box-service-health: fix host mismatch in template

service.status runs bl-local (.timer -> user bus, .service -> system bus).
board.service/caddy.service are inactive on bl by design -- they run on the
VM and are covered by box-health-check.sh over the SSH chain.
response-harvester.timer lives on the bl user bus, not the VM.

jobs/box-service-health.json

@@ -11,7 +11,7 @@
   },
   "name": "box-service-health",
   "on_failure": "alert",
-  "prompt_template": "Box service and data health check.\nJob ID: {job_id}\nTime: {datetime}\n\nExecute the service check on the VM:\n[TOOL service.status {\"unit\": \"board.service\"}]\n\nCheck caddy and harvester timer status:\n[TOOL service.status {\"unit\": \"response-harvester.timer\"}]\n\nIf healthy, end with [RESULT {job_id}] OK.\nIf any service fails, reply with [RESULT {job_id}] FAIL <summary>.",
+  "prompt_template": "Box service and data health check.\nJob ID: {job_id}\nTime: {datetime}\n\nVM check: run /srv/box/bin/box-health-check.sh on the VM (dev-operator-646@34.139.37.135) via your operator SSH chain. Expect zero failures: services 3/3 (board.service, caddy.service, box-request-sweeper.timer), data 3/3, HTTP checks all OK. The request-store WARN is known and not a failure.\n\nbl check: the result harvester lives on bl, not on the VM:\n[TOOL service.status {\"unit\": \"response-harvester.timer\"}]\nExpect active.\n\nNote: board.service and caddy.service are inactive on bl by design (they run on the VM) -- do not check them with the tool; the VM health script covers them.\n\nIf healthy, end with [RESULT {job_id}] OK.\nIf any service fails, reply with [RESULT {job_id}] FAIL <summary>.",
   "schedule": "7,22,37,52 * * * *",
   "sidechat": {
     "create": true,