video-remix
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill acts as an orchestrator, using subprocess.run and subprocess.Popen in scripts/_common.py to execute sibling skills such as video-scene-analysis, avatar-invent, and avatar-reel-composer. These calls are restricted to local script paths and utilize argument lists to prevent shell injection.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface through the analysis of external video content.
- Ingestion points: reference.mp4 and content.json (processed in scripts/analyze_video.py).
- Boundary markers: Instructions are passed to Gemini via structured JSON prompts; however, there are no specific 'ignore embedded instructions' delimiters for the video content itself.
- Capability inventory: Execution of sibling scripts via subprocesses and file writing in the runs/ directory.
- Sanitization: Filenames are generated using slugify, and subprocess calls use list-based arguments to prevent command injection.
- [EXTERNAL_DOWNLOADS]: The skill specifies standard Python dependencies (google-genai, Pillow, numpy) in its requirements.txt file, all of which are trusted libraries.
- [SAFE]: Credential discovery in scripts/_common.py follows a secure pattern by looking for API keys in environment variables and local sibling skill configurations instead of hardcoding secrets.
Audit Metadata