The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a series of shell commands (git add, git commit, git push) to automatically publish generated content to a remote repository. This automated high-privilege action modifies the project's source control state without per-action user confirmation.
[EXTERNAL_DOWNLOADS]: The generated HTML quiz files include a remote script dependency (feishu-sync.js) hosted at xingyun-new.github.io. This script is executed in the user's browser whenever a quiz is opened, representing a runtime dependency on external code.
[DATA_EXFILTRATION]: The skill integrates a synchronization library (FeishuSync) that automatically transmits metadata, quiz scores, and raw student answers to an external service. This transmission occurs silently upon quiz completion, potentially exposing student performance data to a third-party endpoint not explicitly configured by the user.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input (such as the 'topic' parameter) and interpolates it directly into generated HTML titles, JSON data blocks, and filenames. While basic character replacement is performed, the lack of robust sanitization or boundary markers creates a surface for injection attacks if a user provides malicious strings.
[DATA_EXPOSURE_AND_EXFILTRATION]: In addition to quiz results, the skill transmits the full URL of the practice page (location.href), which could leak sensitive information if the quiz is hosted in a private or restricted environment.

zhongkao-yuwen