orchestrator: restore opencode web launcher

cc-ci-plan/JOURNAL.md

@@ -543,3 +543,18 @@ session cc-ci-orchestrator-stale can be killed; recipe-mirrors org still private
   (c) ghost PR debug: plan-ghostpr-debug-fix.md + memory [[ghost-pr-debug]].
 - NOT switching the upgrade to sequential (operator: concurrency is fine; the leak is the issue).
   Duplicate ghost subagent from the interrupt churn — told the upgrader to TaskStop one.
+
+## 2026-06-12 15:43 UTC — opencode web restored + OpenAI launcher added
+- Re-enabled `opencode-web.service` in `nix/hosts/cc-ci-orchestrator-hetzner/configuration.nix`
+  (`wantedBy = [ "multi-user.target" ]`) and persisted the broader PATH that the old runtime
+  override had been providing. `nixos-rebuild switch --flake .#cc-ci-orchestrator-hetzner`
+  brought the service back and it now passes `curl http://127.0.0.1:4096/global/health`.
+- Added `cc-ci-plan/launch-opencode.sh` as a separate entrypoint that keeps the default
+  Claude `launch-orchestrator.sh` untouched. It rebuilds the host flake if `opencode-web`
+  is not enabled, starts the service if needed, then launches the orchestrator with
+  `LOOP_BACKEND=opencode`, `LOOP_MODEL=openai/gpt-5.4`, and default session name
+  `cc-ci-orchestrator-oc`.
+- Patched `launch-orchestrator.py` so opencode launches can force the requested model even
+  though `opencode attach` has no `--model` flag: it now injects `OPENCODE_CONFIG_CONTENT`
+  for the session. Verified live: `cc-ci-orchestrator-oc` tmux session running on
+  `backend=opencode model=openai/gpt-5.4`, visible through the shared web server.