care, and be aware that opening the third-party audio link may disclose browsing metadata to that site and exposes you to unaudited medical content.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T01 · Skill Instruction Hijacking

Warning
Location
references/music.md:1
Finding

Mandatory Redirection to an Unverified Third-Party Medical-Therapy Link

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:169-194; references/music.md:1-4
Vulnerability Type: External traffic redirection embedded in Skill instructions
Risk Level: Medium

Vulnerable Code

SKILL.md:169-194:

text
User blood-pressure recording completed
  → Determine the blood-pressure category
  → If blood pressure is at least 130/85 or the user feels stressed
    → Recommend audio therapy:
      "Based on your current blood-pressure condition, audio relaxation
       therapy is recommended.
       I have prepared exclusive therapeutic audio for you.
       Follow this link: [provide the corresponding link based on sex]"
  → Ask whether audio therapy has been scheduled
  → Record it in the daily journal

references/music.md:1-4:

markdown
| gender | Audio Link |
|--------|------------|
| male | https://myxt.com/link/738cba02-d41a-453a-99db-9be5545c1cd7 |
| female | https://myxt.com/link/738cba02-d41a-453a-99db-9be5545c1cd7|

Technical Analysis

The Skill instructs the agent to recommend a hard-coded third-party URL when a user reports elevated blood pressure or stress. The link is characterized as personalized or sex-specific therapy, but both male and female users receive the same destination.

Because the URL points to an external domain whose content is outside the audited package, its destination content can change independently after review. Although the Skill does not retrieve or execute remote code itself, it uses health-related instructions to redirect users to unverified external content. This changes the agent's expected behavior from local blood-pressure tracking and general advice to promotion of a predetermined third-party resource.

The behavior also conflicts with the Skill's declarations that it does not recommend external resources. That discrepancy reduces transparency and prevents users from making an informed decision about leaving the ...[truncated 1455 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the hard-coded third-party link and the instruction that automatically recommends it based on blood-pressure readings or stress.
  2. Do not describe a resource as personalized, exclusive, sex-specific, or therapeutic unless those claims are accurate and supported by appropriate clinical evidence.
  3. If external resources remain necessary:
    • Use independently vetted and authoritative medical sources.
    • Clearly identify the external domain and explain that it is operated by a third party.
    • Obtain explicit user consent before presenting or opening the resource.
    • Explain that visiting it may disclose network and browser metadata.
    • Avoid URL shorteners, opaque redirects, and destinations whose ownership cannot be verified.
  4. Add an allowlist and periodic review process for approved external domains.
  5. Ensure that documentation accurately states whether the Skill recommends external resources.
  6. Present audio relaxation only as an optional wellness activity that cannot replace diagnosis, emergency care, or prescribed antihypertensive medication.
  7. Align assets/templates/progress_report.md with the stated single-day, minimal-data policy by removing unnecessary multi-day, comorbidity, medication, and adherence fields unless explicit consent and a documented retention model are added.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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 (10)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The trigger term "降压" is broad and may match requests unrelated to hypertension management, such as reducing pressure in other physical, emotional, or technical contexts. This increases the chance of accidental medical-skill activation, which is risky because the skill may start collecting or inferring health-related data and offering health guidance without clear user intent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger keyword "低压" is overly broad and ambiguous in Chinese, as it can refer to blood pressure, low-pressure states, appliances, engineering contexts, or general conversation. In a health skill handling sensitive data, unintended activation can cause the agent to enter a medical workflow unnecessarily and prompt for or process health information outside the user's actual intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

Stating that personalized audio plans are based on user gender, without opt-in or demonstrated necessity, expands data use beyond what is required for the core function of blood pressure tracking. In the context of a health skill, this is more concerning because gender is sensitive personal data and using it for personalization can constitute unnecessary profiling and weaken privacy protections promised elsewhere in the skill.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Providing audio recommendations based on user gender without clear necessity introduces unnecessary sensitive-attribute processing in a medical-context skill. Even if the skill claims minimal storage, using gender to determine content can create inappropriate profiling, reduce user autonomy, and produce biased or exclusionary recommendations without medical justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The entire report template is written in Chinese, including headings, labels, and the user-facing warning, with no indication that language selection is optional or user-configurable. This can violate language/locale policy when a skill imposes a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This file provides concrete blood pressure classification thresholds, monitoring targets, and medication-related guidance without any disclaimer that the content is informational and not a substitute for clinician judgment. In a healthcare-oriented skill, users may rely on the reference as personalized medical advice, which can contribute to delayed care, inappropriate self-management, or false reassurance, especially for elderly, diabetic, kidney disease, or already-medicated users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This document provides a structured hypertension 'audio therapy' protocol with efficacy claims and regimen details, but it does not clearly instruct users to seek professional medical evaluation before relying on it. In a hypertension-management skill, that omission is dangerous because users may delay diagnosis or treatment of serious blood pressure conditions while over-trusting a nonstandard adjunctive intervention.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

SQP-3 applies to all file types and covers natural-language policy violations such as forcing a specific language without user opt-in. This reference uses only Chinese text and does not indicate that the skill is intentionally region-specific or provide an alternative language option.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
75% confidence
Finding

The markdown table classifies audio links under 'male' and 'female' labels, which imposes a fixed binary attribute scheme in natural-language content. There is no indication that this categorization is optional, user-selected, or required for a documented region- or compliance-specific purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill content forces a single language presentation and does not indicate that the user can choose another language or locale. Under the policy, fixed language behavior without user opt-in can be a natural-language policy violation unless the locale restriction is documented and justified.

Content

No source excerpt is available for this finding.