feat: add guided exploration and share user-started MCP backend

apps/admin_console/services/task_queue_service.py

@@ -25,16 +25,10 @@ import time
 from typing import Any
 import uuid
 
-try:
-    from admin_console.core.state import state
-    from admin_console.database.repositories.session_repository import session_repo
-    from admin_console.services import worker_process_io
-    from admin_console.services.media_service import media_service
-except ImportError:
-    from apps.admin_console.core.state import state
-    from apps.admin_console.database.repositories.session_repository import session_repo
-    from apps.admin_console.services import worker_process_io
-    from apps.admin_console.services.media_service import media_service
+from apps.admin_console.core.state import state
+from apps.admin_console.database.repositories.session_repository import session_repo
+from apps.admin_console.services import worker_process_io
+from apps.admin_console.services.media_service import media_service
 
 from artemis.config import (
     PAUSE_FILE,
@@ -585,6 +579,12 @@ class TaskQueueService:
             env["ARTEMIS_SESSION_ID"] = str(sess_id)
         env["ARTEMIS_TASK_INGRESS"] = str(task_item.get("ingress", "frontend"))
         env["ARTEMIS_TASK_WORKER"] = "1"
+        from artemis.exploration.store import explorations
+
+        exploration_run = explorations.run(str(sess_id)) if sess_id else None
+        env.pop("ARTEMIS_EXPLORATION_ID", None)
+        if exploration_run and exploration_run["kind"] == "exploration":
+            env["ARTEMIS_EXPLORATION_ID"] = exploration_run["exploration_id"]
         overrides = task_item.get("runtime_overrides") or {}
         if overrides.get("recording") is not None:
             env["ARTEMIS_TASK_RECORDING_CONFIG"] = json.dumps(overrides["recording"])
@@ -669,6 +669,11 @@ class TaskQueueService:
         batch = store.get("batch", job["batch_id"]) if job and job.get("batch_id") else None
         limit = task_timeout(task_item, job, batch)
         budget = ExecutionBudget(limit)
+        from artemis.exploration.store import explorations
+
+        exploration_run = explorations.run(sid)
+        is_exploration = bool(exploration_run and exploration_run["kind"] == "exploration")
+        exploration_finalized = False
         cleanup_limit = (task_item.get("runtime_overrides") or {}).get(
             "worker_cleanup_timeout_seconds"
         ) or settings.ARTEMIS_WORKER_CLEANUP_TIMEOUT_SECONDS
@@ -681,6 +686,9 @@ class TaskQueueService:
                     return waiter.result()
                 outcome = read_worker_outcome(sid)
                 if outcome:
+                    if exploration_run and not exploration_finalized:
+                        explorations.finish_run(sid, {**outcome, "status": outcome["task_status"]})
+                        exploration_finalized = True
                     finalized_at = finalized_at if finalized_at is not None else time.monotonic()
                     if time.monotonic() - finalized_at >= cleanup_limit:
                         logger.warning(
@@ -692,6 +700,9 @@ class TaskQueueService:
                         return await waiter
                     continue  # Final result is authoritative; only bound resource cleanup now.
                 cls._sync_control_states()
+                if is_exploration:
+                    # The worker owns its pausable budget so it survives a UI server restart.
+                    continue
                 run = state.active_runs.get(run_key, {})
                 remaining = budget.update(
                     excluded=bool(run.get("parked")) or read_state(sid).get("status") == "queued"
@@ -1077,9 +1088,35 @@ class TaskQueueService:
                         f"[QueueWorker] Could not persist failure status for [{sess_id}]"
                     )
         finally:
-            await cls._finish_output_forwarder(output_task)
-            # 5. Clean up the finished task and release this run's scheduling slot
-            cls._release_run_slot(sess_id, run_key, proc)
+            try:
+                await cls._finish_output_forwarder(output_task)
+                if sess_id:
+                    from artemis.exploration.store import explorations
+                    from artemis.runtime.task_result import read_worker_outcome
+
+                    exploration_run = explorations.run(str(sess_id))
+                    if exploration_run and exploration_run.get("outcome") is None:
+                        from artemis.runtime.case_result import read_case_result
+
+                        outcome = read_worker_outcome(str(sess_id)) or {}
+                        session = session_repo.get_session_by_id(str(sess_id)) or {}
+                        status = (
+                            outcome.get("task_status") or session.get("status") or "interrupted"
+                        )
+                        explorations.finish_run(
+                            str(sess_id),
+                            {
+                                **outcome,
+                                "status": status,
+                                "test_result": outcome.get("test_result")
+                                or read_case_result(str(sess_id), status),
+                            },
+                        )
+            except Exception:
+                logger.exception("Cannot finalize exploration metadata for %s", sess_id)
+            finally:
+                # Metadata/storage failure must never leak a scheduling slot.
+                cls._release_run_slot(sess_id, run_key, proc)
 
     @classmethod
     def _find_duplicate_submission(