Security audit

语音车辆控制助手

Security checks across malware telemetry and agentic risk

Overview

This instruction-only vehicle control skill is coherent, but it can trigger real-world car actions without enough explicit confirmation and safety boundaries.

Review carefully before connecting this to a real vehicle. Use it only with a trusted, scoped vehicle API; require explicit confirmation for every control action and scene; separate status-only access from control access; and verify that vehicle state checks, audit logging, rollback/stop behavior, and token protection are enforced outside the skill.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The trigger scope is broad for a skill that can issue safety-critical vehicle commands such as engine start, door unlock, and window control. Without clear activation boundaries, ordinary conversation or loosely related vehicle speech could be misinterpreted as an execution request, increasing the risk of unintended remote actions.

Missing User Warnings

High

Confidence: 93% confidence
Finding: The skill enables remote control of safety-critical vehicle functions but does not prominently warn users that it can perform consequential physical actions. This omission can cause users to invoke it casually or without understanding the operational and safety implications, especially in voice-driven contexts where accidental activation is more likely.

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The trigger phrases include broad natural-language terms such as '露营', '过夜', 'camping', and 'overnight', which can be spoken in casual conversation and may unintentionally activate a scene that changes locks, windows, climate, and seat positions. In a voice-controlled vehicle context, accidental activation can directly affect occupant safety, security posture, and vehicle state without sufficiently deliberate user intent.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill supports multi-step vehicle control sequences that can alter navigation, windows, locks, climate, lighting, and seating, yet the documentation does not warn users that one spoken request may execute several state-changing actions. In a vehicle environment, lack of clear warnings increases the risk of unsafe or surprising automation, especially where commands may continue on failure and include delayed actions.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The protocol exposes safety-critical actions such as engine start/stop, door unlock, horn, lights, and other physical vehicle controls without documenting any confirmation, authorization step-up, occupancy/driving-state checks, or user-facing safety warnings. In a voice-controlled vehicle skill, this omission can enable accidental, unauthorized, or context-inappropriate execution of commands with real-world physical and security consequences.

VirusTotal

60/60 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.