# Roadmap: SummerCMS (Go)

## Overview

v1 ports the Płytarium (fonoteka) headless PHP backend to a single Go binary without vue-fonoteka-app or fonoteka-mcp noticing. The journey starts with the smallest possible kernel plus a day-one parity harness, proves the whole stack on one real endpoint (`GET /_fonoteka/api/v1/genres`) before any further kernel design, then broadens outward: full data-layer fidelity, the three-auth-group HTTP layer, the user plugin, OAuth2.1 for MCP/ChatGPT, the admin schema pipeline and its Vue SPA, jobs/realtime/search infrastructure, the 154-route API surface (split into two delivery slices), domain jobs and external integrations, and finally a cutover phase where the parity harness is green on all 154 routes and both real clients run unchanged against the Go backend.

## Phases

**Phase Numbering:**

- Integer phases (1, 2, 3): Planned milestone work
- Decimal phases (2.1, 2.2): Urgent insertions (marked with INSERTED)

Decimal phases appear between their surrounding integers in numeric order.

- [x] **Phase 1: Framework kernel foundation** - Config, plugin registry, event bus, container and CLI skeleton boot the `summer` binary (completed 2026-09-16)
- [x] **Phase 2: API parity harness bootstrap** - Fixture recorder + replay-and-diff harness against the live PHP backend, built on the Phase 1 command kernel (completed 2026-09-17)
- [x] **Phase 3: First vertical slice — genres end to end** - `GET /_fonoteka/api/v1/genres` passes the parity diff through every layer (completed 2026-09-17)
- [x] **Phase 4: CLI scaffolding, i18n and mail** - Scaffolding commands, translated/pluralized strings, mail templates (completed 2026-09-18)
- [x] **Phase 5: Data layer full fidelity** - All 25 models and their squashed migrations with fillable/hidden/cast/soft-delete discipline (completed 2026-09-18)
- [x] **Phase 6: HTTP routing, auth groups and rate limiting** - Three auth groups, named rate buckets, OAuth-safe middleware structure (completed 2026-09-21)
- [x] **Phase 7: User plugin and authentication** - Registration, login, JWT, organizations, personal tokens, must-change-password (completed 2026-09-22)
- [x] **Phase 8: OAuth2.1 authorization server** - direct standard-library `wristband` server for fonoteka-mcp and the ChatGPT connector (completed 2026-09-23)
- [ ] **Phase 9: Backend admin authentication and schema pipeline** - Admin roles, fields.yaml/columns.yaml, relation manager
- [ ] **Phase 10: Admin Vue SPA** - Login, navigation, lists, forms and relation manager for five controllers
- [ ] **Phase 11: Jobs, realtime and search infrastructure** - River, Centrifugo and Typesense sync brought up before the API phases that need them
- [ ] **Phase 12: Płytarium API — Collections and Albums** - Core content endpoints ported with byte-level parity
- [ ] **Phase 13: Płytarium API — wishlist, notifications, CSV, credentials, public routes** - Remaining core API surface
- [ ] **Phase 14: Domain jobs and external integrations** - CSV/Discogs jobs, wishlist digest, reindex, Discogs client, AI recognition, feedback, sitemap
- [ ] **Phase 15: Cutover** - Parity harness green on all 154 routes, both real clients run unchanged

## Phase Details

### Phase 1: Framework kernel foundation

**Goal**: The `summer` binary boots from layered YAML config, plugins self-register through one required interface plus optional capability interfaces, a typed event bus and service registry are available, and a CLI command framework with a dev watch loop exists — built only as far as the first vertical slice will need it, per the interleaved-not-sequential kernel approach.
**Mode:** mvp
**Depends on**: Nothing (first phase)
**Repos:** summercms.go
**Requirements**: KERN-01, KERN-02, KERN-03, KERN-04, KERN-05, KERN-06, KERN-07, KERN-08, KERN-09, CLI-01
**Success Criteria** (what must be TRUE):

  1. `summer build` regenerates the blank-import plugin list and produces a binary that boots from a layered YAML config (base files, env overlay directory, per-plugin namespace, environment variables) with dot-path access.
  2. A throwaway plugin registered via the Plugin interface has its Register phase run for all plugins before any Boot phase runs, in `Requires()`-topological order, verified by a test with reordered input.
  3. A plugin can type-assert an optional capability interface (e.g. `HasModels`) and skip integration with another plugin that isn't registered, with no hard import.
  4. The typed event bus supports fire-and-forget, fire-and-collect, and fire-until-handled dispatch, exercised by unit tests; per-request state travels only through `context.Context`, never a package-level global.
  5. The `summer` CLI discovers a registered command and renders rich output (spinner, progress, table, prompts) with non-TTY degradation, and a dev watch loop rebuilds and restarts the binary on source change.

**Plans**: 4 plans

Plans:
**Wave 1**

- [x] 01-01-PLAN.md — Build and boot the separate hello app through the shared command kernel

**Wave 2** *(blocked on Wave 1 completion)*

- [x] 01-02-PLAN.md — Add layered config, optional plugin composition, services and typed events

**Wave 3** *(blocked on Wave 2 completion)*

- [x] 01-03-PLAN.md — Add plugin scaffolding, rich output and the watch rebuild loop

**Wave 4** *(blocked on Wave 3 completion)*

- [x] 01-04-PLAN.md — Complete unit, integration and race-enabled verification

### Phase 2: API parity harness bootstrap

**Goal**: A fixture recorder captures request/response pairs from the running PHP backend — including real Nuxt and MCP flows — and a replay-and-diff harness with a normalizer for nondeterministic fields can run against any HTTP backend. The recorder and replayer are `summer parity:*` commands on the Phase 1 command kernel (bonfire); recording itself only needs a running PHP backend, but the tooling waits for Phase 1.
**Mode:** mvp
**Depends on**: Phase 1
**Repos:** summercms.go, fonoteka.go
**Requirements**: QA-01, QA-02, QA-03
**Success Criteria** (what must be TRUE):

  1. The fixture recorder captures a request/response pair from the live PHP backend for at least one route and stores it as a replayable fixture.
  2. The replay-and-diff harness runs a recorded fixture against an arbitrary `httptest.Server` backend and reports a byte-level diff.
  3. The harness's normalizer and assertions explicitly catch the parity classes named in research (nil vs `[]`, date format, tri-state booleans, envelope/conditional keys) on a synthetic test case, not just status codes.
  4. `go vet` and `go test ./...` are green, and the harness's own integration tests run against testcontainers Postgres.

**Plans**: 5 plans

Plans:
**Wave 1**

- [x] 02-01-PLAN.md — Record and replay one fixture through the `summer` CLI

**Wave 2** *(blocked on Wave 1 completion)*

- [x] 02-02-PLAN.md — Capture safe proxy sessions and manifest routes with strict diffs

**Wave 3** *(blocked on Wave 2 completion)*

- [x] 02-03-PLAN.md — Record all 154 PHP routes and real Nuxt/MCP flows

**Wave 4** *(blocked on Wave 3 completion)*

- [x] 02-04-PLAN.md — Wire honest Go replay with testcontainers Postgres

**Wave 5** *(blocked on Waves 1–4 completion)*

- [x] 02-05-PLAN.md — Complete contract, security and integration tests

### Phase 3: First vertical slice — genres end to end

**Goal**: `GET /_fonoteka/api/v1/genres` passes the parity diff end to end (config → DB → plugin registry → JWT-guarded routing → GORM → JSON), proving every kernel layer together before any further kernel abstraction. Touches the JWT auth guard on the request path — treat as security-load-bearing and apply the security-review agent even though the guard is a throwaway seeded-token implementation at this stage.
**Mode:** mvp
**Depends on**: Phase 1, Phase 2
**Repos:** summercms.go, fonoteka.go
**Requirements**: DATA-01, DATA-02, HTTP-01, HTTP-02, QA-04
**Success Criteria** (what must be TRUE):

  1. GORM connects to Postgres over one shared `*sql.DB` (pgx stdlib); a gormigrate migration set runs up and down for a single plugin with per-plugin version tracking, and `AutoMigrate` is not the schema source.
  2. A plugin registers a route group on `net/http` ServeMux with typed params, and a request is served through the full middleware pipeline (recover, CORS, locale, auth group, must-change-password, org context, rate limit, handler).
  3. A JWT-guarded `GET /_fonoteka/api/v1/genres` route returns a real `Genre` row read from Postgres.
  4. The response is diffed byte-for-byte against a fixture recorded from the live PHP backend using the Phase 2 harness, and the diff is green.

**Plans**: 4 plans

Plans:

**Wave 1**

- [x] 03-01-PLAN.md — Boot the Postgres-backed JWT genre route through both plugins

**Wave 2** *(blocked on Wave 1 completion)*

- [x] 03-02-PLAN.md — Port active-collection counts, validation and Polish ordering

**Wave 3** *(blocked on Wave 2 completion)*

- [x] 03-03-PLAN.md — Add typed params, isolated rollback and the first real parity pass

**Wave 4** *(blocked on Wave 3 completion)*

- [x] 03-04-PLAN.md — Complete unit, integration and security verification

### Phase 4: CLI scaffolding, i18n and mail

**Goal**: Scaffolding commands generate compiling stubs for every plugin artifact type, and plugins can register translated, CLDR-pluralized, namespaced strings and mail templates rendered through a pluggable driver interface.
**Mode:** mvp
**Depends on**: Phase 1
**Repos:** summercms.go
**Requirements**: CLI-02, I18N-01, I18N-03
**Success Criteria** (what must be TRUE):

  1. `summer make:plugin`, `make:model`, `make:migration`, `make:command`, `make:job` and `make:admin-controller` each generate stubs that compile and pass `go vet`.
  2. A translation key `vendor.plugin::group.key` resolves for pl and en, including a CLDR plural form, loaded from per-plugin per-locale YAML files with parameter substitution.
  3. A plugin registers a mail template and layout by dotted name with the per-locale suffix convention, and it renders via `html/template` through a driver interface (SMTP via go-mail) in a test send.

**Plans**: 4 plans

Plans:

**Wave 1**

- [x] 04-01-PLAN.md — Generate compiling plugin artifacts, registry and model import checks

**Wave 2** *(blocked on Wave 1 completion)*

- [x] 04-02-PLAN.md — Load namespaced translations, plurals, parameters and locale fallbacks

**Wave 3** *(blocked on Wave 2 completion)*

- [x] 04-03-PLAN.md — Register, render and send plugin mail through pluggable drivers

**Wave 4** *(blocked on Wave 3 completion)*

- [x] 04-04-PLAN.md — Verify scaffolding, translation and mail with unit and SMTP integration tests

### Phase 5: Data layer full fidelity

**Goal**: All 25 Płytarium models and their squashed migration set are ported with matching relations, casts, hooks, and the fillable/hidden/encrypted-cast mass-assignment and serialization discipline. This is security-load-bearing — mass-assignment boundaries and encrypted-at-rest credential casts are named security invariants in the PHP source — apply the security-review agent and the DTO-vs-model convention from the first model onward.
**Mode:** mvp
**Depends on**: Phase 1, Phase 3
**Repos:** summercms.go, fonoteka.go
**Requirements**: DATA-03, DATA-04, DATA-05, DATA-06, DATA-07, DATA-08, DATA-09, DATA-10, DATA-11, CLI-03
**Success Criteria** (what must be TRUE):

  1. All 25 models exist with matching table names, columns, indexes and defaults, and every Go migration runs up and down individually; the final schema matches PHP's (migrations are squashed per final-state table, not a 1:1 port of PHP's 38 files — the historical migration count is not a target). `summer migrate:rollback --plugin=fonoteka` rolls back only that plugin's last migration.
  2. A many-to-many relation with pivot business columns (`album_artists.sort_order`, `CollectionEditor.role/granted_at/granted_by`) round-trips correctly for a 3+ artist album; belongsTo/hasOne/hasMany relations return ordered results.
  3. The fill boundary of the PHP write services (at minimum Album, Collection and the four credential models) is fuzzed against real Postgres with random extra and server-owned keys, asserting nothing outside the allow-list is persisted (service-level, this phase); a request-DTO-level fuzz over every write endpoint is Phase 12's criterion (the HTTP layer does not exist until Phase 6/12). Serialization honors the hidden deny-list with an explicit per-call override.
  4. The money cast round-trips the PHP ceiling and blank-string cases as a fixed 4-decimal JSON string (never `float64`), and an encrypted-at-rest credential column is AES-GCM encrypted at rest and hidden from serialization.
  5. Paginated responses use the exact `{data, meta{current_page,last_page,per_page,total}}` envelope with no `links` key; another plugin extends a model's lifecycle through the GORM callback registry and a companion migration without editing the owning plugin's file; soft-deletable + uniquely-keyed tables pass a delete-then-recreate test.

**Plans**: 11 plans

Plans:
**Wave 1**

- [x] 05-01-PLAN.md — Verify models-leaf rule, restructure plugins into Winter layout, ship lagoon write/read-path primitives

**Wave 2** *(blocked on Wave 1 completion)*

- [x] 05-02-PLAN.md — Album/Collection slice: migrations, models, casts, validation engine, write services

**Wave 3** *(blocked on Wave 2 completion)*

- [x] 05-03-PLAN.md — Secrets slice: encrypted cast, key management, organisations, credentials/OAuth tables
- [x] 05-04-PLAN.md — Attachments: system_files, blob storage, thumbnails, delete lifecycle

**Wave 4** *(blocked on Wave 3 completion)*

- [x] 05-05-PLAN.md — Remaining models, D-02 schema-diff proof, test-only DATA-11 fixture plugin, CLI-03 verification

**Wave 5** *(blocked on Wave 4 completion)*

- [x] 05-06-PLAN.md — Fuzz tests, hidden-marshal coverage, attachment smoke test, security review

### Phase 6: HTTP routing, auth groups and rate limiting

**Goal**: The three mutually exclusive auth groups (JWT, personal token, public/onboarding) share handlers with correct route subsets, named rate-limit buckets are ported 1:1, and OAuth/RFC routes are structurally exempted from any house envelope or error middleware. Security-load-bearing — auth guard registry, rate limiting and the SSRF-guarded outbound fetch helper all live here; apply the security-review agent.
**Mode:** mvp
**Depends on**: Phase 3, Phase 5
**Repos:** summercms.go, fonoteka.go
**Requirements**: HTTP-03, HTTP-04, HTTP-05, HTTP-06, HTTP-07, HTTP-08, HTTP-09
**Success Criteria** (what must be TRUE):

  1. The same handler serves both a JWT-authenticated request under `/_fonoteka/api/v1` and a personal-token request under `/api/v1/fonoteka`, with public/onboarding groups reachable without auth; unknown and malformed ids on ownership-scoped resources both return 404.
  2. All five named rate-limit buckets are enforced with the documented keys and limits, including a route stacking two limiters.
  3. Response conventions hold under test: empty arrays serialize as `[]`, timestamps carry `+00:00`, tri-state booleans keep `null`, conditional keys are omitted not nulled; an OAuth-group route carries no house envelope or error middleware, verified by route-registration inspection.
  4. The guarded outbound fetch helper rejects a non-allow-listed host and enforces a byte cap and timeout on a user-supplied cover URL fetch (manual cover URL, Discogs cover).
  5. OpenAPI is generated from swaggo/swag annotations on handlers and `openapi-typescript` produces valid TypeScript types from it; CORS and JSON body-size limits match the PHP deployment.

**Plans**: 10 plans

Plans:
**Wave 1** *(parallel)*

- [x] 06-01-PLAN.md — Auth-groups slice: router verb/factory growth, bouncer guard registry, real inv_token guard + inv.scope, genres shared under both auth groups
- [x] 06-04-PLAN.md — SSRF-guarded outbound fetch helper (framework primitive, independent of the other three plans)

**Wave 2** *(blocked on 06-01)*

- [x] 06-02-PLAN.md — Rate limiting: fixed-window Store/Limiter, trusted-proxy client IP, five fonoteka buckets, PublicShareHeaders, remaining route groups declared

**Wave 3** *(blocked on 06-02)*

- [x] 06-03-PLAN.md — Contract surface: raw-group enforcement + route table + route:list, wire response helpers, swag/openapi-typescript pipeline, path-scoped CORS, body limits (blocking human-verify checkpoint for production body-size numbers), oauth group declared raw

**Wave 4** *(blocked on 06-01..06-04)*

- [x] 06-05-PLAN.md — Full unit coverage across both repos, full route-table mutual-exclusivity test, 06-SECURITY-REVIEW.md

**Wave 5** *(gap closure; blocked on 06-05)*

- [x] 06-06-PLAN.md — Repair personal-token middleware order and prove unauthenticated request 61 is rate-limited

**Wave 6** *(gap closure; parallel, blocked on 06-06)*

- [x] 06-07-PLAN.md — Make limiter admission atomic and remove attacker-controlled Host from inline keys
- [x] 06-08-PLAN.md — Reject private IPv4 embedded in NAT64 and 6to4 dial addresses
- [x] 06-09-PLAN.md — Buffer route responses so partial-write panics yield clean raw/house 500s
- [x] 06-10-PLAN.md — Restore exact no-newline InvScope 401/403 wire bodies

**Wave 7** *(gap closure; blocked on 06-07..06-10)*

- [x] 06-11-PLAN.md — Re-run Phase 6 security gates and refresh the stale threat verdict

### Phase 7: User plugin and authentication

**Goal**: The user plugin is ported with registration, login, JWT issue/refresh, organizations, personal API tokens and the must-change-password lock. Security-load-bearing — password auth, token scope ceilings and the session lock all live here; apply the security-review agent.
**Mode:** mvp
**Depends on**: Phase 6
**Repos:** summercms.go, fonoteka.go
**Requirements**: AUTH-01, AUTH-02, AUTH-03, AUTH-04, I18N-02
**Success Criteria** (what must be TRUE):

  1. A user can register, log in, log out, reset a forgotten password by email, verify email, and receive/refresh a JWT (golang-jwt) with the same claims and cookie behavior the Nuxt app expects.
  2. Organization fields appear on the user payload through a fire-and-collect event the fonoteka plugin populates, without the user plugin importing fonoteka.
  3. A personal API token is created with a read|write|ai scope ceiling, listed, and revoked; a scope-checking middleware rejects an out-of-scope request.
  4. The must-change-password flag returns 423 on the authenticated surface except the locale and password-change routes, and locale resolves per request from the user's persisted `preferred_locale` with header fallback even while the lock is active.

**Plans**: 8 plans

Plans:
**Wave 1**

- [x] 07-01-PLAN.md — bouncer JWT lifecycle, password hashing, I18N-02 locale-from-principal, lagoon.Validate extensions

**Wave 2** *(blocked on 07-01)*

- [x] 07-02-PLAN.md — User/Throttle schema and the core session loop: login/logout/fetch/refresh/register

**Wave 3** *(blocked on 07-02)*

- [x] 07-03-PLAN.md — Account management: forgot/reset password, activation, update, change-password, avatar, mail
- [x] 07-04-PLAN.md — Personal API tokens (mint/list/revoke), me/locale, 423-exempt route-table proof

**Wave 4** *(blocked on 07-03, 07-04)*

- [x] 07-05-PLAN.md — Parity evidence: record and replay the 15 /_user/api/v1 routes and the nuxt-auth flow

**Wave 5** *(blocked on 07-05)*

- [x] 07-06-PLAN.md — Full unit coverage, 07-VALIDATION.md sign-off

**Wave 6** *(gap closure; blocked on 07-06)*

- [x] 07-07-PLAN.md — Re-record the logout blacklist under a persistent PHP cache, fix the already-activated HTML-500 quirk, and flip all 15 /_user/api/v1 routes plus both nuxt flows to ported

**Wave 7** *(gap closure; blocked on 07-07 UAT)*

- [x] 07-08-PLAN.md — Publish the uploads bucket on serve and Handler so assembled avatar POST is 200

### Phase 8: OAuth2.1 authorization server

**Goal**: A direct standard-library OAuth2.1-style authorization server (`wristband`) implements the RFC 8414/6749/7591/8707 contract needed by fonoteka-mcp and the ChatGPT connector unchanged. The backend preserves its exact Basic invalid-client challenge and existing personal-token 401, while fonoteka-mcp retains ownership of RFC 9728 protected-resource metadata and its rich Bearer challenge. Security-load-bearing — bearer tokens, PKCE, replay-family revocation, and constant-time secret comparison all live here; apply the security-review agent.
**Mode:** mvp
**Depends on**: Phase 6, Phase 7
**Repos:** summercms.go, fonoteka.go
**Requirements**: AUTH-05, AUTH-06, AUTH-07
**Success Criteria** (what must be TRUE):

  1. RFC 8414 metadata is served unenveloped at its well-known path; authorize with PKCE and a consent screen works, and the token endpoint issues/refreshes authorization_code and refresh_token flows, all as unwrapped RFC 6749 bodies with the PHP cache headers (`Cache-Control: no-store`, `Pragma: no-cache`).
  2. RFC 7591 dynamic client registration and RFC 8707 resource parameter tolerance both work against a real client registration call.
  3. The token endpoint returns exactly `WWW-Authenticate: Basic realm="OAuth"` on `invalid_client`, the backend personal-token 401 remains unchanged with no added challenge, and fonoteka-mcp's own rich Bearer challenge plus protected-resource metadata are verified through its actual discovery flow, not just a Go unit test.
  4. Connected apps can be listed and revoked; `OAuthClient`/`OAuthAuthCode`/`OAuthRefreshToken` models persist correctly; fonoteka-mcp completes its install and auth flow unchanged; client-secret comparison uses `crypto/subtle.ConstantTimeCompare`.

**Plans**: 10 plans
**Research flag:** yes

Plans:

**Wave 1**

- [x] 08-01-PLAN.md — Mount exact connector-visible metadata and establish fail-closed RED verification

**Wave 2** *(blocked on 08-01)*

- [x] 08-02-PLAN.md — Deliver persistent connector-visible DCR with corrected schema and transaction-scoped stores

**Wave 3** *(blocked on 08-02)*

- [x] 08-03-PLAN.md — Create durable PKCE-bound authorize requests on the assembled raw route surface

**Wave 4** *(blocked on 08-03)*

- [x] 08-04-PLAN.md — Implement atomic PKCE-bound authorization-code exchange

**Wave 5** *(blocked on 08-04)*

- [x] 08-05-PLAN.md — Wire JWT consent and prove the unchanged Nuxt UI contract

**Wave 6** *(blocked on 08-05)*

- [x] 08-06-PLAN.md — Rotate refresh grants, kill replayed lineages, and manage connected apps

**Wave 7** *(parallel; blocked on 08-06)*

- [x] 08-07-PLAN.md — Provision confidential clients through the exact operator command
- [x] 08-08-PLAN.md — Serve the MCP personal-token bootstrap on the existing token surface

**Wave 8** *(blocked on 08-07 and 08-08)*

- [x] 08-09-PLAN.md — Replay PHP OAuth flows and assemble the self-validating final unchanged-MCP gate

**Wave 9** *(blocked on 08-09; blocking security checkpoint)*

- [x] 08-10-PLAN.md — Close 103-method coverage, independent security review, and the final fail-closed gate

### Phase 9: Backend admin authentication and schema pipeline

**Goal**: Backend admin users with roles are separate from frontend users and gate navigation and controller access; `fields.yaml`/`columns.yaml` drive a JSON form/list schema, including a first-class relation-manager schema replacing the one `partial` field. Security-load-bearing — separate admin authentication and permissions registry live here; apply the security-review agent. This is also the least-precedented design surface in the research (one real relation-manager usage in the PHP source, no direct library equivalent).
**Mode:** mvp
**Depends on**: Phase 5, Phase 6
**Repos:** summercms.go, fonoteka.go
**Requirements**: AUTH-08, ADMIN-01, ADMIN-02, ADMIN-03, ADMIN-04, ADMIN-05
**Success Criteria** (what must be TRUE):

  1. A backend admin user with a role logs in separately from frontend users, and navigation/controller access is gated by the permissions registry.
  2. `fields.yaml` for a real controller parses (goccy/go-yaml) into a JSON form schema covering text, textarea, checkbox, switch, dropdown (model-method options) and relation (nameFrom, emptyOption), with span/tabs/context/attributes layout hints.
  3. `columns.yaml` parses into a JSON list schema with searchable/sortable/relation columns and datetime/switch renderers.
  4. The relation-manager schema supports search/link/unlink/manage-or-view lists for Collections' editors tab, replacing the `partial` field entirely.
  5. Admin CRUD endpoints expose `listExtendQuery`/`formExtendQuery`/`formBeforeCreate`/`formBeforeUpdate`/`relationExtendManageQuery` hooks, bulk delete runs each record's lifecycle hooks, and the Settings model binds to a settings screen through the same schema pipeline.

**Plans**: 2/12 plans executed
**Research flag:** yes

Plans:

**Wave 1**
- [x] 09-01-PLAN.md — Prove the architecture with one production end-to-end Genre list

**Wave 2** *(blocked on Wave 1 completion)*
- [x] 09-02-PLAN.md — Complete backend identity lifecycle and operator provisioning

**Wave 3** *(blocked on Wave 2 completion)*
- [ ] 09-03-PLAN.md — Compile the typed form-schema contract and Winter scaffolding

**Wave 4** *(blocked on Wave 3 completion)*
- [ ] 09-04-PLAN.md — Compile the list contract and the allowlisted query engine

**Wave 5** *(blocked on Wave 4 completion)*
- [ ] 09-05-PLAN.md — Deliver schema-projected CRUD and transactional bulk deletion

**Wave 6** *(blocked on Wave 5 completion)*
- [ ] 09-06-PLAN.md — Port the Albums admin surface and the collection boundary

**Wave 7** *(blocked on Wave 6 completion)*
- [ ] 09-07-PLAN.md — Port the Artists backend controller
- [ ] 09-08-PLAN.md — Complete the Genres controller with form/list parity
- [ ] 09-09-PLAN.md — Port the Styles controller and typed provider/filter behavior
- [ ] 09-10-PLAN.md — Deliver Collections and the typed relation manager

**Wave 8** *(blocked on Wave 7 completion)*
- [ ] 09-11-PLAN.md — Complete permissions, navigation, and singleton settings

**Wave 9** *(blocked on Wave 8 completion)*
- [ ] 09-12-PLAN.md — Close with security, PostgreSQL, OpenAPI, and acceptance gates

### Phase 10: Admin Vue SPA

**Goal**: A minimal Vue 3 + TypeScript admin SPA renders login, permission-gated navigation, lists, forms and the relation manager for Albums, Artists, Collections, Genres and Styles, typed from the generated OpenAPI document.
**Mode:** mvp
**Depends on**: Phase 9
**Repos:** summercms.go, fonoteka.go
**Requirements**: ADMIN-06
**Success Criteria** (what must be TRUE):

  1. An admin logs in through the SPA and sees only the navigation items their permissions allow.
  2. Each of the five controllers (Albums, Artists, Collections, Genres, Styles) renders a working list and form generated from its JSON schema.
  3. The Collections form's relation manager lets an admin search, link and unlink an editor.
  4. API calls in the SPA use TypeScript types generated from the OpenAPI document, with no hand-maintained duplicate type.

**Plans**: TBD
**UI hint**: yes

### Phase 11: Jobs, realtime and search infrastructure

**Goal**: River jobs run on the correct dual-driver split, Centrifugo publishing and channel authorization match the existing server, and Typesense sync stays a re-gated pre-filter — all brought up before the API phases that depend on them (album search needs Typesense sync, CSV import needs River jobs, notifications need the realtime publisher). River's dual-driver split and the Centrifugo/Typesense contracts are the least-implemented-and-verified parts of this research pass.
**Mode:** mvp
**Depends on**: Phase 5, Phase 7
**Repos:** summercms.go, fonoteka.go
**Requirements**: JOBS-01, CLI-04, CLI-06, RT-01, RT-02, RT-03, SRCH-01
**Success Criteria** (what must be TRUE):

  1. River runs on the shared `*sql.DB` (`riverdatabasesql`, transactional enqueue) with a separate `riverpgxv5` client for LISTEN/NOTIFY dispatch, verified by a timed test that job pickup is not poll-interval latency; job outcomes (complete, fail, skip) are queryable through a job manager service.
  2. The queue worker runs via `summer queue:work` and the scheduler runs recurring commands via `summer schedule:run`.
  3. The websockets plugin issues connection and subscription JWTs and publishes to the existing Centrifugo server with the same secret, claims and channel names, served at `GET /api/realtime/token`.
  4. A channel-namespace authorizer registry re-validates on every subscribe; a broadcastable model interface with bulk-write suppression emits exactly one summary event for a bulk operation.
  5. Typesense sync is scoped by `collection_id` behind a settings kill-switch and degrades gracefully without DB/config.

**Plans**: TBD
**Research flag:** yes

### Phase 12: Płytarium API — Collections and Albums

**Goal**: Collections and Albums endpoints are ported with byte-compatible request/response shapes, including active-context switching, editor invitations, ratings, reservations, cover handling and search.
**Mode:** mvp
**Depends on**: Phase 5, Phase 6, Phase 7, Phase 11
**Repos:** fonoteka.go
**Requirements**: API-01, API-02
**Success Criteria** (what must be TRUE):

  1. Collections CRUD, the `me/context` active-collection switch (returning an opaque channel name), editor invitation/acceptance, share-link regeneration and public token views all pass the parity diff.
  2. Albums CRUD, ratings, reservations, photo upload, manual cover URL and Discogs cover price all pass the parity diff.
  3. Album search treats Typesense results as a pre-filter re-gated in SQL, verified by a security test that a stale/mis-scoped search document cannot leak an unauthorized result.
  4. Artists/genres/styles lookup endpoints used by the Albums UI pass the parity diff.
  5. A request-DTO-level fuzz over every write endpoint asserts unknown and server-owned keys are never persisted (inherits the HTTP half of Phase 5 criterion 3; the HTTP layer does not exist until Phase 6/12).

**Plans**: TBD

### Phase 13: Płytarium API — wishlist, notifications, CSV, credentials, public routes

**Goal**: The remaining core API surface — wishlist, notifications, CSV import/export, per-user/org credentials, and onboarding/public/invitation routes — is ported with byte-compatible shapes and their own public rate-limit buckets.
**Mode:** mvp
**Depends on**: Phase 11, Phase 12
**Repos:** fonoteka.go
**Requirements**: API-03, API-04, API-05, API-06, API-07
**Success Criteria** (what must be TRUE):

  1. Wishlist items, subscriptions, `public-wishlist/{token}` views and purchase/digest triggers pass the parity diff.
  2. Notifications list/mark-read/prune endpoints pass the parity diff (the realtime token endpoint itself is owned by the websockets plugin, ported in Phase 11).
  3. CSV import runs as a multi-step session (store, show/poll, mapping patch, per-row edit, commit, cancel) and CSV export works on both authenticated groups, all passing the parity diff.
  4. Per-user and per-org Discogs/AI credentials CRUD store encrypted values, honor the org-lock flag, and resolve env-to-org-to-user correctly.
  5. Onboarding, public and invitation-inspection routes are reachable without auth and enforce their own public rate-limit buckets.

**Plans**: TBD

### Phase 14: Domain jobs and external integrations

**Goal**: The domain-specific River jobs (CSV import write, Discogs match, wishlist digest), the reindex command, the Discogs client and AI cover recognition are ported on top of the Phase 11 jobs/realtime/search infrastructure and the Phase 13 API surface they serve.
**Mode:** mvp
**Depends on**: Phase 11, Phase 13
**Repos:** fonoteka.go (plus summercms.go only if a framework helper is needed)
**Requirements**: JOBS-02, JOBS-03, SRCH-02, INTG-01, INTG-02, API-08, CLI-05
**Success Criteria** (what must be TRUE):

  1. The CSV import write job and the self-redispatching Discogs match job (240s timeout, self-redispatching with a delay on a Discogs rate-limit error) both complete correctly.
  2. The wishlist digest job coalesces a 30-minute window and deletes its queue row on completion.
  3. The `reindex` command asserts zero `collection_id`-0 documents before and after and can drop the legacy index.
  4. The Discogs client enforces its proactive rate threshold, bounded wait budget, retry-after fallback and host-locked cover fetch.
  5. AI cover recognition works through both Anthropic and OpenAI-compatible adapters with per-credential model/base-URL overrides.
  6. Feedback submissions and sitemap output work; the ported `oauth-client`/`prune-notifications`/`reindex` commands all run correctly.

**Plans**: TBD

### Phase 15: Cutover

**Goal**: The parity harness is green on all 154 routes and `vue-fonoteka-app` and `fonoteka-mcp` run unchanged against the Go backend in daily use — the project's definition of done.
**Mode:** mvp
**Depends on**: Phase 2, Phase 8, Phase 9, Phase 10, Phase 12, Phase 13, Phase 14
**Repos:** fonoteka.go, summercms.go
**Requirements**: API-09, QA-05
**Success Criteria** (what must be TRUE):

  1. All 154 routes are registered on the correct auth groups with identical paths, methods and status codes.
  2. The parity harness runs green across recorded fixtures for all 154 routes.
  3. `vue-fonoteka-app` runs unchanged against the Go backend for a full manual session (browse, edit, upload, invite, OAuth-connect an MCP client).
  4. `fonoteka-mcp` completes its install/auth flow and a representative set of tool calls unchanged against the Go backend.

**Plans**: TBD

## Progress

**Execution Order:**
Phases execute in numeric order: 1 → 2 → 3 → 4 → 5 → 6 → 7 → 8 → 9 → 10 → 11 → 12 → 13 → 14 → 15
(Phase 2 depends on Phase 1's command kernel; the two are no longer parallel.)

| Phase | Plans Complete | Status | Completed |
|-------|----------------|--------|-----------|
| 1. Framework kernel foundation | 4/4 | Complete   | 2026-09-16 |
| 2. API parity harness bootstrap | 5/5 | Complete   | 2026-09-17 |
| 3. First vertical slice — genres end to end | 4/4 | Complete   | 2026-09-17 |
| 4. CLI scaffolding, i18n and mail | 4/4 | Complete   | 2026-09-18 |
| 5. Data layer full fidelity | 6/6 | Complete   | 2026-09-18 |
| 6. HTTP routing, auth groups and rate limiting | 14/14 | Complete   | 2026-09-21 |
| 7. User plugin and authentication | 8/8 | Complete    | 2026-09-23 |
| 8. OAuth2.1 authorization server | 10/10 | Complete   | 2026-09-23 |
| 9. Backend admin authentication and schema pipeline | 2/12 | In Progress|  |
| 10. Admin Vue SPA | 0/TBD | Not started | - |
| 11. Jobs, realtime and search infrastructure | 0/TBD | Not started | - |
| 12. Płytarium API — Collections and Albums | 0/TBD | Not started | - |
| 13. Płytarium API — wishlist, notifications, CSV, credentials, public routes | 0/TBD | Not started | - |
| 14. Domain jobs and external integrations | 0/TBD | Not started | - |
| 15. Cutover | 0/TBD | Not started | - |