Security audit
Security checks for vulnerabilities and agentic risk
This is a book-analysis helper with no executable payload; users should only accept its optional follow-up skill creation if they want persistent book-specific guidance.
Install this if you want a structured book-reading assistant. Be aware it may offer to create a separate book-specific reference skill after a session; approve that only when you want reusable guidance saved for future agent use.
The trigger condition is broad enough that any user providing arbitrary text for 'deep analysis' may unintentionally activate this skill, even when the request is not about book digestion. This can cause prompt hijacking at the routing layer, misapply the wrong behavioral policy, and interfere with safer or more appropriate skills for the actual task.
The instruction '中文优先 — 用户用中文交流，用中文回应' imposes a language behavior rule rather than offering a language choice. While partially conditioned on the user's language, it does not clearly allow user override or present language/locale as an explicit option.
The skill directs the assistant to create a new independent skill artifact under research/ after completing the reading workflow, which expands behavior from content analysis into persistent artifact generation. This can lead to unauthorized capability creation, instruction persistence, or scope creep, especially if book content or user prompts are transformed into reusable decision-support skills without explicit separate consent and safeguards.
No suspicious patterns detected.
