Data Agents — Setup Wizard

Goal: produce a correct data-agents.config.yaml at the root of the current project so Debbie, Vera and Max know where the code lives and how this team works. Be quick and friendly. Detect as much as you can, and ask only what you can't infer.

Step 1 — Where is the repo?

• If $ARGUMENTS is a path, use it as repo.path.
• Otherwise ask: "Where does your data platform code live? Paste a path, or say 'here' for this repo." Accept relative or absolute paths. Check that the path exists before continuing.
• If a data-agents.config.yaml already exists, read it, show the current values in a short table, and ask what to change. Don't start over.

Step 2 — Detect the stack (read-only)

Scan repo.path (skip .git, node_modules, .venv, venv) and infer values from these signals:

Also list the top-level folders with a one-line guess of each one's purpose. They become code_dirs candidates.

Step 3 — Confirm in one go

Show a compact table: setting | detected value | evidence (file). Then ask, in a single question, only for what's still missing:

1. Folders that must never be edited (handed-over or vendored code)?
2. Folders agents may write new work into?
3. Who approves plans? (default: "the user")
4. Any architecture docs they should always read?

Use the AskUserQuestion tool if it's available, and keep it to 4 questions or fewer.

Step 4 — Write the config

Start from the template, using the first of these that exists:

1. ${CLAUDE_PLUGIN_ROOT}/templates/data-agents.config.yaml (plugin install)
2. .claude/skills/setup/data-agents.config.yaml (script install)
3. otherwise recreate the structure: sections project, repo, platform, conventions, data_quality, ml, workflow, context_files. Fill in the confirmed values, leave unknowns as "" / [], and keep the comments. Write it to ./data-agents.config.yaml.

Step 5 — Recommend power-ups (don't install without a yes)

Based on the detected stack, suggest the matching external skills. Agents get sharper when these are present:

Step 6 — Finish

Print:

🎉 Squad ready for <project.name or folder name>
Repo:      <repo.path>   Stack: <compute> + <orchestrator> + <table_format>
Try:
  "Debbie, build an incremental silver table for <a real table you found>"
  "Vera, profile <a real table> and propose DQ rules"
  "Max, frame a baseline model for <a plausible target>"

Use real table or file names you found during detection in the examples.