The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions in SKILL.md and documentation files such as ooxml.md and docx-js.md requiring the agent to read the documentation in its entirety (e.g., 'MANDATORY
READ ENTIRE FILE'). These are standard context-enforcement patterns used to ensure the agent understands the complex OOXML schema and library APIs before performing edits and are not bypass attempts.
[COMMAND_EXECUTION]: The skill safely executes local system utilities via subprocess.run using list-based arguments. ooxml/scripts/pack.py uses soffice for document integrity validation, while ooxml/scripts/validation/redlining.py uses git diff for precise verification of tracked changes. Conversion workflows described in SKILL.md also utilize pandoc and pdftoppm correctly.
[EXTERNAL_DOWNLOADS]: Legitimate dependencies are identified in the setup instructions, including pandoc, the docx NPM package, LibreOffice, Poppler, and defusedxml. These are standard tools required for the skill's document manipulation and conversion features.
[SAFE]: The skill consistently implements secure XML handling by using the defusedxml library across its Python components, effectively mitigating common XML-based vulnerabilities such as XML External Entity (XXE) and DoS attacks.

docx