finger-frame-ai
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses subprocess.run and subprocess.Popen in scripts/process.py and scripts/composite.py to call ffmpeg and ffprobe for video compression and compositing. Arguments are passed as lists to prevent shell injection vulnerabilities.
- [EXTERNAL_DOWNLOADS]: The skill downloads the MediaPipe hand landmarker model from Google's official storage services and provides instructions to install the vendor's AdaL CLI via a shell script hosted at https://adal.sylph.ai/install.sh.
- [REMOTE_CODE_EXECUTION]: The orchestration script scripts/process.py uses os.execve to ensure it runs within a dedicated virtual environment, a safe practice for managing Python dependencies in a skill.
- [SAFE]: Gemini API keys are handled securely via environment variables or interactive getpass prompts, ensuring sensitive credentials are not exposed in logs or history.
- [PROMPT_INJECTION]: The skill processes user-supplied video files which may contain instructions targeting the LLM. Ingestion points: Video files are uploaded to the Gemini API in scripts/stylize.py. Boundary markers: No delimiters are used to wrap video content in the prompt. Capability inventory: The skill can write files and execute local media commands. Sanitization: Input file paths are validated for existence before use.
Audit Metadata