何小鹏思维视角 v2

Security checks across malware telemetry and agentic risk

Overview

This is a text-only perspective skill for smart-car and autonomous-driving analysis, with no code, credentials, network access, or system-level behavior.

Safe to install as a writing and strategy-analysis aid. Be aware that it may activate on broad smart-car or autonomous-driving phrases, and its outputs should be treated as a simulated public-information perspective, not real quotes, endorsements, or verified business or technical advice.

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 (2)

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The trigger list includes broad domain phrases such as “互联网造车”, “自动驾驶落地”, and “智能汽车产品定义”, which are likely to appear in ordinary discussion and can activate the skill unintentionally. This creates prompt-routing risk: the assistant may switch into a constrained persona or apply this skill when the user did not explicitly request it, degrading instruction fidelity and potentially overriding safer or more appropriate behavior.

Natural-Language Policy Violations

Medium

Confidence: 88% confidence
Finding: The skill specifies first-person persona responses and a fixed Chinese-language style without giving the user a language-choice or non-persona fallback. If activated unexpectedly, this can force outputs into an unwanted language and identity frame, increasing the chance of user deception, reduced transparency, and failure to follow user preferences or higher-priority system behavior.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal