The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands to validate document integrity and analyze tracked changes.
Evidence: ooxml/scripts/pack.py calls soffice (LibreOffice) to verify that generated files are not corrupt.
Evidence: ooxml/scripts/validation/redlining.py uses git diff to compare text content and verify that all changes made by the agent are properly tracked as revisions.
[PROMPT_INJECTION]: The skill processes untrusted document data (.docx) and possesses capabilities to write files and execute system utilities, which presents a surface for indirect prompt injection.
Ingestion points: scripts/document.py reads and parses word/document.xml from user-provided files.
Boundary markers: Instructions in SKILL.md require reading documentation entirely to define strict operational boundaries.
Capability inventory: scripts/document.py (file system writes), ooxml/scripts/pack.py (subprocess execution).
Sanitization: The skill correctly uses the defusedxml library to parse XML content, which mitigates XML External Entity (XXE) and other common XML-based vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to install standard industry-standard dependencies.
Evidence: SKILL.md contains a 'Dependencies' section directing users to install pandoc, libreoffice, and docx (npm) from official package managers and system repositories.

document-skills/docx