refactor: enhance engine recovery logic by introducing conditional checks for health status before attempting recovery

Author: AsieduDevelopmentHub

dashboard/lib/providers/engine_provider.dart

@@ -308,7 +308,24 @@ class EngineProvider extends ChangeNotifier {
         DateTime.now().difference(lastKill) < _bundledEngineKillCooldown) {
       return;
     }
-    unawaited(_recoverStalledLiveChannel());
+    unawaited(_maybeRecoverStalledLiveChannel());
+  }
+
+  /// Only treat missing live frames as a hung engine if HTTP health still says the
+  /// engine is up. If the process was killed externally, health is down — let the
+  /// normal reconnect path run instead of [recoverOwnedAfterStall], which uses
+  /// [forceRestart] and can burn through the launcher's restart budget.
+  Future<void> _maybeRecoverStalledLiveChannel() async {
+    if (_recoveringLive) return;
+    try {
+      final j = await _service.getHealth();
+      if (j == null || j['engine'] != true) {
+        return;
+      }
+    } catch (_) {
+      return;
+    }
+    await _recoverStalledLiveChannel();
   }
 
   /// HTTP/WS up but no live telemetry (hung orchestrator): restart bundled engine

dashboard/lib/services/engine_bundled_launcher.dart

@@ -93,7 +93,9 @@ class EngineBundledLauncher {
     // installer, or a different launch path). If HTTP health succeeds, align disk and
     // connect — do not spawn a second engine. Stall recovery uses [forceRestart] and
     // skips this path.
-    if (!forceRestart && await _strictHealth(cfg.host, cfg.port)) {
+    final httpHealthy =
+        !forceRestart && await _strictHealth(cfg.host, cfg.port);
+    if (httpHealthy) {
       final dirty = cfg.status != EngineStatus.running ||
           cfg.lastError.isNotEmpty;
       if (dirty) {
@@ -114,6 +116,18 @@ class EngineBundledLauncher {
       );
     }
 
+    // Disk still says "running" (e.g. Task Manager kill — Python never rewrote the
+    // file). HTTP is down; clear stale state so we skip the long "wait for existing"
+    // window on a dead PID.
+    if (!forceRestart && !httpHealthy && cfg.status == EngineStatus.running) {
+      cfg = cfg.copyWith(
+        status: EngineStatus.stopped,
+        lastError: '',
+        pid: 0,
+      );
+      await EngineConfigStore.writeAtomic(cfg);
+    }
+
     if (cfg.status == EngineStatus.failed && !userRetry) {
       return EngineBootstrapOutcome(
         success: false,