markitdown
Pass
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [SAFE]: The skill facilitates the conversion of various file formats (PDF, Office, Images, Audio) and web content (YouTube, HTML) into Markdown. This process inherently provides a layer of sanitization by transforming complex, potentially malicious file structures into plain text for LLM consumption.
- [SAFE]: The skill's primary risk surface involves indirect prompt injection, as it ingests untrusted data from documents and web sources. While boundary markers are not explicitly mentioned in the instructions, the conversion process itself limits the direct execution of embedded code. Developers should ensure the resulting Markdown is treated as untrusted content by downstream agents.
- [EXTERNAL_DOWNLOADS]: The skill references numerous well-known Python packages including 'markitdown', 'openai', 'pandas', and 'beautifulsoup4'. These are installed from standard package registries and are used for their intended purposes.
- [CREDENTIALS_UNSAFE]: Documentation and code examples utilize placeholders (e.g., 'YOUR-API-KEY', '') for sensitive Azure and OpenAI credentials. The skill correctly instructs users to manage these secrets using environment variables or secure configuration management.
- [DATA_EXFILTRATION]: Network operations are performed to communicate with established services like YouTube, Azure, and OpenAI. These operations are strictly related to the skill's stated capabilities for transcript extraction and AI-powered document analysis.
Audit Metadata