The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/convert.py orchestrates document conversion by executing external tools including pandoc and markitdown via subprocess.run. Similarly, scripts/validate_output.py invokes pdftotext. These calls are implemented using argument lists rather than shell strings, which mitigates command injection risks by preventing shell interpretation of filenames or other parameters.
[SAFE]: No hardcoded credentials, unauthorized network operations, or persistence mechanisms (like shell profile modification or crontab entries) were detected. The skill's operations are strictly limited to document processing and local file management (image extraction and markdown output).
[SAFE]: The skill presents a surface for indirect prompt injection because it processes untrusted data from document files (PDF, DOCX, etc.).
Ingestion points: scripts/convert.py processes user-supplied document files provided via command-line arguments.
Boundary markers: None identified in the prompt templates or conversion logic.
Capability inventory: Subprocess execution for conversion tools (subprocess.run), file system operations (shutil.move, zipfile.extract), and file writing (Path.write_text).
Sanitization: Standard for document parsers; content is extracted and cleaned for formatting artifacts (e.g., removing Pandoc attributes), but the skill does not sanitize for embedded instructions intended for the agent.

doc-to-markdown