markitdown
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external document formats (PDF, Word, HTML, etc.) which are untrusted data sources. Maliciously crafted documents could contain instructions aimed at influencing the behavior of an AI agent processing the converted content.
- Ingestion points: The skill ingests data through the
md.convert()method which accepts file paths, URLs, or bytes as shown inSKILL.md,examples.md, andpatterns.md. - Boundary markers: While the core library does not natively enforce boundary markers, the skill documentation in
patterns.mdrecommends implementing validation and sandboxing as mitigation strategies. - Capability inventory: The skill enables file system access (reading documents, writing markdown) and network operations (API calls to Azure Document Intelligence and OpenAI) as detailed in
examples.mdandpatterns.md. - Sanitization: The skill provides a robust
SecureConverterpattern inpatterns.mdthat demonstrates MIME type validation, file size limits, and input sanitization to reduce attack surface. - [EXTERNAL_DOWNLOADS]: The skill provides installation instructions for the
markitdownpackage from Microsoft's official GitHub repository and PyPI, and it references well-known services such as Azure and OpenAI. These are recognized as trusted or well-known technological sources.
Audit Metadata