The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured metadata in STORYBOARD.md and audiomap.json files to orchestrate video generation. This creates a surface where instructions embedded in music file metadata or malicious storyboard modifications could influence sub-agent behavior.
Ingestion points: STORYBOARD.md (read by orchestrator and frame-workers) and audiomap.json (derived from audio analysis).
Boundary markers: Employs Markdown headers (## Frame N) to segment frame-specific data.
Capability inventory: Includes the ability to execute CLI tools, Python scripts, and perform filesystem writes for HTML generation.
Sanitization: Relies on structural parsing in storyboard.mjs but lacks specific sanitization for prompt-breaking phrases.
[COMMAND_EXECUTION]: The audio analysis engine in scripts/analyze-beatgrid.py executes the ffmpeg system binary via Python's subprocess.run. The implementation correctly uses a list of arguments rather than a shell string, mitigating common command injection risks.
[EXTERNAL_DOWNLOADS]: The skill references the gsap and three.js libraries from the jsdelivr CDN within its generated HTML composition files and template documents.

music-to-video