The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to call soffice (LibreOffice) for document validation and git diff for comparing changes during redlining validation. These operations are essential for the skill's primary function and target well-known, trusted system utilities.- [DATA_EXPOSURE_&_EXFILTRATION]: Analysis of the Python scripts and XML templates confirms that no sensitive data is accessed or exfiltrated. The implementation consistently uses defusedxml for parsing, which is the industry standard for preventing XML-based data exposure vulnerabilities.- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process potentially untrusted DOCX files, which creates an indirect prompt injection surface.
Ingestion points: External data enters the agent context through pandoc markdown conversion and raw XML unpacking in ooxml/scripts/unpack.py.
Boundary markers: While explicit boundaries aren't added to the converted text, the agent is instructed to use the output for specific document tasks.
Capability inventory: The skill has the ability to read/write files and execute system commands (soffice, git).
Sanitization: The skill effectively sanitizes XML input via the defusedxml library. The risk of processing malicious instructions within document text is handled by the agent's core safety layers.- [PROMPT_INJECTION]: The skill uses clear, technical instructions for document processing. The use of 'MANDATORY
READ ENTIRE FILE' directives is intended to ensure correct syntax usage for complex OOXML operations and does not attempt to bypass agent safety guidelines.

docx