Elderly Facial Asymmetry / Mouth-Corner Deviation Detection | 老年人面部不对称/口角歪斜识别

• Outcome: Review
• Latest audit: May 25, 2026, 11:18 AM
• Version:
  1.0.0

Overview

This health-monitoring skill has a real facial-analysis purpose, but it handles sensitive face/health data with broad cloud, identity, token-storage, and mismatched backend behaviors that need human review before use.

Install only after reviewing the publisher and privacy terms. This skill can transmit elderly facial images/videos and identifiers to LifeEmergence/SMYX cloud services, create or reuse account tokens, store credentials locally, and expose historical reports. Do not use it for urgent medical decisions or with protected health data unless you have consent, retention/deletion controls, and confidence that the backend API actually returns validated facial-asymmetry results.

SkillSpector (26)

By NVIDIA

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: Requiring the skill to search local configuration files for an open-id and use user identifiers before analysis expands the scope from image analysis into credential and account-material access. In practice this can lead to unauthorized use of locally stored secrets or identifiers, linking sensitive facial-health data to accounts without clear user-driven authentication boundaries.

Description-Behavior Mismatch

Medium

• Confidence: 88% confidence
• Finding: A tool presented as a one-time screening capability also performs cloud history querying and report presentation, which broadens the data lifecycle and retention surface. For biometric and health-adjacent data, this increases privacy risk because users may not realize records are being stored remotely and can be enumerated later.

Context-Inappropriate Capability

Medium

• Confidence: 93% confidence
• Finding: Automatically saving uploaded face images or videos to local storage creates unnecessary retention of highly sensitive biometric and health-related data. If the host is shared, compromised, or insufficiently protected, these files can be exposed long after the analysis completes.

Description-Behavior Mismatch

High

• Confidence: 96% confidence
• Finding: The API request for an elderly facial asymmetry/stroke-screening skill injects a pet-related parameter (`petType`) that is unrelated to the documented medical purpose. This strongly suggests code reuse or misrouting to an incorrect backend workflow, which can cause requests to be processed under the wrong model or tenant context and lead to incorrect medical risk outputs, data handling errors, or privacy/compliance issues for sensitive elder facial data.

Intent-Code Divergence

Medium

• Confidence: 92% confidence
• Finding: The inline comment explicitly states 'add pet type parameter,' which contradicts the skill's stated function as an elderly stroke-screening aid. In a health-monitoring context, such contradictory logic is a red flag because it indicates mismatched business logic that can degrade model correctness, misclassify sensitive inputs, or route medical data through unintended processing paths.

Description-Behavior Mismatch

Low

• Confidence: 83% confidence
• Finding: The script exposes a '--list' function that retrieves prior analysis records via skill.get_output_analysis_list(), but this capability is not reflected in the stated skill functionality and has no visible authorization checks in this file. In a health-monitoring context, undisclosed access to historical facial-analysis results can expose sensitive medical inferences or user activity data.

Intent-Code Divergence

High

• Confidence: 98% confidence
• Finding: The documented API endpoint and response schema describe a generic remote video-analysis service that returns TCM-style constitution and organ-condition outputs, which is materially inconsistent with the skill’s stated purpose of facial asymmetry and stroke-risk screening. In a health-monitoring context, this mismatch can mislead integrators and users into trusting medically irrelevant outputs, causing unsafe decisions, delayed care, and undisclosed transmission of sensitive elderly facial video to a third-party service.

Description-Behavior Mismatch

High

• Confidence: 97% confidence
• Finding: The response examples return constitution, organ-condition, complexion, and lifestyle suggestions rather than asymmetry scores, facial landmark metrics, or stroke-screening indicators promised by the skill metadata. This discrepancy is dangerous because downstream consumers may build alerting or caregiving workflows on invalid semantics, leading to false reassurance or inappropriate medical escalation for a vulnerable elderly population.

Context-Inappropriate Capability

Medium

• Confidence: 74% confidence
• Finding: The delete(cameraSn) method performs a deletion operation based solely on a camera serial number, which introduces device-administration behavior not clearly necessary for a facial asymmetry screening feature. In a home-elderly-care context involving always-on cameras and sensitive health data, unnecessary device-level destructive actions increase the risk of unauthorized record removal, operational disruption, or misuse if this API is exposed through a broader agent surface.

Description-Behavior Mismatch

Medium

• Confidence: 89% confidence
• Finding: The code accepts arbitrary http/https URLs and forwards them to the backend analysis service, even though the skill description frames the feature as using a fixed home camera/local capture workflow. This broadens the trust boundary: an attacker or careless integrator could cause analysis of third-party remote content, potentially enabling unauthorized surveillance inputs, backend SSRF-like fetch behavior in the downstream service, or ingestion of data outside the expected consent model.

Description-Behavior Mismatch

Low

• Confidence: 78% confidence
• Finding: The skill exposes report listing and export-link generation capabilities beyond the manifest's narrow screening workflow. In a health-monitoring context, expanding functionality to enumerate historical reports and retrieve export URLs increases the chance of exposing sensitive medical inferences or identifiers if access control is weak elsewhere in the stack.

Context-Inappropriate Capability

Medium

• Confidence: 84% confidence
• Finding: The script accepts arbitrary remote URLs and passes them into downstream analysis without any visible allowlist, scheme restriction, or network-safety validation. In a health-monitoring skill intended for fixed home-camera/local use, this expands the trust boundary and could enable SSRF-like backend fetching, unintended access to internal resources, or ingestion of attacker-controlled media.

Context-Inappropriate Capability

Medium

• Confidence: 88% confidence
• Finding: This module exposes broad generic CRUD and raw HTTP helper methods that are not constrained to the stated facial-asymmetry screening purpose. In a health-monitoring skill handling sensitive elderly biometric/medical data, such unrestricted request primitives can be repurposed to access unrelated endpoints, exfiltrate data, or perform unauthorized actions if other parts of the skill can influence the URL or request parameters.

Description-Behavior Mismatch

Medium

• Confidence: 88% confidence
• Finding: This module implements a generic local database layer with create, update, delete, and schema-mutation behavior for user records, which is unrelated to the declared facial-asymmetry screening purpose. In a health-monitoring skill, hidden persistence of user/account data expands the data-collection surface and creates privacy and compliance risk because operators may not expect identity storage at all.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The User model stores identity and authentication-style fields including username, email, token, and open_token, none of which are necessary for simple facial asymmetry screening as described. Storing such credentials or session artifacts in a local SQLite database materially increases the harm from device compromise, especially in an elderly-care setting handling sensitive health-adjacent data.

Description-Behavior Mismatch

High

• Confidence: 97% confidence
• Finding: The utility layer silently performs external account lookup/creation via /sys/phoneLogin, acquires tokens, and persists them locally through the DAO, which is outside the stated purpose of facial asymmetry screening. In a health-monitoring context, this creates undisclosed identity linkage and credential handling, expanding the data collection and attack surface significantly beyond what a user would reasonably expect.

Context-Inappropriate Capability

Medium

• Confidence: 91% confidence
• Finding: The code handles account-balance failure states and returns recharge instructions unrelated to the medical screening function, indicating hidden monetization and account coupling inside a general request path. This is dangerous because it mixes healthcare-facing behavior with payment/account mechanics, increasing the chance of deceptive UX, unauthorized billing flows, or user confusion during potentially urgent stroke-related alerts.

Vague Triggers

Medium

• Confidence: 90% confidence
• Finding: The trigger conditions are broad enough to auto-activate on uploaded elderly face images or loosely related keywords, increasing the chance of processing sensitive biometric data without specific, informed user intent. In a medical-screening context, accidental invocation can result in unintended collection, transmission, and storage of private health-adjacent information.

Missing User Warnings

High

• Confidence: 98% confidence
• Finding: The skill lacks a clear upfront warning that sensitive facial images, inferred health-risk information, and related identifiers may be transmitted to and stored/queryable from cloud services. Because this involves biometric and medical-adjacent data for elderly users, the absence of prominent notice and consent materially increases privacy, compliance, and trust risks.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The tool requires an open_id that may be a username or phone number and stores it in a process-wide constant without any privacy warning, minimization, or masking. Because this skill processes health-related facial asymmetry data for elderly users, linking identifiable user data to potential stroke-screening results increases privacy and compliance risk if logs, memory, or downstream components are exposed.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The documentation instructs clients to upload videos or provide public video URLs together with an API key, but it omits any privacy, consent, storage, retention, or security guidance for highly sensitive biometric and health-related data. In this skill’s context—continuous monitoring of elderly people in homes or care facilities—such omissions materially increase the risk of covert collection, overexposure via public URLs, and noncompliant handling of protected data.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The function reads local file contents into memory and sends them, or forwards a remote video URL, to an analysis API without any user-facing notice in this file about transmission of facial/video health data. Because this skill processes elderly facial imagery for stroke-related screening, the data is highly sensitive biometric/health information, making silent upload or forwarding especially risky from a privacy and compliance perspective.

Missing User Warnings

Medium

• Confidence: 86% confidence
• Finding: Initializing the DAO automatically creates local storage and executes ALTER TABLE against the database without any operator confirmation, migration control, or safety checks. Unannounced schema mutation can corrupt deployments, create unauthorized local data stores, and silently expand persisted personal data in a context that should be tightly controlled due to elderly health monitoring.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: When debug mode is enabled, HTTPConnection and urllib3 debug logging can emit full request/response details, which may include authentication headers, identifiers, and health-related payloads. For a system processing elderly facial analysis data, this can leak sensitive medical and account information into logs accessible to operators or other processes.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The request helper automatically attaches user identifiers, tenant metadata, API keys, and authorization tokens to outbound requests without any visible consent or purpose limitation. In a healthcare monitoring skill, undisclosed transmission of identity and auth data is especially risky because it may expose patient-linked information to external services and broaden the blast radius if endpoints or logs are compromised.

Static analysis

Install untrusted source

Warn

• Finding: Install source points to URL shortener or raw IP.

Dep not found on registry

Critical

• Finding: 1 package(s) referenced in dependency files do not exist on their public registries: yaml (pypi)

VirusTotal

VirusTotal findings are pending for this skill version.