The Agent Skills Directory

[COMMAND_EXECUTION]: The md-to-docx.py script uses subprocess.run to invoke the pandoc document converter and the system's open command. These calls are constructed using argument lists, which is a standard security practice to prevent shell command injection.
[DYNAMIC_EXECUTION]: The scripts md-to-docx.py and extract-styles.py utilize importlib.util to dynamically load a local helper script (create-reference-template.py). This load is restricted to the skill's own local directory via Path(__file__).parent.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown content, presenting a theoretical indirect injection surface. * Ingestion points: Markdown source files and manifest YAML files read in md-to-docx.py. * Boundary markers: None; content is converted rather than interpreted as instructions. * Capability inventory: File system writes (DOCX generation), ZIP archive manipulation, and execution of the pandoc binary via subprocess.run. * Sanitization: Path traversal guards are implemented in the resolve_path function to ensure files are accessed only within the intended directories.

md-to-docx-template