The Agent Skills Directory

[DATA_EXFILTRATION]: Path Traversal Vulnerability. The Python function create_poetry_doc in SKILL.md uses the title variable directly to construct the output filename: f'{title}_排版.docx'. A malicious user could provide a title containing directory traversal sequences (such as ../../target_file) to attempt to write files into sensitive directories outside of the intended workspace.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests untrusted user content (poetry, titles, authors) and uses it within document generation logic. There are no boundary markers or instructions provided to the agent to treat this data as potentially untrusted.
Ingestion points: SKILL.md workflow variables title, author, and lines which are populated by user input.
Boundary markers: Absent from the instructions and the code template.
Capability inventory: The skill utilizes the Write and Bash tools, providing the necessary permissions to perform file system operations.
Sanitization: No sanitization or path validation is performed on the user-provided title before it is used to save files.

wps-poetry