The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/convert.py and scripts/doc_to_md.py to execute internal helper scripts and the pandoc system utility. These calls use argument lists rather than shell strings, which mitigates standard shell injection risks.
[DATA_EXFILTRATION]: scripts/pdf_to_md_mineru.py facilitates the upload of PDF content to the MinerU cloud service (mineru.net) for OCR processing. While this is an intended feature requiring an API token, it involves sending user document data to an external third-party domain.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge for untrusted external content to enter the agent's context.
Ingestion points: scripts/convert.py, scripts/web_to_md.py, scripts/pdf_to_md.py, scripts/doc_to_md.py, scripts/excel_to_md.py, scripts/ppt_to_md.py, and scripts/subtitle_to_md.py all ingest external files or URLs.
Boundary markers: The skill does not append explicit boundary markers or warnings to the converted Markdown output to signal the presence of untrusted content to downstream LLM consumers.
Capability inventory: The associated agent environment possesses capabilities for file system access (read/write) and network operations (via requests and curl_cffi).
Sanitization: The conversion process extracts text and formatting but does not filter for or sanitize malicious natural language instructions embedded within the source documents.

markdown-conversion