making-videos

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads the Kokoro ONNX model files and voice weights from GitHub to provide local, high-quality text-to-speech functionality without cloud service dependencies.
  • [REMOTE_CODE_EXECUTION]: The skill uses package managers like npm or pnpm to install the Remotion renderer and its dependencies from its internal directory during the first run to establish the rendering environment.
  • [COMMAND_EXECUTION]: The skill utilizes system tools including ffmpeg and ffprobe for audio processing and video encoding. It also uses uv to manage and execute the Python script for voice synthesis.
  • [DYNAMIC_EXECUTION]: The skill dynamically imports Remotion libraries from the local node_modules directory after installation to prepare and render the video composition.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests an external JSON script containing narration and visual instructions.
  • Ingestion points: The script.json file is provided as input to the render.mjs script.
  • Boundary markers: The skill includes a validate function to check for correct scene types, required fields, and character limits for visual elements.
  • Capability inventory: The skill possesses capabilities for command execution (ffmpeg, uv, npm), file system access (writing videos and cache), and network connectivity for model acquisition.
  • Sanitization: Content from the script is used solely for text-to-speech generation and visual text rendering within a controlled React composition.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:50 PM