Update API open items after hardening and teardown validation

Codex/API/OPEN_ITEMS.md

@@ -7,21 +7,27 @@ Only keep unfinished API work here.
 - service discovery migration: audit edge publish/DNS/Cloudflare/Technitium, Prometheus SD, dev IDE wildcard, and post-provision hot paths for direct host assumptions
 - provisioning validation follow-up where API behavior is involved
 - verify Portal compatibility after API JWT issuer/audience tightening, especially refresh flow and hosted IDE token flow
+- migrate Portal delete calls to `DELETE /api/servers/:id`; `DELETE /api/containers/:vmid` currently remains as owned-user compatibility plus internal-token automation, but should return to internal-only once Portal is updated
 - verify canonical and compatibility file routes still behave identically across list/stat/read/download/delete/put/revert/upload paths after helper extraction
 - align merged live-status readiness fields so Portal-facing `agentReady` semantics fully match semantic `/ready`
-- verify orphan/manual delete behavior after retirement of legacy PortPool release logic
 - decide whether the remaining Portal-side `/api/agent/:serverId/:action` bridge should be deleted outright or formally kept as compatibility-only Portal-owned behavior
 - live-verify Velocity bridge lifecycle callbacks after `ZPACK_PROXY_STATUS_ENDPOINT` is set: confirm `registered_with_proxy`, `proxy_ping_ok`, and `proxy_ping_failed` land in `ContainerInstance.payload.proxy` and surface through `GET /api/servers/:id/status`
 - Portal integration follow-up for host/LXC lifecycle state: consume `GET /api/servers/:id/host/status`, list-level `hostStatus` / `powerState` / `hostOperation`, and `202` host action responses for game and dev containers
 - remove or downgrade the temporary `MaxListenersExceededWarning` tracer in `src/app.js` after outbound Axios socket listener warnings are confirmed quiet in runtime logs
 - verify Proxmox node resolution against all active container ranges; recent local smoke checks showed some DB VMIDs not present in `/cluster/resources` or on the configured node
+- add minimal test/CI coverage for auth boundaries and teardown behavior:
+  - normal users cannot reach admin-only routes
+  - missing internal token fails closed for internal-only routes
+  - owned-user delete archives then removes the active instance
+  - non-owner delete fails
+  - Stripe webhook raw-body handling remains intact
 
 ## Cleanup / consolidation priorities
 - fold repeated ownership/auth/IP-guard patterns into small concrete helpers without hiding route intent
 - split oversized route/service files by responsibility without changing route contracts
 - keep backup/restore status shaping and async-dispatch logic explicit, but remove duplicated mapping/normalization paths where possible
 - keep stream-vs-JSON forwarding rules centralized in one place and avoid route-local reimplementation
-- keep archived legacy flows out of the live tree unless they are intentionally revived and revalidated against the current schema/contracts
+- keep legacy flows out of the live tree unless they are intentionally revived and revalidated against the current schema/contracts
 
 ## Completed and moved out of active cleanup
 - Node/runtime pinning is no longer an open cleanup-only item; Node 24 pinning is now treated as current repo state
@@ -29,7 +35,10 @@ Only keep unfinished API work here.
 - initial file-proxy route deduplication has been completed; only compatibility verification and follow-on cleanup remain open
 - Prisma config migration is no longer an open item
 - baseline proxy cookie/log hardening is no longer an open item
-- worker-era provisioning and detached legacy port reservation have been archived rather than treated as active API surfaces
+- worker-era provisioning and detached legacy port reservation have been removed from the live tree rather than treated as active API surfaces
+- initial control-plane hardening has landed: shared admin/internal-token middleware, admin-only audit/global instance list, internal-token guards on raw control-plane routes, and fail-closed monitoring/SD token behavior outside explicit development/test
+- teardown workflow has been extracted into a service and live-verified against one game VMID and one dev VMID; both archived with `reason=user_delete` and active rows were removed
+- repo hygiene pass removed checked-in key/token/artifact/legacy clutter and tightened ignore rules
 
 ## Cleanup rule
 - prefer behavior-preserving folding over broad refactors