audio-to-video

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local command-line tools to handle media files. Specifically, scripts/generate.ts invokes ffprobe to determine the duration of input audio files, and scripts/split_collage.sh uses magick (ImageMagick v7) to crop and split collage images into individual frames. These are standard operations for a media generation skill.
  • [EXTERNAL_DOWNLOADS]: The skill interacts with fal.ai, a well-known AI service provider. It uploads user assets (audio and images) to fal.ai's storage service to facilitate video generation and subsequently downloads the resulting MP4 video file via the fetch API.
  • [INDIRECT_PROMPT_INJECTION]: The scripts/analyze_audio.ts script processes audio files and user-provided request strings by sending them to the Gemini API. While it uses specific sections in the prompt to isolate user input, the resulting storyboard content (which includes generated prompts) is used for downstream video generation. Ingestion points: User-provided specificRequest and audio file content in analyze_audio.ts. Boundary markers: The script wraps the user request in a marked section within the Gemini prompt. Capability inventory: File writing and network calls to generative APIs. Sanitization: No explicit sanitization or escaping of the user-provided request string is observed.
  • [DATA_EXPOSURE]: The skill scripts reference a hardcoded environment file path at /Users/aviz/.claude/skills/image-generation/scripts/.env. This demonstrates a dependency on the author's local file structure and indicates that the skill is intended to share configuration with another specific skill in the author's environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 02:38 PM