MulmoClaude Collections

A curated registry of MulmoClaude collections — schema-driven data apps you can discover and import from inside MulmoClaude.

A collection is a small bundle: a schema.json that declares a data model and its UI (tables, calendars, kanban, forms), optional custom HTML views, optional action templates, and optional seed records. The MulmoClaude host renders the whole app from that schema — no per-collection host code.

For users

In MulmoClaude, open /collections → Discover. The app reads this registry's published index.json and lists every collection with its icon, fields, views, author, and (if provided) a screenshot. Import one and it appears at /collections/<slug>, with any bundled seed data populated on first import.

For contributors

See CONTRIBUTING.md. In short: add collections/<your-github-login>/<slug>/ (with SKILL.md, schema.json, meta.json), run npm run validate and npm run build-index, and open a PR. You may only publish under your own GitHub-login namespace (CI enforces it).

collections/<author>/<slug>/
  SKILL.md  schema.json  meta.json
  screenshot.png?  views/*.html?  templates/*.md?  seed/items/*.json?

How it works

• scripts/build-index.mjs walks collections/ and regenerates index.json — the single file the MulmoClaude backend fetches (one GET) to render the Discover catalog — plus a per-collection manifest.json listing the bundle files the host fetches when importing. The build-index workflow regenerates index.json and publishes it (and the JSON Schemas) to GitHub Pages on every push to main, so the backend reads a stable URL without bot-commits to main. index.json is therefore generated, not committed (it's gitignored); the committed bundle artifacts are the manifest.json files, which the host fetches from git by raw URL.

  Enable Pages once: Settings → Pages → Source: GitHub Actions.

• scripts/validate.mjs checks every collection (run by pr-validate on each PR): metadata, semver, author/PR-author identity, host-equivalent schema validity, seed records (secrets hard-fail, PII warn), and a custom-view CSP lint.
• Contracts live in schema/: meta.schema.json and index.schema.json (the public, versioned index contract).

Commands

npm run validate         # validate every collection
npm run build-index      # regenerate index.json (gitignored) + each manifest.json
npm run check-manifests  # fail if any committed manifest.json is stale
npm test                 # run the unit tests

License

Registry tooling: MIT (see LICENSE). Each collection declares its own license in its meta.json.