Sandbox
@SJY051/music-composition

Music composition skill for SKILL.md agents

This repository packages a music-composition skill for agents that read the SKILL.md format. It combines operating instructions, genre and technique references, cheat sheets, and validation notes so the agent can make concrete musical decisions instead of giving generic advice.

82 stars13 forksPythonUpdated 4mo ago
Who it's for

Builders who want their agent to help with music composition, arrangement, and reference research.

What it delivers

You can turn vague music requests into usable chords, melodies, forms, and revision notes.

What it does

Workflow routing

`references/00-navigation.md` points the agent to the smallest useful set of reference files.

Composition references

The `references/` tree covers harmony, melody, rhythm, form, orchestration, songwriting, genres, and instrument idiom.

Quick-use assets

`assets/` contains cheat sheets, templates, brainstorm cards, and diagnostic checklists for fast use in sessions.

Research support

The skill includes research files for trend digging, reference tracks, and copyright guardrails.

Validation and sanity checks

`VALIDATION-v1.0.md` and `scripts/music_theory_sanity_check.py` track release readiness and theory regressions.

How to get it

  1. 1The simplest way is via the skills CLI (requires Node.js)
    npx skills add SJY051/music-composition
  2. 2For a manual install, clone the repository and point your agent at it (or copy SKILL.md…
    git clone https://github.com/SJY051/music-composition.git

README

English | 한국어

Music Composition Agent Skill

Version Docs License Code License Compatibility

A modular knowledge and workflow skill for an AI music-composition assistant.

This repository is designed to help an agent turn vague creative requests — “make the chorus hit harder,” “give me a darker K-pop bridge,” “reharmonize this like jazz but not too dense,” “find reference tracks for my playlist” — into concrete musical decisions: chords, melody shapes, grooves, forms, arrangement moves, diagnostic feedback, and research workflows.

Current release: v1.0
Release date: 2026-04-27


Installation

The simplest way is via the skills CLI (requires Node.js):

npx skills add SJY051/music-composition

This installs to your current project's agent directory (e.g., .claude/skills/). Add -g for user-level (global) install, or --all -y to skip prompts.

For a manual install, clone the repository and point your agent at it (or copy SKILL.md and the relevant subdirectories into your agent's skill folder, such as ~/.claude/skills/ or .claude/skills/ in a project):

git clone https://github.com/SJY051/music-composition.git

Compatibility

The skill follows the open Agent Skills standard — the SKILL.md + YAML frontmatter format used by both Claude (Claude Code, Claude Agent SDK, Claude.ai skills) and OpenAI (Responses API tools[].environment.skills). It works on any harness that reads this format.


Does it work?

Two benchmarks ship with v1.0:

Approach 1 (system-prompt injection, GLM-5.1, blind eval). With SKILL.md prepended to the system prompt, the skill version wins 6 of 8 representative composition prompts in an author-bias-controlled blind comparison. The strongest signal is the boundary test (P8) — the skill version explicitly declines an out-of-scope mixing question while the baseline answers freely.

GLM-5.1 blind evaluation — wins per condition

Approach 2 (native lazy loading, Claude Opus 4.7). With the skill installed via Claude Desktop's Skills feature, the model reads references on demand. Most striking: pansori OST advice cites yoseong, toeseong, chuseong with cents-and-milliseconds implementation values that exist only inside references/genres/korean-traditional.md — direct evidence the model used the skill's full content, not just SKILL.md.

P8 boundary respect across both benchmarks

Together: the skill content shapes behavior even on weaker setups, and content depth unlocks on stronger setups. Full methodology and per-prompt analysis: benchmarks/v1.0-eval.md, benchmarks/v1.0-claude.md.


What this skill is for

Use this skill when an AI agent needs to help with:

  • composition brainstorming
  • chord progressions and harmonic direction
  • melody, motif, phrase, and topline writing
  • song forms, transitions, and narrative structure
  • rhythm, groove, meter, and feel
  • orchestration, arrangement, voicing, and instrument idiom
  • genre-aware writing across pop, jazz, classical, game, media, regional, traditional, and hybrid styles
  • critique and revision of a work in progress
  • teaching music concepts in a practical composition-first way
  • current music trend research and reference-track digging
  • user-agent collaborative workflows across multiple turns

The skill is not a DAW, audio engine, MIDI generator, transcription engine, legal service, or substitute for a human specialist in culturally specific traditions. It is a structured reference and workflow layer for an AI assistant.


Core design philosophy

The skill is built around a few practical principles:

  1. Composition is decision-making, not rule-following.
    Music theory is treated as a map of possible effects, not a list of commandments.

  2. Give playable, actionable output.
    Prefer concrete examples such as chord charts, scale-degree melodies, groove sketches, voicing suggestions, and revision blocks.

  3. Preserve user agency.
    Offer options, explain tradeoffs, and let the user choose a direction.

  4. Use only the references needed.
    The agent should usually start with SKILL.md, route through references/00-navigation.md, and load only the most relevant 1–3 files.

  5. Treat current music trends as research tasks.
    Static genre files provide stable craft conventions. Recent artists, charts, platform norms, and scene-specific references should be checked through active web research.

  6. Avoid shallow cultural shortcuts.
    Tradition-specific and region-specific music should not be reduced to generic “flavor.” The skill encourages naming the source tradition, function, language, rhythm system, instrumentation, and collaboration needs.


Repository structure

.
├── SKILL.md
├── MAINTENANCE.md
├── RELEASE-ROADMAP.md
├── KNOWN-LIMITATIONS.md
├── RELEASE-NOTES-v1.0.md
├── VALIDATION-v1.0.md
├── assets/
├── references/
│   ├── 00-navigation.md
│   ├── creative-workflows/
│   ├── form/
│   ├── fundamentals/
│   ├── genres/
│   ├── harmony/
│   ├── instrument-idiom/
│   ├── melody/
│   ├── orchestration/
│   ├── production-aware/
│   ├── research/
│   ├── rhythm-groove/
│   ├── songwriting/
│   ├── techniques/
│   └── validation/
└── scripts/
    └── music_theory_sanity_check.py

Important files

FilePurpose
SKILL.mdRoot operating instructions for the agent. Start here.
references/00-navigation.mdRouting map for choosing the right reference files.
assets/Short cheat sheets and response templates for quick use.
references/creative-workflows/Brainstorming, revision, answer calibration, and user-agent collaboration.
references/research/Web trend research, reference-track digging, streaming-service-aware research, and style/copyright guardrails.
references/genres/Genre and regional starter guides.
references/instrument-idiom/Composition-facing playability and idiomatic writing guides.
references/validation/Release readiness, smoke tests, correctness passes, and packaging checklists.
scripts/music_theory_sanity_check.pyRegression checker for links, known theory errors, risky terminology, snapshot notes, and required files.

Suggested agent workflow

For most user requests:

1. Read SKILL.md.
2. Use references/00-navigation.md to identify the smallest useful document set.
3. Load only the relevant reference files, usually 1–3.
4. Generate concrete musical output.
5. Explain why the choices work.
6. Offer a small number of next-step options.

For example:

User requestLikely routing
“Give me a sad but not cliché progression.”harmony/functional-harmony.md, harmony/modal-harmony.md, assets/progressions-catalog.md
“The chorus does not hit.”assets/diagnostic-checklists.md, songwriting/hooks-and-memorability.md, production-aware/energy-and-dynamics.md
“Make this more playable on guitar.”instrument-idiom/guitar.md, plus the relevant harmony or arrangement file
“Brainstorm three directions.”creative-workflows/musical-brainstorming.md, assets/musical-brainstorming-cards.md
“Find current J-pop references.”research/web-music-trend-research.md, research/reference-track-digging.md, assets/web-search-cheatsheet.md
“Use my playlist as a reference.”research/user-listening-context-and-streaming-services.md, assets/trend-and-reference-matrices.md

Currentness policy

Some files include a snapshot note. These files contain information that can become outdated, especially around:

  • current genre trends
  • charts and streaming-platform behavior
  • artist references
  • regional scenes
  • AI-assisted music tools
  • copyright and policy developments

The snapshot note does not mean the file is permanently current. It means the file was last reviewed as of that snapshot date. For fresh claims, use the research documents and verify current sources.

Useful research files:

references/research/web-music-trend-research.md
references/research/reference-track-digging.md
references/research/user-listening-context-and-streaming-services.md
references/research/regional-trend-evolution-analysis.md
references/research/style-reference-and-copyright.md
assets/web-search-cheatsheet.md

Style, originality, and copyright guardrails

This skill is intended to help generate original musical ideas. It should not be used to copy protected expression from existing songs or artists.

When a user asks for a specific artist or song style, the agent should translate the reference into craft variables such as:

  • tempo range
  • groove type
  • harmonic density
  • form
  • vocal range and phrasing style
  • arrangement density
  • instrumentation
  • energy curve
  • production-era cues

The agent should avoid reproducing or imitating protected elements such as:

  • melody
  • lyrics
  • distinctive riffs
  • samples
  • signature producer tags
  • highly specific vocal identity
  • recognizable arrangement sequences from a particular recording

See:

references/research/style-reference-and-copyright.md
KNOWN-LIMITATIONS.md

Cultural and regional specificity

The skill contains guides for traditional, regional, and hybrid styles. These are composition-facing starter references, not replacements for cultural expertise, native-language knowledge, or practitioner collaboration.

When working with culturally specific material, prefer:

specific tradition / region / language / function / rhythm / instrument idiom / performance context

Instead of vague labels such as:

generic world flavor / ethnic color / Asian flavor / Middle Eastern flavor

Relevant files include:

references/genres/korean-traditional.md
references/genres/folk-roots-and-traditions.md
references/genres/mena-pop.md
references/genres/south-asian-film-pop.md
references/genres/regional-scene-starters.md
references/research/regional-trend-evolution-analysis.md

Validation

The v1.0 package passed the release validation recorded in:

VALIDATION-v1.0.md
references/validation/first-release-readiness.md
references/validation/phase-b-correctness-pass.md
references/validation/phase-c-smoke-test-results.md

To run the sanity checker:

python scripts/music_theory_sanity_check.py

The checker looks for issues such as:

  • broken internal Markdown references
  • known high-risk music-theory regressions
  • risky cultural shortcut terminology
  • missing snapshot notes in current-sensitive files
  • required release and validation files
  • selected interval, mode, chord-symbol, and altered-dominant sanity checks

This checker is a regression tool, not a complete proof of musical correctness.


Maintenance

Before changing the repository:

  1. Read MAINTENANCE.md.
  2. Update references/00-navigation.md when adding or moving files.
  3. Keep high-use assets concise and accurate.
  4. Use snapshot notes for current-sensitive files.
  5. Run scripts/music_theory_sanity_check.py.
  6. Record substantial validation work in the relevant validation document.

Recommended versioning:

Change typeSuggested version
Typo, link, or small validation-script fixv1.0.x
P0/P1 correctness hotfixv1.0.x hotfix
New genre, workflow, or research documentsv1.1
Larger parser/evaluation harness improvementsv1.1 or v1.2
Major structure change or file reorganizationv2.0 candidate

Known limitations

See KNOWN-LIMITATIONS.md for the full list.

In brief, this skill does not directly provide:

  • audio transcription
  • DAW control
  • MIDI export
  • audio mixing or mastering
  • legal advice
  • guaranteed current trend knowledge without web research
  • exhaustive expertise in every local or traditional music practice

License

This repository uses a split license:

  • Skill documents, references, assets, release notes, and validation documents are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).
  • Code in scripts/ is licensed under the MIT License.

See LICENSE.md for details.

Files in the repo

Repository payload17 top-level entries
  • assets
  • benchmarks
  • references
  • scripts
  • .gitignore
  • KNOWN-LIMITATIONS.md
  • LICENSE.md
  • MAINTENANCE.md
  • RC1-v1.0-decision-memo-2026-04-27.md
  • README.ko.md
  • README.md
  • RELEASE-NOTES-RC1.md
  • RELEASE-NOTES-v1.0.md
  • RELEASE-ROADMAP.md
  • SKILL.md
  • VALIDATION-RC1.md
  • VALIDATION-v1.0.md

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More skills

Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.

117k
1 add
Vincentwei1021/
anything2explainer

Topic in, narrated explainer video out. A Claude Code / Codex skill that turns any topic into a black-canvas motion-graphics explainer video with TTS voiceover, subtitles and a chapter progress bar. Chinese or English; every frame drawn in code with Remotion.

666

Open-source AI job search: scan job portals, evaluate listings into a structured A-H report with a global 1-5 score, tailor your CV, track applications — runs locally in your AI coding CLI (Claude Code, Codex, OpenCode, Antigravity…)

71k

Taste-Skill - gives your AI good taste. stops the AI from generating boring, generic slop

86k