The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/docxlib/render.py to invoke LibreOffice for document rendering. This is implemented securely by passing arguments as a list and, on macOS, utilizing a shell wrapper (/bin/sh -c 'exec "$@"') that avoids string interpolation risks. Input paths are validated to be within allowed directories before being used in commands.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion point for document data, creating an attack surface for indirect prompt injection.
Ingestion points: scripts/docxlib/core.py extracts text and structure from .docx files.
Boundary markers: Data is passed to the agent as structured JSON via the CLI.
Capability inventory: The skill can perform file system operations (write/move), execute rendering commands, and the environment supports custom Python script execution.
Sanitization: Employs secure XML parsing (no entities, no network) to prevent XXE, though prose text is not filtered for potential instructions.
[EXTERNAL_DOWNLOADS]: The scripts/docx.sh script provides a mechanism to set up a Python virtual environment and install necessary dependencies (python-docx, lxml, PyMuPDF) from standard package registries.

docx