The Agent Skills Directory

[DATA_EXFILTRATION]: The skill mandates loading a local file at ~/.claude/PAI/SKILL.md as a prerequisite for operation. This file is described as containing sensitive personal data, including contact lists, voice IDs, and security protocols.
[PROMPT_INJECTION]: The skill is designed to ingest and process external Word documents by converting them to Markdown or unpacking their XML structure. This creates a surface for indirect prompt injection, where malicious instructions embedded in a processed document could be interpreted and executed by the agent.
[COMMAND_EXECUTION]: The skill utilizes several system-level utilities via shell execution and Python's subprocess.run module. These include pandoc for document conversion, soffice (LibreOffice) for PDF generation and validation, and pdftoppm for image extraction.
[SAFE]: The skill follows security best practices by using the defusedxml library for all XML parsing operations to prevent XML External Entity (XXE) attacks.

Docx