The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses uvx to dynamically install and run packages from the Python Package Index (PyPI), specifically markitdown, httpx, and yt-dlp. These are well-known, reputable tools from established developers.
[COMMAND_EXECUTION]: The skill executes shell commands to run its internal Python scripts and the markitdown CLI tool. These operations are directly related to the document conversion functionality described in the documentation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from URLs and various file formats, which serves as an ingestion point for indirect prompt injection. Instructions embedded in these files could potentially influence the agent's behavior after conversion.
Ingestion points: URL content fetched via scripts/web-to-md and scripts/mhtml-to-md, YouTube transcripts via scripts/youtube-transcript, and local file parsing via uvx markitdown.
Boundary markers: None identified. The converted text is output as raw Markdown without markers to separate it from agent instructions.
Capability inventory: The skill has file reading (scripts/mhtml-to-md), file writing (scripts/web-to-md via the output flag), and network access capabilities across all scripts.
Sanitization: No semantic sanitization is performed on the extracted content.
Mitigation: To reduce risk, the converted content should be wrapped in unique delimiters with instructions for the agent to treat the enclosed text as data only.

convert-to-markdown