# 2026-05-13 — AZ-512 admin classes CRUD prerequisite (cross-workspace)

> **PARTIALLY RESOLVED 2026-05-13T03:51+02:00** — prerequisite ticket **AZ-513** was filed on the admin/ workspace (Jira Task, parent epic AZ-509, "Blocks" link to AZ-512). Matching task spec written to `admin/_docs/02_tasks/todo/AZ-513_classes_crud_routes.md`. AZ-512 carries a comment pointing at AZ-513. Replay obligation below now waits on AZ-513 shipping (admin/ side work), not on the autodev session itself.

## Summary

AZ-512 (Admin edit detection class) hit its spec-defined Cross-Workspace Verification BLOCKING gate during cycle 3 batch 15 implementation in the UI workspace. The `admin/` sibling service (Azaion.AdminApi) does not expose `/classes` routes at all. This leftover records (a) the deferred AZ-512 work in the UI, and (b) a separately-noted pre-existing bug discovered during verification.

## Timestamp

`2026-05-13T02:12:00+02:00` (Europe/Paris) — entry created by autodev cycle 3 batch 15 BLOCKING gate.

## What was blocked

1. **AZ-512 implementation (UI workspace)** — the inline edit form + `PATCH /api/admin/classes/{id}` wiring on `src/features/admin/AdminPage.tsx`. Task parked in `_docs/02_tasks/backlog/AZ-512_admin_edit_detection_class.md`.

2. ~~**Cross-workspace prerequisite ticket**~~ — **FILED as AZ-513 on 2026-05-13** with user-confirmed epic linkage (AZ-509). Spec at `admin/_docs/02_tasks/todo/AZ-513_classes_crud_routes.md`. "Blocks" link AZ-513 → AZ-512 created in Jira. Comment on AZ-512 references AZ-513. Pending: admin/ team picks up and ships AZ-513.

## Prerequisite payload (for the user to file)

**Suggested ticket summary**: `[admin/] Add /classes CRUD routes (POST + PATCH + DELETE) to Azaion.AdminApi`

**Suggested description**:

> The UI workspace (`ui/src/features/admin/AdminPage.tsx`) calls three /classes endpoints today, but only the read path is served (and it is served by the `annotations` service, not `admin`):
>
> - `POST /api/admin/classes` — UI calls this to add a new detection class (`handleAddClass`). Today: 404. Pre-existing bug.
> - `DELETE /api/admin/classes/{id}` — UI calls this to delete a class (`handleDeleteClass`). Today: 404. Pre-existing bug.
> - `PATCH /api/admin/classes/{id}` — UI does NOT call this today; AZ-512 (UI workspace) needs to call it to deliver the in-place edit affordance promised by Architecture Vision principle P12. Currently the route does not exist either.
>
> nginx.conf in the ui workspace routes `/api/admin/` to `http://admin:8080/`, so the path inside the admin service is `/classes` and `/classes/{id}`. The admin service's `Program.cs` exposes only `/login`, `/users*`, `/resources*` today (search 2026-05-13 in this UI workspace's chat transcript).
>
> The UI is the authoritative wire-shape contract via `ui/src/api/endpoints.test.ts` — `endpoints.admin.classes()` and `endpoints.admin.class(id)` pin the URLs.
>
> **Acceptance**:
>
> 1. `POST /classes` accepts `{ name, shortName, color, maxSizeM }` (and any of `photoMode`, etc. that the live backend already supports for ADD), returns the created class object on 200/201.
> 2. `DELETE /classes/{id}` deletes the class by id, returns 200/204.
> 3. `PATCH /classes/{id}` accepts a partial-merge body `{ name?, shortName?, color?, maxSizeM? }`, returns the updated class object on 200. Send-complete-body semantics are also fine — the UI sends every field per AZ-512 spec Risk 2 mitigation.
> 4. All three routes guarded by the same auth middleware as `/users` (admin role required).
> 5. After this ticket lands, AZ-512 (UI workspace) un-blocks and the existing add+delete affordances start working end-to-end.
>
> **Story points**: 3 (single Program.cs file, 3 minimal-API handlers, an `IDetectionClassService` injected like `IUserService` is today).

**Suggested epic**: whatever the admin/ workspace's "API contract / CRUD coverage" epic is — to be decided by the user when filing.

## Reason for blockage

Spec-defined BLOCKING gate. The AZ-512 task spec explicitly forbids inventing a workaround that bypasses the missing endpoint:

> *"Do not invent a workaround that bypasses the missing endpoint."*

The spec's three options at the gate:

- **A**: File a hard-prerequisite ticket on the `admin/` workspace, pause AZ-512 until it lands.
- **B**: Implement only the UI form, MSW-stubbed in tests, mark Step 11 blocked-on-admin/PATCH, ship draft PR.
- **C**: Drop AZ-512 from cycle 3, defer to a future cycle.

User was prompted via AskQuestion in the same chat turn; user skipped the prompt. The autodev defaulted to **A** (most conservative; spec-aligned; respects workspace boundary).

## Replay obligation

This entry is NOT auto-replayable from the UI workspace alone — it requires (a) cross-workspace ticket creation that the UI's autodev should not do unilaterally, and (b) actual implementation work on the admin/ workspace which is owned by a separate Cursor workspace per `.cursor/rules/coderule.mdc`.

When AZ-512 batch 15 is re-attempted (next `/autodev` invocation that covers cycle 3 leftovers, or any cycle that re-prioritises P12), the leftovers replay step should:

1. Re-run the verification: `grep -E "MapPost|MapPatch|MapDelete" /Users/.../suite/admin/Azaion.AdminApi/Program.cs | grep classes`.
2. If routes exist → move `_docs/02_tasks/backlog/AZ-512_*.md` back to `_docs/02_tasks/todo/`, update this leftover with the resolution, and proceed with batch 15.
3. If routes still missing → leave the leftover as-is, surface to the user that the prerequisite is still outstanding.

## Side note (separate concern, do not bundle)

While verifying the gate, I noticed that `AdminPage.tsx` already calls `POST /api/admin/classes` (handleAddClass) and `DELETE /api/admin/classes/{id}` (handleDeleteClass) today, neither of which is served by the admin/ service. So the existing add+delete buttons on the Detection Classes table are broken end-to-end against the live admin/ service in production. This is a **pre-existing bug**, NOT introduced by AZ-512 or any cycle 3 work. It should be tracked as its own UI-workspace ticket once the admin/ work is filed (the same admin/ ticket above will likely fix the production behaviour for free, but a UI-side test would confirm the wire-up post-fix).