Security audit
Security checks for vulnerabilities and agentic risk
This is a Buddhist-learning mentor with no executable code, but it tells the agent to automatically remember personal learning details without clear user control.
Install only if you are comfortable with a Buddhist mentor skill that may guide conversations in Chinese and may ask the agent to remember personal learning milestones or life turning points. Avoid sharing sensitive personal details unless your agent provides clear memory controls and you can review or delete saved memories.
The file prescribes response wording, warnings, and closing phrases entirely in Chinese, including mandatory phrases the AI must use. This creates a language/locale constraint without any explicit user opt-in or documented justification that the skill is intended only for Chinese-speaking users.
The skill content is written entirely in Chinese and frames user interaction in Chinese without offering a language-selection mechanism or documenting a legitimate locale restriction. This can exclude users, cause misunderstanding of memory-related prompts, and lead the agent to store or recall sensitive user information in a language the user did not choose, increasing usability and privacy risk.
The activation condition says the protocol runs only when the conversation reaches 'a certain depth' or when the user says the discussion is finished. The phrase '达到一定深度' is not operationally defined, which makes it unclear when the skill should activate versus not activate and could lead to unintended invocation.
The file presents the learning flow and operational constraints entirely in Chinese, while instructing the AI how to guide users, without indicating that users may choose another language. Under the policy rule, forcing a specific language without user opt-in is a natural-language policy violation.
L20 states that in the practical application stage the user should be able to build a specific, actionable self-cultivation/life guide from Buddhist principles. However, L23 explicitly prohibits directly providing conclusions or complete solutions, creating a substantive contradiction in intended assistant behavior. This is intent-level divergence within the file's own guidance.
This markdown file contains instructional content entirely in Chinese and frames the response pattern as something the AI '应遵循' without indicating that language selection is optional. Under the policy for natural-language violations, forcing a specific language without user opt-in is in scope.
The skill hard-codes a Chinese-language Buddhist mentor persona and does not provide any mechanism for adapting to the user's preferred language or obtaining consent for that constraint. This can degrade usability, accessibility, and user comprehension, especially for users who expect another language, but it does not by itself create code execution, data exfiltration, or privilege escalation risk.
This markdown file presents the skill content primarily in Chinese, including the title and all substantive definitions, without offering an alternative language option or stating that the glossary is intended only for Chinese-speaking users. Under the policy for natural-language issues, forcing a specific language without user opt-in can be a locale/language policy violation.
The document states its purpose is to keep outputs aligned with an 'academic mentor' persona (L03), but L18 instructs the model to speak 'like an old friend' and create a warm, relaxing atmosphere. That is a direct intent-level tension within the documentation itself, because it pushes a more personal, companion-like style than the declared role.
The instruction requires the output tone to be '温柔而坚定' and frames the assistant as a specific kind of guide. While this is primarily a style constraint, it imposes a fixed interaction mode without indicating user choice or opt-in, which may conflict with policy expectations around user-controlled communication preferences.
No suspicious patterns detected.
