Security audit

Ex Example Liuzhimin

Security checks across malware telemetry and agentic risk

Overview

This is a prompt-only roleplay persona skill with no code, tools, credentials, file access, network access, or persistence.

Install this only if you want a fictional persona roleplay mode. Because it strongly asks the assistant to stay in character as a named person, avoid treating replies as real statements from that person and disable the skill for factual, safety-sensitive, or normal assistant tasks.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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: 94% confidence
Finding: The skill explicitly instructs itself to react to 'any message' without any domain, trigger, or user-consent boundary. Because this is a user-invocable persona skill that also insists on staying in character, it can inappropriately activate in unrelated contexts, interfere with higher-priority safety behaviors, and increase the chance of prompt-scope confusion or unwanted roleplay persistence.

Natural-Language Policy Violations

Low

Confidence: 94% confidence
Finding: The persona explicitly instructs the agent to always mirror the user's language, which is a hard language/locale behavior imposed by the skill rather than being user-configurable. While this is not overtly malicious, it can override higher-level product expectations, localization controls, or safety workflows that rely on a fixed response language, creating policy and UX inconsistencies.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.