da-sc-sdk

A headless SDK for working with DA Structured Content — schema-constrained JSON documents serialized as a defined HTML wire format. It exposes a small, focused API covering the operations supported today.

The SDK is pure ESM and ships with no DOM dependencies and no I/O. UI rendering, transport, and persistence are intentionally left to the consuming application.

In exchange, every tool built on the SDK shares the same DA Structured Content behavior — defined here once and used across all consumers, instead of being re-implemented (and inevitably drifting) in each tool.

Status

0.x — the public API may change before 1.0.0.

Install

npm install @adobe/da-sc-sdk

import {
  createEngine,
  convertJsonToHtml,
} from "@adobe/da-sc-sdk";

Quick start

import { createEngine, convertJsonToHtml } from "@adobe/da-sc-sdk";

const schema = {
  type: "object",
  title: "Project",
  required: ["name"],
  properties: {
    name: { type: "string", title: "Name" },
    tags: {
      type: "array",
      title: "Tags",
      items: { type: "string", title: "Tag" },
    },
  },
};

const engine = createEngine({
  schema,
  document: { metadata: { schemaName: "project" }, data: {} },
  onChange: () => {
    /* state changed — read engine.getState() */
  },
});

engine.setField("/data/name", "Alice");
engine.addItem("/data/tags");
engine.setField("/data/tags/0", "demo");

const { html } = convertJsonToHtml({ json: engine.getState().document });
// `html` is ready to POST. Persistence is your job.

Runnable scripts in examples/.

API

Five top-level functions:

| Function | Purpose | Returns | | ------------------- | ---------------------------------------------------- | --------------------------------- | | createEngine | Stateful editing (UI editors, agents holding state). | Engine | | validateSchema | Lint a schema before binding to it. | { valid, schemaIssues } | | validateData | Check data against a schema. | { valid, errors, schemaIssues } | | convertJsonToHtml | Convert JSON → DA wire-format HTML. | { html } \| { error } | | convertHtmlToJson | Convert DA HTML → JSON. | { json } \| { error } |

createEngine compiles a schema + initial document into a stateful, mutable handle.

What it offers:

• Pointer-addressed mutations (setField, addItem, insertItem, removeItem, moveItem) that update immutably and re-validate atomically.
• Schema defaults materialized into the initial document.
• onChange fires exactly once per real mutation. Setting a field to its current value is a no-op and does not fire onChange.

Use it when you edit a document over time — interactive editors, agents holding state across turns, anything where the document survives between operations.

The handle exposes six methods. All synchronous, all addressed by RFC 6901 JSON Pointer:

| Method | Purpose | | ----------------------------- | ------------------------------------ | | getState() | Read the current state snapshot. | | setField(pointer, value) | Set a primitive value at pointer. | | addItem(pointer) | Append to the array at pointer. | | insertItem(pointer) | Insert before the item at pointer. | | removeItem(pointer) | Remove the item at pointer. | | moveItem(pointer, from, to) | Reorder an array item. |

Every real mutation triggers onChange; the return value is the engine's current state after the call. onChange does not fire at construction — read the initial state via getState(). To start over with a different (schema, document), create a new engine.

Usage

Validate

import { validateSchema, validateData } from "@adobe/da-sc-sdk";

validateSchema({ schema });
// → { valid, schemaIssues }

validateData({ schema, data });
// → { valid, errors, schemaIssues }

errors is a pointer-keyed map; errors[pointer]?.message is O(1). Each entry exposes keyword, instancePath, params, and message fields. valid is true only when both schemaIssues and errors are empty — you can't claim a document is valid against a broken schema.

Convert (JSON ↔ HTML)

import { convertJsonToHtml, convertHtmlToJson } from "@adobe/da-sc-sdk";

const { html } = convertJsonToHtml({
  json: { metadata: { schemaName: "project" }, data: { name: "Alice" } },
});

const { json } = convertHtmlToJson({ html });

convertHtmlToJson returns { error: "<reason>" } on empty or malformed input.

Edit (with persistence)

The engine has no transport and no save status. Wire your own persistence on top of onChange:

let engine;
let lastValues;
engine = createEngine({
  schema,
  document,
  onChange: () => {
    const next = engine.getState().document;
    if (next === lastValues) return; // skip non-mutation transitions
    lastValues = next;
    // convert via convertJsonToHtml, POST, IndexedDB-cache — whatever fits
  },
});
lastValues = engine.getState().document; // onChange does NOT fire at init

docs/headless-consumer.md has a ~40-line single-flight save pattern with status tracking.

Error handling

The SDK reports validation, conversion, and mutation failures through return values rather than exceptions. Truly invalid arguments (wrong types, malformed pointers) may still surface as runtime errors.

• Validators (validateSchema, validateData) — check result.valid. When false, inspect result.schemaIssues for schema-level problems and result.errors for data-level problems (the latter is a { [pointer]: ValidationError } map).
• Converters (convertJsonToHtml, convertHtmlToJson) — destructure with a fallback: const { html, error } = convertJsonToHtml(...). The error field is set on empty or malformed input; otherwise the success field (html or json) is set.
• Engine — if the schema is malformed, createEngine still returns a handle but getState().schemaIssues will be non-empty and mutations no-op. Check schemaIssues.length === 0 before relying on the engine.

TypeScript

Type declarations ship in index.d.ts — no separate @types/* install needed.

import type {
  Engine,
  EditorState,
  Document,
  JsonPointer,
  SchemaIssue,
  ErrorsByPointer,
  ValidationError,
} from "@adobe/da-sc-sdk";

Runtime support

• ES2022 + ESM. No Node built-ins, no DOM.
• Node ≥18 and modern browsers.
• One runtime dependency: hast-util-from-html.

Design

• Headless. No UI, no DOM, no rendering. UIs subscribe to onChange.
• No I/O. No fetch, no file system, no DA endpoints. Callers wire their own transport.
• Five named exports. The package.json exports map blocks deep imports. The five exports are the entire public API and the intended semver boundary.
• Strict schema subset. A strict subset of JSON Schema 2020-12 — see docs/schema-spec.md. Anything outside that contract surfaces in schemaIssues.
• Not for general JSON Schema validation.

Development

Inside the repo:

• npm install — install dev dependencies
• npm test — run the test suite (Web Test Runner)
• npm run lint / npm run lint:fix — lint
• npm run build — produce dist/index.js

Internals

For contributors and anyone going deeper than the public API — layout, lifecycle, schema/model builders, and the headless consumer pattern — see docs/.

License

Apache-2.0 © Adobe Inc.