The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file uses markers like 'MANDATORY' and 'NEVER' to instruct the agent to read documentation thoroughly. These are operational meta-instructions and do not attempt to bypass safety filters.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface because it processes untrusted Office documents and has access to shell-based validation tools. Ingestion points: ooxml/scripts/unpack.py and scripts/document.py extract and load XML content. Boundary markers: Absent for raw XML processing. Capability inventory: subprocess.run calls exist in ooxml/scripts/pack.py and ooxml/scripts/validation/redlining.py to invoke LibreOffice and Git for validation. Sanitization: Handled by secure XML parsing practices using defusedxml and lxml.

docx