markitdown
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes a wide range of untrusted external content, which presents an attack surface for instructions embedded within data.\n
- Ingestion points: Processes various formats including PDF, DOCX, XLSX, PPTX, HTML, ZIP, and YouTube transcripts as defined in SKILL.md.\n
- Boundary markers: There are no explicit instructions or delimiters mentioned to ensure the agent ignores instructions found within the converted text.\n
- Capability inventory: Includes file reading, network access for web content, and optional external API integrations (OpenAI, Azure).\n
- Sanitization: No mention of specific sanitization to filter out malicious prompts from the extracted Markdown text.\n- [DYNAMIC_EXECUTION]: The skill mentions an extensible plugin system that can be enabled using
enable_plugins=True. Loading external or custom plugins represents a dynamic code execution vector.\n- [EXTERNAL_DOWNLOADS]: The installation process involves downloading themarkitdownlibrary and its associated dependencies from Python package registries.\n- [COMMAND_EXECUTION]: The skill documentation includes examples of command-line interface usage and references a batch conversion script (scripts/batch_convert.py) for processing local directories.
Audit Metadata