The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes command-line utilities to perform its core functions. It executes ffmpeg for audio decoding and normalization, python3 for beat-grid analysis, and the hyperframes CLI for project initialization and rendering. These commands are executed locally within the project environment and are essential for the video generation process.
[EXTERNAL_DOWNLOADS]: The skill instructions facilitate the installation of industry-standard Python libraries including librosa, numpy, and soundfile from official registries. It also loads common frontend libraries such as GSAP and Three.js, along with fonts, from established and well-known content delivery networks (jsDelivr and Google Fonts).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted music files and user-supplied assets, creating a potential surface for indirect instruction through file metadata or file paths.
Ingestion points: The audio analyzer (analyze-beatgrid.py) reads external music tracks to generate a timing map, and the storyboard parser (storyboard.mjs) processes asset paths provided by the user.
Boundary markers: The workflow implements a robust human-in-the-loop checkpoint system. Step 3 (Plan approval) and Step 6 (Render approval) explicitly require user verification before the agent proceeds to high-capability tasks like full video rendering.
Capability inventory: The skill scripts perform local file operations and execute subprocesses to coordinate the project assembly and rendering.
Sanitization: Complex input parsing is delegated to established technical libraries (FFmpeg and Librosa) which perform validation of media file integrity and structures.

music-to-video