Pages

@webx-ui/module-pages is the section where the site's pages are edited, and webx-ui/module-pages on the server is the tree behind it. This page is both, because neither is useful alone.

Very little is written in either half, and that is the point. The tree is webx-ui/nested-set, the address is the registry of webx-ui/routing, the content is blocks, the draft and the history are module-admin, what a page says about itself is module-seo, the languages are webx-ui/localization. The module is the first entity that puts all of them together on something real — and therefore the first honest test of each.

Install

pnpm add @webx-ui/module-pages
composer require webx-ui/module-pages
php artisan migrate

import { createAdmin } from '@webx-ui/module-admin'
import { pages } from '@webx-ui/module-pages'
import '@webx-ui/module-pages/style.css'

createAdmin({
  modules: [pages()],
})

The section appears in the Content group, first in it, once both halves are there. Permissions: pages.view opens the section and the picker a link field would use, pages.manage writes. The migration creates the pages table and the home page in it.

A site that installs this module gives / to it and deletes its own route for the front page — including the welcome one from the Laravel skeleton. As long as the application answers at / itself, the registry holds that address reserved and the home page never gets a row in it.

The home page is the root

The tree has exactly one root, it is the home page, and it is always there. Its slug is empty in every language, so the path it formats to is '' — which is what the registry calls the front page — and its child about answers at /about rather than /home/about.

Its content is edited like any other page's. Its structure is not: it cannot be moved, deleted or given an address, and a second root is refused. The panel is told which of those apply by the node's capabilities rather than by a "this is the root" flag, so the next page that has to stay put — a search page, a 404 — reuses the rule instead of adding a check:

$page->capabilities();          // ['move' => true, 'delete' => true, 'address' => true]
Page::home()->capabilities();   // all three false

Addresses

A page's address is the slugs of its ancestors and then its own, which makes moving a page under another one the same gesture as changing its address:

RouteTypes::register(new RouteType(
    type: 'page',
    model: Page::class,
    formatter: TreePath::class,
    handler: PageHandler::class,
    onConflict: OnConflict::Fail,
));

Fail rather than Suffix: a page address is chosen deliberately, so a taken one is an error on the slug field and not a quiet about-2. Moving a branch rewrites every address below it and leaves a 301 alias on each of the old ones — the registry does that by itself, and both the panel and the agent tools say how many pages it happened to.

title and slug are translatable, and the address is per language. A page with no slug in a language has no address in that language and does not open there. That is deliberate: an address built from another language's slugs would serve English words in front of content nobody translated.

Drafts, publishing, history

Every save from the panel is a save into the draft; the site keeps showing what it showed until somebody publishes:

$page->saveDraft(['title' => ['en' => 'About'], 'blocks' => [...]]);
$page->status();      // 'draft' — never published
$page->publish();     // now the site shows it
$page->status();      // 'published', or 'modified' once there is a draft again
$page->unpublish();

Three states and no fourth: never published · on the site · on the site with edits waiting. A page that was never published is a 404 to everybody, and the same page under a preview token of module-blocks is shown as it will be — that decision belongs to the handler, not to the registry, which is why unpublishing does not release the address.

The history is the publications. The autosaves the editor writes every couple of minutes are insurance, not history: a ring of unnumbered copies of the draft, and a list that mixed them in would be a list nobody reads. Restoring an old version puts it in the draft rather than on the site — publishing it is the same separate step it always is.

Deleting takes the branch

$about->delete();          // $about and everything under it
$about->restoreBranch();   // exactly what went down with it, and nothing else

One node at a time rather than one statement, because it is the delete event that releases an address: a branch turned off in bulk would leave every address in it held by a page nobody can see. Which page's deletion trashed a node is kept in trashed_with, so a restore does not bring back something that was deleted earlier and separately.

Addresses come back with the pages — and if one was taken while the page was in the bin, the restore is refused. A page quietly restored to a different address is worse than one that says the place is occupied.

The list

One level of the tree at a time. The home page is pinned above the list and its children are the level: every page of the site is inside it, so drawing it as a branch would give every row a step of indentation that says nothing.

• Children arrive when a branch is opened — a catalogue of a few hundred pages is never fetched whole.
• Searching puts the tree away and answers with a flat list of matches, each with its address underneath: a branch drawn for the sake of one match deep inside it tells the reader nothing. The box looks in every language the site has, not in the one the panel is open in — the list shows the title a page carries, so a page named in English alone is on the screen of a Russian panel and is found by that English name.
• The bin is a filter, not a section. It lists the pages somebody deleted; what went down with a page comes back with it.
• Two ways to move a page: drag it, or use "Move…" and pick the page it goes inside — which is the one that works on a touch screen and in a big catalogue.

On a phone the section stops pretending to be a tree. Cards below a table have neither indentation to read nor a chevron to open, so at that width it asks the server for a flat list and prints each page's address under its name — which is what identifies a page anyway.

The editor

Four tabs: Content · Settings · SEO · History. The form is a described screen, pages.form, registered by the package, so a module or a project adds to it with a patch rather than a fork — the SEO card arrives that way, from module-seo and not from the site:

app(ScreenRegistry::class)->extend('pages.form', [
    ['op' => 'add', 'target' => 'seo', 'node' => [...]],
]);

What a save carries is decided by that tree and checked by ScreenValues, so a key the screen does not name is dropped and a refused value lands under the field it belongs to.

The panel sends back the revision it read the page at — a short hash of the content, because a draft has no version number of its own — and a save whose revision is no longer the current one is answered with a 409 carrying the page as it now is, instead of being written over whoever saved in between. Two writers who saved the same thing did not conflict, and the check says so. The same guard covers an agent, which is the case it was really written for.

GET    /api/cms/pages                            a level of the tree, search, filters, the bin
POST   /api/cms/pages                            a new page under a parent
GET    /api/cms/pages/{id}                       values, the trail, a preview link, the revision
PUT    /api/cms/pages/{id}                       the draft: { values, revision }
POST   /api/cms/pages/{id}/move                  { target, zone }
POST   /api/cms/pages/{id}/publish               · /unpublish · /duplicate
DELETE /api/cms/pages/{id}                       to the bin, with the branch · POST /restore
GET    /api/cms/pages/{id}/versions              the publications, newest first
POST   /api/cms/pages/{id}/versions/{n}/restore  an old one becomes the draft

The view

The handler hands the page to webx-pages.view, which defaults to pages.show, as $page:

@php($content = $page->renderBlocks())
<!doctype html>
<html>
<head>
    @webxSeo($page)
    @webxBlocks
</head>
<body>{!! $content !!}</body>
</html>

Until the site has written that view, the package prints its own — the blocks, the SEO head and nothing else — so a fresh installation serves a page rather than an error.

For an agent: MCP

The section is also a set of tools. With webx-ui/mcp installed (it comes with this module), the panel serves one MCP server at /api/cms/mcp:

php artisan vendor:publish --tag=sanctum-migrations && php artisan migrate   # once per site
php artisan webx:mcp:token admin@example.com --name=claude

The agent then acts as that administrator, within the token's scopes: pages:read for the tools that look, pages:write for the ones that change. On the machine the site runs on, php artisan mcp:start webx is the same server over stdio.

Tool	What it does
pages_tree	The tree, one level of it, a search, or the bin: addresses, status, who touched it last
pages_get	One page in full: the trail, the values, the revision, a preview link
pages_create	A new page under a parent, as a draft
pages_update	The values into the draft, guarded by the revision
pages_move	Somewhere else in the tree, and it says how many addresses changed
pages_publish	The draft onto the site · pages_unpublish takes it off
pages_delete	To the bin with its branch · pages_restore brings the branch back

Every tool that changes something accepts dry_run: true and then reports what it would do without doing it. A page is named by its id or by its address — "/catalog/shoes", and "/" for the home page — because that is what a site is talked about in, and what the sitemap hands over. Text fields answer with every language at once: an agent that got one title has no way of knowing whether the others exist.

Content does not travel through these tools. Blocks are blocks_edit_content, which names the node it changes and leaves the rest of the page alone; sending a blocks key to pages_update is refused with that sentence rather than ignored. There is one way to change the content of a page, and a second one would be the way that overwrites twenty blocks to fix a heading.

Before writing, an agent reads pages://sitemap: every page of the site nested the way it is nested, with its address in each language, its status and what may be done to it. One prompt, build_page, packages the loop — read the map and the block catalogue, create the page, fill it with blocks_edit_content, look at the preview, write the SEO card, and leave it as a draft for a person.

Config

config/webx-pages.php:

Key	Default	What it is
view	pages.show	The view the handler prints a page with; the package's own until it exists

What is deferred

• Permissions beyond pages.view and pages.manage: publishing and deleting as rights of their own arrive with the wider conversation about roles.
• sitemap.xml belongs to module-seo, which needs every kind of entity and not only pages.
• Scheduled publishing needs a scheduler running on the site.
• A diff between two versions, bulk operations over a selection, copying a branch whole.
• The site's menu — a candidate for its own package, not a field of a page.