The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocesses to call trusted external utilities including pandoc, soffice (LibreOffice), and git. These calls are essential for the skill's primary document processing and validation functions and are executed with strictly controlled arguments.
[DATA_EXFILTRATION]: All document processing and editing occurs within secure temporary directories. No network operations to external domains were detected, and the logic is confined to local file system operations with no risk of sensitive data exposure.
[PROMPT_INJECTION]: The skill is designed to process external document content, which naturally includes a surface for indirect prompt injection. This risk is mitigated through the use of defusedxml for secure parsing and mandatory XSD schema validation, ensuring structural integrity and protecting against malicious XML payloads.

docx