music-video

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as YouTube search results and audio analysis, to generate visual prompts via Gemini. These prompts are then directly interpolated into shell commands. Evidence: Step 3 and Step 5 in SKILL.md use LLM-generated strings as arguments for npx ts-node calls. Ingestion: YouTube search queries and audio-to-text transcription. Capability: Execution of shell commands. Sanitization: No explicit escaping or validation of the generated strings identified.
  • [DYNAMIC_EXECUTION]: The skill generates and executes code at runtime based on templates. Evidence: Step 7 in SKILL.md instructs the agent to create a temporary React composition file (.tsx) in the Remotion directory and execute it using npx remotion render. It also executes inline Python scripts using shell heredocs.
  • [COMMAND_EXECUTION]: The skill makes extensive use of system tools for media processing and file management. Evidence: yt-dlp, ffmpeg, and imagemagick (via magick and split_collage.sh) are used for downloading, audio editing, and image manipulation. It also performs file operations (cp, rm) in specific local user directories.
  • [EXTERNAL_DOWNLOADS]: The skill downloads media content from external platforms. Evidence: Use of yt-dlp to search for and download audio from YouTube based on artist and song name.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 02:38 PM