Security audit
Security checks for vulnerabilities and agentic risk
This skill coherently recommends a next book using limited relevant context and has no executable code, persistence, or hidden high-impact behavior.
Before installing, note that the skill may answer in Chinese by default and may use relevant long-term memory or reading-history tool data if available. Its own instructions limit that use and tell the agent not to expose unrelated memories or save new memory without explicit permission.
The response contract hard-codes Chinese output (e.g. '最推荐：《书名》— 作者') without any requirement to detect or honor the user's language preference. This can cause the skill to answer in an unexpected language, reducing usability and potentially leading users to misunderstand recommendation rationale, caveats, or uncertainty disclosures.
The skill metadata hard-codes Chinese in both the short description and default prompt, which can override or bias the agent's response language without an explicit user request. This can cause user-confusion, mismatched expectations, and prompt-level policy drift, especially in multilingual contexts where the user did not opt into Chinese output.
