The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for installing the 'markitdown' tool via Python package managers (uv, pipx) and executing it through the command line to process files. It also includes a local bash script (scripts/batch_convert.sh) to automate the conversion of entire directory trees.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data—such as PDFs, Word documents, and web content—into Markdown for the agent to read. This establishes an attack surface where malicious instructions hidden in the source documents could influence the agent's behavior.
Ingestion points: Converted content from local files, binary streams, and external URLs (e.g., YouTube transcripts, HTML pages) enters the agent's context (SKILL.md, scripts/batch_convert.sh).
Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions to wrap the converted output.
Capability inventory: The agent has access to 'Bash', 'Read', 'Write', and 'Glob' tools, allowing it to perform file operations and execute shell commands based on the processed content.
Sanitization: No explicit sanitization or filtering of the converted Markdown content is performed before it is presented to the agent.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external sources, such as YouTube URLs for transcripts and Bing search results. It also supports optional features that send data to external APIs, including Google's Web Speech API for audio transcription, OpenAI for image captioning, and Microsoft Azure for cloud-based OCR.

markitdown