The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires several well-known external tools and libraries to function correctly.
SKILL.md specifies dependencies: pandoc (via apt), docx (via npm), libreoffice (via apt), poppler-utils (via apt), and defusedxml (via pip).
[COMMAND_EXECUTION]: The skill executes shell commands to perform document conversion and validation tasks.
ooxml/scripts/pack.py uses subprocess.run to invoke the soffice (LibreOffice) binary for validating repacked documents.
ooxml/scripts/validation/redlining.py uses subprocess.run to invoke git diff for comparing text content between document versions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted document content provided by users, which represents an indirect prompt injection surface.
Ingestion points: Document content is read via pandoc (converting to markdown) or by unpacking the ZIP structure and parsing XML files.
Sanitization: The skill mitigates technical XML risks (like XXE) by consistently using the defusedxml library for parsing.
Capability inventory: The skill can read/write files and execute specific binaries (soffice, git, pandoc).
[SAFE]: The skill follows security best practices for document manipulation, such as working in temporary directories and providing clear instructional boundaries for the AI agent.

docx