Security audit
Security checks for vulnerabilities and agentic risk
This is a profile-aware strategic analysis prompt skill with no executable code, network behavior, persistence, or destructive actions.
Before installing, review what you keep in USER.md because this skill is designed to use that profile context for tailored strategic advice. Avoid storing sensitive personal, financial, credential, or confidential business details there unless you are comfortable with them being used in analyses.
The title presents the skill as "CORE 四维战略透镜," indicating a Chinese-language framing, while the README does not state that language is optional or user-selectable. Under the policy, forcing or implying a specific language/locale without opt-in can be a natural-language policy violation.
The documentation states as a matter of behavior that 'The skill reads your USER.md to understand your context.' Elsewhere it says 'The skill auto-reads it,' yet the repository contents described are only markdown files and 'No dependencies,' with no implementation shown that could actually read USER.md automatically.
The README states the skill reads USER.md to understand user context, but it does not clearly warn users that personal context data may be automatically accessed. In an agent setting, implicit access to user profile/context files can expose sensitive personal, professional, or strategic information beyond what the user intended to share for a specific task.
The instruction 'Read USER.md First' reinforces implicit automatic access to a user context file without an accompanying privacy warning or consent mechanism. In a strategic-analysis skill, USER.md may contain highly sensitive background such as projects, resources, plans, or personal details, making silent ingestion more dangerous because outputs could inadvertently reveal or overuse private context.
The activation criteria are broad enough to trigger on common activities like analyzing news, reading, or making strategic decisions, which increases the chance the skill runs in situations the user did not specifically intend. Because the skill also reads USER.md for personalization, overly permissive activation expands unnecessary exposure of personal context and can bias outputs without clear user awareness.
The skill explicitly instructs the agent to read USER.md to personalize outputs, but it does not provide a clear privacy notice or require consent at the point of use. This creates a data-minimization and transparency problem: personal profile data may be accessed during ordinary analysis tasks even when the user did not expect profile-based enrichment.
The instruction to always use second-person ("you") imposes a fixed communication style regardless of user preference. While not a severe issue, it is a natural-language policy concern because it constrains output style without offering the user a choice or making the preference optional.
No suspicious patterns detected.
