markdown-converter
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill facilitates the execution of the
markitdowncommand-line utility. This tool reads local files (PDF, Office, etc.) and outputs text content, which is the intended behavior of a conversion utility. - [EXTERNAL_DOWNLOADS]: The skill uses
uvxto fetch and execute themarkitdownpackage from the Python Package Index (PyPI). Since PyPI is a well-known service and the package is maintained by a trusted organization (Microsoft), this is a safe way to handle dependencies. - [PROMPT_INJECTION]: As a data-processing skill, it has an inherent surface for indirect prompt injection. Malicious instructions embedded within processed files (e.g., hidden text in a PDF or transcription from a YouTube URL) could be parsed into the agent's context and attempt to override behavior.
- Ingestion points: Processes various untrusted formats including PDF, Word, Excel, ZIP, and content from YouTube URLs.
- Boundary markers: The instructions do not currently include markers or delimiters to isolate the converted text from the agent's control instructions.
- Capability inventory: Shell execution (
uvx), file reading/writing, and network access for external URL processing. - Sanitization: The skill relies on the internal parsing logic of the
markitdownutility without additional sanitization steps provided in the prompt instructions.
Audit Metadata