qwen-mm-plugins-omni-chatcut-movie-commentary
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a bundled Python script
inspect_movie_commentary_state.pyto verify project manifest integrity and determine current workflow status based on existing filesystem artifacts. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source movie files which are analyzed by multimodal models (
omni_call) to generate textual watch notes. These notes subsequently guide the narration and edit planning. Evidence Chain: 1. Ingestion points:source_movievideo files and generatedplan/watch_notes/markdown files. 2. Boundary markers: While the workflow specifies 'neutral factual prompts', it lacks explicit boundary markers or isolation instructions to prevent the model from obeying text or instructions embedded within the video content itself. 3. Capability inventory: The skill possesses capabilities for file system operations, execution of local scripts, and synthesis of media via TTS and FFmpeg. 4. Sanitization: No specific sanitization or filtering is applied to the video content or the resulting multimodal observations to mitigate adversarial instructions.
Audit Metadata