The Agent Skills Directory

[SAFE]: The skill facilitates state tracking via Markdown files. It includes a traversal check to ensure file operations remain within the repository boundaries. Write access is gated by mandatory user confirmation.- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it reads data from repository files (notably during the '恢复讨论' phase). While boundary markers are absent, the impact is minimized because restoration is read-only, and any subsequent state changes require explicit user adjudication and confirmation. No specific sanitization of the file content is mentioned.

checkpoint