setup-artist（布景艺术家）

Setup Artist

Scaffold a complete artist workspace so agents can start working immediately.

Prerequisites

• The sandbox has already been set up (see setup-sandbox skill)
• An artist folder exists at orgs/{org}/artists/{artist-slug}/ with a RECOUP.md marker file
• The RECOUP.md file contains the artist's name, slug, and Recoup ID (created by setup-sandbox)

Folder Structure

{artist-slug}/
├── RECOUP.md
├── README.md
├── .env.example
├── .env
├── context/
│   ├── artist.md
│   ├── audience.md
│   ├── era.json
│   ├── tasks.md
│   └── images/
│       └── README.md
├── memory/
│   ├── README.md
│   └── MEMORY.md
├── songs/
│   └── README.md
├── releases/
│   └── README.md
├── content/
│   ├── README.md
│   ├── images/
│   └── videos/
├── config/
│   ├── README.md
│   └── SERVICES.md
├── library/
│   └── README.md
└── apps/
    └── README.md

Steps

Step 1: Read RECOUP.md and create the directory structure

1. Navigate to the artist folder and read RECOUP.md to get the artist's name, slug, and ID:

cd orgs/{org}/artists/{artist-slug}
cat RECOUP.md

1. Create the directory structure:

mkdir -p {context/images,memory,songs,releases,content/images,content/videos,config,library,apps}

Step 2: Update RECOUP.md

Update the status field from not-setup to active and replace the body with a brief description:

---
artistName: {Artist Name}
artistSlug: {artist-slug}
artistId: {uuid-from-recoupable}
status: active
---

# {Artist Name}

Connects this workspace to the Recoupable platform. See `README.md` for the full directory guide and setup checklist.

Step 3: Create context files

Create each file from the templates in references/context-files.md. The essential files:

Step 4: Create memory system

Create two files:

• memory/README.md — Full instructions for agents on how to use the memory system, including the scope concept. See references/memory-system.md.
• memory/MEMORY.md — Nearly empty starting point with frontmatter and guidelines comment (including scope rules).

The memory system uses three scopes to prevent knowledge bloat:

• permanent — true regardless of era (goes in MEMORY.md)
• era — true for the current release cycle (goes in MEMORY.md, tagged with era)
• session — about a specific piece of content or task (goes in log/ only, never MEMORY.md)

Agents should ask the user about scope before saving feedback to long-term memory.

Step 5: Create services and environment files

Services are tracked in config/, not pre-filled at setup. Create:

Do NOT pre-fill service entries. Services are added when the agent has real information — a handle, an API key, a confirmed account. The old approach of creating a massive JSON file with every possible service set to not-setup creates noise, not value.

Step 6: Create README files for remaining directories

Each directory needs a README.md explaining its purpose. See references/directory-readmes.md for templates.

Step 7: Create root README

Create README.md at the artist root with:

• Artist name as heading
• Directory structure table
• Context files table
• Config & services table
• Setup checklist

See references/root-readme.md for the template.

Step 8: Fill in what you can

If you have information about the artist (from the user, from research, or from the Recoup platform):

1. Fill context/artist.md with as much identity/brand info as possible
2. Fill context/audience.md with audience insights
3. Set context/era.json with the current release phase

Don't fabricate information. Leave placeholders for anything you don't know.

Step 9: Commit

git add -A
git commit -m "setup: create {artist-name} artist workspace"
git push origin main

Naming Conventions

• Directories and slugs: lowercase-kebab-case (e.g. gatsby-grace, a-thing-called-love)
• Audio files: Match the folder slug (e.g. songs/a-thing-called-love/a-thing-called-love.mp3)
• Context files: Use the names exactly as specified — agents and shared tools expect them

Principles

• Start lean. Only create what's needed. Agents and pipelines will create additional files (like content/videos/shortform/) as they run.
• Placeholders over empty. Use {placeholder} syntax for unknown values — it's better than blank fields.
• Don't pre-fill what you don't know. A file full of not-setup and null isn't a placeholder — it's clutter. Services, accounts, and configs should be added when they're real.
• README everything. Every directory gets a README so agents know what belongs there.
• Don't duplicate. Songs live in songs/, releases reference them by slug. Content goes in content/, not copied elsewhere.
• Scope your memories. Not all knowledge lasts forever. Tag era-specific memories, keep session feedback in logs, and ask before promoting to long-term memory.