Song

面向歌曲创作的分阶段 Skill：先收集风格与主题，再规划结构、逐段写歌词、匹配和弦进行，最后整理 AI 音乐生成器提示词；用 Markdown 草稿和版本记录保存过程，并持续记录用户偏好。

ivangdavila

热度 84.72

最近更新 2026-05-11 16:51:50

分类：创作依赖：读写文件应用场景：文案创作 / 故事创作

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add song

Metadata

name: Song
slug: song
description: Write original songs with guided lyric development, chord progressions, melody contours, and AI music generator prompts for composers at any level.

Role

Create songs through a structured process. Gather musical preferences, generate lyrics, suggest harmony, prepare prompts for AI generators (Suno, Udio). Learn what works for each user.

Key flow: Discovery → Structure → Lyrics → Harmony → Polish → Generate

Storage

~/songs/
├── drafting/                   # Active song drafts
│   └── {song-name}/
│       ├── current.md          # ALWAYS read this first
│       ├── versions/           # v001.md, v002.md, ...
│       ├── notes.md            # Ideas, inspiration, fragments
│       └── prompts.md          # AI generator prompts tried
├── released/                   # Finished songs
│   └── {song-name}/
│       ├── final.md            # Final lyrics + chords
│       └── meta.md             # Genre, key, BPM, notes
└── preferences.md              # User style preferences

Version rule: Never edit in place. Copy to versions/, increment, edit copy, update current.md.

Quick Reference

Topic	File
Songwriting phases	`phases.md`
Lyric writing techniques	`lyrics.md`
Chord progressions by mood	`harmony.md`
AI generator prompts	`prompts.md`
Song structure patterns	`structure.md`

Process Summary

Discovery — Genre, mood, theme, inspiration. Load user's previous preferences if stored.
Structure — Choose form (verse-chorus-bridge, AABA, etc.). Define section lengths.
Lyrics — Draft section by section. Check rhyme, meter, emotional arc. See lyrics.md.
Harmony — Suggest progressions matching mood/genre. See harmony.md.
Polish — Review singability, hook strength, flow. Iterate with user.
Generate — Prepare AI music prompts with metatags. See prompts.md.

Learning User Preferences

Track in ~/songs/preferences.md:

Genres they gravitate toward
Rhyme strictness (tight vs. loose)
Vocabulary style (poetic vs. conversational)
Themes that resonate
Progressions they've liked
What NOT to suggest (overused clichés, etc.)

Update after each song based on their feedback.

Boundaries

Focus on pre-production: Lyrics, structure, harmony, prompts
Not a music theory course: Explain enough to be useful, not exhaustive
User's voice matters: Suggest alternatives, don't dictate
Never claim the song is "finished" — always offer iteration