The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local system utilities to perform document processing and validation tasks.
ooxml/scripts/pack.py: Invokes soffice (LibreOffice) in headless mode to validate documents by converting them to HTML.
ooxml/scripts/validation/redlining.py: Uses git diff to perform character-level comparisons between original and modified document text.
SKILL.md: Instructs the agent to use pandoc for text extraction, soffice for PDF conversion, and pdftoppm for image generation.
[PROMPT_INJECTION]: The skill processes external document files, which presents a surface for indirect prompt injection attacks (Category 8).
Ingestion points: Untrusted content is read from .docx, .pptx, and .xlsx files during the unpacking and editing processes (ooxml/scripts/unpack.py, scripts/document.py).
Boundary markers: The skill does not programmatically wrap extracted document text in delimiters or provide explicit instructions to ignore embedded commands.
Capability inventory: The skill possesses file system write permissions (doc.save) and the ability to execute external binaries (soffice, git).
Sanitization: The skill employs defusedxml for secure XML parsing to prevent XML External Entity (XXE) vulnerabilities and applies html.escape to metadata such as author names.

docx