@skriva/payload-plugin

@skriva/payload-plugin

Payload CMS 3 plugin that lets Skriva publish generated articles into any collection of any Payload site. Skriva discovers your collection's schema at runtime — no hardcoded field mapping, and Skriva never needs to know your content model up front.

Requirements

• Payload ^3.0.0 with @payloadcms/richtext-lexical (peer dependencies — already present in any Payload 3 project)
• Node.js 18+

Install

pnpm add @skriva/payload-plugin
# or: npm install @skriva/payload-plugin

// payload.config.ts
import { contentIngest } from '@skriva/payload-plugin'

export default buildConfig({
  // ...
  plugins: [
    contentIngest({
      targetCollection: 'posts',       // where to publish (required)
      mediaCollections: ['media'],     // relation/upload fields to these are skipped (text-only v1)
      // authCollection: 'users',      // default; useAPIKey is enabled on it
      // blockAllowlist: ['hero', 'content', 'cta'], // for block-based collections
      // relationDefaults: { author: '<docId>' },    // fills relations the AI can't pick
    }),
  ],
})

Connect to Skriva

1. In the Payload admin, open a document in your authCollection (default users) and enable its API key.
2. In Skriva, open your site's Connection page, choose destination "Payload", and paste the site's base URL and the API key.
3. Hit "Test connection" — Skriva reads /manifest and confirms the publish target.

Note: rotating PAYLOAD_SECRET invalidates all Payload API keys — regenerate the key and update it in Skriva afterwards.

How it works

All configuration that is shaped like your CMS (target collection, media collections, block allowlist, relationship defaults) lives here, in your payload.config. Skriva only stores a base URL and an API key, sends markdown plus scalar values, and the plugin owns the conversion to Lexical.

The plugin registers three endpoints under /api/content-ingest:

Endpoint	Purpose
GET /manifest	Health check + the configured publish target (drives "Test connection").
GET /schema	The target collection's field tree as JSON. Relationship/upload fields pointing at mediaCollections are flagged isMedia; other relations are enriched with candidate options (cap 100); the block palette is restricted to blockAllowlist.
POST /publish	Idempotent upsert into the target collection. richText slots arrive as markdown and are converted to Lexical with each field's own editor config (convertMarkdownToLexical); relationship defaults are filled server-side. Drafts supported via _status / draft: true.

Develop

pnpm install   # devDeps only; peers are not auto-installed
pnpm test      # vitest — covers the pure schema walker
pnpm build     # tsc → dist

The pure serializeFields walker is unit-tested here; the endpoint handlers and markdown→Lexical conversion are integration-tested against a real Payload instance.

License

MIT