markdown
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill uses
uvx markitdownto run the conversion tool. This command downloads and executes themarkitdownpackage at runtime from a well-known repository maintained by Microsoft. - [COMMAND_EXECUTION]: The skill executes shell commands using
uvxand pipes (e.g.,cat input.pdf | uvx markitdown) to perform file conversions. This is the intended behavior for the tool's functionality. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8).
- Ingestion points: Processes untrusted external files (PDF, Word, Excel, etc.) and remote URLs (Web pages, YouTube) provided by the user or found in the environment (SKILL.md).
- Boundary markers: None identified. The skill instructions do not specify using delimiters or 'ignore embedded instructions' warnings when processing converted content.
- Capability inventory: The skill has the ability to read arbitrary local files and fetch content from the network via URLs.
- Sanitization: There is no mention of escaping, filtering, or validating the content of the converted documents before it is returned to the agent's context.
Audit Metadata