Turn every code commit into a song
A beginner-friendly guide to building AI skills, extracted from the Spring into AI competition.
Start with a fun concept. Ask: "What if [mundane thing] could [unexpected output]?"
Break your idea into smaller, testable components:
Verify each piece works in isolation before combining:
Map out how everything connects:
Start a new AI session to write documentation:
Make it reproducible for others:
AI skills use "progressive disclosure" — agents only load full instructions when needed. This keeps sessions fast and contexts manageable.
How it works: Discovery → Activation → Execution. The agent knows the skill exists, loads instructions when triggered, then executes only what's required.
The Git Jukebox skill demonstrates agentic AI in action — AI that doesn't just respond to prompts but actively builds tools for itself. This workflow pattern applies beyond music generation: data pipelines, code review automation, documentation generation, and more.
Explore related resources, the original write-up, and other AI experiments.
Git Jukebox: Every commit is a song
The full Substack write-up with audio examples, screenshots, and the backstory of building this skill via Discord while away from the computer.
Read on SubstackCompetition Hub
Weekly AI agent skill competitions. See all submissions, participant projects, and learn how to build your own agentic tools.
View CompetitionGitHub Repository
Full source code for the Git Jukebox skill, including the ElevenLabs integration and git hook setup instructions.
View on GitHubAI Music Skills Overview
The official competition submission page with links, context, and related projects from other participants.
View SubmissionAgent Skills Documentation
Learn how AI skills work: progressive disclosure, activation patterns, and how to build your own reusable agent capabilities.
Explore SkillsAdvisory Hour Newsletter
Weekly experiments in AI, agentic workflows, and leadership in the AI era. Real results, no hype.
Subscribe