"Configure and operate Muzic MusicAgent safely: install
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill music-agent-workflows --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Music Agent Workflows?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-music-agent-workflows)More formats (shields.io, HTML) on the badges page.
---
name: music-agent-workflows
description: "Configure and operate Muzic MusicAgent safely: install
prerequisites, manage model and cache layout, validate config, launch CLI or
Gradio, and troubleshoot tool selection and secrets."
metadata:
disco-role: operating
disable-model-invocation: true
license: MIT
---
# Music Agent Workflows
Use this sub-skill when the request is about MusicAgent itself: setup, configuration, model and download layout, startup, tool enablement, or runtime troubleshooting.
## Start here
- [Setup and configuration](references/setup-and-configuration.md)
- [Tool and runtime map](references/tool-and-runtime-map.md)
- [Troubleshooting](references/troubleshooting.md)
- [Config validator](scripts/validate_musicagent_config.py)
## What it covers
- Docker vs Conda/pip installation trade-offs.
- Linux system packages needed for audio, notation, and model download support.
- `config.yaml` fields, `local_fold`, `src_fold`, `log_file`, and secret handling.
- CLI vs Gradio launch paths and their credential differences.
- Task planning, tool selection, and model/tool enablement boundaries.
- Safe config validation without importing the source repo.
## What it does not cover
- ROC, DiffSinger, DDSP, and other tool internals.
- Generic provider or gateway policies outside MusicAgent.
- Training or algorithm-level guidance for the wrapped models.
## Operating notes
- Keep API keys, tokens, and `.env` contents out of version control.
- Run MusicAgent from the project directory that contains `config.yaml` so relative paths resolve as expected.
- Not every planner task has a loaded pipe. Some task names are aspirational and will fall through to an "unloaded models" error until the plugin layer is extended.
- If a task fails, check `disabled_tools`, then the `local_fold` cache, then the required system package or credential.
- Use the bundled validator before trying to launch the agent.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!