Open-source music-skill lets Claude compose music note by note, fully headless

BandicootNo3152 · reddit · 2026-11-11

A Reddit user open-sourced music-skill (MIT, GitHub: owoevan-dev/music-skill), a Claude skill for fully headless AI composition: no audio-generation model involved — the AI writes every note itself and renders it, with a bundled sound library (including the open-sourced ex-commercial synth Surge XT). After setup, an agent can compose on demand, outputting MP3, MIDI, per-part stems, and a project file where any note can be edited and re-rendered. Output is completely unrestricted, even for commercial sale.

The v0.6.0 demo features 7 pieces, each from a single prompt with no editing, spanning Claude Fable 5.2, Opus 5.5, Sonnet 5.5, GPT-6 Astra, and GPT-6 Luna across orchestral, city pop, lo-fi, baroque fugue, and 8-bit styles. The author says they know little about music and wrote the skill almost entirely with Claude, originally to score a visual novel. Limits: significant disk space (synthesis, not generation), no vocals, render time scales with length/effort, and quality below commercial libraries. Fits naturally into AI video/game workflows, including seamless loops.

Related event: Open-Source music-skill Lets Claude Compose Music Note by Note(2 posts)→

Original post →

More from coding & agent

coding & agent channel →