[fmontes]

Spec-Kit PR 2, part 2 of 3. Based on part 1 (PR #38324), so the diff shows only this part. It holds what were stack layers 3, 4 and 5: every editor entry point in the page editor (UVE), fields and block editors asking their host instead of opening panels or navigating, and the last admin surfaces (legacy JSP portlets, deep links, Pages, Publishing Queue) with the admin's legacy iframe wrappers deleted. The implementation was first published as seven stacked PRs; it was folded to three on 2026-10-29. The commit history keeps the layer boundaries, so the Commits tab reads as three reviews in order.

Refs #38303

Review by layer

Layer	Commits	What to look at
3. UVE (US1, US5)	68c1303df0..d4eef8a38c, 12 commits, Red gates on ce2793bbd5 and be367367da	libs/portlets/edit-ema/portlet: editor, shell, actions handler, utils; deletion of dot-ema-dialog, its store and DotEmaWorkflowActionsService
4. Fields and block editors (US4)	d4eef8a38c..05d6e17b9f, 11 commits, Red gate on 9691f02d4c	EditContentHost.openContent, the two hosts, Relationship and block editor fields, new-block-editor toolbar, block-editor bubble menu, apps/dotcms-block-editor/src/main.ts; deletion of dot-create-content-dialog
5. Admin surfaces (US2b, US6)	05d6e17b9f..8338af145d, 13 commits, Red gate on cae822ce36	apps/dotcms-ui: custom-event handler, deep-link component and routes, Pages, open-edit-content-panel.ts; libs/portlets/dot-publishing-queue; libs/edit-content-bridge; deletion of dot-contentlet-editor/, dot-contentlets/, the Pages guards

Each layer ends with a style(...) commit that only runs prettier on files the layer committed unformatted. Review layer 5 as a range, not commit by commit: 41086d6499 carries staged folder deletions and does not compile on its own (the references go in the next commit); the final tree is what every check below ran on.

Layer 3: the page editor (User Stories 1 and 5)

Every editor entry point in the UVE (pencil, full editor, URL-map edit, Edit Source fallback, palette drop, container "+", page Properties, translate, SDK editContentlet() and createContentlet()) sets one DotEditContentRequest bound through injectEditContentPanel. Legacy screens that are not editors (content selector, reorder menu) open in the generic legacy-screen panel; the form selector opens as panel content. The page refreshes on every saved while the panel stays open; a palette create inserts once; URL-mapped saves load the new URL; requests carry the experiment variant and the site. dot-ema-dialog, its store and DotEmaWorkflowActionsService are deleted.

Decisions the tests encode: languageChanged(id) calls pageLoad({ language_id: '3' }) as a string because DotPageAssetParams.language_id is a string; a page's saved loads the page's URL_MAP_FOR_CONTENT, else its url, when it differs from the current page URL; the cancel event is NG_CUSTOM_EVENTS.CANCEL_SAVING_MENU_ORDER; edit requests take languageId from the contentlet, falling back to the page language; create requests carry no title (the opener heads them with the type name).

One design revision after the Red gate: translating a page opens the legacy editor panel, not the generic legacy-screen panel, because the generic panel lost the translate wizard, the compare view and the dirty prompt (be367367da, 76de1faa8a, 2bd58f9f57).

Layer 4: fields and block editors (User Story 4)

• EditContentHost.openContent(event) is the one way a field asks for other content. The full-screen host (RouterEditContentHost) always opens a panel through DotEditContentOpenerService with the shell's injector. The panel host (OverlayEditContentHost) replaces its content in place with a breadcrumb for an edit coming from a Relationship field, and stacks a second panel for everything else (any create, a block editor's edit), opened with the panel's injector so a re-render of the form underneath never closes the panel on top.
• DotOpenContentEvent.from?: { inode; title }: the field names the breadcrumb's origin, because inside a panel only the field knows what it is showing (an unsaved translation names its source version). Absent means a fresh trail. Recorded in contracts/open-content-from-fields.md and data-model.md.
• Relationship field: openRelated and "create new" both call host.openContent. The create reply is an upsert by identifier: the first save adds the row, later saves refresh it, so the list never shows the created item twice (FR-030). The field's own side-panel code, the dynamic import of the centered dialog, the flag read and the store's withFlags slice are gone. "Create new" stays disabled for a related type on the legacy editor.
• Block editor field binds (contentletEdit) on both editors and hands { inode, contentType } to the host with source: 'block-editor'.
• New block editor toolbar: editContentlet() only emits. Deleted: navigateToContentEditor, the unsaved-changes confirm and the editor's own ConfirmationService provider and <p-confirmdialog>, ContentletEditUrlService, breadcrumb.utils.ts.
• Old block editor bubble menu gains contentletEdit = output<{ inode; contentType }>(); goToContentlet() only emits it. The native confirm, the content type lookup, the relationshipReturnValue localStorage write, generateBackUrl, extractInodeFromUrl and the window.parent navigation are deleted. DotBlockEditorComponent re-emits it through a new @Output() contentletEdit (this lib keeps its decorator style; it is on its way out).
• Legacy JSP bridge (apps/dotcms-block-editor/src/main.ts): a capture-phase contentletEdit listener on document, filtered to the two tags this bundle registers, re-dispatches ng-event { name: 'edit-contentlet', data: { inode, contentType } }. Capture because Angular Elements dispatches outputs as non-bubbling CustomEvents on the element. The legacy editor panel already maps that event to openContent (layer 1, unit-tested).
• Deleted: libs/edit-content/src/lib/components/dot-create-content-dialog/ and its export.

EditContentDialogData, the layout's #interceptDirtyClose / #dialogRef and the overlay host's optional DynamicDialogRef stay until part 3, where they go with the Query Tool move.

Layer 5: the admin surfaces (User Stories 2b and 6)

• Legacy JSP portlets (content search, Site Browser, Sites, Calendar, publishing list): DotCustomEventHandlerService routes edit-contentlet, edit-host, create-contentlet and create-host to DotEditContentOpenerService.open. The request names the portlet that was current when the event arrived; every save and the close reload that portlet's list. MainComponentLegacyComponent registers its own injector with the handler (setInjector) so panels opened from JSP events find the root MessageService and <p-toast>.
• The admin app reaches the opener through a dynamic import (apps/dotcms-ui/src/app/api/util/open-edit-content-panel.ts). app.routes.ts lazy-loads @dotcms/edit-content, so a static import anywhere in the app is forbidden by the Nx module-boundary rule and would pull the library into the main bundle. The helper awaits import('@dotcms/edit-content'), skips opening if the caller's DestroyRef is already destroyed, and resolves the opener from the caller's injector. The main bundle grew by 209 bytes.
• Deep links (/c/:id/:asset, /c/:id/new/:contentType, /pages/:asset, /pages/new/:contentType): DotEditContentDeepLinkComponent opens the panel on init, reloads the parent portlet on save, returns to the parent portlet URL with query params preserved when the panel closes, and closes the panel without navigating when the route is left. DotPortletDetailComponent keeps the workflow-task branch; the Pages guards and DotCreateContentletResolver are gone.
• Pages portlet: Edit opens through the opener with portletId: 'pages' and the component's injector; save and close refresh the listing with the current filters and page (getPages() with no arguments). The dead DotPageStore provider is dropped. The store's state is typed (currentUser: DotCurrentUser | null) so getPages is a known method for the strict gate (467460d179).
• Publishing Queue: both dialogs open contentlet rows through the opener with the dialog's injector; save and close reload the dialog's own contents (FR-017); window.open, the URL resolvers and $assetEditUrls are gone, replaced by one isEditableContentlet predicate. Both DialogService.open configs set focusTrap: false so the panel over the modal owns focus.
• Form bridge (@dotcms/edit-content-bridge): pushFormBridge() returns a frame and popFormBridge(frame?) removes exactly that frame, so a panel closed under a panel stacked over it no longer pops the stacked panel's bridge.
• Deleted: view/components/dot-contentlet-editor/ (wrapper, edit, create, add, reorder, DotContentletEditorService, resolver, routes, specs), portlets/dot-porlet-detail/dot-contentlets/, the <dot-create-contentlet> mount and onCustomEvent in the admin root, the Pages guards. dot-iframe-dialog stays for dot-workflow-task-detail.

Tests (TDD, gates T051, T061, T073, T088 and T098 approved on the Red commits)

Command (from core-web/)	Result
pnpm nx test portlets-edit-ema-portlet	green (3 pre-existing skips)
pnpm nx test edit-content	2603 passed, 0 failed
pnpm nx test new-block-editor	356 passed, 0 failed
pnpm nx test block-editor	92 passed, 35 failed, all pre-existing: the project is tagged skip:test, never runs in CI, and the same cases fail with the layer 2 version of the spec
pnpm nx test dotcms-ui	2323 passed, 0 failed
pnpm nx test portlets-dot-publishing-queue-portlet	290 passed, 0 failed
pnpm nx test edit-content-bridge	148 passed, 0 failed
pnpm nx affected -t lint --base=origin/38303-sidepanel-everything --exclude=tag:skip:lint	0 errors
pnpm nx format:check against part 1	clean
strict gate against part 1, and whole stack against the spec branch	PASS
pnpm nx build dotcms-ui, pnpm nx build dotcms-block-editor	OK

Manual-only exceptions (agreed at T073 and T098)

1. "Edit contentlet" from a block editor inside the legacy JSP editor, itself in a panel: an e2e test would have to drive TipTap inside a Dojo iframe inside a drawer. The bridge is a short dispatch; the panel's mapping of the resulting event is unit-tested; the hop is checked by hand (quickstart US4 step 5). The legacy editor page loads /dotcms-block-editor/main.js from the backend's webapp, so the bundle has to be rebuilt in the checkout the backend mounts.
2. Focus inside the legacy editor's iframe over the Publishing Queue's modal, and Escape pressed inside that iframe: checked by hand with the issue's six-point checklist in quickstart US6.

Checklist

• Tests: every Red case from the five gates passes; see the table.
• Translations: none.
• Security Implications Contemplated: the block editors stop writing to localStorage and stop navigating the parent window; the JSP bridge forwards only inode and contentType from events whose target is one of the two editor tags; the Publishing Queue stops opening new tabs; the legacy screens load same-origin admin URLs through the generic panel's guard.

Additional Info

• Stack: spec 38303-sidepanel-everything (docs(edit-content): spec for opening every content edit and create in the side panel (#38303) #38305) → feat(edit-content): side-panel opener foundation, and Content Drive opens content through it (#38303) #38324 (layers 1 and 2) → this (fmontes/38303-5-admin-surfaces) → feat(query-tool): the Query Tool opens content in the side panel; flag and dialog-era code removed; e2e specs (#38303) #38345 (layers 6 and 7). The intermediate layer branches still exist and anchor the compare links above; their PRs (feat(uve): open every editor and legacy screen through the side panel (#38303) #38330, feat(edit-content): fields ask their host to open content; block editors stop navigating (#38303) #38332) are closed.
• Variant: the UVE sends variantName on every request; the backend's by-action fire endpoint does not read it yet, so the first edit of still-DEFAULT content from an experiment page needs a backend follow-up.

🤖 Generated with Claude Code

[claude]

Claude finished @fmontes's task in 2m 41s —— View job

---

Code Review

• Gather context (diff vs origin/fmontes/38303-4-fields-block-editors)
• Review all 10 changed files
• Post review feedback

This PR is test + spec only — the 9 .spec.ts files are the TDD Red tests for US6/R11/R12 (commit 4a7e19ef), plus one contract doc update (edit-request.md) that adds portletId to every admin surface's edit request. There is no production code in the diff, so there's no runtime behavior to break here; I checked the tests for internal correctness, broken references, and whether they faithfully encode the contract.

New Issues

No issues found.

Notes (non-blocking)

• Red tests reference not-yet-existing production — by design. dot-edit-content-deep-link.component.spec.ts imports ./dot-edit-content-deep-link.component, which doesn't exist yet, and several specs probe for a not-yet-added setInjector(...) on DotCustomEventHandlerService / MainComponentLegacyComponent. These are guarded with typeof ... === 'function' assertions (main-legacy.component.spec.ts:244, dot-custom-event-handler.service.spec.ts), so they fail at a meaningful assertion rather than silently — correct for Red. The missing component file will make that spec fail at module resolution rather than at an assertion; that's acceptable for TDD but worth confirming the implementation PR adds it.
• Cross-boundary imports are correctly annotated — the @nx/enforce-module-boundaries disables in the deep-link and page-actions specs carry a reason (opener is the dependency + lazy-loaded routes). Good.
• Verified consistency I was worried about:
  • PORTLET_ID = '123-567' in dot-custom-event-handler.service.spec.ts matches MockDotRouterService.currentPortlet.id (dot-router-service.mock.ts:11) ✅
  • DotEditContentRequest / DotEditContentOpenOptions / DotEditContentOpenerService are all exported from the @dotcms/edit-content barrel (index.ts:12,15,16 — the model via export *), so the dotcms-ui specs resolve them ✅
  • The vi.mock('@dotcms/edit-content', …) stubs in the two publishing-queue dialog specs only replace the opener class (to dodge the TipTap/y-protocols/awareness inline-deps issue); DotSidePanelRef still comes from the real @dotcms/ui, and the erased type imports are fine ✅
• Spec/test alignment — the edit-request.md additions (portletId: 'pages', 'publishing-queue', legacy "portlet current when the event arrived", deep-link "parent portlet" + folderInode from ?folder) line up exactly with the assertions across the specs. The one surface that intentionally opens without a content type (edit-host → opener fetches it) is asserted at dot-custom-event-handler.service.spec.ts (request.contentType undefined), consistent with the contract row.

Looks good to proceed to the implementation (Green) phase.

· fmontes/38303-5-admin-surfaces

[fmontes]

Green half landed and the PR is out of draft. Nine commits on top of the Red tests (4c6ff75f65 … 8e4ea6c842), reviewed against the approved specs and the contracts.

What I checked on my own run, from core-web/:

Check	Result
pnpm nx test dotcms-ui	2323 passed, 0 failed (223 files)
pnpm nx test portlets-dot-publishing-queue-portlet	290 passed, 0 failed
pnpm nx test edit-content	2603 passed, 0 failed
pnpm nx test edit-content-bridge	148 passed, 0 failed
affected lint	0 errors across 38 projects, no new eslint-disable outside specs
strict gate vs origin/fmontes/38303-4-fields-block-editors	PASS
pnpm nx build dotcms-ui	OK, main +209 bytes against the layer 4 build

Three things worth a reviewer's eye:

1. How the admin app reaches the opener. apps/dotcms-ui/src/app/api/util/open-edit-content-panel.ts awaits import('@dotcms/edit-content') and resolves the opener from the caller's injector, skipping the open if that injector's DestroyRef is already destroyed. This keeps the lazy boundary the routes already have, which the Nx module-boundary rule enforces, and keeps the library out of the main bundle. Decision taken at gate T088.
2. The form bridge now pops by frame. pushFormBridge() returns a frame and popFormBridge(frame) removes exactly that frame, so a panel closed under a stacked one no longer kills the stacked panel's bridge. Four new cases in angular-form-bridge.spec.ts exercise the real stack. Decision taken at gate T098.
3. The deep-link component closes its panel without navigating when the route is left, so leaving through the menu does not bounce back to the parent portlet. One case added for a route left before the editor finished loading.

History note: 1a062cdb9c and 41086d6499 carry folder deletions that were staged alongside their own change, and 41086d6499 does not compile on its own (the references go in 1907e4e127). The final tree is what the checks ran on; squash-merge makes it moot.

Manual passes still owed on :4200: quickstart US2 steps 3–6 (JSP portlets, deep links, Pages) and US6 with the six-point Publishing Queue checklist, where focus inside the editor iframe over the modal and Escape inside it are the agreed manual-only checks.

[fmontes]

Two commits added at the top of this layer, both housekeeping: a formatting-only style(...) commit for files this and lower layers committed with line wrapping CI's nx format:check rejects (the lower layers got the same treatment, one commit each), and fix(dotcms-ui): type the Pages store state so getPages is a known method, which fixes the state type that had collapsed the Pages store to an index signature; without it the whole-stack strict gate failed on the plain getPages() call. No behavior change.