markitdown
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install the 'markitdown' package from Microsoft's official GitHub repository and PyPI. These are recognized as trusted and well-known sources.
- [PROMPT_INJECTION]: The instructions contain behavioral steering patterns, specifically directing the agent that 'scientific schematics should be generated by default' and to 'always consider adding scientific diagrams.' These instructions attempt to override the agent's neutral processing of user requests by mandating specific tool usage and output formats.
- [COMMAND_EXECUTION]: The skill provides numerous examples for executing shell commands using the 'Bash' tool, including package installation via 'pip', document conversion via the 'markitdown' CLI, and the execution of a local Python script 'scripts/generate_schematic.py'.
- [INDIRECT_PROMPT_INJECTION]: The primary function of the skill is to ingest and process untrusted external data (PDFs, images, YouTube transcripts, etc.). This data is converted to text and placed directly into the agent's context. Because the skill has access to powerful tools like 'Bash' and 'Write', malicious instructions embedded in the processed files could potentially execute unauthorized commands or modify the file system.
Audit Metadata