doc-to-markdown
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/convert.pyusessubprocess.runto execute external conversion tools likepandocandmarkitdown. This is a standard orchestration pattern for this type of utility and is implemented safely using argument lists rather than shell interpolation.\n - Evidence:
subprocess.run(["markitdown", str(file_path)], ...)andsubprocess.run(["pandoc", ...], ...)inscripts/convert.py.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external document formats (PDF, DOCX) which may contain malicious instructions intended for the AI agent that subsequently processes the converted markdown output.\n - Ingestion points: Document files processed by
scripts/convert.pyandscripts/extract_pdf_images.py.\n - Boundary markers: None explicitly implemented for the final markdown output beyond standard markdown delimiters.\n
- Capability inventory: The skill performs file reads, executes local CLI tools via subprocess, and writes files to disk.\n
- Sanitization: The post-processing logic focus on formatting cleanup (e.g., CJK spacing, grid table fixes) but does not specifically sanitize for prompt injection strings.
Audit Metadata