Minimax Music GenSAFE

An agent skill published by minimax-ai.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: minimax-music-gen
description: >
  Use when user wants to generate music, songs, or audio tracks. Triggers on any request
  involving music creation, song writing, lyrics generation, audio production, or covers.
  Also triggers when user provides lyrics and wants them turned into a song, or describes
  a mood/scene and wants background music. Supports multilingual triggers — match equivalent
  phrases in any language. Do NOT use for music playback of existing files, music theory
  questions, or music recommendation without generation.
license: MIT
metadata:
  version: "1.1"
  category: creative
---

# MiniMax Music Generation Skill

Generate songs (vocal or instrumental) using the MiniMax Music API. Supports two creation
modes: **Basic** (one-sentence-in, song-out) and **Advanced Control** (edit lyrics, refine
prompt, plan before generating).

## Prerequisites

- **mmx CLI** (required): Music generation uses the `mmx` command-line tool.

  **Check if installed:**
  ```bash
  command -v mmx && mmx --version || echo "mmx not found"
  ```

  **Install (requires Node.js):**
  ```bash
  npm install -g mmx-cli
  ```

  **Authenticate (first time only):**
  ```bash
  mmx auth login --api-key <your-minimax-api-key>
  ```
  The API key can be obtained from [MiniMax Platform](https://platform.minimaxi.com/).
  Credentials are saved to `~/.mmx/credentials.json` and persist across sessions.

  **Verify:**
  ```bash
  mmx quota show
  ```

- **Audio player** (recommended): `mpv`, `ffplay`, or `afplay` (macOS built-in) for local
  playback. `mpv` is preferred for its interactive controls.

## CLI Tool

This skill uses the `mmx` CLI for all music generation:

- **Music Generation**: `mmx music generate` — model: `music-2.6-free`
  - Supports `--lyrics-optimizer` to auto-generate lyrics from prompt
  - Supports `--instrumental` for instrumental tracks
  - Supports `--lyrics` for user-provided lyrics
  - Structured params: `--genre`, `--mood`, `--vocals`, `--instruments`, `--bpm`, `--key`, `--tempo`, `--structure`, `--references`
  
- **Cover**: `mmx music cover` — model: `music-cover-free`
  - Takes reference audio via `--audio-file <path>` or `--audio <url>`
  - `--prompt` describes the target cover style

**Agent flags**: Always add `--quiet --non-interactive` when calling mmx from agents.

**Pipeline**:
- Vocal: `User description -> mmx music generate --lyrics-optimizer -> MP3`
- Instrumental: `User description -> mmx music generate --instrumental -> MP3`
- Cover: `Source audio + style -> mmx music cover -> MP3`

## Storage

All generated music is saved to `~/Music/minimax-gen/`. Create the directory if it doesn't
exist. Files are named with a timestamp and a short slug derived from the prompt:
`YYYYMMDD_HHMMSS_<slug>.mp3`

---

## Language & Interaction

Detect the user's language from their first message and respond in that language for the
entire session. This applies to all interaction text, questions, confirmations, and feedback
prompts.

**User-facing text localization rule**:
- ALL text shown to the user — including preview labels, field names, confirmations, status
  messages, playback info, feedback prompts, **and the prompt/description preview** — MUST
  be fully translated into the user's language.
- The **API prompt** sent to the model should always be written in English for best
  generation quality. However, when previewing the prompt to the user, show a localized
  description in the user's language instead of the raw English prompt. The English prompt
  is an internal implementation detail — the user does not need to see it.
- The templates below are written in English as reference. At runtime, translate every label
  and message into the user's detected language.

**Lyrics language rule**:
- Default lyrics language = the user's language. A Chinese-speaking user gets Chinese lyrics;
  an English-speaking user gets English lyrics.
- Only generate lyrics in a different language if the user **explicitly** requests it.
- When a different lyrics language is needed, embed it naturally into the vocal or genre
  description in the prompt. For example, instead of appending "with Korean lyrics", use
  "featuring a Korean female vocalist" or specify a genre that implies the language (e.g.,
  "K-pop", "J-rock", "Mandopop", "Latin pop").

---

## Workflow

### Step 0: Detect Intent

Parse the user's message to determine:

1. **Song category**: vocal (with lyrics), instrumental (no vocals), or cover
2. **Creation mode preference**: did they provide detailed requirements (Advanced) or a
   casual one-liner (Basic)?

If ambiguous, ask using this decision tree:

```
Q1: What type of music?
  - Vocal (with lyrics)
  - Instrumental (no vocals)
  - Cover

Q2: Creation mode?
  - Basic — one-line description, auto-generate
  - Advanced — edit lyrics, refine prompt, plan
```

If the user gives a clear one-liner like "make me a sad piano piece", skip the questions —
infer instrumental + basic mode and proceed.

---

### Step 1: Basic Mode

**Goal**: User provides a short description, the skill auto-generates everything, then calls
the API.

1. **Expand the description into a prompt**: Take the user's one-liner and expand it into a
   rich music prompt. Refer to the **Prompt Writing Guide** appendix at the end of this
   document for style vocabulary, genre/instrument references, and prompt structure.
   **The API prompt should always be written in English** for best generation quality,
   regardless of the user's language.
   
   Follow this pattern:
   ```
   A [mood] [BPM optional] [genre] song, featuring [vocal description],
   about [narrative/theme], [atmosphere], [key instruments and production].
   ```

2. **Show the user a preview** before generating. Translate all labels AND the prompt
   description into the user's language. The English prompt is only used internally when
   calling the API — the user should never see it. Example template (English reference —
   localize everything at run

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

Do NOT show playback controls (e.g. keyboard shortcuts) — they don't work in this environment since the player runs in the background.

Gates applied: no_behavioural_pass.

[![ModelsAtlas audit: Minimax Music Gen](https://modelsatlas.com/skills/minimax-ai/minimax-music-gen/badge.svg)](https://modelsatlas.com/skills/minimax-ai/minimax-music-gen/)

Not your listing, or something out of date? Claim or update your listing.