markitdown

Pass

Audited by Gen Agent Trust Hub on Jun 27, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external content (PDFs, Office docs, web pages) which can contain instructions designed to influence an LLM's behavior when the resulting Markdown is processed.
  • Ingestion points: Document conversion methods throughout the skill.
  • Boundary markers: None explicitly mentioned for the generated output.
  • Capability inventory: File system reading and network API requests to OpenAI and Azure.
  • Sanitization: No sanitization of ingested content is described before conversion.
  • [COMMAND_EXECUTION]: The batch processing utility (scripts/batch_convert.py) performs file system operations, including scanning directories and writing output files, based on user-provided path arguments.
  • [REMOTE_CODE_EXECUTION]: The skill documentation outlines an optional plugin system that allows users to register custom Python objects to handle specific file formats, enabling dynamic extension of conversion logic at runtime.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes well-known external services including Azure Document Intelligence, OpenAI, and Google Speech Recognition for enhanced conversion features, and recommends standard packages from official registries.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 27, 2026, 11:42 AM
Security Audit — agent-trust-hub — markitdown