Changelog

The Svelte + Convex profile now matches the current template-svelte-convex-starter baseline: Bun 1.4.2 / Node 22.23.2, Svelte 5.57.0, SvelteKit 2.70.3, TypeScript 6.0.3, Vite 8.2.2, Tailwind 4.3.3, shadcn-svelte 1.5.1, Convex 1.45.0 and convex-svelte 0.14.0.

A phone was still getting a sheet-shaped copy of the desktop sidebar: shadcn's Sidebar renders its list inside a Sheet under md, and the site navbar and docs shell each had their own list-in-a-sheet. That is three different mobile menus with the same shape and none of them thumb-friendly. dashboard-shell 1.2.0 replaces all of it with one component — MobileMenuDrawer: a bottom drawer of thumbnail TILES (icon in a rounded square + label, 4 per row), grouped exactly like the rail, with one level of drill-down for items that have sub-items and a back chevron. The rail itself no longer renders at all under md (DashboardSidebar returns null when shadcn's own useSidebar() reports mobile), so there is no second copy to keep in sync: the dock's Menu tile and the topbar button both open the drawer. The docs shell and the marketing navbar now mount the same drawer — docs clusters and slice categories became tiles, and tapping a category drills into its slices.

Third and last pass of the dashboard merge. A sweep of every slice that renders its own persistent left rail turned up six; five of them are NOT app navigation and were deliberately left alone — user-management's Roles/Teams rails list adapter DATA rows (master-detail, and they already sit inside horizontal tabs), image-editor's rail is a canvas tool palette, reel-editor's files pane is a content browser and its track-head column is one row per timeline track. Only settings qualified: a hard-coded four-section catalog rendered as a second rail. It now follows the admin console's pattern — `SettingsShell nav={false}` renders the active panel only, and `settingsSectionsToNav(onSelect, { activeId, dockCount })` hands the sections to the app sidebar as plain data (structural shape, no cross-slice import). The section catalog moved to lib/nav.ts as the SSOT for both the built-in rail and the mapper, with the old `SettingsSectionId` export path kept alive. Default stays `nav: true`, so every existing consumer renders byte-identically.

Wave 1 built the one dashboard but left the surfaces that actually render on their own hand-rolled chrome — so nothing looked different. This wave migrates them. rr's own /admin operator panel drops its bespoke SidebarProvider + AppSidebar + NavMain (382 lines) and mounts the slice: same rail and breadcrumb topbar on desktop, and on mobile it now gets the bottom dock (Overview / Layouts / Registry / Menu) it never had. The three preview routes that faked a dashboard with `grid-cols-[220px_1fr]` divs now mount the real thing, so what the catalog shows is what `npx rr add` gives you. The admin console gains `nav={false}` + a `sectionsToNav()` mapper: its 26 sections feed the ONE app sidebar instead of stacking a second rail beside it — one navigation, one dock, on every screen. Three orphan sidebars (shared/layout/AppSidebar, site-sidebar and its two group files) had no importers left and are deleted.

The repo carried six overlapping dashboard chromes: a facade slice pointing at superspace-coupled code under template-base (tsc-EXCLUDED, so it never compiled — and it advertised a `sidebar`/`topbar` API the component never had, meaning `npx rr add dashboard-shell` shipped consumers a broken import), a mock desktop/mobile pair that existed only so the preview had something renderable, appshell's single-pane cockpit shell, and the template `_shared` admin chrome in two archetypes (simple + advanced). This wave replaces all of them with one real, in-tree, typechecked slice. ONE `nav` prop is the SSOT: the desktop rail, the mobile sheet and the bottom dock all render from it — flag `dock: true` on the items you want docked, or let it take the first few. Breakpoints are CSS (`md:hidden`) plus shadcn's Sidebar sheet, so the slice never measures the viewport and there's no hydration flash. Slots cover what used to be forked components: sidebarHeader (workspace switcher), sidebarFooter (user menu), actions, topbar (full replace), secondary (the old three-column 'advanced' archetype). Net −3.4k lines.

The site had cacheComponents (PPR) on but barely used it. This wave closes the gaps a ponytail/perf audit surfaced: (1) a route-level loading.tsx paints an instant skeleton for all 63 /preview/** routes (each ships a 0.4–1.1MB slice chunk) instead of blank, plus a preview-shaped skeleton for /tour. (2) recharts (~150kB) is code-split out of the markdown slice — it only loads when a chart fence actually renders, not on every markdown page (ChartBlock now next/dynamic's a ChartCanvas chunk, mirroring the already-lazy MermaidBlock). (3) the slice-files fs walkers get cacheLife('max') + cacheTag('slice-files') so immutable repo files cache for the deploy's life instead of a 5–15min revalidate. (4) 25 zero-interactivity preview pages + the docs footer + code-tab dropped 'use client', leaving the client graph so they PPR-prerender.

New variant mechanism: a slice can declare `variants` in slice.json. `npx rr add <slug> <variant>` installs one variant (its files flatten to the slice root, so imports resolve at @/features/<slug> like any slice); `npx rr add <slug>` installs all variants + a root switcher and you pick via a prop — exactly the shadcn pattern. An optional `shared/` folder is copied for every variant, and each variant can declare its own `convex` roots so it pulls only its own backend. Purely additive: the ~60 non-variant slices install exactly as before, and every old slug keeps working via an alias. Used to collapse the confusingly-named slice families into fewer, clearer slices.

The ideal admin panel lands as a portable composition slice (admin-console): a gated two-column shell over a 26-section registry harvested from ~15 project admin panels, mounting existing rr slices by id via a consumer components map (no slice→slice imports), plus 5 net-new gap sections nothing else covered — Analytics, Audit-log viewer, Navigation config, Leads/CRM inbox, SEO health — each adapter-driven with in-memory mocks so the whole console runs backend-free. Access is injected; gate logic is pure + unit-tested. Convex copy-source ships ac_leads + ac_nav_items with requireAdmin-gated CRUD. Alongside it, the /slices catalog gains a clearer toolbar — a Sort dropdown (Featured / Recently updated / Name A–Z / Version) and a labelled tag-filter row — and rahman-resources publishes 1.15.0 so `npx rr add admin-console` resolves.

Ported from the rahmanef-com web-OS as a brand-free, self-contained slice. A tiny web-page studio: type HTML / CSS / JS and watch it render live in a sandboxed iframe (srcdoc + sandbox=allow-scripts WITHOUT allow-same-origin → opaque origin, the security boundary), then Save to a shareable /p/<slug>. Keeps the Code / Split / Preview view toggle, device-width preview (responsive / tablet / phone), a saved-pages rail, and public/private visibility. The Convex page store + auth/write-key/session bits were dropped in favour of an injected HtmlStudioAdapter (save/load/list/remove); unwired it runs on an in-memory mock so the editor + live preview + saved list are interactive with zero backend.

Ported from the rahmanef-com web-OS as a brand-free, self-contained slice. An owner-gated admin for a curated icon-launcher: add / edit / remove / reorder links (label, lucide icon NAME, url, group, order) that open in a new tab. The backend is injected via a ResourcesAdapter (list/upsert/remove/canManage); unwired it runs on an in-memory mock so the whole CRUD — including reorder — is interactive with zero backend. The Convex resources.* mutations and the auth/sign-in gating were replaced by the adapter plus a simple canManage flag, and the launcher icon map was brand-stripped to generic lucide NAMEs.