# Marketplace Project Editor - Sprint 13 (updated Sprint 18)

## Scope

Sprint 13 introduces first version of Marketplace Project Editor.

Editor only manages bootstrap configuration.

Out of scope:
- products
- categories
- orders
- analytics management

## Architecture

```text
src/app/features/project-editor/
  pages/
  components/
  sections/
  models/
  services/
  facade/
```

Editor uses strongly typed bootstrap models and isolated feature state.
Components do not call APIs directly.

## Facade API

`ProjectEditorFacade` exposes:
- `loadBootstrap()`
- `updateBootstrap()`
- `exportBootstrap()`
- `importBootstrap()`
- `preview()`

Current source of truth:
- existing bootstrap provider via `ConfigService`

Future backend integration:
- replace local import/export/save flow with builder endpoints
- keep component contracts unchanged

## Supported Sections

- General
  - marketplace name
  - domain
  - description
  - default language
  - supported languages
- Branding
  - logo
  - small logo
  - favicon
  - marketplace title
- Theme
  - palette tokens via color pickers
- Header
  - logo/search/categories/languages/cart/profile/wishlist/compare/region toggles
- Footer
  - company information
  - address
  - phone
  - email
  - copyright
  - payment icons
  - social links
  - static pages
- Homepage
  - home page sections list
  - visibility
  - order
  - layout
  - drag-and-drop ordering
- Widgets
  - homepage widget configuration
  - typed shortcuts for hero/categories/product collection widgets
  - JSON fallback for other widget props
- Marketplace Features
  - feature flags
  - search suggestions/history
  - recently viewed
  - reviews/questions/recommendations
- Preview
  - export JSON
  - import JSON
  - runtime preview without full browser refresh
- Languages (Sprint 16)
  - add/remove supported locale
  - set default locale
  - generically syncs translation keys across static pages and navigation labels (`LocaleSyncService`)
- Navigation (Sprint 16)
  - header navigation: add/remove/reorder/edit label/URL/visibility
  - flat footer navigation: same actions
  - grouped footer navigation (column-based) is read-only in this tab for now

## Preview Strategy

Preview updates bootstrap snapshot in memory and re-applies:
- theme
- branding
- runtime state
- route-based page rendering on next Angular navigation

This keeps bootstrap engine intact while enabling fast local preview.

## Widget Configuration

Current widget editor supports explicit fields for:
- Hero
  - layout
  - height
  - overlay
  - autoplay
- Categories
  - layout
  - columns
- Product Collection
  - layout
  - cards per row
  - filters
  - badges
  - rating
  - price

Other widgets use JSON props fallback until dedicated editors are added.

## Draft / Publish (Sprint 16, autosave added Sprint 18)

There is still no backend draft/publish API. This sprint models it client-side
in `ProjectEditorFacade`:
- `status: 'draft' | 'published'` and `dirty` (diffed against the
  last-saved snapshot) live in facade state.
- `save()` snapshots the current in-memory bootstrap as "last saved" and
  timestamps it (`lastSavedAt`).
- `publish()` runs `ProjectValidator`, and if there are no issues, applies
  the bootstrap via `PlatformRuntimeService.reloadFromBootstrap`, marks
  status `published`, and becomes the new `originalBootstrap` snapshot used
  by reset.

**Backend gap, not yet implemented:** real persistence needs
`PUT /builder/bootstrap/draft` and `POST /builder/bootstrap/publish`
endpoints so drafts/publishes survive a reload and are shared across editors.

### Autosave (Sprint 18)

`ProjectEditorDraftStorageService` (`services/project-editor-draft-storage.service.ts`)
persists the full bootstrap draft to `localStorage` (key
`projectEditor.draftBootstrap.v1`, scoped by `tenant.id`) on every
`updateBootstrap()` call, `save()`, and `publish()`. On `loadBootstrap()`, if a
stored draft exists for the same tenant it is loaded instead of the
freshly-fetched bootstrap and `draftRestored` is set true (surfaced in the
save bar as a dismissible notice). The published/loaded bootstrap is never
overwritten automatically — only explicit `publish()` calls change what the
runtime actually serves; the localStorage draft is a separate, purely local
concern that survives refreshes and browser restarts.

Status indicators in `ProjectEditorSaveBarComponent`:
- **Unsaved changes** - shown while `dirty()` is true.
- **Last saved: HH:MM:SS** - shown once not dirty and `lastSavedAt` is set.
- **Draft restored** banner - shown once after a local draft is loaded from
  a previous session, dismissible.

### Reset (Sprint 18)

- **Reset section** - button above the active section (shown only for
  sections with a bootstrap-key mapping in `EDITOR_SECTION_BOOTSTRAP_KEYS`,
  `models/project-editor.model.ts`). Reverts that section's bootstrap keys
  to `originalBootstrap` (the last loaded/published snapshot). Confirmation
  required.
- **Reset draft** - button in the save bar. Reverts the entire bootstrap to
  `originalBootstrap` and clears the persisted local draft. Confirmation
  required.
- Per-field reset is **not implemented** - the bootstrap schema has no
  registry of per-field defaults, so only section- and project-level reset
  exist. Adding field-level reset would require either a default-value
  registry per field or storing per-field undo history; deferred.

## Validation

`ProjectValidator` (`services/project-validator.service.ts`) runs on every
render of the save bar: missing logo, no languages, invalid marketplace URL,
duplicate static-page slugs, empty homepage, a homepage widget with no
`type`, duplicate header navigation links, invalid theme colors. Publish is
blocked while any issue is present; Save is not.

## Rich HTML editing

Static page HTML is edited via `MarketplaceHtmlEditorComponent`
(`components/html-editor/`), a `contentEditable` + toolbar component with no
external dependency. It emits raw HTML on every change and never sanitizes —
sanitization remains a storefront-render concern.

## QR Login Reuse (Sprint 18, corrected)

There is exactly **one** Telegram QR/session backend
(`{authApiUrl}/users/sessions`) and exactly **one** QR login component/UI.
Nothing about the QR flow is duplicated for admin:

- `TelegramSessionApiService` (`services/telegram-session-api.service.ts`) is
  the single place that calls `POST/GET/DELETE {authApiUrl}/users/sessions...`
  and normalizes the response into `AuthSession`. It holds no state and
  writes no cookies - it's a pure API wrapper.
- `QrLoginEngine<TSession>` (`shared/qr-login/qr-login.engine.ts`) is the
  QR/polling/expiry/"return from Telegram app" state machine (extracted from
  the original `TelegramLoginComponent`), driven by a small
  `QrLoginAdapter<TSession>` (`shared/qr-login/qr-login.model.ts`).
- `TelegramLoginComponent` (`components/telegram-login/`) is **the same
  component for both customer and admin login** - not two components. It
  takes a `mode: 'customer' | 'admin'` input; `ngOnInit` picks
  `AuthService` or `AdminAuthService` accordingly and builds the
  `QrLoginAdapter` from whichever one, but the QR image, polling loop,
  timeouts, and dialog markup are identical either way. Customer usage is
  unchanged (`<app-telegram-login />` on the cart page, `mode` defaults to
  `'customer'`); admin usage is `<app-telegram-login mode="admin" />`,
  mounted once globally in `app.html`.

An earlier version of this sprint's work built a separate
`AdminAuthService`/`AdminLoginComponent` pair that called its own
`adminAuthApiUrl` placeholder endpoint. That was wrong: there is no separate
admin backend, and inventing one client-side would have meant testing against
an endpoint that doesn't exist. It was replaced with the shared-API approach
described above.

## Admin Authentication (Sprint 18, corrected)

Only the **storage** is separate between customer and admin - the QR/session
API and UI component are shared (see above), by design, since one Telegram
QR/session backend serves both. What stays separate is everything needed so
that scanning the admin QR can never authenticate the customer session (or
vice versa):

| | Customer (`AuthService`) | Admin (`AdminAuthService`, `core/admin-auth/`) |
|---|---|---|
| Cookie | `webSessionID` (`SameSite=Lax`) | `adminSessionID` (`SameSite=Strict`) |
| Token storage | `web_session_id` (localStorage, anonymous API attribution only, unrelated to auth) | `adminToken` / `adminRefreshToken` (localStorage, reserved for a future JWT pair - unused today) |
| Signals | `session`, `status`, `showLoginDialog` on `AuthService` | `session`, `status`, `showLoginDialog` on `AdminAuthService` |
| Guard | none yet for customer routes | `adminAuthGuard` (`core/admin-auth/admin-auth.guard.ts`) |
| Interceptor | `apiHeadersInterceptor` | `adminAuthHeadersInterceptor` (`core/admin-auth/admin-auth-headers.interceptor.ts`), self-guards on `/admin/` in the request URL, sets `AdminWebSessionID` + `Authorization: Bearer <adminToken>` when present |
| Session/QR API | `TelegramSessionApiService` | same `TelegramSessionApiService` instance/endpoint |
| Login UI | `TelegramLoginComponent` (`mode="customer"`, default) | same `TelegramLoginComponent` (`mode="admin"`) |

**Backend gap this creates, and why it matters:** because admin login goes
through the exact same Telegram session API as customer login, the backend
has **no concept of "this is an admin session"** at the point the QR is
scanned - it's just a regular Telegram user session, identical in shape to a
customer's. The frontend only decides *where to store* the resulting session
id (admin cookie vs. customer cookie); it cannot and does not decide whether
that Telegram user is actually allowed to act as an admin. **Real admin
authorization must be enforced server-side**, at the point admin API calls
are made with the `AdminWebSessionID` header - the backend must check the
authenticated user against an admin/role list and reject non-admins, since
nothing on the frontend prevents any Telegram user from completing the QR
flow while `mode="admin"` is showing. This needs a backend decision (role
check keyed off the session id, or a dedicated admin-scoped token issuance)
before admin login can be considered secure, not just "separate storage."

### Login test mode

`?login=true` and `?adminLogin=true` query params (handled once in
`App.ngOnInit` via `openLoginDialogsFromTestModeQueryParams()`, `app.ts`) call
`AuthService.requestLogin()` / `AdminAuthService.requestLogin()` respectively,
for manual testing. This only sets the same signal a normal "please log in"
action would set - it does not bypass authentication or change any other
behavior, so it is safe in all environments. `TelegramLoginComponent` in
customer mode is currently mounted only on the cart page, so `?login=true`
only shows a dialog there; the admin-mode instance is mounted globally so
`?adminLogin=true` works from any route.

### Ed25519 prep

`core/admin-auth/ed25519-verification.model.ts` defines
`Ed25519VerificationService` (abstract, injectable) with
`requestChallenge()` / `verify(signedResponse)` and the
`Ed25519Challenge` / `Ed25519SignedResponse` / `Ed25519VerificationResult`
shapes (nonce, timestamp, payload, public key, signature). No crypto is
implemented. The current DI binding,
`NoopEd25519VerificationService` (registered in `app.config.ts`), fails
closed (throws) rather than silently accepting anything, so it's safe to wire
into a real login path today - it will error loudly instead of pretending to
verify a signature. Swap the DI binding for a real implementation once the
backend ships challenge/verify endpoints; nothing else needs to change.

## Known gaps / deferred (Sprint 18)

Full field-by-field coverage of every supported bootstrap property (with
bilingual EN/RU labels, description, and validation state per field) was not
completed in this pass - the bootstrap schema is large (theme typography/
spacing/shadows, full company address, per-locale footer copyright, grouped
footer navigation editing, sidebar navigation, per-page SEO map, catalog/
product-page/user-experience sub-fields, permissions, API endpoints) and
several concepts named in the sprint brief (payments, delivery/shipping,
checkout, unified search config) have **no corresponding model in
`shared/models/config` at all** - they would need new bootstrap schema before
an editor could expose them. See the section-by-section gap list gathered
during Sprint 18 investigation for the full inventory; treat as a follow-up
sprint rather than something silently skipped.

## Constraints

- runtime bootstrap engine not replaced
- configuration stays source of truth
- no hardcoded marketplace values
- no business domain management mixed into editor