The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents how to install the markitdown package from PyPI, a well-known service. This is a standard procedure for using the tool.
[REMOTE_CODE_EXECUTION]: The skill describes how to use a legitimate CLI tool and Python API. It does not provide any commands that download and execute arbitrary code from untrusted sources.
[DATA_EXFILTRATION]: The skill mentions optional features like Azure Document Intelligence and OpenAI for image descriptions, which involve sending data to well-known cloud providers. The instructions explicitly warn the user not to send sensitive documents to these external services without confirmation and to manage API keys via environment variables rather than hardcoding them.
[COMMAND_EXECUTION]: The skill includes shell commands (pip install, markitdown, find | while read) that are standard for document processing workflows and do not exhibit malicious intent.
[PROMPT_INJECTION]: No prompt injection or behavior override patterns were detected in the instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes external documents (PDFs, Office files, URLs). While this is an ingestion surface for potential indirect prompt injection, the skill is focused on conversion rather than autonomous execution of the content. Standard boundary markers and safe practices (e.g., advising the user to verify output) are noted.

markitdown