The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/render_docx.py uses subprocess.run to interface with libreoffice and poppler for document conversion. These calls are implemented securely by passing commands as argument lists rather than shell strings, which effectively prevents shell injection vulnerabilities.
[PROMPT_INJECTION]: The skill processes external DOCX files, which introduces a surface for indirect prompt injection if the files contain instructions designed to manipulate the AI agent's behavior.
Ingestion points: Content from untrusted documents is read and processed in scripts/render_docx.py and then presented to the agent via rendered images.
Boundary markers: The instructions do not specify any delimiters or safety markers to isolate document content from the agent's control logic.
Capability inventory: The skill allows for local file creation and execution of system-level utilities for conversion.
Sanitization: While input file paths are normalized, the script does not perform any sanitization of the document content prior to rendering or review.
[EXTERNAL_DOWNLOADS]: The documentation specifies the installation of well-known and legitimate dependencies including python-docx, pdf2image, libreoffice, and poppler using standard package managers (pip, apt, brew).

doc