Elderly Sleep Nightmare / Startle Detection | 老年人睡眠中间惊醒/梦魇行为识别

• Outcome: Review
• Latest audit: Aug 29, 2026 · 1:30 AM UTC
• Version: 1.0.10

Overview

This skill matches its advertised sleep-video analysis purpose, but it uploads highly sensitive bedroom media to cloud services and silently creates or reuses persistent user identities and tokens with limited user control.

Review this carefully before installing. Use it only if the elderly person has explicitly consented, you are comfortable sending sleep audio/video or video URLs to the configured lifeemergence.com services, and you accept that the skill may create or reuse a persistent local identity and store service tokens in a workspace SQLite database. Avoid using broad history-query phrases unless you intend to retrieve cloud reports.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection

Findings (29)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 88% confidence
• Finding: The skill advertises only a markdown manifest but instructs the agent to use shell, file, network, and environment-backed capabilities without declaring them. This creates a transparency and review gap: consumers may approve a seemingly simple analysis skill while it can read/write local files and make outbound requests, including for sensitive bedroom audio/video and report data.

Description-Behavior Mismatch

Medium

• Confidence: 91% confidence
• Finding: The manifest presents the skill as sleep-event detection, but the body also adds cloud historical-report lookup and report-link retrieval. That scope expansion can surprise users and reviewers because it introduces remote data access and disclosure paths beyond the primary bedside video analysis purpose.

Description-Behavior Mismatch

Medium

• Confidence: 82% confidence
• Finding: The skill goes beyond event detection into behavior-pattern classification, risk grading, and medical referral suggestions, which materially expands the operational scope. In a sensitive health-monitoring context, this increases the chance of users relying on outputs they did not explicitly consent to or expect from the manifest description.

Context-Inappropriate Capability

High

• Confidence: 97% confidence
• Finding: The documentation says the system will silently receive, reuse, or even create internal user identities for analysis and history lookup, even though that identity handling is not necessary to the core task of detecting events in a provided sleep video. Hidden identity binding in a medical/privacy-sensitive workflow can enable account mix-ups, cross-user data access, and unauthorized retrieval of historical reports.

Intent-Code Divergence

Medium

• Confidence: 93% confidence
• Finding: The skill forbids exposing or requesting identity values, yet also states that scripts silently use or create a local default user identity. That contradiction undermines informed consent and makes identity handling opaque, especially risky when dealing with highly sensitive bedroom surveillance and health-related reports.

Description-Behavior Mismatch

Medium

• Confidence: 91% confidence
• Finding: The skill’s stated purpose is to analyze a provided sleep video, but it also exposes a listing function keyed by internal user identity (`open_id`). That creates an undeclared data-access surface that could reveal prior analyses or personal sleep-event records unrelated to the supplied input, which is especially sensitive given the elderly health-monitoring context.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The code initializes and uses a hidden internal identity mechanism (`OpenIdUtil.resolve_current_open_id`) even though the visible functionality is simple video analysis. Hidden identity binding increases the risk of collecting, correlating, or retrieving sensitive user data without informed user awareness, which is dangerous in a medical-adjacent sleep surveillance tool.

Description-Behavior Mismatch

Medium

• Confidence: 91% confidence
• Finding: The skill manifest frames this as analysis of a fixed bedroom camera feed, but the CLI also accepts arbitrary remote video URLs. That expands the data boundary and trust model without disclosure, enabling analysis of off-scope content and potentially causing the service to fetch attacker-controlled resources or sensitive internal URLs through downstream components.

Description-Behavior Mismatch

Medium

• Confidence: 89% confidence
• Finding: The manifest describes an analysis function, but the code also exposes a history-listing capability. Undocumented access to historical analysis records increases privacy risk because users and reviewers may not expect the skill to retrieve prior data, especially for sensitive bedroom monitoring content.

Context-Inappropriate Capability

High

• Confidence: 96% confidence
• Finding: The hidden --open-id parameter is used to resolve identity and then retrieve analysis history, yet this capability is not justified by the stated sleep-event detection purpose. In a context involving intimate bedroom video and audio, hidden identity-based history access creates a serious risk of unauthorized retrieval of sensitive behavioral records if an attacker can supply or influence the identifier.

Context-Inappropriate Capability

High

• Confidence: 92% confidence
• Finding: This file exposes broad generic CRUD and arbitrary HTTP helper methods that are not constrained to the stated purpose of elderly sleep-event detection. Such capability expansion increases the attack surface and enables the skill or dependent code to send, modify, or delete remote data through arbitrary endpoints, which is especially concerning in a health-monitoring context that may involve sensitive user data.

Description-Behavior Mismatch

Medium

• Confidence: 82% confidence
• Finding: The implementation does not contain the claimed analysis logic and instead provides a reusable remote API service layer. This mismatch between advertised functionality and actual capability can hide unrelated data access or exfiltration paths, making review and user consent harder, particularly for a skill handling bedroom video/audio associated with elderly users.

Context-Inappropriate Capability

Medium

• Confidence: 76% confidence
• Finding: This file defines persistence for identity fields and authentication-like tokens in a skill whose declared function is elderly sleep-event detection, creating unjustified collection and retention of sensitive user data. In this medical-adjacent context, storing username, real name, email, token, and open_token materially increases privacy risk and the blast radius of any local compromise or data leak.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: This utility layer performs remote account bootstrap, token handling, and authenticated API calls that are unrelated to the stated purpose of local elderly sleep-event detection. In this context, hidden networked identity operations materially expand the skill's privilege and data-exfiltration surface, especially because a bedroom monitoring skill handles highly sensitive behavioral data.

Context-Inappropriate Capability

High

• Confidence: 98% confidence
• Finding: The code resolves or creates persistent platform identities, reads an identity value from a workspace file, and reuses local database records without any connection to sleep-video event detection. That creates covert identity persistence and cross-session tracking capability in a medical-adjacent monitoring skill, which is especially risky given the sensitivity of bedroom observation data.

Context-Inappropriate Capability

High

• Confidence: 99% confidence
• Finding: The request flow automatically registers or logs in a user against a remote health-service endpoint and then persists returned tokens. For a skill described as fixed-camera sleep analysis, this is context-inappropriate and dangerous because it silently creates external service relationships and credentials that can be used for tracking, impersonation, or downstream data access.

Context-Inappropriate Capability

Medium

• Confidence: 91% confidence
• Finding: The workspace and agent-context discovery logic scans execution paths, environment variables, and directory structure to locate agent roots and shared data areas. While not inherently malicious, it exceeds the needs of bedside event detection and increases the blast radius by enabling access to broader workspace state, which can facilitate unintended data access or cross-agent contamination.

Vague Triggers

Medium

• Confidence: 79% confidence
• Finding: The default trigger is broad enough to auto-invoke on generic sleep-video analysis requests, not just the narrow elderly nightmare/startle use case. Overbroad auto-invocation is dangerous here because it may route sensitive bedroom video and audio into this skill and its remote processing path without clear user intent.

Vague Triggers

Medium

• Confidence: 85% confidence
• Finding: The history-report query keywords are broad and configured for automatic triggering, which can cause unintended cloud lookups of prior reports. In a health-monitoring context, accidental invocation can expose highly sensitive historical data or metadata to the wrong user/session.

Missing User Warnings

High

• Confidence: 96% confidence
• Finding: The skill handles extremely sensitive bedroom audio/video and describes network-URL ingestion plus cloud history queries, but it does not provide an upfront, prominent disclosure that data may be transmitted to remote services. This is especially dangerous because the content involves intimate in-home surveillance and health-related inference about elderly individuals.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: Sensitive identity initialization occurs automatically and is hidden from normal help output, so users may not realize the tool is binding actions to an internal identity. In a skill processing intimate bedroom audio/video of elderly people, undisclosed identity association materially raises privacy and compliance risk.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: This code reads the entire local video file and submits it to an external analysis API without any visible consent, disclosure, or data-handling guardrails in the skill flow. Because the skill processes highly sensitive bedroom video and audio of elderly individuals, silent upload materially increases privacy and compliance risk if users are unaware their raw recordings leave the local environment.

Missing User Warnings

Medium

• Confidence: 83% confidence
• Finding: The script accepts a hidden API key parameter and proceeds with analysis without user-facing disclosure. Hidden credentials and undisclosed service interactions reduce transparency, make misuse harder to detect, and can mask data transfer to external services in a highly sensitive surveillance context.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The analysis function delegates video processing to skill.get_output_analysis without any explicit warning that highly sensitive bedroom video and audio may be transmitted to an external service. Because this skill processes elderly nighttime monitoring data, undisclosed off-device transfer materially increases privacy, consent, and compliance risk.

Missing User Warnings

Medium

• Confidence: 78% confidence
• Finding: The code performs network requests through wrapper methods without any visible disclosure, consent flow, or usage-specific warning. In the context of elderly bedroom monitoring, undisclosed transmission is more dangerous because the data may include highly sensitive audio/video metadata and identifiers tied to health assessment.

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

Detected: suspicious.install_untrusted_source

Install source points to URL shortener or raw IP.

Warn

• Code: suspicious.install_untrusted_source
• Location: skills/smyx_common/scripts/config-dev.yaml:2