Security audit
Security checks for vulnerabilities and agentic risk
This is a local Chinese medical exam study assistant whose storage and export behavior is disclosed and aligned with its tutoring purpose.
Before installing, expect this skill to keep local study history, wrong answers, preferences, and document-cache metadata in its memory folder, and to write an exported wrong-question file when requested. Avoid storing sensitive personal notes in shared or cloud-synced locations, and ask the publisher to add clear inspect, clear, and export-confirmation controls. Also note the current assistant.js appears to reference an undefined OUTPUT_DIR, so export and initialization may need a publisher fix before reliable use.
The skill declares a narrow tutoring purpose, but the document also describes persistent local storage of progress, user profile data, wrong-question history, document cache, and local file export behavior. This is a transparency and consent problem: users may provide sensitive study notes or personal data without realizing it will be retained or written to disk beyond the immediate session.
Referenced artifact was not completely inspected
在 `assistant.js` 中修改：
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
};
// 返回指令，让 OpenClaw 主程序处理
return instruction;
}
/**
The skill states that it uses persistent memory to store study progress, but it does not warn users that their learning history, mistakes, and related profile data may be retained across sessions. Even if the data is not highly regulated, undisclosed retention can expose personal habits, educational records, and potentially sensitive notes to other local users or later compromise.
The documentation describes exporting wrong-question data to a Word document without warning that study content will be written to a local path. That can unintentionally expose personal notes, exam preparation history, or copyrighted question material through insecure directories, shared machines, or cloud-synced folders.
The file’s natural-language interface, usage examples, and outputs are entirely specified in Chinese, and the export routine later hard-codes the zh-CN locale. There is no indication that users may opt into another language or that the skill is intentionally restricted to a China-specific deployment context.
The export operation writes a study-history file containing wrong-answer records, dates, and review metadata to disk without an explicit confirmation or warning at the point of export. In a shared host environment, this can create unintended disclosure of sensitive learning-history data and surprise persistence outside the user's immediate awareness.
The manifest describes the assistant as generating daily review plans, evaluating answers, and managing a wrong-question set. However, the CLI also advertises and accepts a separate "sync-docs" action, which is a filesystem-oriented document synchronization capability not mentioned in the manifest description.
The natural-language description forces a specific language/locale experience by presenting the skill entirely in Chinese without offering user choice. Under the policy, language constraints should be optional or explicitly justified as region-specific.
The inline comment states that the skill '专注于复习计划生成' (focuses on review plan generation), which narrows the apparent intent of the module. In practice, the file also evaluates answers, records wrong questions, and exports the wrong-question set, so the comment actively misstates the module’s implemented scope.
Detected: suspicious.dangerous_exec