On the server
A stored document is JSON sent by the browser, so the server must check it before saving. Two entry points have no Svelte dependency and are meant for server code:
@atom-forge/brickee/model — StoredBlock, MD, MAX_INDENT, MAX_DEPTH, isMd, acceptsId, inlineToPlainText, blocksToPlainText;
@atom-forge/brickee/schema — blocksSchema and blocksSchemaFor, built with zod (an optional peer dependency: install it to use this entry point).
Validating a document
import { blocksSchemaFor } from '@atom-forge/brickee/schema';
import { plugins } from '$lib/blocks';
const contentSchema = blocksSchemaFor(plugins);
export function parseContent(input: unknown) {
const result = contentSchema.safeParse(input);
if (!result.success) throw new Error('Invalid content');
return result.data;
}
blocksSchema checks the shape of any document; blocksSchemaFor(plugins) adds the containers' accepts lists. It needs only type and accepts of each plugin, so a server that does not want to import the plugin components can pass those alone:
const contentSchema = blocksSchemaFor([
{ type: 'note', accepts: ['md:*', 'youtube'] },
{ type: 'youtube' }
]);
What the schema checks:
| Rule |
Limit |
| Block keys |
type, indent, text, props, blocks only (strict objects). |
type |
^[a-z][a-z0-9:-]{0,39}$ |
indent |
An integer, 0–6. |
text |
At most 100 000 characters. |
props |
Keys up to 60 characters, JSON values. |
blocks |
At most 500 nested blocks per container, one level deep only. |
| Document |
At most 2000 top-level blocks and 1 000 000 characters as JSON. |
accepts (blocksSchemaFor) |
Every nested block is one its container takes. |
What it does not check: whether a type is known, whether a unique block appears once, and what the props contain. Add your own refinements for the rules your application relies on — for example, to reject unknown block types:
import type { StoredBlock } from '@atom-forge/brickee/model';
import { blocksSchemaFor } from '@atom-forge/brickee/schema';
import { plugins } from '$lib/blocks';
const BUILT_IN = ['md:p', 'md:h1', 'md:h2', 'md:h3', 'md:h4', 'md:h5', 'md:h6', 'md:ul', 'md:ol', 'md:todo', 'md:quote', 'md:code', 'md:hr'];
const known = new Set([...BUILT_IN, ...plugins.map((p) => p.type)]);
const allKnown = (list: StoredBlock[]): boolean => list.every((b) => known.has(b.type) && allKnown(b.blocks ?? []));
export const contentSchema = blocksSchemaFor(plugins).refine(allKnown, 'Unknown block type.');
Plain text
blocksToPlainText(blocks) collects every text of a document in order, separated by blank lines: block texts with their marks and link targets removed, string props that contain whitespace (titles, captions — not ids or colours), table rows, and nested blocks. Use it for search indexes, leads and meta descriptions. inlineToPlainText(md) does the same for a single inline markdown string.
import { blocksToPlainText, type StoredBlock } from '@atom-forge/brickee/model';
export const description = (doc: StoredBlock[]) => blocksToPlainText(doc).slice(0, 160);
Rendering
Preview is an ordinary Svelte component and can be server-rendered; the core renderer and the bundled previews use no browser APIs. Only the markdown export path (toMarkdown, toStored) needs the browser.
On the server
A stored document is JSON sent by the browser, so the server must check it before saving. Two entry points have no Svelte dependency and are meant for server code:
@atom-forge/brickee/model—StoredBlock,MD,MAX_INDENT,MAX_DEPTH,isMd,acceptsId,inlineToPlainText,blocksToPlainText;@atom-forge/brickee/schema—blocksSchemaandblocksSchemaFor, built withzod(an optional peer dependency: install it to use this entry point).Validating a document
// content.server.ts import { blocksSchemaFor } from '@atom-forge/brickee/schema'; import { plugins } from '$lib/blocks'; const contentSchema = blocksSchemaFor(plugins); export function parseContent(input: unknown) { const result = contentSchema.safeParse(input); if (!result.success) throw new Error('Invalid content'); return result.data; // StoredBlock[] }blocksSchemachecks the shape of any document;blocksSchemaFor(plugins)adds the containers'acceptslists. It needs onlytypeandacceptsof each plugin, so a server that does not want to import the plugin components can pass those alone:const contentSchema = blocksSchemaFor([ { type: 'note', accepts: ['md:*', 'youtube'] }, { type: 'youtube' } ]);What the schema checks:
type,indent,text,props,blocksonly (strict objects).type^[a-z][a-z0-9:-]{0,39}$indenttextpropsblocksaccepts(blocksSchemaFor)What it does not check: whether a type is known, whether a unique block appears once, and what the props contain. Add your own refinements for the rules your application relies on — for example, to reject unknown block types:
import type { StoredBlock } from '@atom-forge/brickee/model'; import { blocksSchemaFor } from '@atom-forge/brickee/schema'; import { plugins } from '$lib/blocks'; const BUILT_IN = ['md:p', 'md:h1', 'md:h2', 'md:h3', 'md:h4', 'md:h5', 'md:h6', 'md:ul', 'md:ol', 'md:todo', 'md:quote', 'md:code', 'md:hr']; const known = new Set([...BUILT_IN, ...plugins.map((p) => p.type)]); const allKnown = (list: StoredBlock[]): boolean => list.every((b) => known.has(b.type) && allKnown(b.blocks ?? [])); export const contentSchema = blocksSchemaFor(plugins).refine(allKnown, 'Unknown block type.');Plain text
blocksToPlainText(blocks)collects every text of a document in order, separated by blank lines: block texts with their marks and link targets removed, string props that contain whitespace (titles, captions — not ids or colours), table rows, and nested blocks. Use it for search indexes, leads and meta descriptions.inlineToPlainText(md)does the same for a single inline markdown string.import { blocksToPlainText, type StoredBlock } from '@atom-forge/brickee/model'; export const description = (doc: StoredBlock[]) => blocksToPlainText(doc).slice(0, 160);Rendering
Previewis an ordinary Svelte component and can be server-rendered; the core renderer and the bundled previews use no browser APIs. Only the markdown export path (toMarkdown,toStored) needs the browser.