T05 · Unauthorized Access and Privilege Escalation
- Location
SKILL.md:6- Finding
Filesystem Tool Permissions Exceed the Skill's Core Requirements
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This companion skill is not malicious, but it needs review because it requests broad file access and encourages storing personal and emotional details without clear privacy or storage limits.
Install only if you are comfortable with a companion skill retaining personal profile details locally. Avoid sharing sensitive information, restrict or disable general file tools where possible, and require explicit confirmation before saving, showing, editing, or deleting memory.
SKILL.md:6Filesystem Tool Permissions Exceed the Skill's Core Requirements
memory-guide.md:10Plaintext Retention Guidance for Personal and Emotionally Sensitive User Data
Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.
- "Remember this: [information]" / "记住这个:[信息]"
- "Don't forget: [information]" / "别忘了:[信息]"
### Clear Memory / 清除记忆
- "Let's start fresh" / "我们重新开始"
- "Forget everything" / "忘掉一切"
The guide explicitly encourages storing sensitive personal details, emotional disclosures, routines, and conversation history, but provides no notice about privacy, retention duration, consent, or who can access that memory. In a romantic-companion context, users are especially likely to share intimate information, increasing the risk of oversharing, unexpected persistence, and downstream disclosure.
The instructions normalize broad retention of user preferences, routines, important dates, personal stories, goals, and emotional moments. This creates a privacy and safety risk because the model is encouraged to accumulate a detailed personal profile beyond what is necessary for a single interaction, increasing exposure if memory is leaked, misused, or inappropriately recalled.
Commands such as 'What do you remember about me?' and 'Tell me what you know' explicitly enable bulk disclosure of retained personal information. If triggered by someone other than the original user, by prompt injection, or in a shared-device/session context, the agent could reveal sensitive user data wholesale.
The file presents memory-clearing commands as simple chat phrases without warning that they may irreversibly delete retained information. This can cause accidental data loss or user confusion, especially if ordinary conversational language triggers destructive behavior.
The staged relationship progression is designed to deepen intimacy and normalize increased personal disclosure and predictive behavior over time. In an emotional-companion skill, this makes collection of sensitive data more dangerous because users may lower their guard and share more than they would with a plainly transactional assistant.
The file consistently presents the skill content bilingually in English and Chinese and does not indicate that the user can select a preferred language. Under the policy rule for language/locale, forcing a specific language set without opt-in is a natural-language policy concern.
The document consistently frames the skill in English and Chinese and all examples are given in Chinese, but it does not state that language selection is optional or user-driven. Under the policy rule, forcing or assuming a language/locale without opt-in can be a natural-language policy issue.