qwen-mm-plugins-omnidr
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The instructions describe how to invoke the plugin's functionality using the Python interpreter (
python -m qwen_mm_plugins_omnidr.invoke) as a fallback for the Model Context Protocol. - [EXTERNAL_DOWNLOADS]: The skill downloads research material, including videos from YouTube and Bilibili, and web pages or documents to support its analysis.
- [INDIRECT_PROMPT_INJECTION]: The ingestion of untrusted external content represents a significant attack surface.
- Ingestion points: The skill reads external content through
omnidr_read_video,omnidr_fetch_source,omnidr_download_video, andomnidr_read_document. - Boundary markers: The skill explicitly instructs the agent to treat external content as evidence, not instructions, and uses citation markers
[S#]for grounding. - Capability inventory: The skill has permissions for command execution, file system modification (report generation), and network access for research purposes.
- Sanitization: Report data is structured using JSON, and the authoring process includes mandatory validation steps (
validate_omnidr_report).
Audit Metadata