# Implementation Status

Last updated: 2026-09-09

## Resume Here

Before changing code, read this document, `AGENTS.md`, and the linked source-of-truth documentation. Then confirm the current baseline with:

```sh
npm test
```

For persistence or Redis-backed changes, also run the isolated integration suite after setting `TEST_DATABASE_URL` to the `*_test` database and `TEST_REDIS_URL` to the test Redis instance:

```sh
TEST_DATABASE_URL='postgresql://...' TEST_REDIS_URL='redis://...' npm run test:integration
```

The current local container runtime is OrbStack. If Docker commands point at a stale host, run:

```sh
unset DOCKER_HOST
docker context use orbstack
```

## Completed

### Workspace And Database

- Npm workspace with the API and shared contracts/validation packages.
- PostgreSQL Compose service on local port `5433`; API port `3000` is registered in `.local/project-ports.json`.
- Redis Compose service on local port `6380` supports distributed login throttling.
- Drizzle schema and migrations:
  - `0000_core_inventory`: users, sellers, categories, condition profiles, conditions, products, and physical inventory items.
  - `0001_auth_sessions`: hashed server-side sessions, expiry, revocation, IP address, and user agent.
  - `0002_session_csrf_tokens`: session-bound hashed CSRF tokens.
  - `0003_security_audit_events`: private append-only authentication security events.
  - `0004_password_reset_tokens`: hashed, one-time password recovery tokens.
  - `0005_mfa_methods`: protected TOTP method records and lifecycle state.
  - `0006_listings`: listing records and lifecycle enum.
  - `0007_listing_image_count`: persisted image prerequisite for publication.
  - `0008_inventory_images`: item-owned image metadata and processing state.
  - `0009_remove_listing_image_count`: publication no longer stores a listing-level image snapshot.
- Isolated integration tests require a database name ending in `_test` and truncate only that database's tables.
  - Local-only `npm run provision:admin` creates the first Admin User and linked Seller atomically. It requires `INITIAL_ADMIN_EMAIL`, `INITIAL_ADMIN_PASSWORD`, and `INITIAL_SELLER_NAME`, hashes the password with scrypt, and refuses once any User exists.

### Domain And Persistence

- Product identity requires name, category, and seller.
- Products receive seller-scoped, normalized slugs; duplicate names receive a numeric suffix.
- Each Inventory Item is a separate physical copy.
- Inventory condition must belong to the Product category.
- Inventory lifecycle supports `ACTIVE`, `SOLD`, and `ARCHIVED`; sold and archived items are locked against ordinary edits.
- Repositories for products, inventory items, conditions, users, and sessions use Drizzle/PostgreSQL.
- Listing persistence supports draft creation, seller-scoped lookup, status updates, and inventory-item history.

### API And Authentication

- Fastify app bootstrap and `GET /health` response.
- Admin Product endpoints:
  - `GET /api/v1/admin/products`
  - `POST /api/v1/admin/products`
- Admin Inventory endpoint:
  - `POST /api/v1/admin/inventory`
- Authentication endpoints:
  - `POST /api/v1/auth/login`
  - `POST /api/v1/auth/logout`
  - `GET /api/v1/auth/me`
- Login uses asynchronous scrypt password verification. Session tokens are random, stored only as SHA-256 hashes, and never returned in JSON.
- Session cookie settings are `HttpOnly`, `SameSite=Strict`, path `/`, and `Secure` in production.
- Login is limited to five attempts per client IP every 15 minutes and returns `429 AUTH_RATE_LIMITED` with `Retry-After` once exhausted. Redis atomically coordinates the limit across API processes.
- Login sets a separate same-site CSRF cookie. Authenticated state-changing routes require its matching `x-csrf-token` header, validated against the session-bound hash.
- `GET /api/v1/auth/sessions` returns safe metadata for the current user's active sessions and marks the current session. `DELETE /api/v1/auth/sessions/:id` revokes an owned session; `DELETE /api/v1/auth/sessions` revokes all other sessions while retaining the current one. Revoking the current session clears both cookies.
- Active sessions renew `last_activity_at` after five minutes, avoiding a database write on every authenticated request.
- Authentication events are retained separately from inventory activity. `LOGIN_SUCCESS`, `LOGIN_FAILURE`, `SESSION_CREATED`, and `SESSION_REVOKED` include the known actor where available, plus safe request context. Passwords and session/CSRF tokens or hashes are never recorded.
- Password recovery uses generic `POST /api/v1/auth/forgot-password` responses to prevent account enumeration. Opaque tokens are delivered only through the SMTP adapter, stored only as hashes, expire after one hour, are consumed once, and revoke existing sessions after a password reset.
- API startup requires `SMTP_URL`, `SMTP_FROM`, and `PASSWORD_RESET_URL`; configure these local values before starting the server.
- Existing admin routes validate the authenticated session and derive seller ownership server-side.
- Request schemas live in `packages/validation`; admin response DTOs live in `packages/contracts`.
- MFA enrollment, verification, disable, protected secret handling, login verification, and audit events are implemented.
- Admin listing creation and lifecycle routes are implemented:
  - `POST /api/v1/admin/listings`
  - `GET /api/v1/admin/listings`
  - `GET /api/v1/admin/listings/:id`
  - `POST /api/v1/admin/listings/:id/publish`
  - `POST /api/v1/admin/listings/:id/unpublish`
  - `POST /api/v1/admin/listings/:id/pending`
  - `POST /api/v1/admin/listings/:id/archive`
- Listing publication is seller-scoped, CSRF-protected, and gated by persisted identity, price, physical inventory, condition, and image prerequisites.
- Listing publication rejects physical inventory in `SOLD` or `ARCHIVED` status with a conflict response.
- Listing reads are seller-scoped and return only admin listing DTO fields.
- Item-owned image metadata can be added through a seller-scoped, CSRF-protected admin route; publication uses the persisted image count.
- Public search is available at `GET /api/v1/public/search` through a normalized SearchService boundary with pagination, price filters, text matching, and sold-item visibility controls.
- Public search also supports active category and condition lookup options, URL-backed filters, and allowlisted recent/price sorting.
- Public search responses include `items` plus `pagination` metadata (`page`, `limit`, `total`, and `totalPages`); pending listings are visible by default and sold listings require an explicit filter.
- Public product pages are available at `GET /api/v1/public/products/:slug`; responses use a public DTO and require an active physical item with an available or pending listing.
- Public listing collection and detail reads are available at `GET /api/v1/public/listings` and `GET /api/v1/public/listings/:id`; collection responses include pagination metadata and detail reads hide internal inventory identity.
- Public listing detail responses include safe condition labels, latest market value, and public image URLs; unavailable image states are rendered accessibly in the gallery.
- Public inventory collection and detail reads are available at `GET /api/v1/public/inventory` and `GET /api/v1/public/inventory/:id`; responses join only active physical items to available or pending listings and exclude private item notes.
- The public seller directory is available at `GET /api/v1/public/sellers`; it returns only sellers with visible inventory and exposes display identity without user or contact fields.
- The web workspace now has a Vue/Vite public catalog shell with Vue Router, a catalog API service, search composable, loading/error/empty states, and responsive search-first styling.
- The web catalog now provides keyboard-accessible listing cards with stable `/listings/:id` links, a public listing detail page, and service-level detail loading/error handling.
- Catalog search and pagination are synchronized to the browser URL (`q` and `page`) with accessible previous/next controls; the web port `5173` is declared in the project port registry.
- The web catalog includes a stable `/products/:slug` product detail route with API-backed loading, not-found/error states, and semantic product facts.
- The visual design foundation is centralized in web tokens with light/dark/system runtime theme support; the fixture composition has moved to Storybook at port `6006` with the accessibility addon and component stories for available, pending, and sold states.
- `ImageStorage`, traversal-safe local storage, multipart upload transport, binary signature validation, and worker-backed derivative processing are implemented.
- Derivative orchestration now defines thumbnail, medium, and large WebP storage outputs behind an injected renderer; the concrete worker-backed renderer remains.
- The worker owns an image derivative job runner with explicit processing, completion, and failure outcomes backed by persistent PostgreSQL job records.
- Image metadata limits are enforced in the service boundary: supported MIME types, 10 MiB default file size, and 10,000-pixel maximum dimensions.
- Acquisition, sales, dashboard, pricing observations, market valuations, audited sale voiding, and manual market-value overrides are implemented with seller-scoped API and admin UI workflows.
- Seller-scoped bulk archive is implemented through the CSRF-protected `POST /api/v1/admin/bulk` route; it validates active inventory lifecycle transitions and never substitutes for the dedicated sale workflow.

### Verified Baseline

| Suite | Current coverage |
| --- | --- |
| Domain | Product requirements and inventory lifecycle rules |
| Application | Product slug allocation, condition compatibility, password verification/recovery, session issuance, activity renewal, management, and audit events |
| HTTP | Product and inventory creation, validation, authentication/recovery, CSRF, rate limiting, and session management |
| Integration | PostgreSQL catalog/session/audit/password recovery persistence and Redis-shared login throttling |

Verified locally: `npm test` passes 163 tests with no failures (22 domain, 78 application, and 63 HTTP); worker tests, web tests, web build, Storybook build, Storybook interaction/accessibility checks, integration tests, and authenticated E2E workflows also pass when their required local services and credentials are available.

## Next Milestone

Complete production configuration and operational maturity around the implemented vertical slices.

1. Configure and validate an approved external pricing provider and provider-specific mapping data.
2. Expand broader accessibility/manual coverage and address any remaining integration edge cases.

Acceptance criteria: production provider credentials are secret-managed, migrations and backups are exercised in the target environment, and CI runs integration, authenticated E2E, web, build, Storybook build, and Storybook interaction/accessibility gates.

CI also runs the worker suite, production dependency audit, and deployment/restore script contract checks.

## Overall Roadmap

Work through one bounded vertical slice at a time. The linked documentation remains authoritative when this roadmap is less detailed.

Completed vertical slices include authentication, listings and image processing, acquisitions, sales and audited corrections, dashboard metrics/alerts, pricing observations and quality handling, provider mappings, durable pricing refresh jobs, public search/detail/image states, bulk operations, exports, settings, Saved Views, admin workflows, CI integration/E2E gates, production Docker packaging, rollback/health checks, PostgreSQL backup verification, restore drill scripts, image-storage restore support, and deployment smoke checks.

Remaining roadmap areas are approved external provider credentials/adapters, broader accessibility/manual coverage, and deployment-specific NAS/TLS validation.

## Known Gaps

- No initial User/Seller is provisioned in the development database yet. Run `npm run provision:admin` after setting its required local `.env` variables.
- Listings are partially implemented: draft creation, seller-scoped reads, publication gating, lifecycle writes, archive behavior, sold/archived inventory checks, and item-owned image metadata are complete; storage-backed processing remains.
- The provider abstraction, fixture provider, HTTP provider adapter, Product mapping workflow, durable refresh queue, worker execution, polling contract, retry policy, and completed-job purge operation are implemented. Production readiness still requires approved external provider credentials and provider-specific mapping data.
- API and worker production startup now reject fixture pricing, missing provider endpoints, and non-HTTPS provider endpoints before opening runtime connections.
- Pricing refresh requests now enqueue seller-scoped jobs and expose a safe `GET /api/v1/admin/jobs/:id` polling contract instead of doing provider work inline.
- Product-level provider mappings now persist seller-scoped `PRODUCT`/`SEARCH` relationships and support active listing and controlled disabling.
- Pricing refresh enqueue now resolves its provider from an active Product mapping and rejects unmapped Products explicitly.
- Local development and CI can enable a deterministic fixture provider for end-to-end refresh coverage; production provider adapters remain explicitly configured deployment boundaries.
- Pricing refresh jobs have a bounded three-attempt retry policy with exponential backoff. The worker schedules completed-job cleanup using configurable retention days and cleanup interval settings.
- Bulk archive, asynchronous bulk pricing refresh, and Product-level bulk tag updates are implemented through the seller-scoped API boundary. Inventory `SOLD` remains intentionally exclusive to the audited sale workflow.
- Admin inventory JSON export is implemented at `GET /api/v1/admin/exports/inventory.json` and serializes seller-scoped admin DTOs as a downloadable attachment.
- Admin inventory CSV export is implemented at `GET /api/v1/admin/exports/inventory.csv` with escaped spreadsheet-safe fields and seller-scoped admin DTOs.
- Admin inventory Excel-compatible export is implemented at `GET /api/v1/admin/exports/inventory.xls` using the XML workbook format and seller-scoped admin DTOs.
- Seller-scoped admin settings are implemented through `GET/PATCH /api/v1/admin/settings` with strict category allowlists and server-side partial-update merging.
- Seller-scoped saved-view CRUD is implemented through the documented admin API with validated filter maps and repository-level ownership checks.
- Large inventory JSON exports can be enqueued through `POST /api/v1/admin/exports/inventory.json/jobs`, polled through `GET /api/v1/admin/exports/jobs/:id`, and downloaded after completion; the worker owns claim, completion, and failure processing.
- Completed asynchronous inventory JSON exports can be downloaded through `GET /api/v1/admin/exports/jobs/:id/download`.
- PostgreSQL restore verification and image snapshot restore scripts are implemented and covered by shell tests as part of the deployment drill workflow.
- `npm audit --omit=dev` currently reports zero production vulnerabilities. Re-run the audit after dependency changes and review development-dependency findings separately before production releases.
- Authenticated E2E coverage now verifies both the core catalog workflow and the Settings, Saved Views, and export-controls workflow.
- Production dependency audit currently reports zero vulnerabilities with `npm audit --omit=dev`.
- Public catalog sorting now supports recently added, recently updated, name A-Z, name Z-A, price low-to-high, and price high-to-low through a URL-synchronized atomic sort control. Broader inventory filter parity and market-value sorting remain.
- Public catalog filtering now supports URL-synchronized set, event, and tag text filters through a reusable atomic text-filter control and the shared search API.

## Local Commands

```sh
npm install
npm run db:up
npm run db:migrate
npm run provision:admin
npm test
```

The Compose database requires a local `.env` created from `.env.example`. To prepare the test database once after bringing PostgreSQL up:

```sh
docker compose exec -T postgres sh -c 'createdb -U "$POSTGRES_USER" pokemon_inventory_test'
```

Then migrate it using its `TEST_DATABASE_URL` value as `DATABASE_URL`, and run integration tests with both `TEST_DATABASE_URL` and `TEST_REDIS_URL` set. See [README.md](../README.md) for the environment-file prerequisites.

## Suggested Resume Prompt

`Read docs/IMPLEMENTATION-STATUS.md and continue the next bounded milestone. Verify the existing baseline first, follow AGENTS.md, and keep the change focused with tests.`