The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with external system tools, specifically soffice (LibreOffice) and pdftoppm (Poppler), to perform document conversion and rendering.
The bundled helper script scripts/render_docx.py uses subprocess.run() with the command passed as a list of arguments, which is a recommended security practice to prevent shell injection vulnerabilities.
The SKILL.md file provides shell command templates for manual execution that include shell variables (e.g., $INPUT_DOCX, $OUTDIR). The security of these commands relies on the agent safely handling variable interpolation during execution.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves processing and rendering user-provided .docx files, which presents a surface for indirect prompt injection or the exploitation of vulnerabilities in third-party document parsers.
Ingestion points: The skill accepts and processes .docx files provided by the user via the python-docx library and the LibreOffice rendering engine.
Boundary markers: The instructions do not define explicit boundaries or warnings for the agent to distinguish between its instructions and the content within the documents it processes.
Capability inventory: The skill has the capability to read from and write to the file system (specifically in /tmp/docs and output/doc) and execute system commands for document processing.
Sanitization: The skill does not implement specific sanitization or filtering of the content or metadata within the documents before rendering or processing them.

doc