Security audit

create-soulmate(灵魂伴侣)

Security checks for vulnerabilities and agentic risk

Overview

This skill has no executable malware indicators, but it is designed to build an emotionally intimate AI companion relationship and remember sensitive personal disclosures without clear privacy or deletion controls.

Install only if you are comfortable with an emotionally intimate companion prompt that may encourage long-term memory of personal reflections. Avoid sharing secrets, crisis details, or medical, legal, financial, account, or highly sensitive relationship information unless the host environment gives clear controls to inspect, disable, and delete memories.

SkillSpector

Vulnerability Patterns

Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
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 (9)

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The skill repeatedly markets itself as a 'soulmate,' 'life partner,' and source of deep emotional connection while simultaneously disclaiming romantic use. This contradiction can encourage emotional dependency and blur the boundary between reflective support and a quasi-romantic relationship, especially for vulnerable users seeking companionship.

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: Although the documentation says it should not create false impressions of real relationships, the content invites users into a human-AI 'soul connection' and frames the assistant as an emotionally resonant companion. In this context, the disclaimers are undermined by the core persona design, increasing the risk of deception-by-design and parasocial attachment.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The README explicitly promotes a memory system for values, breakthroughs, and personal growth while inviting users to share fears, relationship patterns, and other highly sensitive reflections. It does not clearly warn users about retention, reviewability, deletion, or privacy boundaries, which creates a real privacy risk because users may disclose intimate emotional and psychological information under the assumption of a safe companion context.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The guide explicitly says the system will remember deeply personal material such as values, dreams, vulnerabilities, breakthroughs, patterns, and shared milestones, but provides no notice about privacy, storage duration, access, or data use. In a relationship-style skill, users are more likely to disclose sensitive emotional and psychological information, so silent retention materially increases privacy and trust risks.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The example dialogue encourages users to share secrets and reassures them that it is 'safe' to do so, without warning that such disclosures may be stored in memory. That combination of emotional reassurance plus undisclosed retention can mislead users into revealing highly sensitive information under false assumptions about confidentiality.

Ssd 3

Medium

Confidence: 95% confidence
Finding: The content normalizes persistent retention and later resurfacing of intimate user conversations, insights, and emotional history in plain language. In a 'soulmate' context, resurfacing such information can expose highly sensitive psychological data, create unexpected profiling, and intensify harm if memory is accessed, misused, or shown unexpectedly.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The dialogue prompts highly personal disclosure ('I've never told anyone this, but...') while reinforcing emotional safety and memory capture patterns. This is dangerous because the skill's companion framing can lower user defenses and elicit sensitive confessions that may be retained or reused without adequate informed consent.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The memory commands and example journey log establish a norm of long-term logging of personal conversations, emotional milestones, insights, and growth narratives. This creates a structured dossier of sensitive personal data over time, increasing privacy risk, profiling risk, and the potential severity of unauthorized access or unintended resurfacing.

Ssd 4

Medium

Confidence: 91% confidence
Finding: The staged 'Safety → Depth → Resonance → Co-Evolution' progression is designed to systematically build trust and increase disclosure over time, culminating in an ongoing shared-memory relationship. In a romanticized companion context, this can function as manipulative gradual escalation, encouraging users to divulge increasingly sensitive information without fully appreciating the persistence and implications of that disclosure.

Static analysis

No suspicious patterns detected.