Scaffolding skill

The new-nebari-app Claude Code skill teaches coding agents to generate apps in the exact layout this pack expects — real code plus a nebari-app.yaml launch manifest — and to launch them through the MCP server. Together they make “build me a dashboard, then launch it” a single conversation.

Install

Copy the skill from the repository into your project (or user) skills directory:

# project-level
cp -r skill/new-nebari-app .claude/skills/

# or user-level
cp -r skill/new-nebari-app ~/.claude/skills/

It triggers on requests like “create a nebari app”, “scaffold a static site for the cluster”, “scaffold a python app for nebari”, or “launch it” in a directory containing nebari-app.yaml.

What it scaffolds

docs-site/                     py-app/
  nebari-app.yaml                nebari-app.yaml
  index.html                     pixi.toml
  styles.css                     app.py

Apps get real files (never hand-written inline YAML). The starters follow the platform conventions: static assets served as-is, Python apps launched by a pixi task that serves on 0.0.0.0:8080, and no auth code in the app — the gateway handles SSO.

The manifest

nebari-app.yaml sits next to the code and maps 1:1 onto the App resource:

name: docs-site
namespace: apps
displayName: "Docs Site"
source:
  type: files
  files: { path: "." }
access:
  public: true
  subdomain: docs-site

Manifest source types are files, git, and pvc. files is an authoring convenience: a directory path next to the manifest — on launch, the agent reads those files and passes them as the app’s inline source (same rules as uploads). The full schema and the manifest → launch_app mapping live in the skill’s references/manifest.md.

”Launch it”

When the user asks, the agent:

1. reads nebari-app.yaml,
2. calls describe_cluster (and authenticate if prompted),
3. maps the manifest onto launch_app — idempotent, so iterating is safe,
4. polls get_app_status until Running (checking get_app_logs on failure),
5. reports the app URL.

The committed nebari-app.yaml also doubles as the GitOps artifact for teams who keep their apps in version control.