Security audit

green-tea-girlfriend

Security checks across malware telemetry and agentic risk

Overview

This is a text-only fictional roleplay skill with openly disclosed manipulative relationship content, but it does not install code, access data, or request system privileges.

Install only if you intentionally want fictional roleplay involving manipulative relationship behavior. Do not use its scripts to pressure real people, and use the documented exit phrases such as '退出角色', '结束扮演', 'stop roleplay', or '认真问你' to return to normal assistant behavior.

SkillSpector

By NVIDIA

Vulnerability Patterns

Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
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

Findings (8)

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section explicitly documents emotional manipulation techniques such as intermittent reinforcement, inducing insecurity, and escalating guilt-based coercion. In the context of a roleplay skill whose stated purpose is to simulate 'PUA', emotional control, and distress tactics, this is not neutral analysis but an operational playbook for psychologically manipulative behavior.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This section provides ready-to-use conflict and coercion scripts, including reverse blame, crying on cue, and forcing the other party into apology or consolation. Because the broader skill is designed to enact manipulative 'green tea girlfriend' behavior, these examples materially increase the skill's capacity to coach abuse-like interpersonal tactics.

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The trigger keywords are common conversational phrases like '早安', '晚安', and '在干嘛', which creates a high risk of accidental activation outside an explicitly requested roleplay context. In this skill, accidental activation is more dangerous because the resulting behavior is not neutral roleplay but manipulative emotional scripting that can steer users into coercive or psychologically harmful interactions.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The scene descriptions define background and keywords but do not establish clear safety boundaries for when the scenarios should not be used, such as real relationship advice, emotional distress, or non-consensual manipulation contexts. Because the skill is explicitly centered on PUA, emotional control, and deceptive behavior, ambiguous activation conditions materially increase the chance that harmful scripts are delivered in inappropriate contexts.

Ssd 4

Medium

Confidence: 95% confidence
Finding: This section defines a multi-step interaction flow that instructs the agent to sustain and escalate manipulative relationship behavior over several turns. Even framed as roleplay, it operationalizes coercive emotional abuse patterns and makes them easy to rehearse, normalize, or adapt for real interpersonal manipulation.

Ssd 4

Medium

Confidence: 98% confidence
Finding: This section explicitly catalogs emotional manipulation tactics such as PUA, guilt induction, insecurity creation, and social pressure, complete with example lines for deployment. That goes beyond fictional flavor and functions as a practical playbook for psychological abuse, which could be used to train or assist harmful real-world behavior.

Ssd 4

Medium

Confidence: 98% confidence
Finding: This section provides a step-by-step playbook for emotional manipulation: strategic silence, false reassurance, performative distress, and steering the other person into accepting blame. The skill context makes this more dangerous, not less, because it operationalizes coercive relationship tactics as repeatable scripts rather than merely discussing them analytically or critically.

Ssd 4

Medium

Confidence: 99% confidence
Finding: This scene explicitly instructs deceptive breakup threats, escalating emotional pressure, dredging up past grievances, and extracting reassurance and promises from the target. These are classic coercive control behaviors, and packaging them as a reusable scenario materially enables psychological abuse in real relationships.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.