The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external documents (PDF, DOCX, etc.) and URLs, which creates an attack surface for indirect prompt injection.
Ingestion points: External data enters the agent context through the convert_to_markdown MCP tool, the native Read tool, and the WebFetch tool.
Boundary markers: The skill explicitly mandates the use of boundary markers (--- EXTERNAL CONTENT [source: <tool>] ---) to delimit untrusted content.
Capability inventory: The skill utilizes file-reading and network-fetching capabilities to ingest data for conversion.
Sanitization: The conversion protocol includes a mandatory sanitization step that requires stripping system/instruction tags and truncating content to 2000 words per section.
[EXTERNAL_DOWNLOADS]: The skill provides the user with an installation snippet for the markitdown-mcp tool using uvx. This is a transparent recommendation for the user to extend the agent's capabilities and requires manual configuration of the .mcp.json file.
[SAFE]: The skill implements a structured and documented approach to handling external file formats. It includes error handling, format detection, and clear instructions for the agent to inform the user when conversion capabilities are missing, rather than attempting to guess or fabricate content.

document-conversion