The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted document content which could contain malicious instructions designed to influence agent behavior (indirect prompt injection).
Ingestion points: ooxml/scripts/unpack.py and scripts/document.py extract and parse XML content from user-supplied .docx files.
Boundary markers: The skill does not implement delimiters or specific instructions to ignore embedded commands within the extracted document text.
Capability inventory: The skill ships with scripts that can execute system commands (soffice and git) to validate or compare documents.
Sanitization: The skill employs the defusedxml library in its main editing utilities to prevent XML-based attacks such as XML External Entity (XXE) injection, although some validation scripts use standard library xml.etree.ElementTree or lxml which are less secure by default.
[COMMAND_EXECUTION]: The skill code spawns subprocesses to interact with system-level document tools.
Evidence: ooxml/scripts/pack.py invokes soffice (LibreOffice) to validate the integrity of repacked Office files.
Evidence: ooxml/scripts/validation/redlining.py invokes git diff to perform word-level comparisons between document versions.

docx