The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess.run to execute system utilities for document processing.
In ooxml/scripts/pack.py, it invokes soffice (LibreOffice) to validate documents by converting them to HTML.
In ooxml/scripts/validation/redlining.py, it invokes git diff for word-level comparisons.
These calls are securely implemented using argument lists rather than shell strings, which prevents shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided .docx files by converting them to markdown or extracting XML content, creating an ingestion surface for untrusted data.
Ingestion points include ooxml/scripts/unpack.py and instructions in SKILL.md to use pandoc.
The skill mitigates XML-based vulnerabilities by using the defusedxml library for all XML parsing operations.
[PROMPT_INJECTION]: The SKILL.md file contains meta-instructions directing the agent to avoid using range limits when reading its documentation files (docx-js.md and ooxml.md).
While this overrides normal tool usage parameters, it is a task-specific optimization meant to ensure the agent has the full API context for document manipulation.

docx