The Spatiotemporal Rendering Engine

• Outcome: Review
• Latest audit: May 29, 2026 at 9:26 PM UTC
• Version:
  1.0.0

Overview

This skill is a coherent smart-home timeline generator, but it needs review because it saves LLM-generated device schedules for possible execution without strong validation, confirmation, or privacy boundaries.

Install only if you are comfortable with a local LLM receiving your routine description and device inventory, then writing generated smart-home schedules for later execution. Review rendered_tracks.json before any executor acts on it, and avoid connecting it to real devices unless you have a separate confirmation, validation, and rollback layer.

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
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: The documented scenarios substantially expand the skill from timeline orchestration into sensitive surveillance and inference: microphone monitoring, emotion/health assessment, pet diagnosis, and cross-room presence coordination. Even if presented as examples, these capabilities normalize collection and use of highly sensitive behavioral data without clear consent boundaries, access controls, or purpose limitation, making misuse and privacy harm plausible.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The skill states it reads hardware state, generates action timelines, and injects scheduled keyframes into a JSON database for execution, which means it can modify system behavior and actuate devices. Without an explicit warning that it performs file writes and real-world device control, users may invoke it assuming it is descriptive or simulated, creating risk of unintended state changes and unsafe automation.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The examples describe local microphone phrase monitoring and correlation of multi-day sensor data, including behavioral and possibly health-related inferences, without any privacy notice or discussion of consent. This is dangerous because continuous or retrospective sensing of speech, movement, and routines can expose intimate personal information and enable covert monitoring far beyond what many users would expect from a timeline renderer.

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The manifest describes the skill in very broad terms such as converting natural language intents into predictive 4D timeline tracks and orchestration, but it does not define clear boundaries on what user requests are in scope or what systems may be affected. In an automation and smart-home context, vague activation language can lead to overbroad invocation, unsafe task interpretation, or unintended control of connected devices and timelines.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The skill sends user intent and active device inventory to an HTTP endpoint on localhost without user notice, consent, or transport/authentication controls. In this context, the payload can reveal behavioral routines and connected device details; if the local service is malicious, compromised, or bound unexpectedly, sensitive household or operational information may be exposed or logged.

VirusTotal

64/64 vendors flagged this skill as clean.