The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes external system utilities for document validation and processing tasks in a controlled manner.
The ooxml/scripts/pack.py script executes the soffice (LibreOffice) command to convert generated documents to HTML for automated validation.
The ooxml/scripts/validation/redlining.py script uses git diff to compare text content and verify that all edits are correctly tracked.
These subprocess calls use a list of arguments rather than shell strings, and do not use shell=True, significantly reducing command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes complex external data from .docx files, presenting a surface for indirect prompt injection.
Ingestion points: ooxml/scripts/unpack.py and scripts/document.py read content from document archives and XML files.
Boundary markers: Data is handled within its structured XML context, though no explicit 'ignore instructions' warnings are added before processing text content.
Capability inventory: The skill has the ability to read and write to the file system and execute specific document-handling utilities.
Sanitization: All XML processing is performed using the defusedxml library, which protects against XML-based attacks such as XXE (External Entity) and document expansion bombs.

docx