The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands to process and validate Office documents. Specifically, ooxml/scripts/pack.py runs soffice (LibreOffice) to verify document integrity through HTML conversion, and ooxml/scripts/validation/redlining.py executes git diff for character-level text comparison. These operations are restricted to intended functionality and target local temporary files.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external .docx files, which could contain instructions designed to influence the agent's behavior.
Ingestion points: Untrusted data is ingested when parsing XML components from document archives in ooxml/scripts/unpack.py, scripts/document.py, and scripts/utilities.py.
Boundary markers: The instructions do not define specific delimiters or directives to ignore instructions embedded within the text content of documents.
Capability inventory: The skill can perform file system operations and execute specific local commands (soffice, git) via scripts such as ooxml/scripts/pack.py and ooxml/scripts/validation/redlining.py.
Sanitization: The skill mitigates XML-based attacks like XXE by using defusedxml in its primary parsing components. Document text is extracted and treated as data for editing or verification rather than executable content.

docx