music-skill
Don't generate it. Compose it: your agent writes every note, then plays, mixes and masters the piece — and you keep the MIDI.
Project intro
AI music tools hand you an audio file you cannot edit: one wrong note means regenerating the whole piece. music-skill is an agent skill that writes original instrumental music the way a composer and a small studio would — every note written by the agent, then checked, performed on sampled instruments, mixed and mastered.
Each piece starts from twelve words drawn at random, linked with your request into one scene; the agent picks the title, the form and the sounds, writes every note and revises its own work. You get the MP3 and WAV, the MIDI, a stem per part and a project file in which any note can be changed and rendered again. No audio-generation model is involved: the sounds come from recorded instruments and synthesizers, each named with its licence.
The proof is six example pieces, one by each of six models, each made in one take from one message with nobody editing the notes: an orchestral film climax, a folk battle reel, a baroque fughetta, a looping lo-fi beat, a slow ballad and an 8-bit game theme. On a 24-thread laptop each one rendered from scratch in 25 to 156 seconds.
It has 257 stars on GitHub with 32 forks, was created on 2 November 2026 and was last pushed on 17 November 2026. It is MIT licensed. You need Python 3.10 or newer and about 3.7 GB of sound banks; the compositions and recordings are CC0.
Actions
Tech stack
Project info
- Project category
- Music & Audio
- Project status
- Online
- Collaboration intent
- None