
Music Caption Rewriter
SkillMiniMax 官方音乐技能:把一句简单的曲风描述(可附带分段歌词标签)扩写成 Music 3 需要的结构化描述,含整体风格、人声细节和分段编曲,用于生成视频配乐或完整歌曲。只读本地模板、不调用接口;真正出曲需另行部署开源的 Music 3 模型(需两块 CUDA GPU)。
MiniMax Music 3 — Agent Skills
This directory contains agent skills that extend AI coding agents (Claude Code, Cursor, Codex, etc.) with MiniMax Music 3 capabilities. Skills are plain-text, self-contained instruction packs — no external APIs, no runtime dependencies, no model weights required.
Available Skills
| Skill | What it does | Invocation |
|---|---|---|
music-caption-rewriter | Turns a brief music description and optional tagged lyrics into a detailed Music 3.0 Structured Caption (Global Metadata, Vocal Details, Arrangement) for precise generation control. | $music-caption-rewriter |
Installation
Option 1: Install with the skills CLI (recommended)
From the repository root:
npx skills add MiniMax-AI/MiniMax-Music3 --skill music-caption-rewriter
This downloads the skill into your agent's skills directory and makes it available as $music-caption-rewriter.
Option 2: Manual install
Copy the skill folder into your agent's skills directory:
# Claude Code (user-level)
cp -r skills/music-caption-rewriter ~/.claude/skills/
# Cursor (project-level)
cp -r skills/music-caption-rewriter .cursor/skills/
# Codex (user-level)
cp -r skills/music-caption-rewriter ~/.codex/skills/
The directory name must stay music-caption-rewriter — agents locate skills by folder name.
Basic Usage
Invoke $music-caption-rewriter in a supported coding agent with:
- Caption (required) — a natural-language music description, e.g. "A warm acoustic pop song with intimate female vocals, fingerpicked guitar, and a gradual emotional build into a wide final chorus."
- Lyrics (optional) — lyric text containing bracketed section tags such as
[Verse],[Chorus],[Bridge],[Instrumental]. - Constraints (optional) — desired length, output format, exclusions, or creative direction.
Example prompt:
Use
$music-caption-rewriterto turn my music description and optional tagged lyrics into a concise, section-aware Music 3.0 caption.
Output
The skill returns a structured caption with exactly three sections:
- Global Metadata — genre, subgenre, tempo, emotional progression, and production profile.
- Vocal Details — lead configuration, timbre, register, delivery, harmony/backing vocals, and restrained vocal effects.
- Arrangement — a section-by-section timeline with instrument lifecycles, groove development, and transitions.
JSON or JSONL output is available on request.
The rewritten caption is designed to be passed directly to MiniMax Music 3 — as the instructions field of the generation API — alongside the original lyrics in the input field. See the top-level README for the full workflow.
How It Works
The skill is built on a fully static, text-based library with progressive disclosure — the agent routes to the right style family first and only opens the templates it actually needs:
SKILL.md Skill instructions and workflow
├── agents/openai.yaml Agent metadata (display name, default prompt)
├── references/
│ ├── genre-router.md Entry point: maps genres to style families
│ └── index-*.md 18 family indexes with compact style cards
└── templates/ 1,000 full caption templates
The genre router maps your genre/mood cues to one of 18 style families (pop & ballad, hip-hop & rap, metal & heavy rock, cinematic orchestral, East Asian modern, jazz & swing, and more). The agent selects up to three references with distinct roles — Foundation, Modifier, and Arrangement — then synthesizes a new, original caption around your brief. Lyric text is never reproduced or rewritten; only bracketed section tags act as musical directives.
Repository Layout
skills/
├── README.md This file
└── music-caption-rewriter/
├── SKILL.md Skill instructions
├── agents/openai.yaml Agent metadata
├── references/ Genre router + family indexes
└── templates/ Caption template library
Maintenance
- The library is intentionally 100% text-based — no scripts, databases, embeddings, or external services.
- To add a template: add one complete caption file under
templates/, add one compact card to exactly one family index linked from the genre router, and update that index's family count. - Keep template IDs matching their filenames and their referenced paths existing.
相關資源

H3 Prompt Writing
SkillMiniMax 官方为海螺 H3 视频模型准备的提示词技能:把视频创意改写成 H3 要求的结构化提示词,覆盖文生视频、首尾帧和全参考模式,含画面、音效与配乐描述。纯文档、不调用接口;成片需在海螺网页、MiniMax API 或本地部署的模型上生成。
MiniMax-AI/MiniMax-H312k
MiniMax Skills
PluginMiniMax 官方技能包插件,提供前端、全栈、Android、iOS、Flutter、React Native 与 GLSL 着色器开发指南,另含 PDF / PPT / Excel / Word 文档与图像、视频、音乐生成技能。多媒体技能需要 MiniMax API Key,项目仍处于 Beta。
MiniMax-AI/skills13.9k
ffmpeg-skill
Skill本地 FFmpeg 视频剪辑技能:用自然语言让 agent 剪切拼接、去静音、改竖屏、加字幕、统一响度,并按 TikTok、Reels、YouTube 规格导出,全程离线、无需 API Key。需要安装 ffmpeg 和 Python 3,也可作为 MCP 使用。
kajisho5/ffmpeg-skill8.1k
Seedance 2.0 Prompt Writing Skills
Skill教 agent 为字节跳动即梦 Seedance 2.0 写视频提示词:讲清输入限制、@ 引用素材的写法和镜头语言,附产品广告、短剧、舞蹈、配乐风景等模板,并覆盖电商、科普教学等场景,有中英文两版。只负责写提示词,生成视频要用自己的即梦账号在平台上完成。
dexhunter/seedance2-skill4.5k
Scenario Agent Skills
SkillScenario 官方技能集:通过 Scenario MCP 生成可直接交付的图片、视频、配音配乐、贴图与 3D 素材,覆盖产品图、广告、分镜和游戏美术;另含驱动 Blender、Maya、ZBrush 等软件的专家技能。需要 Scenario 账号(有免费额度)。
scenario-labs/skills3.9k
claude-code-video-toolkit
Skill面向 Claude Code 的视频制作工作区:从写稿、AI 配音、配乐、生成画面到用 Remotion 渲染 MP4,内置讲解短视频、产品演示等模板。完整流程需克隆仓库并安装 Node.js 18+,AI 生成环节要自备云 GPU(如 Modal 的免费额度)。
digitalsamba/claude-code-video-toolkit2.7k