@rynx-ai/runtime 0.1.11-beta.20 → 0.1.11-beta.22

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/host.js CHANGED
@@ -247,7 +247,13 @@ export class LocalAgentHost {
247
247
  // multi-connection model). Defaults to an ExternalWsChannel client attached to
248
248
  // the backend's app-server; tests inject a fake.
249
249
  forwarderClientFactory;
250
- observerReconnectTimeoutMs;
250
+ // Builds Omnigent's short-lived cold-resume preload connection. This must be
251
+ // distinct from both the app-server owner/injector and the background
252
+ // observer: the owner's initialize handshake is the readiness probe, then a
253
+ // freshly initialized connection performs the one resume and closes.
254
+ preloadClientFactory;
255
+ /** Builds one short-lived message/interrupt client. */
256
+ injectionClientFactory;
251
257
  // Keyed by `codexBackendKey` — the bare runtime id for budget-less agents, or
252
258
  // a `${runtime}::${retryHash}` composite for a per-agent retry budget.
253
259
  backends = new Map();
@@ -281,7 +287,7 @@ export class LocalAgentHost {
281
287
  /** Short-lived dedupe for managed fork notifications delivered after the
282
288
  * `thread/fork` response. Values are expected source Provider thread ids. */
283
289
  managedForkThreadStarts = new Map();
284
- constructor({ config, commandRunner, sessionStore = new FileCodexSessionStore(resolveCodexSessionStorePath(config)), allowedRoots = resolveAllowedRoots(config), appServerClient, now = () => Date.now(), backendIdleTtlMs = 300_000, forwarderClientFactory, observerReconnectTimeoutMs = 15_000, sessionId, runtimeHomeSessionId, }) {
290
+ constructor({ config, commandRunner, sessionStore = new FileCodexSessionStore(resolveCodexSessionStorePath(config)), allowedRoots = resolveAllowedRoots(config), appServerClient, now = () => Date.now(), backendIdleTtlMs = 300_000, forwarderClientFactory, preloadClientFactory, injectionClientFactory, sessionId, runtimeHomeSessionId, }) {
285
291
  this.config = config;
286
292
  this.sessionId = sessionId ?? "__default__";
287
293
  this.runtimeHomeSessionId = runtimeHomeSessionId ?? this.sessionId;
@@ -292,10 +298,10 @@ export class LocalAgentHost {
292
298
  this.injectedAppServerClient = appServerClient;
293
299
  this.clock = now;
294
300
  this.backendIdleTtlMs = backendIdleTtlMs;
295
- this.observerReconnectTimeoutMs = observerReconnectTimeoutMs;
296
- this.forwarderClientFactory =
297
- forwarderClientFactory ??
298
- ((appServerUrl) => new CodexAppServerClient({ channel: new ExternalWsChannel(appServerUrl) }));
301
+ const defaultAttachedClientFactory = (appServerUrl) => new CodexAppServerClient({ channel: new ExternalWsChannel(appServerUrl) });
302
+ this.forwarderClientFactory = forwarderClientFactory ?? defaultAttachedClientFactory;
303
+ this.preloadClientFactory = preloadClientFactory ?? defaultAttachedClientFactory;
304
+ this.injectionClientFactory = injectionClientFactory ?? defaultAttachedClientFactory;
299
305
  }
300
306
  getBackend(runtime, budget) {
301
307
  this.reapIdleBackends();
@@ -412,36 +418,7 @@ export class LocalAgentHost {
412
418
  if (live.settledInteractions.has(interactionId)) {
413
419
  return { disposition: "already_resolved" };
414
420
  }
415
- if (live.interactionRecoveries.has(interactionId)) {
416
- return {
417
- disposition: "invalid",
418
- message: "interaction delivery is recovering; retry shortly",
419
- };
420
- }
421
- const submit = (client) => {
422
- // Store before entering the client: a transport can report a synchronous
423
- // write failure, and failover must already have the exact accepted answer
424
- // available for replay. A synchronous terminal event may delete it again.
425
- const hadPrevious = live.interactionSubmissions.has(interactionId);
426
- const previous = live.interactionSubmissions.get(interactionId);
427
- live.interactionSubmissions.set(interactionId, resolution);
428
- const result = client.resolveInteraction(interactionId, resolution);
429
- if (result.disposition !== "applied") {
430
- if (hadPrevious && previous)
431
- live.interactionSubmissions.set(interactionId, previous);
432
- else
433
- live.interactionSubmissions.delete(interactionId);
434
- }
435
- return result;
436
- };
437
- const owner = live.interactionOwners.get(interactionId);
438
- if (owner)
439
- return submit(owner);
440
- const injected = submit(live.injectClient);
441
- if (injected.disposition !== "not_found") {
442
- return injected;
443
- }
444
- return submit(live.forwarderClient);
421
+ return live.forwarderClient.resolveInteraction(interactionId, resolution);
445
422
  }
446
423
  /**
447
424
  * The command to run in a session's live terminal so it co-drives the codex
@@ -596,7 +573,7 @@ export class LocalAgentHost {
596
573
  snapshotSkills = await this.prepareExecutionSkills(execution);
597
574
  }
598
575
  catch (err) {
599
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} skill preparation failed: ${err instanceof Error ? err.message : String(err)}`);
576
+ this.liveStartupErrors.set(localThreadId, nativeLiveFailure(runtime, "native_skill_preparation_failed", "skill preparation failed", err));
600
577
  console.error(`[session-snapshot] session=${localThreadId} skill materialization failed: ${err instanceof Error ? err.message : String(err)}`);
601
578
  return false;
602
579
  }
@@ -609,7 +586,7 @@ export class LocalAgentHost {
609
586
  }
610
587
  catch (err) {
611
588
  void snapshotSkills.skillsCleanup();
612
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} skill persistence failed: ${err instanceof Error ? err.message : String(err)}`);
589
+ this.liveStartupErrors.set(localThreadId, nativeLiveFailure(runtime, "native_skill_persistence_failed", "skill persistence failed", err));
613
590
  console.error(`[session-snapshot] session=${localThreadId} skill persistence failed: ${err instanceof Error ? err.message : String(err)}`);
614
591
  return false;
615
592
  }
@@ -627,37 +604,45 @@ export class LocalAgentHost {
627
604
  void snapshotSkills.skillsCleanup();
628
605
  return false;
629
606
  };
630
- const injectClient = this.getBackend(runtime, execution.budget ?? undefined).appServerClient;
631
- if (!injectClient) {
632
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} app-server is unavailable for this Session`);
607
+ const appServerOwner = this.getBackend(runtime, execution.budget ?? undefined).appServerClient;
608
+ if (!appServerOwner) {
609
+ this.liveStartupErrors.set(localThreadId, nativeLiveFailure(runtime, "native_app_server_unavailable", "app-server is unavailable for this Session"));
633
610
  console.error(`[codex-live] session=${localThreadId} runtime=${runtime} no app-server client`);
634
611
  return abandon();
635
612
  }
636
613
  try {
637
- await injectClient.ensureInitialized();
614
+ await appServerOwner.ensureInitialized();
638
615
  }
639
616
  catch (err) {
640
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} app-server initialization failed during its 10s readiness phase: ${err instanceof Error ? err.message : String(err)}`);
617
+ this.liveStartupErrors.set(localThreadId, nativeLiveFailure(runtime, "native_app_server_start_failed", "app-server initialization failed during its 10s readiness phase", err));
641
618
  console.error(`[codex-live] session=${localThreadId} runtime=${runtime} app-server init failed: ${err instanceof Error ? err.message : String(err)}`);
642
619
  return abandon();
643
620
  }
644
- const appServerUrl = injectClient.terminalRemoteUrl();
621
+ const appServerUrl = appServerOwner.terminalRemoteUrl();
645
622
  if (!appServerUrl) {
646
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} app-server started without a Terminal remote endpoint`);
623
+ this.liveStartupErrors.set(localThreadId, nativeLiveFailure(runtime, "native_app_server_endpoint_missing", "app-server started without a Terminal remote endpoint"));
647
624
  console.error(`[codex-live] session=${localThreadId} runtime=${runtime} app-server has no remote url`);
648
625
  return abandon(); // needs the ws app-server (live-terminal mode)
649
626
  }
650
627
  // reference implementation multi-connection: a SEPARATE forwarder connection to the SAME
651
628
  // app-server, which `thread/resume`s to subscribe to the thread's item/turn
652
629
  // notifications (the backend client injects; this one only observes).
630
+ // Fresh sessions must connect this listener before the TUI launches so its
631
+ // one-shot `thread/started` cannot race discovery. A known-thread cold resume
632
+ // already has its durable id: like Omnigent's `_codex_forward_known_thread`,
633
+ // its observer starts in the background only after the replacement TUI has
634
+ // launched and is never part of the resume admission boundary.
653
635
  const forwarderClient = this.forwarderClientFactory(appServerUrl);
654
- try {
655
- await forwarderClient.ensureInitialized();
656
- }
657
- catch (err) {
658
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} observer could not attach to the ready app-server: ${err instanceof Error ? err.message : String(err)}`);
659
- console.error(`[codex-live] session=${localThreadId} runtime=${runtime} forwarder init failed: ${err instanceof Error ? err.message : String(err)}`);
660
- return abandon();
636
+ const coldResume = Boolean(record?.codexSessionId);
637
+ if (!coldResume) {
638
+ try {
639
+ await forwarderClient.ensureInitialized();
640
+ }
641
+ catch (err) {
642
+ this.liveStartupErrors.set(localThreadId, nativeLiveFailure(runtime, "native_fresh_listener_failed", "fresh-thread listener could not attach to the ready app-server", err));
643
+ console.error(`[codex-live] session=${localThreadId} runtime=${runtime} fresh-thread listener init failed: ${err instanceof Error ? err.message : String(err)}`);
644
+ return abandon();
645
+ }
661
646
  }
662
647
  const model = execution.model ?? "";
663
648
  const reasoningEffort = execution.reasoningEffort ?? undefined;
@@ -674,7 +659,8 @@ export class LocalAgentHost {
674
659
  markTerminalReady = resolve;
675
660
  });
676
661
  const live = {
677
- injectClient,
662
+ appServerOwner,
663
+ appServerUrl,
678
664
  forwarderClient,
679
665
  forwarder: null,
680
666
  runtime,
@@ -698,16 +684,10 @@ export class LocalAgentHost {
698
684
  subscribing: false,
699
685
  rotationPending: false,
700
686
  stopped: false,
701
- observerAvailable: true,
687
+ observerAvailable: !coldResume,
702
688
  skillsCleanup: snapshotSkills.skillsCleanup,
703
- interactionOwners: new Map(),
704
- interactionStandbys: new Map(),
705
- interactionSubmissions: new Map(),
706
689
  canonicalInteractions: new Set(),
707
690
  settledInteractions: new Set(),
708
- interactionRecoveries: new Map(),
709
- interactionFailedClients: new Map(),
710
- disconnectedClients: new Set(),
711
691
  };
712
692
  let currentSessionId = localThreadId;
713
693
  let pendingRotationEvents = null;
@@ -795,168 +775,81 @@ export class LocalAgentHost {
795
775
  for (const se of n.next(agentEvent))
796
776
  emitCurrent(se);
797
777
  };
798
- const clearRecovery = (interactionId) => {
799
- live.interactionRecoveries.delete(interactionId);
800
- };
801
- const clearInteractionRouting = (interactionId) => {
802
- clearRecovery(interactionId);
803
- live.interactionOwners.delete(interactionId);
804
- live.interactionStandbys.delete(interactionId);
805
- live.interactionSubmissions.delete(interactionId);
806
- live.interactionFailedClients.delete(interactionId);
807
- };
808
- const markInteractionClientFailed = (interactionId, client) => {
809
- const failed = live.interactionFailedClients.get(interactionId) ?? new Set();
810
- failed.add(client);
811
- live.interactionFailedClients.set(interactionId, failed);
812
- };
813
- const interactionClientUnavailable = (interactionId, client) => live.disconnectedClients.has(client) ||
814
- live.interactionFailedClients.get(interactionId)?.has(client) === true;
815
- const allInteractionClientsUnavailable = (interactionId) => [live.injectClient, live.forwarderClient].every((client) => interactionClientUnavailable(interactionId, client));
816
- const finishRecovery = (interactionId, event) => {
817
- clearInteractionRouting(interactionId);
818
- forwardInteraction(event);
819
- };
820
- const beginRecovery = (interactionId, event) => {
821
- live.interactionOwners.delete(interactionId);
822
- live.interactionRecoveries.set(interactionId, { event });
823
- };
824
- const promoteInteractionOwner = (interactionId, replacement) => {
825
- const recovery = live.interactionRecoveries.get(interactionId);
826
- live.interactionOwners.set(interactionId, replacement);
827
- const submission = live.interactionSubmissions.get(interactionId);
828
- if (!submission) {
829
- clearRecovery(interactionId);
830
- return true;
831
- }
832
- const replayed = replacement.resolveInteraction(interactionId, submission);
833
- // A write failure can synchronously re-enter the listener and remove this
834
- // owner. Only clear recovery when the replacement still owns the request.
835
- if (replayed.disposition === "applied" &&
836
- live.interactionOwners.get(interactionId) === replacement) {
837
- clearRecovery(interactionId);
838
- return true;
839
- }
840
- if (live.interactionOwners.get(interactionId) === replacement) {
841
- live.interactionOwners.delete(interactionId);
842
- }
843
- if (recovery &&
844
- live.canonicalInteractions.has(interactionId) &&
845
- !live.settledInteractions.has(interactionId) &&
846
- !live.interactionRecoveries.has(interactionId)) {
847
- live.interactionRecoveries.set(interactionId, recovery);
848
- }
849
- return false;
850
- };
851
- const bindInteractionClient = (client) => {
852
- client.setInteractionListener((event) => {
853
- const interactionId = event.type === "requested"
854
- ? event.request.interactionId
855
- : event.interactionId;
856
- const owner = live.interactionOwners.get(interactionId);
857
- if (event.type === "requested") {
858
- // Receiving a native request proves this connection is alive and can
859
- // be retried even if an earlier copy failed on it.
860
- live.disconnectedClients.delete(client);
861
- const failed = live.interactionFailedClients.get(interactionId);
862
- failed?.delete(client);
863
- if (failed?.size === 0)
864
- live.interactionFailedClients.delete(interactionId);
865
- if (live.settledInteractions.has(interactionId))
866
- return;
867
- const recovery = live.interactionRecoveries.get(interactionId);
868
- if (recovery) {
869
- promoteInteractionOwner(interactionId, client);
870
- return;
871
- }
872
- if (owner && owner !== client) {
873
- const standbys = live.interactionStandbys.get(interactionId) ?? new Set();
874
- standbys.add(client);
875
- live.interactionStandbys.set(interactionId, standbys);
876
- return;
877
- }
878
- live.interactionOwners.set(interactionId, client);
879
- forwardInteraction(event);
880
- return;
881
- }
882
- const recoverableOwnerFailure = event.type === "cancelled" &&
883
- (event.reason === "app_server_disconnected" || event.reason === "response_write_failed");
884
- if (recoverableOwnerFailure)
885
- markInteractionClientFailed(interactionId, client);
886
- if (!owner || owner !== client) {
887
- const standbys = live.interactionStandbys.get(interactionId);
888
- standbys?.delete(client);
889
- if (standbys?.size === 0)
890
- live.interactionStandbys.delete(interactionId);
891
- const recovery = live.interactionRecoveries.get(interactionId);
892
- if (recoverableOwnerFailure &&
893
- live.canonicalInteractions.has(interactionId) &&
894
- allInteractionClientsUnavailable(interactionId)) {
895
- finishRecovery(interactionId, recovery?.event ?? event);
896
- }
897
- return;
898
- }
899
- if (recoverableOwnerFailure) {
900
- const standbys = live.interactionStandbys.get(interactionId);
901
- while (standbys && standbys.size > 0) {
902
- const replacement = standbys.values().next().value;
903
- if (!replacement)
904
- break;
905
- standbys.delete(replacement);
906
- if (interactionClientUnavailable(interactionId, replacement))
907
- continue;
908
- if (promoteInteractionOwner(interactionId, replacement)) {
909
- if (standbys.size === 0)
910
- live.interactionStandbys.delete(interactionId);
911
- return;
912
- }
913
- if (live.settledInteractions.has(interactionId))
914
- return;
915
- }
916
- if (!live.settledInteractions.has(interactionId)) {
917
- if (allInteractionClientsUnavailable(interactionId)) {
918
- finishRecovery(interactionId, event);
919
- }
920
- else {
921
- beginRecovery(interactionId, event);
922
- }
923
- return;
924
- }
925
- }
926
- clearInteractionRouting(interactionId);
927
- forwardInteraction(event);
928
- });
778
+ const bindObserverClient = (client) => {
779
+ client.setInteractionListener(forwardInteraction);
929
780
  client.setConnectionListener((state) => {
930
- if (state === "connected") {
931
- // Observer transport recovery alone does not prove that an in-flight
932
- // interaction request was replayed onto this connection. Its request
933
- // listener clears disconnectedClients when that duplicate arrives.
934
- if (client !== live.forwarderClient)
935
- live.disconnectedClients.delete(client);
781
+ if (state === "connected" || live.stopped)
936
782
  return;
783
+ live.observerAvailable = false;
784
+ for (const interactionId of [...live.canonicalInteractions]) {
785
+ forwardInteraction({
786
+ type: "cancelled",
787
+ interactionId,
788
+ reason: "app_server_disconnected",
789
+ });
937
790
  }
938
- live.disconnectedClients.add(client);
939
- if (client === live.forwarderClient && !live.stopped) {
940
- live.observerAvailable = false;
941
- this.armObserverReconnectDeadline(live);
942
- void this.reconnectForwarder(live);
943
- }
944
- for (const [interactionId, recovery] of [...live.interactionRecoveries]) {
945
- if (allInteractionClientsUnavailable(interactionId)) {
946
- finishRecovery(interactionId, recovery.event);
947
- }
948
- }
791
+ this.failObserverLifecycle(localThreadId, live, nativeLiveFailure(runtime, "native_observer_disconnected", "observer connection closed"));
949
792
  });
950
793
  };
951
794
  const closeCanonicalInteractions = () => {
952
795
  for (const interactionId of live.canonicalInteractions) {
953
796
  rememberSettledInteraction(interactionId);
954
- clearInteractionRouting(interactionId);
955
797
  }
956
798
  live.canonicalInteractions.clear();
957
799
  };
800
+ const completeCurrentTurn = (usage, reason) => {
801
+ closeCanonicalInteractions();
802
+ if (!normalizer)
803
+ return;
804
+ const completedResponseId = currentResponseId;
805
+ if (usage) {
806
+ for (const se of normalizer.next({ type: "turn_completed", usage }))
807
+ emitCurrent(se);
808
+ }
809
+ for (const se of normalizer.next({ type: "done" })) {
810
+ if (reason === "superseded" && se.type === "session.status")
811
+ continue;
812
+ emitCurrent(se);
813
+ }
814
+ clearPendingInputsForResponse(completedResponseId);
815
+ normalizer = null;
816
+ currentResponseId = null;
817
+ };
818
+ const failCurrentTurn = (error) => {
819
+ closeCanonicalInteractions();
820
+ if (!normalizer)
821
+ return;
822
+ const failedResponseId = currentResponseId;
823
+ const responseStopped = error.message === "Codex turn was interrupted";
824
+ const providerName = runtime === "traex" ? "Traex" : "Codex";
825
+ const message = responseStopped
826
+ ? "Response stopped"
827
+ : runtime === "traex"
828
+ ? error.message.replace(/^Codex\b/, providerName)
829
+ : error.message;
830
+ for (const se of normalizer.fail({
831
+ code: responseStopped
832
+ ? "response_stopped"
833
+ : runtime === "traex" ? "traex_error" : "codex_error",
834
+ message,
835
+ source: "execution",
836
+ })) {
837
+ emitCurrent(se.type === "session.status"
838
+ ? { ...se, status: "failed", note: message }
839
+ : se);
840
+ }
841
+ clearPendingInputsForResponse(failedResponseId);
842
+ normalizer = null;
843
+ currentResponseId = null;
844
+ };
958
845
  const sink = {
959
846
  onTurnStart: (turnId) => startNormalizer(turnId),
847
+ onTurnObserved: (turnId) => {
848
+ const n = startNormalizer(turnId);
849
+ for (const event of n.next({ type: "turn_started", ...(turnId ? { turnId } : {}) })) {
850
+ emitCurrent(event);
851
+ }
852
+ },
960
853
  onUserMessage: (content) => {
961
854
  const normalizedContent = typeof content === "string"
962
855
  ? [{ type: "input_text", text: content }]
@@ -990,45 +883,28 @@ export class LocalAgentHost {
990
883
  ...(statusKind ? { statusKind } : {}),
991
884
  });
992
885
  },
993
- onTurnEnd: (usage) => {
994
- closeCanonicalInteractions();
995
- if (!normalizer)
886
+ // A newer authoritative turn/started completes the old canonical
887
+ // response without publishing a false idle edge in a running→running
888
+ // transition.
889
+ onTurnEnd: completeCurrentTurn,
890
+ onRecoveredTurnStatus: (status, turnId, error) => {
891
+ const responseId = turnId ? `resp_codex_${turnId}` : undefined;
892
+ if (normalizer && (!responseId || currentResponseId === responseId)) {
893
+ if (status === "failed")
894
+ failCurrentTurn(error ?? new Error("Codex turn failed"));
895
+ else
896
+ completeCurrentTurn();
996
897
  return;
997
- const completedResponseId = currentResponseId;
998
- if (usage) {
999
- for (const se of normalizer.next({ type: "turn_completed", usage }))
1000
- emitCurrent(se);
1001
898
  }
1002
- for (const se of normalizer.next({ type: "done" }))
1003
- emitCurrent(se);
1004
- clearPendingInputsForResponse(completedResponseId);
1005
- normalizer = null;
1006
- currentResponseId = null;
1007
- },
1008
- onTurnError: (error) => {
1009
- closeCanonicalInteractions();
1010
- if (!normalizer)
1011
- return;
1012
- const failedResponseId = currentResponseId;
1013
- const responseStopped = error.message === "Codex turn was interrupted";
1014
- const providerName = runtime === "traex" ? "Traex" : "Codex";
1015
- const message = responseStopped
1016
- ? "Response stopped"
1017
- : runtime === "traex"
1018
- ? error.message.replace(/^Codex\b/, providerName)
1019
- : error.message;
1020
- for (const se of normalizer.fail({
1021
- code: responseStopped
1022
- ? "response_stopped"
1023
- : runtime === "traex" ? "traex_error" : "codex_error",
1024
- message,
1025
- source: "execution",
1026
- }))
1027
- emitCurrent(se);
1028
- clearPendingInputsForResponse(failedResponseId);
1029
- normalizer = null;
1030
- currentResponseId = null;
899
+ emitCurrent({
900
+ type: "session.status",
901
+ sessionId: currentSessionId,
902
+ ...(responseId ? { responseId } : {}),
903
+ status,
904
+ ...(error ? { note: error.message } : {}),
905
+ });
1031
906
  },
907
+ onTurnError: failCurrentTurn,
1032
908
  // A resumed TUI may rebroadcast `thread/started`; binding is idempotent.
1033
909
  shouldIgnoreThreadStarted: (threadId, forkedFromId) => this.shouldIgnoreManagedForkThreadStarted(live, threadId, forkedFromId),
1034
910
  onThreadStarted: (threadId, forkedFromId) => this.onLiveThreadStarted(live, localThreadId, threadId, forkedFromId),
@@ -1089,7 +965,7 @@ export class LocalAgentHost {
1089
965
  live.rotationPending = false;
1090
966
  for (const event of queued)
1091
967
  emit(event);
1092
- void this.subscribeUntilReady(live, threadId);
968
+ void this.subscribeUntilReady(newSessionId, live, threadId);
1093
969
  })
1094
970
  .catch(() => {
1095
971
  pendingRotationEvents = null;
@@ -1097,8 +973,7 @@ export class LocalAgentHost {
1097
973
  live.stopped = true;
1098
974
  });
1099
975
  };
1100
- bindInteractionClient(injectClient);
1101
- bindInteractionClient(forwarderClient);
976
+ bindObserverClient(forwarderClient);
1102
977
  this.liveSessions.set(localThreadId, live);
1103
978
  forwarder.start();
1104
979
  // Existing bindings resume through the public app-server API. A fresh thread
@@ -1108,57 +983,53 @@ export class LocalAgentHost {
1108
983
  // turn-less app-server thread resumable by the TUI.
1109
984
  try {
1110
985
  if (record?.codexSessionId) {
1111
- let resumeError;
1112
- let resumedThreadId;
1113
- const attempts = 20;
1114
- for (let attempt = 0; attempt < attempts; attempt += 1) {
986
+ // Match Omnigent's cold-resume preload exactly: one history-free
987
+ // `thread/resume` on a short-lived, separately initialized connection.
988
+ // The backend owner's handshake above is the app-server readiness
989
+ // probe; reusing that first connection here races Codex's startup state
990
+ // and can return `Not initialized`. A failure stays attached to the
991
+ // original binding and is reported by this phase; it is not polled.
992
+ const preloadClient = this.preloadClientFactory(appServerUrl);
993
+ try {
994
+ let resumed;
1115
995
  try {
1116
- const resumed = await injectClient.threadResume({
996
+ await preloadClient.ensureInitialized();
997
+ resumed = await preloadClient.threadResume({
1117
998
  threadId: record.codexSessionId,
1118
999
  ...threadWorkspaceParams(runtime, workspace, sandbox),
1119
1000
  approvalPolicy,
1120
- // The injection connection only needs to bind/subcribe. Codex
1121
- // 0.146 can ignore initialTurnsPage and return the whole rollout,
1122
- // so explicitly suppress history here; it is never replayed.
1123
1001
  excludeTurns: true,
1124
- initialTurnsPage: {
1125
- limit: 1,
1126
- sortDirection: "desc",
1127
- itemsView: "summary",
1128
- },
1129
1002
  });
1130
- resumedThreadId = resumed.threadId;
1131
- break;
1132
1003
  }
1133
- catch (error) {
1134
- resumeError = error;
1135
- if (!isRetryableThreadResumeError(error))
1136
- throw error;
1137
- if (attempt + 1 < attempts) {
1138
- await new Promise((resolve) => setTimeout(resolve, 150));
1139
- }
1004
+ finally {
1005
+ await preloadClient.stop().catch((error) => {
1006
+ console.warn(`[codex-live] session=${localThreadId} runtime=${runtime} preload client cleanup failed: ${error instanceof Error ? error.message : String(error)}`);
1007
+ });
1140
1008
  }
1009
+ forwarder.noteThreadBound(resumed.threadId);
1010
+ this.onLiveThreadStarted(live, localThreadId, resumed.threadId);
1141
1011
  }
1142
- if (resumedThreadId) {
1143
- this.onLiveThreadStarted(live, localThreadId, resumedThreadId);
1144
- }
1145
- else if (resumeError &&
1146
- isThreadNotReadyError(resumeError) &&
1147
- !record.parentSessionId &&
1148
- !record.runtimeHomeOwnerSessionId) {
1149
- // A native thread that still has no rollout after the bounded retry
1150
- // window has never materialized a Turn. It is safe to forget this
1151
- // ordinary binding and let the TUI create a fresh native thread.
1152
- // Fork bindings stay strict because their Provider-native ancestry
1153
- // must not be silently replaced.
1012
+ catch (error) {
1013
+ if (!isThreadNotReadyError(error) ||
1014
+ record.parentSessionId ||
1015
+ record.runtimeHomeOwnerSessionId) {
1016
+ throw error;
1017
+ }
1018
+ // Preserve the existing turn-less binding recovery without polling:
1019
+ // an ordinary native id with no rollout/turn has no Provider history
1020
+ // to retain. Fork bindings remain strict, and `thread not found` is
1021
+ // not classified here because it does not prove the history is empty.
1154
1022
  await this.sessionStore.delete(localThreadId);
1155
1023
  live.threadId = null;
1024
+ try {
1025
+ await forwarderClient.ensureInitialized();
1026
+ }
1027
+ catch (listenerError) {
1028
+ throw nativeLiveFailure(runtime, "native_fresh_listener_failed", "fresh-thread listener could not attach after the turn-less binding was replaced", listenerError);
1029
+ }
1030
+ live.observerAvailable = true;
1156
1031
  live.markTerminalReady(true);
1157
- }
1158
- else {
1159
- // A delayed/missing native index or a fork binding remains explicit:
1160
- // neither is proof that replacing Provider history is safe.
1161
- throw resumeError;
1032
+ console.warn(`[codex-live] session=${localThreadId} runtime=${runtime} replaced turn-less native binding ${record.codexSessionId} with fresh discovery`);
1162
1033
  }
1163
1034
  }
1164
1035
  else {
@@ -1173,7 +1044,10 @@ export class LocalAgentHost {
1173
1044
  live.stopped = true;
1174
1045
  forwarder.stop();
1175
1046
  void forwarderClient.stop().catch(() => undefined);
1176
- this.liveStartupErrors.set(localThreadId, `${runtime === "traex" ? "Traex" : "Codex"} native thread binding failed: ${err instanceof Error ? err.message : String(err)}`);
1047
+ const failure = err instanceof CodexRuntimeError
1048
+ ? err
1049
+ : nativeLiveFailure(runtime, "native_resume_failed", "existing native session could not be resumed", err);
1050
+ this.liveStartupErrors.set(localThreadId, failure);
1177
1051
  console.error(`[codex-live] session=${localThreadId} runtime=${runtime} thread bind failed: ${err instanceof Error ? err.message : String(err)}`);
1178
1052
  return abandon();
1179
1053
  }
@@ -1213,7 +1087,57 @@ export class LocalAgentHost {
1213
1087
  })
1214
1088
  .catch(() => undefined);
1215
1089
  live.markReady(); // thread id known → injection can turn/start
1216
- void this.subscribeUntilReady(live, threadId).then(live.markTerminalReady);
1090
+ if (live.observerAvailable) {
1091
+ void this.subscribeUntilReady(localThreadId, live, threadId).then(live.markTerminalReady);
1092
+ }
1093
+ else {
1094
+ // Known-thread cold resume has already completed the dedicated preload. Its
1095
+ // observer is deliberately post-TUI background work, not Terminal
1096
+ // readiness. RunnerSession calls startLiveCodexObserver after launch.
1097
+ live.markTerminalReady(true);
1098
+ }
1099
+ }
1100
+ /** Start the known-thread observer after the replacement TUI has launched.
1101
+ * Fresh sessions already connected their discovery listener before launch;
1102
+ * this is therefore an idempotent no-op for them and for a healthy observer. */
1103
+ startLiveCodexObserver(localThreadId) {
1104
+ const live = this.liveSessions.get(localThreadId);
1105
+ if (!live ||
1106
+ live.stopped ||
1107
+ live.observerAvailable ||
1108
+ !live.threadId)
1109
+ return;
1110
+ void (async () => {
1111
+ try {
1112
+ await live.forwarderClient.ensureInitialized();
1113
+ if (live.stopped)
1114
+ return;
1115
+ live.observerAvailable = true;
1116
+ // Omnigent starts subscription as a sibling task after the transport
1117
+ // connects. A no/empty rollout remains event-driven retryable; any
1118
+ // other subscription rejection is diagnostic only and must not tear
1119
+ // down the otherwise usable Terminal/injection lifecycle.
1120
+ void this.subscribeUntilReady(localThreadId, live, live.threadId);
1121
+ }
1122
+ catch (error) {
1123
+ this.failObserverLifecycle(localThreadId, live, nativeLiveFailure(live.runtime, "native_observer_connect_failed", "observer could not attach to the ready app-server", error));
1124
+ }
1125
+ })();
1126
+ }
1127
+ /** Omnigent treats the forwarder as a required component of one native
1128
+ * lifecycle: if its transport dies, it closes the app-server instead of
1129
+ * accepting turns that can no longer reach the canonical mirror. */
1130
+ failObserverLifecycle(localThreadId, live, error) {
1131
+ if (live.stopped)
1132
+ return;
1133
+ live.startupError = error;
1134
+ this.liveStartupErrors.set(localThreadId, error);
1135
+ live.forwarder.failOpenTurn(error);
1136
+ const appServerOwner = live.appServerOwner;
1137
+ this.stopLiveCodexSession(localThreadId);
1138
+ void appServerOwner.stop().catch((stopError) => {
1139
+ console.warn(`[codex-live] session=${localThreadId} runtime=${live.runtime} app-server cleanup after observer failure failed: ${stopError instanceof Error ? stopError.message : String(stopError)}`);
1140
+ });
1217
1141
  }
1218
1142
  shouldIgnoreManagedForkThreadStarted(live, threadId, forkedFromId) {
1219
1143
  const pending = live.managedFork;
@@ -1237,37 +1161,36 @@ export class LocalAgentHost {
1237
1161
  * Subscribe the forwarder connection to a thread (reference implementation's
1238
1162
  * `_subscribe_until_ready`). A fresh TUI thread has no rollout until its first
1239
1163
  * turn, so `thread/resume` is retried: park until the forwarder observes the
1240
- * thread active, then retry. Resume only fetches the newest summarized Turn:
1241
- * recovery reconciles the exact active Turn's terminal status and never replays
1242
- * historical items. Once resume succeeds, subsequent turns arrive live.
1164
+ * thread active, then retry. Like Omnigent, the first attempt always uses
1165
+ * `excludeTurns`: a known-session cold resume therefore never reconstructs
1166
+ * historical running/completed state. Only a fresh thread whose first attempt
1167
+ * failed as not-ready retries without `excludeTurns`, backfilling the newly
1168
+ * materialized first turn and de-duplicating it against live notifications.
1243
1169
  */
1244
- async subscribeUntilReady(live, threadId) {
1170
+ async subscribeUntilReady(localThreadId, live, threadId) {
1171
+ let sawNotReady = false;
1245
1172
  while (!live.stopped) {
1246
1173
  try {
1247
1174
  const resp = await live.forwarderClient.threadResume({
1248
1175
  threadId,
1249
1176
  ...threadWorkspaceParams(live.runtime, live.workspace, live.sandbox),
1250
1177
  approvalPolicy: live.approvalPolicy,
1251
- initialTurnsPage: {
1252
- limit: 1,
1253
- sortDirection: "desc",
1254
- itemsView: "summary",
1255
- },
1178
+ ...(!sawNotReady ? { excludeTurns: true } : {}),
1256
1179
  });
1257
- // Codex 0.146.1 accepts initialTurnsPage but can still return the entire
1258
- // rollout in chronological order. Never infer newest from array position:
1259
- // recovery may settle only the exact active turn id already owned here.
1260
- const activeTurnId = live.forwarder.currentTurnId();
1261
- const activeTurn = activeTurnId && Array.isArray(resp.thread.turns)
1262
- ? resp.thread.turns.find((turn) => (turn.id ?? turn.turnId) === activeTurnId)
1263
- : undefined;
1264
- live.forwarder.reconcileActiveTurn(activeTurn);
1180
+ if (sawNotReady) {
1181
+ const turns = Array.isArray(resp.thread.turns)
1182
+ ? resp.thread.turns
1183
+ : [];
1184
+ live.forwarder.replayBackfill(turns);
1185
+ }
1265
1186
  return true; // subscribed — live item/turn notifications now flow to the forwarder
1266
1187
  }
1267
1188
  catch (error) {
1268
1189
  if (!isThreadNotReadyError(error)) {
1269
- return false; // other failure — injection still works via the backend client
1190
+ console.warn(`[codex-live] session=${localThreadId} runtime=${live.runtime} observer failed to subscribe to thread ${threadId}: ${error instanceof Error ? error.message : String(error)}`);
1191
+ return false;
1270
1192
  }
1193
+ sawNotReady = true;
1271
1194
  // Park until the thread goes active (its first turn materializes the
1272
1195
  // rollout); a short poll covers the flush race after "active".
1273
1196
  await new Promise((resolve) => {
@@ -1279,54 +1202,6 @@ export class LocalAgentHost {
1279
1202
  }
1280
1203
  return false;
1281
1204
  }
1282
- /** Restore the independent observer after an unexpected exit. The active turn
1283
- * remains open during the bounded grace so resume can reconcile its exact id. */
1284
- reconnectForwarder(live) {
1285
- if (live.observerReconnect)
1286
- return live.observerReconnect;
1287
- const reconnect = (async () => {
1288
- while (!live.stopped && live.threadId) {
1289
- try {
1290
- await live.forwarderClient.ensureInitialized();
1291
- if (await this.subscribeUntilReady(live, live.threadId)) {
1292
- live.observerAvailable = true;
1293
- this.clearObserverReconnectDeadline(live);
1294
- return;
1295
- }
1296
- }
1297
- catch {
1298
- // Retry below. Injection uses its independent client and remains usable.
1299
- }
1300
- await new Promise((resolve) => {
1301
- const timer = setTimeout(resolve, 250);
1302
- timer.unref?.();
1303
- });
1304
- }
1305
- })();
1306
- const settled = reconnect.finally(() => {
1307
- if (live.observerReconnect === settled)
1308
- live.observerReconnect = undefined;
1309
- });
1310
- live.observerReconnect = settled;
1311
- return settled;
1312
- }
1313
- armObserverReconnectDeadline(live) {
1314
- if (live.observerReconnectTimer || !live.forwarder.isTurnOpen())
1315
- return;
1316
- live.observerReconnectTimer = setTimeout(() => {
1317
- live.observerReconnectTimer = undefined;
1318
- if (live.stopped)
1319
- return;
1320
- const provider = live.runtime === "traex" ? "Traex" : "Codex";
1321
- live.forwarder.failOpenTurn(new Error(`${provider} observer did not reconnect within ${this.observerReconnectTimeoutMs}ms`));
1322
- }, this.observerReconnectTimeoutMs);
1323
- live.observerReconnectTimer.unref?.();
1324
- }
1325
- clearObserverReconnectDeadline(live) {
1326
- if (live.observerReconnectTimer)
1327
- clearTimeout(live.observerReconnectTimer);
1328
- live.observerReconnectTimer = undefined;
1329
- }
1330
1205
  /** Await a live session's thread binding. `null` leaves the deadline to the
1331
1206
  * caller; a number keeps the Provider-local bound. Returns false on timeout /
1332
1207
  * no live session. Injection and the runner's `live.ready` gate on this. */
@@ -1350,9 +1225,9 @@ export class LocalAgentHost {
1350
1225
  clearTimeout(timer);
1351
1226
  return ok;
1352
1227
  }
1353
- /** Await the stronger Terminal gate: another app-server connection has
1354
- * successfully resumed the thread, so the detached TUI cannot race rollout
1355
- * discovery or indexing. */
1228
+ /** Await the observer subscription diagnostic. Codex/Traex Terminal startup
1229
+ * deliberately does not gate on this promise: the dedicated preload owns resume,
1230
+ * while the independent observer attaches in the background. */
1356
1231
  async waitTerminalReady(localThreadId, timeoutMs = 20_000) {
1357
1232
  const claude = this.liveClaudeSessions.get(localThreadId);
1358
1233
  // Claude has no separate app-server indexing gate. Its TUI must launch first;
@@ -1376,8 +1251,24 @@ export class LocalAgentHost {
1376
1251
  /** Diagnostic from the provider adapter when native discovery/resume failed. */
1377
1252
  liveSessionError(localThreadId) {
1378
1253
  return this.liveClaudeSessions.get(localThreadId)?.error
1379
- ?? this.liveSessions.get(localThreadId)?.startupError
1254
+ ?? this.liveSessions.get(localThreadId)?.startupError?.message
1255
+ ?? this.liveStartupErrors.get(localThreadId)?.message;
1256
+ }
1257
+ /** Structured phase + concrete cause used by the runner wire protocol. */
1258
+ liveSessionFailure(localThreadId) {
1259
+ const failure = this.liveSessions.get(localThreadId)?.startupError
1380
1260
  ?? this.liveStartupErrors.get(localThreadId);
1261
+ if (failure) {
1262
+ return {
1263
+ message: failure.message,
1264
+ code: failure.code,
1265
+ statusCode: failure.statusCode,
1266
+ };
1267
+ }
1268
+ const claudeError = this.liveClaudeSessions.get(localThreadId)?.error;
1269
+ return claudeError
1270
+ ? { message: claudeError, code: "claude_native_start_failed", statusCode: 503 }
1271
+ : undefined;
1381
1272
  }
1382
1273
  /** Publish the background TUI/thread discovery failure so an executor
1383
1274
  * already waiting in the 60s bridge window exits immediately with the exact
@@ -1386,8 +1277,11 @@ export class LocalAgentHost {
1386
1277
  const live = this.liveSessions.get(localThreadId);
1387
1278
  if (!live || live.threadId || live.stopped)
1388
1279
  return false;
1389
- live.startupError = error.message;
1390
- this.liveStartupErrors.set(localThreadId, error.message);
1280
+ const failure = error instanceof CodexRuntimeError
1281
+ ? error
1282
+ : new CodexRuntimeError(error.message, 503, "native_thread_discovery_failed");
1283
+ live.startupError = failure;
1284
+ this.liveStartupErrors.set(localThreadId, failure);
1391
1285
  live.markStartupFailed();
1392
1286
  return true;
1393
1287
  }
@@ -1470,36 +1364,20 @@ export class LocalAgentHost {
1470
1364
  if (index >= 0)
1471
1365
  live.pendingInjectedInputs.splice(index, 1);
1472
1366
  };
1367
+ const injectionClient = this.injectionClientFactory(live.appServerUrl);
1473
1368
  try {
1474
- // Inject via the BACKEND client (the forwarder connection only observes).
1369
+ await injectionClient.ensureInitialized();
1370
+ // Match Omnigent's executor: each message uses one initialized client
1371
+ // that closes as soon as turn/start or turn/steer is acknowledged.
1475
1372
  if (live.forwarder.isTurnOpen()) {
1476
1373
  const turnId = live.forwarder.currentTurnId();
1477
1374
  if (turnId) {
1478
1375
  injectionMethod = "turn/steer";
1479
- let steered;
1480
- let retryIndex = 0;
1481
- while (!steered) {
1482
- try {
1483
- steered = await live.injectClient.turnSteer({
1484
- threadId,
1485
- expectedTurnId: turnId,
1486
- input: nativeInput,
1487
- });
1488
- }
1489
- catch (error) {
1490
- const delayMs = TURN_STEER_ACTIVATION_RETRY_DELAYS_MS[retryIndex];
1491
- if (delayMs === undefined ||
1492
- !isTransientTurnSteerActivationRace(error) ||
1493
- live.stopped ||
1494
- live.rotationPending ||
1495
- !live.forwarder.isTurnOpen() ||
1496
- live.forwarder.currentTurnId() !== turnId) {
1497
- throw error;
1498
- }
1499
- retryIndex += 1;
1500
- await new Promise((resolve) => setTimeout(resolve, delayMs));
1501
- }
1502
- }
1376
+ const steered = await injectionClient.turnSteer({
1377
+ threadId,
1378
+ expectedTurnId: turnId,
1379
+ input: nativeInput,
1380
+ });
1503
1381
  if (pendingInput.state === "prepublished") {
1504
1382
  // The caller already persisted and published this user input
1505
1383
  // before waiting for the native Terminal to become ready.
@@ -1512,15 +1390,12 @@ export class LocalAgentHost {
1512
1390
  pendingInput.state = "optimistic";
1513
1391
  }
1514
1392
  live.forwarder.noteTurnAccepted(steered.turnId);
1515
- if (!live.observerAvailable) {
1516
- this.armObserverReconnectDeadline(live);
1517
- }
1518
- return "injected";
1393
+ return "steered";
1519
1394
  }
1520
1395
  }
1521
1396
  // Carry the agent-spec model on the turn so a web-injected turn runs the
1522
1397
  // agent's model even if the TUI's config default differs.
1523
- const started = await live.injectClient.turnStart({
1398
+ const started = await injectionClient.turnStart({
1524
1399
  threadId,
1525
1400
  input: nativeInput,
1526
1401
  ...turnWorkspaceParams(live.runtime, live.workspace, live.sandbox),
@@ -1543,9 +1418,6 @@ export class LocalAgentHost {
1543
1418
  // Do not wait for the independent observer connection's `turn/started`:
1544
1419
  // a second message accepted in that window must steer, not double-start.
1545
1420
  live.forwarder.noteTurnAccepted(started.turnId);
1546
- if (!live.observerAvailable) {
1547
- this.armObserverReconnectDeadline(live);
1548
- }
1549
1421
  return "injected";
1550
1422
  }
1551
1423
  catch (error) {
@@ -1556,7 +1428,12 @@ export class LocalAgentHost {
1556
1428
  const startupDetail = live.forwarder.mcpStartupDetail();
1557
1429
  const detail = startupDetail ? `${baseDetail} (${startupDetail})` : baseDetail;
1558
1430
  console.error(`[codex-live] session=${localThreadId} runtime=${live.runtime} injection failed: ${detail}`);
1559
- throw new Error(detail, { cause: error });
1431
+ throw nativeLiveFailure(live.runtime, "native_message_injection_failed", `message injection via ${injectionMethod} failed`, detail);
1432
+ }
1433
+ finally {
1434
+ await injectionClient.stop().catch((stopError) => {
1435
+ console.warn(`[codex-live] session=${localThreadId} runtime=${live.runtime} injection client cleanup failed: ${stopError instanceof Error ? stopError.message : String(stopError)}`);
1436
+ });
1560
1437
  }
1561
1438
  });
1562
1439
  live.injectLock = run.then(() => undefined, () => undefined);
@@ -1607,27 +1484,34 @@ export class LocalAgentHost {
1607
1484
  if (!threadId)
1608
1485
  return pendingMcp.length > 0;
1609
1486
  let handled = pendingMcp.length > 0;
1610
- if (pendingMcp.length > 0) {
1611
- // Codex's TUI cancels a provider-owned MCP startup round with an empty
1612
- // turn id. Best-effort for Traex: it shares the app-server surface, while
1613
- // the active-turn interrupt below remains authoritative if it rejects this.
1614
- try {
1615
- await live.injectClient.turnInterrupt({ threadId, turnId: "" });
1487
+ const interruptClient = this.injectionClientFactory(live.appServerUrl);
1488
+ try {
1489
+ await interruptClient.ensureInitialized();
1490
+ if (pendingMcp.length > 0) {
1491
+ // Codex's TUI cancels a provider-owned MCP startup round with an empty
1492
+ // turn id. Best-effort for Traex: it shares the app-server surface, while
1493
+ // the active-turn interrupt below remains authoritative if it rejects this.
1494
+ try {
1495
+ await interruptClient.turnInterrupt({ threadId, turnId: "" });
1496
+ }
1497
+ catch (error) {
1498
+ console.warn(`[codex-live] session=${localThreadId} runtime=${live.runtime} MCP startup interrupt failed: ${codexRpcError(error, "turn/interrupt")}`);
1499
+ }
1616
1500
  }
1617
- catch (error) {
1618
- console.warn(`[codex-live] session=${localThreadId} runtime=${live.runtime} MCP startup interrupt failed: ${codexRpcError(error, "turn/interrupt")}`);
1501
+ if (turnId) {
1502
+ try {
1503
+ await interruptClient.turnInterrupt({ threadId, turnId });
1504
+ handled = true;
1505
+ }
1506
+ catch {
1507
+ // The startup cancellation above is still a handled Stop operation.
1508
+ }
1619
1509
  }
1510
+ return handled;
1620
1511
  }
1621
- if (turnId) {
1622
- try {
1623
- await live.injectClient.turnInterrupt({ threadId, turnId });
1624
- handled = true;
1625
- }
1626
- catch {
1627
- // The startup cancellation above is still a handled Stop operation.
1628
- }
1512
+ finally {
1513
+ await interruptClient.stop().catch(() => undefined);
1629
1514
  }
1630
- return handled;
1631
1515
  }
1632
1516
  /** Stop + drop a session's live forwarder and its dedicated connection (session
1633
1517
  * close / runner shutdown). The backend inject client is shared — left running. */
@@ -1654,24 +1538,17 @@ export class LocalAgentHost {
1654
1538
  this.liveSessions.delete(localThreadId);
1655
1539
  live.stopped = true;
1656
1540
  if (!live.threadId) {
1657
- live.startupError ??= "native Session stopped before thread discovery completed";
1541
+ live.startupError ??= nativeLiveFailure(live.runtime, "native_thread_discovery_stopped", "Session stopped before native thread discovery completed");
1658
1542
  live.markStartupFailed();
1659
1543
  }
1660
- this.clearObserverReconnectDeadline(live);
1661
1544
  live.releaseActive?.();
1662
- live.injectClient.cancelInteractions("session_stopped");
1545
+ live.appServerOwner.cancelInteractions("session_stopped");
1663
1546
  live.forwarderClient.cancelInteractions("session_stopped");
1664
- live.injectClient.setInteractionListener(null);
1547
+ live.appServerOwner.setInteractionListener(null);
1665
1548
  live.forwarderClient.setInteractionListener(null);
1666
- live.injectClient.setConnectionListener(null);
1549
+ live.appServerOwner.setConnectionListener(null);
1667
1550
  live.forwarderClient.setConnectionListener(null);
1668
- live.interactionOwners.clear();
1669
- live.interactionStandbys.clear();
1670
- live.interactionSubmissions.clear();
1671
1551
  live.forwarder.stop();
1672
- live.interactionRecoveries.clear();
1673
- live.interactionFailedClients.clear();
1674
- live.disconnectedClients.clear();
1675
1552
  live.canonicalInteractions.clear();
1676
1553
  live.settledInteractions.clear();
1677
1554
  void live.forwarderClient.stop().catch(() => undefined);
@@ -1988,6 +1865,17 @@ export class LocalAgentHost {
1988
1865
  emitCurrent(se);
1989
1866
  },
1990
1867
  onInteraction: forwardInteraction,
1868
+ onStatus: (status, blockedOn) => {
1869
+ const responseId = currentResponseId ??
1870
+ live.pendingInjectedInputs.find((entry) => entry.responseId)?.responseId;
1871
+ emitCurrent({
1872
+ type: "session.status",
1873
+ sessionId: currentSessionId,
1874
+ ...(responseId ? { responseId } : {}),
1875
+ status,
1876
+ ...(blockedOn ? { note: blockedOn } : {}),
1877
+ });
1878
+ },
1991
1879
  onTurnEnd: (usage) => {
1992
1880
  if (!normalizer)
1993
1881
  return;
@@ -2028,16 +1916,20 @@ export class LocalAgentHost {
2028
1916
  });
2029
1917
  }
2030
1918
  },
2031
- onTurnError: () => {
1919
+ onTurnError: (error) => {
2032
1920
  if (!normalizer)
2033
1921
  return;
2034
1922
  const rid = currentResponseId;
1923
+ const message = error.message || "Agent turn failed";
2035
1924
  for (const se of normalizer.fail({
2036
1925
  code: "agent_error",
2037
- message: "Agent turn failed",
1926
+ message,
2038
1927
  source: "execution",
2039
- }))
2040
- emitCurrent(se);
1928
+ })) {
1929
+ emitCurrent(se.type === "session.status"
1930
+ ? { ...se, status: "failed", note: message }
1931
+ : se);
1932
+ }
2041
1933
  live.pendingInjectedInputs = live.pendingInjectedInputs.filter((entry) => entry.responseId !== rid);
2042
1934
  normalizer = null;
2043
1935
  currentResponseId = undefined;
@@ -2196,21 +2088,22 @@ export class LocalAgentHost {
2196
2088
  * (they inject via the app-server). Idempotent. */
2197
2089
  attachTerminalInjector(localThreadId, injector) {
2198
2090
  const live = this.liveClaudeSessions.get(localThreadId);
2199
- if (live)
2091
+ if (live) {
2092
+ if (live.injector === injector)
2093
+ return;
2200
2094
  live.injector = injector;
2095
+ live.forwarder.attachStatusSource({ panePid: () => injector.panePid?.() });
2096
+ }
2201
2097
  }
2202
2098
  /** Close an active native response when its terminal or runner disappears. */
2203
2099
  failLiveSession(localThreadId, error) {
2204
2100
  const claude = this.liveClaudeSessions.get(localThreadId);
2205
2101
  if (claude)
2206
- return claude.forwarder.failOpenTurn(error);
2102
+ return claude.forwarder.noteTerminalExit(error);
2207
2103
  const codex = this.liveSessions.get(localThreadId);
2208
2104
  if (!codex)
2209
2105
  return false;
2210
- const failed = codex.forwarder.failOpenTurn(error);
2211
- if (failed)
2212
- this.clearObserverReconnectDeadline(codex);
2213
- return failed;
2106
+ return codex.forwarder.failOpenTurn(error);
2214
2107
  }
2215
2108
  /** List the models a runtime exposes (App Server for codex/traex; static for claude). */
2216
2109
  async listModels(runtime) {
@@ -2240,7 +2133,7 @@ export class LocalAgentHost {
2240
2133
  return { ok: false, reason: "unsupported" };
2241
2134
  }
2242
2135
  try {
2243
- return { ok: true, data: await fn(live.injectClient, record.codexSessionId) };
2136
+ return { ok: true, data: await fn(live.appServerOwner, record.codexSessionId) };
2244
2137
  }
2245
2138
  catch (error) {
2246
2139
  return {
@@ -2312,7 +2205,7 @@ export class LocalAgentHost {
2312
2205
  return { ok: true, data: undefined };
2313
2206
  }
2314
2207
  const runtime = execution.provider;
2315
- const client = live?.injectClient ??
2208
+ const client = live?.appServerOwner ??
2316
2209
  this.getBackend(runtime, execution.budget ?? undefined).appServerClient;
2317
2210
  if (!client) {
2318
2211
  return { ok: false, reason: "unsupported" };
@@ -2387,20 +2280,6 @@ function codexRpcError(error, method) {
2387
2280
  }
2388
2281
  return `Codex ${method} failed: ${error instanceof Error ? error.message : String(error)}`;
2389
2282
  }
2390
- /**
2391
- * Codex 0.146 can acknowledge `turn/start` a few milliseconds before its
2392
- * internal active-turn pointer becomes visible to `turn/steer`. The injection
2393
- * lock still prevents double-starts, but the immediately following steer must
2394
- * briefly retry the same expected turn id. Keep this narrow: all other RPC
2395
- * errors remain terminal and visible to the caller.
2396
- */
2397
- function isTransientTurnSteerActivationRace(error) {
2398
- return error instanceof CodexTransportError &&
2399
- error.code === -32600 &&
2400
- (/no active turn to steer/i.test(error.message) ||
2401
- /expected active turn id\b.+\bfound\b/i.test(error.message));
2402
- }
2403
- const TURN_STEER_ACTIVATION_RETRY_DELAYS_MS = [10, 20, 40, 80, 160, 250, 440];
2404
2283
  export function parseCodexLoginStatus(exitCode, output) {
2405
2284
  const normalized = output.trim();
2406
2285
  const match = normalized.match(/Logged in using (.+)$/im);
@@ -2502,10 +2381,13 @@ export function isThreadNotReadyError(error) {
2502
2381
  const message = (error instanceof Error ? error.message : String(error)).toLowerCase();
2503
2382
  return message.includes("no rollout found") || (message.includes("rollout") && message.includes("empty"));
2504
2383
  }
2505
- /** A persisted thread id that a freshly started app-server cannot load yet.
2506
- * During startup, both errors can be transient while the rollout index catches
2507
- * up. Retry the same id; never use either error as permission to replace it. */
2508
- export function isRetryableThreadResumeError(error) {
2509
- const message = (error instanceof Error ? error.message : String(error)).toLowerCase();
2510
- return message.includes("thread not found") || isThreadNotReadyError(error);
2384
+ /** Build the user-facing native lifecycle failure once, at the phase that owns
2385
+ * it. Downstream layers preserve this code/message verbatim instead of replacing
2386
+ * it with a generic "failed to start; see logs" wrapper. */
2387
+ function nativeLiveFailure(runtime, code, phase, cause, statusCode = 503) {
2388
+ const provider = runtime === "traex" ? "Traex" : "Codex";
2389
+ const detail = cause === undefined
2390
+ ? ""
2391
+ : `: ${cause instanceof Error ? cause.message : String(cause)}`;
2392
+ return new CodexRuntimeError(`${provider} ${phase}${detail}`, statusCode, code);
2511
2393
  }