The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external software to perform PDF conversions.
Evidence: The renderer script in renderer/render.py uses subprocess.run to call soffice (LibreOffice) for converting docx files.
Risk Assessment: The implementation uses list-based arguments instead of shell strings and employs slugification for user-controlled inputs, which effectively mitigates command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes external templates and user-provided input, which represents a potential attack surface for indirect prompt injection.
Ingestion points: User responses during the interview process and template files (.docx, .html, .md) loaded from the project directory.
Capability inventory: File system writes and document rendering via Jinja2 and docxtpl.
Sanitization: The system includes a validation phase that checks for unprocessed control tags and unfilled placeholders in the final output, reducing the risk of accidental or malicious command leakage.

coquill