The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from various sources (PDFs, Word documents, YouTube transcripts, RSS feeds, and Wikipedia pages) and convert it into Markdown for consumption by an LLM. This creates a surface where malicious instructions embedded in those documents or web pages could influence the agent's behavior once the content is converted and placed into its context.
Ingestion points: The conversion methods demonstrated in SKILL.md and references/api-reference.md, such as md.convert(), md.convert_url(), and md.convert_uri(), allow the processing of arbitrary local files and remote web content.
Boundary markers: The provided examples do not demonstrate the use of delimiters or specific instructions to the agent to ignore any embedded prompts within the converted output.
Capability inventory: The skill provides the agent with capabilities to read local files, access the network to fetch URLs, and prepare content specifically for LLM context (e.g., the prepare_for_llm function in SKILL.md).
Sanitization: No evidence of content sanitization or filtering to prevent the propagation of embedded prompt injection attacks is provided in the documentation or helper scripts.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install the markitdown utility and related packages from the official Python Package Index (PyPI) and to clone the source code from Microsoft's official GitHub repository.
Evidence: SKILL.md contains commands such as pip install 'markitdown[all]' and git clone https://github.com/microsoft/markitdown.git. These resources are hosted by a well-known and trusted organization.

markitdown