The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/render_docx.py utilizes subprocess.run to call system utilities like soffice (LibreOffice) and pdftoppm. These calls are implemented using list-based arguments rather than shell strings, which effectively prevents command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill references several standard dependencies including the python-docx and pdf2image Python packages, as well as libreoffice and poppler-utils system packages. These are well-known, legitimate tools for document conversion and manipulation.
[PROMPT_INJECTION]: The skill is designed to ingest and process external .docx files, creating an attack surface for indirect prompt injection. If a processed document contains malicious instructions, the agent might inadvertently follow them during the rendering or review phase.
Ingestion points: scripts/render_docx.py parses the OOXML structure of Word documents from zip archives.
Boundary markers: No explicit delimiters are used in the instructions to separate document content from agent instructions.
Capability inventory: The skill can execute local rendering commands and write files to the output directory.
Sanitization: The script uses path normalization and reads specific, expected components from the DOCX structure.

doc