Update PROJECT_CONTEXT — dev access model pivot, remove Traefik approach, add Headscale path

PROJECT_CONTEXT.md

@@ -29,9 +29,9 @@ System posture: stable, controlled expansion phase.
 | 1001 | zlh-dns | Technitium DNS |
 | 1002 | zlh-proxy | Traefik — core/frontend SSL termination (portal traffic) |
 | 1003 | zlh-artifacts | Runtime binaries + Minecraft server jars (agent install source) |
-| 1004 | zlh-zpack-proxy | Traefik — game and dev server traffic |
+| 1004 | zlh-zpack-proxy | Traefik — game server traffic only |
 | 1005 | zpack-api | Node.js API |
-| 1006 | zlh-zpack-router | Game + dev server router |
+| 1006 | zlh-zpack-router | Game server router |
 | 1100 | zpack-portal | Next.js frontend |
 | 2001 | zlh-back | PBS backup + Backblaze B2 |
 
@@ -72,10 +72,9 @@ access. Pulls runtimes + server jars from zlh-artifacts (VM 1003).
 - Self-update: periodic check + apply
 - Forge/Neoforge: automated 5-step post-install patch sequence
 - Modrinth mod lifecycle: install/enable/disable/delete — fully operational
-- Provenance: `.zlh_metadata.json` — source is `null` if not set, no curated inference currently implemented
+- Provenance: `.zlh_metadata.json` — source is `null` if not set
-- Status transport model remains poll-based (`/status`), not push-based
+- Status transport model: poll-based (`/status`), not push-based
-- `/status` content now includes richer dev/runtime/code-server fields
+- State transitions: `idle`, `installing`, `starting`, `running`, `stopping`, `crashed`, `error`
-- State transitions remain internal to agent: `idle`, `installing`, `starting`, `running`, `stopping`, `crashed`, `error`
 
 ---
 
@@ -90,38 +89,74 @@ access. Pulls runtimes + server jars from zlh-artifacts (VM 1003).
 - code-server port: `6000`
 - agent port: `18888`
 
-Confirmed during current validation:
+Confirmed:
 
-- code-server process launches inside the container
+- code-server process launches and binds to `0.0.0.0:6000`
-- process binds to `0.0.0.0:6000`
+- frontend host/console state updates correctly via API status endpoint
-- Traefik dynamic dev route is created by API during provisioning
-- frontend host/console state now updates correctly after API status endpoint work
-
-Current unresolved item:
-
-- external browser access to code-server through Cloudflare → Traefik → dev container is still being finalized
 
 ---
 
-## API / Proxy Coordination (Current State)
+## Dev Container Access Model
 
-Dev provisioning now includes an additive dev-only routing path.
+Dev containers provide two access methods.
 
-Current behavior:
+### Browser IDE (Primary)
 
-- game publish flow remains untouched
+```
-- if container type is `dev` and `enable_code_server=true`, API:
+Browser
-  - creates Technitium A record
+  ↓
-  - creates Cloudflare A record
+Portal
-  - writes Traefik dynamic config on `zlh-zpack-proxy`
+  ↓
-- remote Traefik file writes use SSH service account `zlh`
+API proxy
-- proxy SSH configuration is now service-account based, not tied to personal user
+  ↓
+container:6000
+```
 
-Frontend/API status behavior:
+URL format: `/dev/<vmid>/ide`
 
-- API still polls agent state
+Benefits:
-- API now exposes server status back to frontend so host/console UI is accurate
+
-- portal is no longer relying on stale DB-only host state for console availability
+- no DNS automation
+- no Traefik routing
+- no TLS per container
+- portal JWT controls access
+- ownership validated by API before proxying
+
+### Local Developer Access (Advanced)
+
+Advanced users connect via **Tailscale / Headscale**.
+
+Benefits:
+
+- SSH
+- VS Code Remote
+- local tooling
+
+Headscale server: `zlh-ctl` (status to be confirmed)
+
+Constraints:
+
+- no exit nodes
+- `magic_dns: false`
+- no DNS takeover on customer machine
+
+### Removed Approach
+
+The previous model used DNS record per container, Traefik dynamic routes,
+and subdomain IDE access. This was removed due to operational complexity,
+TLS issues, and per-container DNS automation overhead.
+
+Code paths to remove from API: `createDevRouting()`, proxy SSH writes,
+Traefik dynamic file creation.
+
+---
+
+## API / Frontend Status
+
+- API polls agent `/status`
+- API exposes polled state back to frontend via server status endpoint
+- Portal no longer relies on stale DB-only state for console availability
+- Game publish flow remains untouched — dev routing changes are additive only
 
 ---
 
@@ -151,10 +186,10 @@ Revenue multiplier: 1 developer → ~10 players → $147.50/mo total.
 
 ## Open Threads
 
-1. External dev IDE access — finalize end-to-end browser reachability for code-server
+1. Implement browser IDE proxy (`/dev/:id/ide`) in API
-2. Curated provenance — tracking install origin (Modrinth, manual, etc.)
+2. Confirm Headscale `zlh-ctl` VM status
-3. Dev routing cleanup — normalize hostname/domain generation and deletion cleanup
+3. Curated provenance — tracking install origin
-4. Optional future: config diff viewer, upload progress UI, upload cancellation, log search
+4. Optional future: config diff viewer, upload progress UI, log search
 
 ---
 
@@ -181,4 +216,5 @@ All at `git.zerolaghub.com/jester/<repo>`
 - Portal never calls agents directly — all traffic through API
 - Upload transport uses raw http.request piping, never fetch()
 - VMs 100, 101, 103, 1000 are legacy/unused — not active production
-- Do not mark unresolved routing or TLS work as complete
+- Do not mark unimplemented work as complete
+- Game publish flow must never be modified by dev routing changes