create-virtual-boyfriend（虚拟男友）

• Outcome: Review
• Latest audit: May 29, 2026 at 9:34 PM UTC
• Version:
  1.0.0

Overview

This is a disclosed virtual companion skill, but it requests broad file read/write/edit tools and encourages persistent personal memory without clear privacy controls.

Install only if you are comfortable with a companion skill that may save personal details. Keep sensitive information out of it, verify where any memory file is stored, and prefer using it only when the platform can constrain Read/Write/Edit access and gives you clear review and deletion controls.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The quick-start section encourages invocation with extremely common phrases like "Hi" and other ordinary chat openers, which can cause accidental or ambiguous triggering in unrelated contexts. Because this is a user-invocable companion skill with broad conversational scope, overly generic activation guidance increases the risk of unintended engagement and prompt routing confusion.

Natural-Language Policy Violations

Medium

• Confidence: 82% confidence
• Finding: The skill instructs the agent to detect the user's first-message language and continue in that language throughout, without confirming the preference or allowing easy switching. This can lock users into an unintended language choice, reduce transparency, and create usability or safety issues if the first message is mixed-language, quoted text, or otherwise ambiguous.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The guide explicitly encourages storing user preferences, habits, important dates, stories, goals, and emotional moments, which are all personal or potentially sensitive data. Because it provides no privacy warning, minimization guidance, consent language, or restrictions on sensitive categories, it normalizes broad retention of intimate user data in a relationship-oriented skill where oversharing is likely.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The example memory file demonstrates persistent storage of detailed personal information and behavioral patterns, such as birthday, work schedule, preferences, and relationship cues, without any warning about sensitivity or safe handling. Concrete examples like this operationalize unsafe data collection by showing authors and users exactly how intimate profile data may be retained.

Ssd 3

Medium

• Confidence: 95% confidence
• Finding: The memory commands instruct the assistant to disclose a broad summary of what it remembers about the user on request, which can expose aggregated personal data and conversation history. In a virtual-boyfriend context, this is more dangerous because the interaction is designed to elicit intimate disclosures, making the recalled summary potentially highly sensitive if shown to the wrong person or invoked without adequate authentication or context checks.

VirusTotal

64/64 vendors flagged this skill as clean.