Security audit
Security checks for vulnerabilities and agentic risk
The skill largely matches its child-safety video-analysis purpose, but it quietly links sensitive child-monitoring footage and report history to a persistent remote identity and cloud account flow.
Review before installing. This skill sends child/home monitoring images or video references to external cloud services and links activity to a persistent user identity that can be silently created and reused. Use it only if you are comfortable with cloud processing, report-history storage/retrieval, and token persistence in the workspace data directory; prefer explicit consent, narrow prompts, trusted video URLs, and clear retention/deletion controls.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
The described skill scope does not mention remote authenticated request handling, local workspace management, user/open-id generation, token management, or automatic account creation, yet those are substantial behaviors. In the context of children's home-camera footage, undeclared identity provisioning and backend account creation create serious privacy, consent, and data-governance concerns.
This utility code performs remote account provisioning and login flows that are unrelated to the stated child dangerous-object detection purpose. It can silently create or reuse identities, obtain tokens, and bind the skill to a remote platform, expanding data exposure and operational scope far beyond what users would expect from a home-safety video analysis skill.
The skill invokes capabilities associated with shell execution, network access, local file read/write, and environment access, yet the manifest does not declare any corresponding tool scope or permission boundaries. This increases the risk of over-privileged execution and makes it harder for reviewers or runtime policy to constrain what the skill can do, especially since it also instructs running local Python scripts and handling remote URLs.
The default trigger condition is broad enough to auto-run on any uploaded child-area monitoring video, even absent a clear request for dangerous-object detection. In a privacy-sensitive context involving children's home footage, overbroad invocation can cause unnecessary analysis, uploads, and retention of data the user did not intend to process with this skill.
The skill scope expands from real-time detection/alerting into cloud history-report querying and report-link retrieval without making that expansion central in the headline purpose. This is dangerous because it introduces additional data access paths for sensitive child-monitoring records and may expose past analyses beyond what a user expects from a live safety skill.
The historical-report trigger list is broad and causes automatic cloud lookups without clear scope limits. That can expose or retrieve sensitive child-safety records based on loose phrasing, increasing the chance of unintended disclosure or access to more report data than the user specifically requested.
Automatic creation or reuse of a local default user identity is not clearly necessary for analyzing a provided video and introduces silent identity binding for child-safety data. This can lead to cross-session linkage, unintended data aggregation, and reduced user control over where minors' analysis records are stored and retrieved.
The API doc includes exporting full reports and sharing snapshots to apps or smart speakers, yet it gives no warning that these artifacts may contain sensitive images, timestamps, and in-home activity involving children. Without guidance on secure sharing, authorization, redaction, and retention, the feature can expose private household surveillance data to unintended recipients or leak sensitive evidence beyond its intended purpose.
The documentation explicitly recommends 24/7 camera collection in home spaces such as children's rooms and describes alerting workflows, but provides no privacy notice, consent requirements, retention limits, or safeguards for surveillance of minors. In this context, the omission is dangerous because the system processes highly sensitive in-home video of children, increasing risks of unlawful monitoring, overcollection, and misuse of intimate behavioral data.
Line L17 returns a fixed Chinese-language string for analysis output. This imposes a specific locale on all users with no visible language selection, fallback, or documented region-specific justification in this file.
The script exposes a hidden --list mode that retrieves a per-user analysis history via skill.get_output_analysis_list(open_id=...), which is outside the manifest’s stated real-time hazard detection/alarm purpose. In a home child-safety surveillance context, retaining and listing prior analyses can reveal sensitive behavioral and monitoring history, creating an unnecessary privacy and data exposure surface.
The script accepts a remote --url video source for analysis but provides no warning that home monitoring footage may be fetched from or transmitted over a network. Because this skill processes videos from children's living spaces, the absence of clear disclosure and transport/privacy constraints materially raises the risk of inadvertent leakage of sensitive footage.
The code resolves and uses an internal open_id identity even though the advertised function is simply analyzing a supplied video. Tying analysis operations to a hidden internal identifier introduces unnecessary identity linkage for highly sensitive child-monitoring data and can enable unauthorized access to user-specific records if identity handling is weak elsewhere.
OpenIdUtil.resolve_current_open_id(args.open_id, use_current=bool(args.open_id)) initializes internal user identity without any visible user-facing warning or consent flow. In the context of child safety monitoring, silently resolving identity increases privacy risk because sensitive household surveillance activity may be linked to a specific account without the operator understanding that this is happening.
Detected: suspicious.install_untrusted_source
