Security audit
Security checks for vulnerabilities and agentic risk
This skill is coherent with its stated purpose but handles highly sensitive elder-monitoring footage through cloud APIs with automatic identity creation and local token persistence that deserve careful review.
Install only if the monitored elder and authorized caregivers have explicitly consented, cloud upload and report retention are acceptable, and local token storage in the workspace data database is understood. Review the provider's access controls, deletion policy, and legal requirements before using it in bedrooms, private nursing rooms, or with audio.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
The reported behavior includes authenticated external requests, local file access, identity creation/resolution, user-database interaction, and token handling, while lacking visible behavior-detection logic. In the context of continuous in-room monitoring of elderly individuals, this discrepancy is especially dangerous because it combines intimate data collection with opaque backend identity and storage operations.
YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).
---
name: "smyx-elderly-loneliness-comfort-analysis"
description: "Using a fixed camera in the home of a solitary-living elderly person or in a private nursing-home room, the system analyzes daily activity video and detects loneliness-related behaviors: prolonged solitude (no social interaction), static gazing (long-time fixation with no purposeful activity), sighing (rapid chest/abdomen rise-fall with exhale), and talking-to-self (mouth activity with no conversation partner). | 通过独居老人家中或养老院单人房的固定摄像头，分析老人日常行为视频，检测孤独相关行为：长时间独处（无社交互动）、静�
The skill involves continuous in-room audio/video monitoring and cloud report querying, yet the description does not foreground a clear warning about those privacy-invasive operations. In a private home or nursing-room setting, inadequate disclosure undermines informed consent and can conceal sensitive collection, processing, and sharing of behavioral data.
The model stores token and open_token fields despite the skill's stated function being behavioral video analysis, creating unjustified credential retention. Persisting authentication material in a local SQLite database is dangerous because theft of the DB can lead to account compromise, lateral movement, or unauthorized API access.
The code performs external login/registration and token provisioning through a remote service, which is outside the declared role of local video-behavior analysis. This creates an undisclosed account lifecycle and external identity linkage, potentially exposing sensitive user context and enabling backend tracking of deployments involving elderly residents.
The helper silently auto-registers or logs in a user account via a network call with registration enabled, using generated or reused identifiers and no visible warning. This is especially dangerous in a surveillance-oriented elderly-care setting because it can create backend accounts for monitored subjects or operators without informed consent and tie sensitive observations to those accounts.
The skill declares no explicit tool scope or permissions even though the documented behavior requires shell execution, network access, local file handling, and likely environment-variable use. In a privacy-sensitive elderly-monitoring context, this lack of least-privilege boundaries makes it harder to review, constrain, and safely deploy the skill, and it can mask unexpected data access or exfiltration paths.
The documentation centers actual operation on a remote/cloud API and cloud history retrieval rather than clearly explaining how behavioral detection is performed. That creates a security-relevant transparency gap: users may assume local analysis while sensitive footage and behavioral summaries are actually sent off-device to third-party services.
A broad default trigger that activates on any uploaded elderly-room video can cause inadvertent processing of highly sensitive in-home footage without sufficiently specific user intent. In this domain, accidental activation is especially harmful because it may initiate cloud upload, identity association, and historical record creation for vulnerable individuals.
The keyword trigger list overlaps with ordinary caregiving discussion and can invoke the skill in contexts where the user did not intend surveillance analysis or report lookup. Given the sensitivity of elderly mental-state inference, overbroad activation increases the risk of privacy-invasive processing and inappropriate recommendations.
This is a true privacy/security issue. The API section documents endpoints for starting analysis, retrieving results, paginating historical reports, and exporting full reports for continuous in-home elderly video/audio monitoring, but it does not place prominent consent, minimization, access-control, retention, and export-sensitivity warnings directly alongside those interfaces. In this context, the omission is dangerous because the skill processes highly sensitive behavioral surveillance data in private spaces and includes report export capability, increasing the risk of unauthorized deployment, overcollection, misuse, and excessive retention by integrators.
The elderly-loneliness analysis entrypoint still accepts and mutates a global pet-related setting, which indicates the skill was repurposed from another domain without fully removing incompatible semantics. In a sensitive healthcare-like context, this can cause incorrect model selection or misleading outputs, undermining trust and potentially affecting decisions about an elderly person's wellbeing.
The CLI presents the tool as loneliness analysis for elderly people, but the implementation wiring still exposes pet-type semantics, creating a mismatch between user expectations and actual behavior. In a surveillance-style health monitoring setting, this is dangerous because operators may rely on outputs that were generated under irrelevant or misconfigured model assumptions.
This code performs highly sensitive video analysis of elderly individuals in private spaces without presenting any privacy warning, consent check, or safeguard before processing. Given the context involves home or private-room surveillance and infers vulnerable-person mental state indicators, silent processing increases the risk of unauthorized monitoring and privacy harm.
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
result_json = JsonUtil.parse(result_json_pure_text, result_json_pure_text)
result_json_common_ai_response = result_json.get("commonAiResponse") if isinstance(result_json,
dict) else result_json
if result_json_common_ai_response:
result_json = result_json_common_ai_response
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
result_json = JsonUtil.parse(result_json_pure_text, result_json_pure_text)
result_json_common_ai_response = result_json.get("commonAiResponse") if isinstance(result_json,
dict) else result_json
if result_json_common_ai_response:
result_json = result_json_common_ai_response
Detected: suspicious.install_untrusted_source