@evident-ai/cli 3.1.1-dev.38348be → 3.1.1-dev.3b21f0c

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/index.js CHANGED
@@ -38,11 +38,6 @@ function getApiUrl() {
38
38
  function getTunnelUrl() {
39
39
  return tunnelOverride ?? process.env.EVIDENT_TUNNEL_URL ?? defaults.tunnelUrl;
40
40
  }
41
- var config = new Conf({
42
- projectName: "evident",
43
- projectSuffix: "",
44
- defaults
45
- });
46
41
  var credentials = new Conf({
47
42
  projectName: "evident",
48
43
  projectSuffix: "",
@@ -267,16 +262,28 @@ async function getToken() {
267
262
  }
268
263
  return null;
269
264
  }
265
+ function toError(err) {
266
+ return err instanceof Error ? err : new Error(String(err));
267
+ }
270
268
  async function deleteToken(options = {}) {
271
269
  const keytar = await getKeytar();
270
+ const failures = [];
272
271
  if (keytar) {
273
272
  if (options.all) {
274
- const all = await keytar.findCredentials(SERVICE_NAME).catch(() => []);
273
+ let accounts = [];
274
+ try {
275
+ accounts = await keytar.findCredentials(SERVICE_NAME);
276
+ } catch (err) {
277
+ failures.push({ type: "enumerate", error: toError(err) });
278
+ }
275
279
  await Promise.all(
276
- all.map(
277
- (entry) => keytar.deletePassword(SERVICE_NAME, entry.account).catch(() => {
278
- })
279
- )
280
+ accounts.map(async (entry) => {
281
+ try {
282
+ await keytar.deletePassword(SERVICE_NAME, entry.account);
283
+ } catch (err) {
284
+ failures.push({ type: "delete", account: entry.account, error: toError(err) });
285
+ }
286
+ })
280
287
  );
281
288
  } else {
282
289
  await keytar.deletePassword(SERVICE_NAME, keychainAccount());
@@ -287,6 +294,7 @@ async function deleteToken(options = {}) {
287
294
  } else {
288
295
  clearCredentials();
289
296
  }
297
+ return { failures };
290
298
  }
291
299
 
292
300
  // src/utils/ui.ts
@@ -404,8 +412,10 @@ async function deviceFlowLogin(options) {
404
412
  }
405
413
  async function tokenLogin() {
406
414
  console.log("Token login mode.");
407
- console.log("Run `evident login` on a machine with a browser to get a token.");
408
- console.log("Manage or revoke existing tokens under Settings \u2192 CLI tokens.");
415
+ console.log("Create a token under Settings \u2192 CLI tokens in the dashboard, then paste it below.");
416
+ console.log(
417
+ "(Alternatively, run `evident login` on a machine with a browser, or set EVIDENT_TOKEN for CI.)"
418
+ );
409
419
  blank();
410
420
  process.stdout.write("Paste token: ");
411
421
  const token = await new Promise((resolve3) => {
@@ -429,13 +439,22 @@ async function tokenLogin() {
429
439
  printError("No token provided.");
430
440
  process.exit(1);
431
441
  }
442
+ await validateAndStoreToken(token);
443
+ }
444
+ async function validateAndStoreToken(token) {
432
445
  const spinner = ora("Validating token...").start();
433
446
  try {
434
- const result = await api.post("/auth/token/validate", { token });
447
+ const result = await api.get("/me", {
448
+ headers: { Authorization: `Bearer ${token}` }
449
+ });
450
+ if (!result.user) {
451
+ throw new Error(
452
+ "This token is not a user login (e.g. a runner key). Paste a CLI token instead."
453
+ );
454
+ }
435
455
  await storeToken({
436
456
  token,
437
- user: result.user,
438
- expiresAt: result.expires_at
457
+ user: { email: result.user.email }
439
458
  });
440
459
  spinner.stop();
441
460
  printSuccess(`Logged in as ${chalk2.bold(result.user.email)}`);
@@ -455,9 +474,22 @@ async function login(options) {
455
474
  }
456
475
 
457
476
  // src/commands/logout.ts
477
+ function describeFailure(failure) {
478
+ if (failure.type === "enumerate") {
479
+ return `could not list stored keychain entries (${failure.error.message})`;
480
+ }
481
+ return `${failure.account} (${failure.error.message})`;
482
+ }
458
483
  async function logout(options = {}) {
459
484
  if (options.all) {
460
- await deleteToken({ all: true });
485
+ const result = await deleteToken({ all: true });
486
+ if (result.failures.length > 0) {
487
+ printError(
488
+ `Failed to fully clear your keychain: ${result.failures.map(describeFailure).join("; ")}. Your local credentials file was cleared, but stale keychain entries may remain \u2014 run \`evident logout --all\` again, or remove them manually from your OS keychain / credential manager.`
489
+ );
490
+ process.exitCode = 1;
491
+ return;
492
+ }
461
493
  printSuccess("Logged out of all endpoints.");
462
494
  return;
463
495
  }
@@ -482,7 +514,9 @@ async function whoami() {
482
514
  blank();
483
515
  console.log(keyValue("Endpoint", apiUrl));
484
516
  console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
485
- console.log(keyValue("User ID", credentials2.user.id));
517
+ if (credentials2.user.id) {
518
+ console.log(keyValue("User ID", credentials2.user.id));
519
+ }
486
520
  if (credentials2.expiresAt) {
487
521
  const expiresAt = new Date(credentials2.expiresAt);
488
522
  const now = /* @__PURE__ */ new Date();
@@ -498,11 +532,474 @@ async function whoami() {
498
532
  blank();
499
533
  }
500
534
 
535
+ // src/lib/auth.ts
536
+ async function getAuthCredentials() {
537
+ const runnerKey = process.env.EVIDENT_RUNNER_KEY;
538
+ const agentKey = process.env.EVIDENT_AGENT_KEY;
539
+ if (runnerKey) {
540
+ return {
541
+ token: runnerKey,
542
+ authType: "agent_key",
543
+ keySource: "runner_key",
544
+ notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
545
+ };
546
+ }
547
+ if (agentKey) {
548
+ return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
549
+ }
550
+ const userToken = process.env.EVIDENT_TOKEN;
551
+ if (userToken) {
552
+ return { token: userToken, authType: "bearer" };
553
+ }
554
+ const keychainCreds = await getToken();
555
+ if (keychainCreds) {
556
+ return {
557
+ token: keychainCreds.token,
558
+ authType: "bearer",
559
+ user: keychainCreds.user
560
+ };
561
+ }
562
+ return null;
563
+ }
564
+ function getAuthHeader(credentials2) {
565
+ if (credentials2.authType === "agent_key") {
566
+ return `SandboxKey ${credentials2.token}`;
567
+ }
568
+ return `Bearer ${credentials2.token}`;
569
+ }
570
+ function isInteractive(jsonOutput) {
571
+ if (jsonOutput) return false;
572
+ if (process.env.CI) return false;
573
+ if (process.env.GITHUB_ACTIONS) return false;
574
+ if (!process.stdin.isTTY) return false;
575
+ return true;
576
+ }
577
+
578
+ // src/commands/agent-lookup.ts
579
+ async function readErrorMessage(response) {
580
+ const text = await response.text().catch(() => "");
581
+ if (!text) return response.statusText || void 0;
582
+ try {
583
+ const data = JSON.parse(text);
584
+ const message = data.message ?? data.error;
585
+ if (typeof message === "string" && message.trim()) {
586
+ return message;
587
+ }
588
+ } catch {
589
+ }
590
+ return text.trim() || response.statusText || void 0;
591
+ }
592
+ function authFailureHint(apiUrl, serverMessage) {
593
+ const reason = serverMessage ? `: ${serverMessage}` : "";
594
+ return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
595
+ }
596
+ async function resolveAgentIdFromKey(authHeader) {
597
+ const apiUrl = getApiUrlConfig();
598
+ try {
599
+ const response = await fetch(`${apiUrl}/me`, {
600
+ headers: { Authorization: authHeader }
601
+ });
602
+ if (response.status === 401) {
603
+ const serverMessage = await readErrorMessage(response);
604
+ return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
605
+ }
606
+ if (!response.ok) {
607
+ const serverMessage = await readErrorMessage(response);
608
+ return {
609
+ error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
610
+ };
611
+ }
612
+ const data = await response.json();
613
+ if (data.auth_type === "agent_key" && data.agent_id) {
614
+ return { agent_id: data.agent_id };
615
+ }
616
+ return {
617
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
618
+ };
619
+ } catch (error2) {
620
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
621
+ return { error: `Failed to resolve runner from key: ${message}` };
622
+ }
623
+ }
624
+ var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
625
+ async function notifyAgentDisconnected(agentId, authHeader) {
626
+ const apiUrl = getApiUrlConfig();
627
+ try {
628
+ const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
629
+ method: "POST",
630
+ headers: { Authorization: authHeader },
631
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
632
+ });
633
+ if (!response.ok) {
634
+ const serverMessage = await readErrorMessage(response);
635
+ return {
636
+ ok: false,
637
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
638
+ };
639
+ }
640
+ return { ok: true };
641
+ } catch (error2) {
642
+ return { ok: false, error: describeBestEffortError(error2) };
643
+ }
644
+ }
645
+ function describeBestEffortError(error2) {
646
+ const name = error2?.name;
647
+ if (name === "TimeoutError" || name === "AbortError") {
648
+ return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
649
+ }
650
+ return error2 instanceof Error ? error2.message : String(error2);
651
+ }
652
+ async function reportMicrovmId(agentId, authHeader, microvmId) {
653
+ try {
654
+ const apiUrl = getApiUrlConfig();
655
+ const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
656
+ method: "POST",
657
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
658
+ body: JSON.stringify({ microvm_id: microvmId }),
659
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
660
+ });
661
+ if (!response.ok) {
662
+ const serverMessage = await readErrorMessage(response);
663
+ return {
664
+ ok: false,
665
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
666
+ };
667
+ }
668
+ return { ok: true };
669
+ } catch (error2) {
670
+ return { ok: false, error: describeBestEffortError(error2) };
671
+ }
672
+ }
673
+ function toReportedWindow(window) {
674
+ if (!window) return null;
675
+ return { utilization: window.utilization, resets_at: window.resetsAt };
676
+ }
677
+ async function reportClaudeUsage(agentId, authHeader, snapshot) {
678
+ try {
679
+ const apiUrl = getApiUrlConfig();
680
+ const response = await fetch(`${apiUrl}/runners/${agentId}/claude-usage`, {
681
+ method: "POST",
682
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
683
+ body: JSON.stringify({
684
+ five_hour: toReportedWindow(snapshot.fiveHour),
685
+ seven_day: toReportedWindow(snapshot.sevenDay)
686
+ }),
687
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
688
+ });
689
+ if (!response.ok) {
690
+ const serverMessage = await readErrorMessage(response);
691
+ return {
692
+ ok: false,
693
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
694
+ };
695
+ }
696
+ return { ok: true };
697
+ } catch (error2) {
698
+ return { ok: false, error: describeBestEffortError(error2) };
699
+ }
700
+ }
701
+ async function getAgentInfo(agentId, authHeader) {
702
+ const apiUrl = getApiUrlConfig();
703
+ try {
704
+ const response = await fetch(`${apiUrl}/runners/${agentId}`, {
705
+ headers: { Authorization: authHeader }
706
+ });
707
+ if (response.status === 401) {
708
+ const serverMessage = await readErrorMessage(response);
709
+ return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
710
+ }
711
+ if (response.status === 403) {
712
+ const serverMessage = await readErrorMessage(response);
713
+ return {
714
+ valid: false,
715
+ error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
716
+ };
717
+ }
718
+ if (response.status === 404) {
719
+ const serverMessage = await readErrorMessage(response);
720
+ return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
721
+ }
722
+ if (!response.ok) {
723
+ const serverMessage = await readErrorMessage(response);
724
+ return {
725
+ valid: false,
726
+ error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
727
+ };
728
+ }
729
+ const agent = await response.json();
730
+ if (agent.agent_type !== "local") {
731
+ return {
732
+ valid: false,
733
+ error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
734
+ };
735
+ }
736
+ return { valid: true, agent };
737
+ } catch (error2) {
738
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
739
+ return { valid: false, error: `Failed to validate runner: ${message}` };
740
+ }
741
+ }
742
+
743
+ // src/commands/status.ts
744
+ var STATUS_TIMEOUT_MS = 1e4;
745
+ function authLabelFor(credentials2) {
746
+ if (credentials2.authType === "agent_key") {
747
+ return credentials2.keySource === "agent_key" ? "runner key (EVIDENT_AGENT_KEY)" : "runner key (EVIDENT_RUNNER_KEY)";
748
+ }
749
+ return "user token";
750
+ }
751
+ function describeFetchError(error2) {
752
+ const name = error2?.name;
753
+ if (name === "TimeoutError" || name === "AbortError") {
754
+ return `timed out after ${STATUS_TIMEOUT_MS}ms waiting for a response`;
755
+ }
756
+ return error2 instanceof Error ? error2.message : String(error2);
757
+ }
758
+ async function checkStatus(jsonMode) {
759
+ const apiUrl = getApiUrlConfig();
760
+ const credentials2 = await getAuthCredentials();
761
+ if (!credentials2) {
762
+ return {
763
+ ok: false,
764
+ endpoint: apiUrl,
765
+ reason: "no_credentials",
766
+ error: "No credentials configured. Set EVIDENT_RUNNER_KEY (or EVIDENT_AGENT_KEY), or EVIDENT_TOKEN, or run `evident login`.",
767
+ exitCode: 1
768
+ };
769
+ }
770
+ if (credentials2.notice && !jsonMode) {
771
+ printWarning(credentials2.notice);
772
+ }
773
+ let response;
774
+ try {
775
+ response = await fetch(`${apiUrl}/me`, {
776
+ headers: { Authorization: getAuthHeader(credentials2) },
777
+ signal: AbortSignal.timeout(STATUS_TIMEOUT_MS)
778
+ });
779
+ } catch (error2) {
780
+ return {
781
+ ok: false,
782
+ endpoint: apiUrl,
783
+ authLabel: authLabelFor(credentials2),
784
+ reason: "unreachable",
785
+ error: `Could not reach ${apiUrl}: ${describeFetchError(error2)}. The credentials were NOT validated.`,
786
+ exitCode: 75
787
+ };
788
+ }
789
+ if (response.status === 401) {
790
+ const serverMessage = await readErrorMessage(response);
791
+ return {
792
+ ok: false,
793
+ endpoint: apiUrl,
794
+ authLabel: authLabelFor(credentials2),
795
+ reason: "unauthorized",
796
+ error: authFailureHint(apiUrl, serverMessage),
797
+ exitCode: 1
798
+ };
799
+ }
800
+ if (response.status >= 500) {
801
+ const serverMessage = await readErrorMessage(response);
802
+ return {
803
+ ok: false,
804
+ endpoint: apiUrl,
805
+ authLabel: authLabelFor(credentials2),
806
+ reason: "unreachable",
807
+ error: `${apiUrl} returned HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}. The credentials were NOT validated.`,
808
+ exitCode: 75
809
+ };
810
+ }
811
+ if (!response.ok) {
812
+ const serverMessage = await readErrorMessage(response);
813
+ return {
814
+ ok: false,
815
+ endpoint: apiUrl,
816
+ authLabel: authLabelFor(credentials2),
817
+ reason: "http_error",
818
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`,
819
+ exitCode: 1
820
+ };
821
+ }
822
+ const data = await response.json();
823
+ return {
824
+ ok: true,
825
+ endpoint: apiUrl,
826
+ authType: data.auth_type,
827
+ authLabel: authLabelFor(credentials2),
828
+ runnerId: data.auth_type === "agent_key" ? data.agent_id : void 0,
829
+ reason: "ok",
830
+ exitCode: 0
831
+ };
832
+ }
833
+ function printJson(result) {
834
+ const payload = {
835
+ ok: result.ok,
836
+ endpoint: result.endpoint
837
+ };
838
+ if (result.authType) payload.auth_type = result.authType;
839
+ if (result.runnerId) payload.runner_id = result.runnerId;
840
+ if (result.reason) payload.reason = result.reason;
841
+ if (result.error) payload.error = result.error;
842
+ console.log(JSON.stringify(payload));
843
+ }
844
+ function printHuman(result) {
845
+ blank();
846
+ console.log(keyValue("Endpoint", result.endpoint));
847
+ if (result.ok) {
848
+ console.log(keyValue("Auth", result.authLabel ?? "\u2014"));
849
+ if (result.runnerId) {
850
+ console.log(keyValue("Runner", result.runnerId));
851
+ }
852
+ console.log(keyValue("Status", "OK \u2014 credentials accepted"));
853
+ blank();
854
+ return;
855
+ }
856
+ if (result.authLabel) {
857
+ console.log(keyValue("Auth", result.authLabel));
858
+ }
859
+ blank();
860
+ printError(result.error ?? "Unknown error");
861
+ }
862
+ async function status(options = {}) {
863
+ const result = await checkStatus(Boolean(options.json));
864
+ if (options.json) {
865
+ printJson(result);
866
+ } else {
867
+ printHuman(result);
868
+ }
869
+ process.exit(result.exitCode);
870
+ }
871
+
872
+ // src/lib/claude-usage.ts
873
+ import { execFileSync } from "child_process";
874
+ import { readFileSync } from "fs";
875
+ import { homedir } from "os";
876
+ import { join } from "path";
877
+ var CLAUDE_USAGE_URL = "https://api.anthropic.com/api/oauth/usage";
878
+ var KEYCHAIN_SERVICE = "Claude Code-credentials";
879
+ function parseClaudeCliCredentials(raw) {
880
+ let parsed;
881
+ try {
882
+ parsed = JSON.parse(raw);
883
+ } catch {
884
+ return null;
885
+ }
886
+ const data = parsed.claudeAiOauth ?? parsed;
887
+ const creds = data;
888
+ if (typeof creds.accessToken !== "string" || typeof creds.expiresAt !== "number") {
889
+ return null;
890
+ }
891
+ return { accessToken: creds.accessToken, expiresAt: creds.expiresAt };
892
+ }
893
+ function readClaudeCliCredentials() {
894
+ if (process.platform === "darwin") {
895
+ try {
896
+ const raw = execFileSync(
897
+ "/usr/bin/security",
898
+ ["find-generic-password", "-s", KEYCHAIN_SERVICE, "-w"],
899
+ { encoding: "utf-8", timeout: 2e3, stdio: ["pipe", "pipe", "ignore"] }
900
+ );
901
+ return parseClaudeCliCredentials(raw);
902
+ } catch {
903
+ return null;
904
+ }
905
+ }
906
+ try {
907
+ const raw = readFileSync(join(homedir(), ".claude", ".credentials.json"), "utf-8");
908
+ return parseClaudeCliCredentials(raw);
909
+ } catch {
910
+ return null;
911
+ }
912
+ }
913
+ var ClaudeUsageError = class extends Error {
914
+ constructor(message, reason) {
915
+ super(message);
916
+ this.reason = reason;
917
+ }
918
+ };
919
+ function isLocalCredentialProblem(err) {
920
+ return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
921
+ }
922
+ function normalizeResetsAt(value) {
923
+ const ms = Date.parse(value);
924
+ return Number.isNaN(ms) ? null : new Date(ms).toISOString();
925
+ }
926
+ function toWindow(value) {
927
+ if (!value || typeof value !== "object") {
928
+ return null;
929
+ }
930
+ const window = value;
931
+ if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
932
+ return null;
933
+ }
934
+ const resetsAt = normalizeResetsAt(window.resets_at);
935
+ if (resetsAt === null) {
936
+ return null;
937
+ }
938
+ return { utilization: window.utilization, resetsAt };
939
+ }
940
+ async function getClaudeUsage() {
941
+ const credentials2 = readClaudeCliCredentials();
942
+ if (!credentials2) {
943
+ throw new ClaudeUsageError(
944
+ "No local Claude Code login found. Run `claude` once to sign in with your Claude subscription.",
945
+ "no_credentials"
946
+ );
947
+ }
948
+ if (credentials2.expiresAt < Date.now()) {
949
+ throw new ClaudeUsageError(
950
+ "Claude Code credentials have expired. Run `claude` to refresh them.",
951
+ "credentials_expired"
952
+ );
953
+ }
954
+ const res = await fetch(CLAUDE_USAGE_URL, {
955
+ headers: {
956
+ Authorization: `Bearer ${credentials2.accessToken}`,
957
+ "Content-Type": "application/json",
958
+ "anthropic-version": "2023-06-01"
959
+ }
960
+ });
961
+ if (!res.ok) {
962
+ throw new ClaudeUsageError(`Claude usage request failed: HTTP ${res.status}`, "request_failed");
963
+ }
964
+ const body = await res.json();
965
+ return {
966
+ fiveHour: toWindow(body.five_hour),
967
+ sevenDay: toWindow(body.seven_day)
968
+ };
969
+ }
970
+
971
+ // src/commands/claude-usage.ts
972
+ function formatWindow(label, window) {
973
+ if (!window) {
974
+ return keyValue(label, "not available for this plan");
975
+ }
976
+ const resetsAt = new Date(window.resetsAt);
977
+ return keyValue(label, `${window.utilization}% used, resets ${resetsAt.toLocaleString()}`);
978
+ }
979
+ async function claudeUsage() {
980
+ try {
981
+ const usage = await getClaudeUsage();
982
+ blank();
983
+ console.log(formatWindow("5-hour session", usage.fiveHour));
984
+ console.log(formatWindow("7-day", usage.sevenDay));
985
+ blank();
986
+ } catch (err) {
987
+ if (err instanceof ClaudeUsageError) {
988
+ printError(err.message);
989
+ process.exit(1);
990
+ }
991
+ throw err;
992
+ }
993
+ }
994
+
501
995
  // src/commands/run.ts
502
- import { homedir as homedir2 } from "os";
503
- import { isAbsolute as isAbsolute2, join as join2, parse, resolve as resolvePath } from "path";
996
+ import { homedir as homedir3 } from "os";
997
+ import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
504
998
  import chalk6 from "chalk";
505
999
 
1000
+ // ../../packages/types/src/agents/index.ts
1001
+ var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
1002
+
506
1003
  // ../../packages/types/src/telemetry/index.ts
507
1004
  var TelemetryEventTypes = {
508
1005
  // Agent activity events (shown in web UI activity log)
@@ -510,7 +1007,10 @@ var TelemetryEventTypes = {
510
1007
  AGENT_DISCONNECTED: "agent.disconnected",
511
1008
  AGENT_MESSAGE_PROCESSING: "agent.message_processing",
512
1009
  AGENT_MESSAGE_DONE: "agent.message_done",
513
- AGENT_MESSAGE_FAILED: "agent.message_failed"
1010
+ AGENT_MESSAGE_FAILED: "agent.message_failed",
1011
+ // A `warn`/`error` runner-side log line forwarded server-side for
1012
+ // observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
1013
+ RUNNER_ACTIVITY: "runner.activity"
514
1014
  };
515
1015
 
516
1016
  // ../../packages/types/src/tunnel/index.ts
@@ -565,6 +1065,13 @@ var isShuttingDown = false;
565
1065
  var FLUSH_INTERVAL_MS = 5e3;
566
1066
  var MAX_BUFFER_SIZE = 50;
567
1067
  var FLUSH_TIMEOUT_MS = 3e3;
1068
+ var authProvider = null;
1069
+ function setTelemetryAuthProvider(provider) {
1070
+ authProvider = provider;
1071
+ }
1072
+ var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
1073
+ var lastFlushFailureLoggedAt = 0;
1074
+ var suppressedFlushFailureCount = 0;
568
1075
  function logEvent(eventType, options = {}) {
569
1076
  const event = {
570
1077
  event_type: eventType,
@@ -599,9 +1106,16 @@ async function flushEvents() {
599
1106
  flushTimeout = null;
600
1107
  }
601
1108
  try {
602
- const credentials2 = await getToken();
603
- if (!credentials2) {
604
- return;
1109
+ const providerContext = authProvider?.();
1110
+ let authHeader;
1111
+ if (providerContext?.authHeader) {
1112
+ authHeader = providerContext.authHeader;
1113
+ } else {
1114
+ const credentials2 = await getToken();
1115
+ if (!credentials2) {
1116
+ return;
1117
+ }
1118
+ authHeader = `Bearer ${credentials2.token}`;
605
1119
  }
606
1120
  const apiUrl = getApiUrlConfig();
607
1121
  const controller = new AbortController();
@@ -616,7 +1130,7 @@ async function flushEvents() {
616
1130
  method: "POST",
617
1131
  headers: {
618
1132
  "Content-Type": "application/json",
619
- Authorization: `Bearer ${credentials2.token}`
1133
+ Authorization: authHeader
620
1134
  },
621
1135
  body: JSON.stringify(request),
622
1136
  signal: controller.signal
@@ -628,8 +1142,15 @@ async function flushEvents() {
628
1142
  clearTimeout(timeout);
629
1143
  }
630
1144
  } catch (error2) {
631
- if (process.env.DEBUG) {
632
- console.error("Telemetry error:", error2);
1145
+ const now = Date.now();
1146
+ if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
1147
+ const message = error2 instanceof Error ? error2.message : String(error2);
1148
+ const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
1149
+ console.error(`Telemetry flush error: ${message}${suffix}`);
1150
+ lastFlushFailureLoggedAt = now;
1151
+ suppressedFlushFailureCount = 0;
1152
+ } else {
1153
+ suppressedFlushFailureCount++;
633
1154
  }
634
1155
  }
635
1156
  }
@@ -698,47 +1219,67 @@ var EventTypes = {
698
1219
  DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
699
1220
  };
700
1221
 
701
- // src/lib/auth.ts
702
- async function getAuthCredentials() {
703
- const runnerKey = process.env.EVIDENT_RUNNER_KEY;
704
- const agentKey = process.env.EVIDENT_AGENT_KEY;
705
- if (runnerKey) {
706
- return {
707
- token: runnerKey,
708
- authType: "agent_key",
709
- keySource: "runner_key",
710
- notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
711
- };
712
- }
713
- if (agentKey) {
714
- return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
715
- }
716
- const userToken = process.env.EVIDENT_TOKEN;
717
- if (userToken) {
718
- return { token: userToken, authType: "bearer" };
1222
+ // src/lib/runner-activity-telemetry.ts
1223
+ var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
1224
+ var SEVERITY_BY_LEVEL = {
1225
+ warn: "warning",
1226
+ error: "error"
1227
+ };
1228
+ var MAX_MESSAGE_LENGTH = 500;
1229
+ var TRUNCATION_MARKER = "\u2026";
1230
+ function redact(message) {
1231
+ return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
1232
+ }
1233
+ function truncate(message) {
1234
+ if (message.length <= MAX_MESSAGE_LENGTH) return message;
1235
+ return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
1236
+ }
1237
+ var RATE_LIMIT_WINDOW_MS = 6e4;
1238
+ var RATE_LIMIT_MAX_EVENTS = 30;
1239
+ var windowStartedAt = 0;
1240
+ var windowCount = 0;
1241
+ var windowDroppedCount = 0;
1242
+ function admitUnderRateLimit(now) {
1243
+ if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
1244
+ if (windowDroppedCount > 0) {
1245
+ console.error(
1246
+ `[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
1247
+ );
1248
+ }
1249
+ windowStartedAt = now;
1250
+ windowCount = 0;
1251
+ windowDroppedCount = 0;
719
1252
  }
720
- const keychainCreds = await getToken();
721
- if (keychainCreds) {
722
- return {
723
- token: keychainCreds.token,
724
- authType: "bearer",
725
- user: keychainCreds.user
726
- };
1253
+ if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
1254
+ windowDroppedCount++;
1255
+ if (windowDroppedCount === 1) {
1256
+ console.error(
1257
+ `[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
1258
+ );
1259
+ }
1260
+ return false;
727
1261
  }
728
- return null;
1262
+ windowCount++;
1263
+ return true;
729
1264
  }
730
- function getAuthHeader(credentials2) {
731
- if (credentials2.authType === "agent_key") {
732
- return `SandboxKey ${credentials2.token}`;
1265
+ function forwardRunnerActivity(entry, context) {
1266
+ try {
1267
+ if (!FORWARDED_LEVELS.has(entry.level)) return;
1268
+ if (!context.agentId || !context.authHeader) return;
1269
+ if (!admitUnderRateLimit(Date.now())) return;
1270
+ const rawMessage = entry.error ?? entry.message ?? "";
1271
+ const message = truncate(redact(rawMessage));
1272
+ logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
1273
+ severity: SEVERITY_BY_LEVEL[entry.level],
1274
+ message,
1275
+ metadata: { source: "cli.run" },
1276
+ agentId: context.agentId
1277
+ });
1278
+ } catch (err) {
1279
+ console.error(
1280
+ `[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
1281
+ );
733
1282
  }
734
- return `Bearer ${credentials2.token}`;
735
- }
736
- function isInteractive(jsonOutput) {
737
- if (jsonOutput) return false;
738
- if (process.env.CI) return false;
739
- if (process.env.GITHUB_ACTIONS) return false;
740
- if (!process.stdin.isTTY) return false;
741
- return true;
742
1283
  }
743
1284
 
744
1285
  // src/lib/opencode/health.ts
@@ -1532,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
1532
2073
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1533
2074
  return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1534
2075
  }
2076
+ function isB2AbandonmentConfirmed(params) {
2077
+ return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
2078
+ }
1535
2079
  function messageError(messages, userMessageId) {
1536
2080
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1537
2081
  const error2 = errorOf(reply);
@@ -1545,6 +2089,57 @@ function messageError(messages, userMessageId) {
1545
2089
  }
1546
2090
  return "The agent run failed.";
1547
2091
  }
2092
+ function isAbortedTerminalReply(messages, userMessageId) {
2093
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
2094
+ const error2 = errorOf(reply);
2095
+ if (error2 == null) return false;
2096
+ if (typeof error2 === "string") return error2.trim() === "Aborted";
2097
+ if (typeof error2 === "object") {
2098
+ const e = error2;
2099
+ if (e.name === "MessageAbortedError") return true;
2100
+ if (e.name === "AbortError") return true;
2101
+ const dataMessage = e.data?.message;
2102
+ const rendered = typeof dataMessage === "string" ? dataMessage : typeof e.message === "string" ? e.message : null;
2103
+ return rendered != null && rendered.trim() === "Aborted";
2104
+ }
2105
+ return false;
2106
+ }
2107
+ function messageFailure(messages, userMessageId) {
2108
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
2109
+ const error2 = errorOf(reply);
2110
+ if (error2 == null || typeof error2 !== "object") return null;
2111
+ const e = error2;
2112
+ const replyProviderId = reply?.info?.providerID ?? null;
2113
+ const replyModelId = reply?.info?.modelID ?? null;
2114
+ if (e.name === "ProviderAuthError") {
2115
+ const data = e.data;
2116
+ const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
2117
+ return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
2118
+ }
2119
+ if (e.name === "APIError") {
2120
+ const data = e.data;
2121
+ const statusCode = data?.statusCode;
2122
+ if (statusCode === 401 || statusCode === 403) {
2123
+ return {
2124
+ kind: "model_auth",
2125
+ providerId: replyProviderId,
2126
+ modelId: replyModelId,
2127
+ reason: "rejected"
2128
+ };
2129
+ }
2130
+ }
2131
+ return null;
2132
+ }
2133
+ function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
2134
+ if (classified != null) return classified;
2135
+ if (hasConfiguredProvider !== false) return null;
2136
+ return {
2137
+ kind: "model_auth",
2138
+ providerId: replyProviderId,
2139
+ modelId: replyModelId,
2140
+ reason: "missing"
2141
+ };
2142
+ }
1548
2143
  function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1549
2144
  if (!messages || messages.length === 0) return false;
1550
2145
  return messages.some(
@@ -2073,13 +2668,56 @@ var RunnerConnection = class {
2073
2668
  }
2074
2669
  };
2075
2670
 
2671
+ // src/lib/tunnel/ready-marker.ts
2672
+ import { writeFileSync } from "fs";
2673
+ function writeTunnelReadyMarker(path, agentId) {
2674
+ try {
2675
+ writeFileSync(path, `${agentId}
2676
+ `);
2677
+ return { ok: true };
2678
+ } catch (error2) {
2679
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2680
+ }
2681
+ }
2682
+
2683
+ // src/lib/claude-usage-reporting.ts
2684
+ var VALID_MODES = ["auto", "on", "off"];
2685
+ function resolveClaudeUsageReportingMode(flagValue, env) {
2686
+ const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
2687
+ if (raw === void 0 || raw === "") {
2688
+ return { mode: "auto", warnings: [] };
2689
+ }
2690
+ const normalized = raw.trim().toLowerCase();
2691
+ if (VALID_MODES.includes(normalized)) {
2692
+ return { mode: normalized, warnings: [] };
2693
+ }
2694
+ const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
2695
+ return {
2696
+ mode: "auto",
2697
+ warnings: [
2698
+ `Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
2699
+ ]
2700
+ };
2701
+ }
2702
+ var BASE_REPORT_DELAY_MS = 10 * 6e4;
2703
+ var REPORT_DELAY_JITTER_FRACTION = 0.2;
2704
+ function nextReportDelayMs(random = Math.random) {
2705
+ const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
2706
+ return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
2707
+ }
2708
+ var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
2709
+ var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
2710
+ function claudeUsageFailureLogLevel(consecutiveFailures) {
2711
+ return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
2712
+ }
2713
+
2076
2714
  // src/lib/channels/driver.ts
2077
- import { homedir } from "os";
2715
+ import { homedir as homedir2 } from "os";
2078
2716
 
2079
2717
  // src/lib/file-push.ts
2080
2718
  import { randomUUID } from "crypto";
2081
2719
  import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
2082
- import { basename, dirname as dirname2, isAbsolute, join, relative, resolve as resolve2, sep } from "path";
2720
+ import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
2083
2721
  var FILE_MODE = 384;
2084
2722
  var DIRECTORY_MODE = 448;
2085
2723
  async function writePushedFile(request) {
@@ -2112,7 +2750,7 @@ async function writePushedFile(request) {
2112
2750
  const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
2113
2751
  dirname2(candidate)
2114
2752
  );
2115
- const realTarget = join(existingAncestor, ...missingSegments, basename(candidate));
2753
+ const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
2116
2754
  const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
2117
2755
  if (allowedDirectory === null) {
2118
2756
  return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
@@ -2148,7 +2786,7 @@ function expandAndValidate(requestedPath, homeDir) {
2148
2786
  if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
2149
2787
  return null;
2150
2788
  }
2151
- const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join(homeDir, requestedPath.slice(2)) : requestedPath;
2789
+ const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
2152
2790
  if (expanded.split(/[/\\]/).includes("..")) {
2153
2791
  return null;
2154
2792
  }
@@ -2221,13 +2859,13 @@ function contains(realDirectory, realTarget) {
2221
2859
  async function createMissingDirectories(existingAncestor, missingSegments) {
2222
2860
  let current = existingAncestor;
2223
2861
  for (const segment of missingSegments) {
2224
- current = join(current, segment);
2862
+ current = join2(current, segment);
2225
2863
  await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
2226
2864
  await chmod(current, DIRECTORY_MODE);
2227
2865
  }
2228
2866
  }
2229
2867
  async function writeAtomically(realTarget, content) {
2230
- const temporaryPath = join(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2868
+ const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2231
2869
  let handle;
2232
2870
  try {
2233
2871
  handle = await open2(temporaryPath, "wx", FILE_MODE);
@@ -2394,8 +3032,8 @@ async function applyOne(options, file) {
2394
3032
  await ack(options, file, "applied");
2395
3033
  return true;
2396
3034
  }
2397
- function durableDownloadCode(status) {
2398
- return status === 413 ? "file_too_large" : "write_failed";
3035
+ function durableDownloadCode(status2) {
3036
+ return status2 === 413 ? "file_too_large" : "write_failed";
2399
3037
  }
2400
3038
  async function downloadContent(options, file, label) {
2401
3039
  try {
@@ -2428,8 +3066,8 @@ async function downloadContent(options, file, label) {
2428
3066
  return { ok: false, terminal: false };
2429
3067
  }
2430
3068
  }
2431
- async function ack(options, file, status, reason) {
2432
- const outcome = `${status}${reason ? ` (${reason})` : ""}`;
3069
+ async function ack(options, file, status2, reason) {
3070
+ const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
2433
3071
  try {
2434
3072
  const res = await options.fetchImpl(
2435
3073
  `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
@@ -2439,7 +3077,7 @@ async function ack(options, file, status, reason) {
2439
3077
  Authorization: options.getAuthHeader(),
2440
3078
  "Content-Type": "application/json"
2441
3079
  },
2442
- body: JSON.stringify(reason ? { status, reason } : { status })
3080
+ body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
2443
3081
  }
2444
3082
  );
2445
3083
  if (!res.ok) {
@@ -2499,6 +3137,8 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
2499
3137
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
2500
3138
  var HEARTBEAT_MS = 6e4;
2501
3139
  var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
3140
+ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
3141
+ var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2502
3142
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
2503
3143
  var MAX_SUPERSEDED_CONVERSATIONS = 256;
2504
3144
  var ChannelAuthError = class extends Error {
@@ -2509,10 +3149,10 @@ var ChannelAuthError = class extends Error {
2509
3149
  };
2510
3150
  var ChannelTerminalError = class extends Error {
2511
3151
  status;
2512
- constructor(message, status) {
3152
+ constructor(message, status2) {
2513
3153
  super(message);
2514
3154
  this.name = "ChannelTerminalError";
2515
- this.status = status;
3155
+ this.status = status2;
2516
3156
  }
2517
3157
  };
2518
3158
  function backoffDelay(attempt, policy) {
@@ -2520,8 +3160,8 @@ function backoffDelay(attempt, policy) {
2520
3160
  const capped = Math.min(policy.maxDelayMs, exp);
2521
3161
  return Math.floor(Math.random() * capped);
2522
3162
  }
2523
- function isRetryableStatus(status) {
2524
- return status === 429 || status >= 500 && status <= 599;
3163
+ function isRetryableStatus(status2) {
3164
+ return status2 === 429 || status2 >= 500 && status2 <= 599;
2525
3165
  }
2526
3166
  var ChannelDriver = class _ChannelDriver {
2527
3167
  agentId;
@@ -2636,6 +3276,24 @@ var ChannelDriver = class _ChannelDriver {
2636
3276
  * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
2637
3277
  */
2638
3278
  readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
3279
+ /**
3280
+ * "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
3281
+ * `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
3282
+ * every ~2s drain until opencode's status becomes readable, but the
3283
+ * server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
3284
+ * on any non-`unresolved` outcome so the set cannot grow beyond the currently
3285
+ * unresolvable rows.
3286
+ */
3287
+ redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
3288
+ /**
3289
+ * First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
3290
+ * `pending` row is invisible to every cron arm (all require `status =
3291
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
3292
+ * nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
3293
+ * takes `dispatch` instead of `unresolved` (reusing the existing knob — see
3294
+ * ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
3295
+ */
3296
+ redriveUnresolvedSince = /* @__PURE__ */ new Map();
2639
3297
  /**
2640
3298
  * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
2641
3299
  * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
@@ -2724,23 +3382,23 @@ var ChannelDriver = class _ChannelDriver {
2724
3382
  * and stops opencode.
2725
3383
  */
2726
3384
  stopped = false;
2727
- constructor(config2) {
2728
- this.agentId = config2.agentId;
2729
- this.port = config2.port;
2730
- this.apiUrl = config2.apiUrl.replace(/\/$/, "");
2731
- this.getAuthHeader = config2.getAuthHeader;
2732
- this.conversationFilter = config2.conversationFilter ?? null;
2733
- this.retry = { ...DEFAULT_RETRY_POLICY, ...config2.retry };
2734
- this.log = config2.log ?? (() => {
3385
+ constructor(config) {
3386
+ this.agentId = config.agentId;
3387
+ this.port = config.port;
3388
+ this.apiUrl = config.apiUrl.replace(/\/$/, "");
3389
+ this.getAuthHeader = config.getAuthHeader;
3390
+ this.conversationFilter = config.conversationFilter ?? null;
3391
+ this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
3392
+ this.log = config.log ?? (() => {
2735
3393
  });
2736
- this.fetchImpl = config2.fetchImpl ?? fetch;
2737
- this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
2738
- this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
2739
- this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2740
- this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2741
- this.now = config2.now ?? (() => Date.now());
2742
- this.fileSyncDirectories = config2.fileSyncDirectories ?? [];
2743
- this.homeDir = config2.homeDir ?? homedir();
3394
+ this.fetchImpl = config.fetchImpl ?? fetch;
3395
+ this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
3396
+ this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
3397
+ this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
3398
+ this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
3399
+ this.now = config.now ?? (() => Date.now());
3400
+ this.fileSyncDirectories = config.fileSyncDirectories ?? [];
3401
+ this.homeDir = config.homeDir ?? homedir2();
2744
3402
  }
2745
3403
  /** The IPv4-loopback base URL for the local `opencode serve`. */
2746
3404
  get opencodeBase() {
@@ -2975,6 +3633,12 @@ var ChannelDriver = class _ChannelDriver {
2975
3633
  skippedAlreadyDispatched += 1;
2976
3634
  continue;
2977
3635
  }
3636
+ if (message.opencode_message_id) {
3637
+ const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
3638
+ if (outcome !== "dispatch") {
3639
+ break;
3640
+ }
3641
+ }
2978
3642
  const options = {
2979
3643
  agent: message.opencode_agent ?? void 0,
2980
3644
  model: message.opencode_model ?? void 0
@@ -3033,36 +3697,264 @@ var ChannelDriver = class _ChannelDriver {
3033
3697
  });
3034
3698
  });
3035
3699
  this.log({
3036
- level: "error",
3037
- message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
3700
+ level: "error",
3701
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
3702
+ conversation_id: conv.id,
3703
+ message_id: message.id
3704
+ });
3705
+ break;
3706
+ }
3707
+ if (opencodeMessageId === null) {
3708
+ this.log({
3709
+ level: "warn",
3710
+ message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
3711
+ conversation_id: conv.id,
3712
+ message_id: message.id
3713
+ });
3714
+ continue;
3715
+ }
3716
+ this.dispatched.add(message.id);
3717
+ this.registerInFlight(conv, sessionId, message, opencodeMessageId);
3718
+ dispatched += 1;
3719
+ void this.postSignal(conv.id, message.id, "dispatched");
3720
+ }
3721
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3722
+ this.log({
3723
+ level: "warn",
3724
+ message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
3725
+ conversation_id: conv.id
3726
+ });
3727
+ }
3728
+ this.ensureWatcherRunning(sessionId);
3729
+ return dispatched;
3730
+ }
3731
+ /**
3732
+ * Poll a session's message list for the re-drive fence (#965), via the
3733
+ * INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
3734
+ * hits the global `fetch` and would bypass the same override every other
3735
+ * opencode poll in this file respects. Mirrors `readoptProcessing`'s own
3736
+ * snapshot fetch (`:3081-3111`). `null` = unreadable (non-OK response,
3737
+ * non-array body, or a network exception) — treated as "can't observe",
3738
+ * never as "confirmed gone".
3739
+ */
3740
+ async pollSessionMessagesForRedrive(conv, message, sessionId) {
3741
+ try {
3742
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
3743
+ if (!res.ok) {
3744
+ this.log({
3745
+ level: "warn",
3746
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status} \u2014 treating as unreadable this tick`,
3747
+ conversation_id: conv.id,
3748
+ message_id: message.id
3749
+ });
3750
+ return null;
3751
+ }
3752
+ const body = await res.json();
3753
+ if (!Array.isArray(body)) {
3754
+ this.log({
3755
+ level: "warn",
3756
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
3757
+ conversation_id: conv.id,
3758
+ message_id: message.id
3759
+ });
3760
+ return null;
3761
+ }
3762
+ return body;
3763
+ } catch (err) {
3764
+ this.log({
3765
+ level: "warn",
3766
+ message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3767
+ conversation_id: conv.id,
3768
+ message_id: message.id
3769
+ });
3770
+ return null;
3771
+ }
3772
+ }
3773
+ /**
3774
+ * The re-drive fence for a `pending` row that already carries a stored
3775
+ * `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
3776
+ * least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
3777
+ * The lifecycle cron can falsely reclaim a `processing` row back to `pending`
3778
+ * mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
3779
+ * without this fence the drain loop would re-`prompt_async` the SAME turn a
3780
+ * second time against live GitHub state. Mirrors `readoptOne`'s job for the
3781
+ * `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
3782
+ * needed here because `refusedSessionId` already handles the one case
3783
+ * (#553 abandoned session) that path exists for.
3784
+ *
3785
+ * Only `ChannelAuthError` propagates; every other failure resolves to
3786
+ * `unresolved` and is retried whole on the next ~2s drain tick.
3787
+ */
3788
+ async resolveRedrive(conv, sessionId, message, refusedSessionId) {
3789
+ const ocId = message.opencode_message_id ?? null;
3790
+ if (refusedSessionId) {
3791
+ this.clearRedriveUnresolved(message.id);
3792
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3793
+ return "dispatch";
3794
+ }
3795
+ const messages = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
3796
+ if (messages == null || messages.length === 0) {
3797
+ return this.resolveRedriveUnresolved(conv, message);
3798
+ }
3799
+ const state = messageRunState(messages, ocId ?? "");
3800
+ if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
3801
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3802
+ if (ongoing === false) {
3803
+ this.log({
3804
+ level: "info",
3805
+ message: `Re-drive: message ${message.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
3038
3806
  conversation_id: conv.id,
3039
3807
  message_id: message.id
3040
3808
  });
3041
- break;
3809
+ this.clearRedriveUnresolved(message.id);
3810
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3811
+ return "dispatch";
3042
3812
  }
3043
- if (opencodeMessageId === null) {
3813
+ }
3814
+ if (state === "done" || state === "failed") {
3815
+ return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
3816
+ }
3817
+ if (state === "running" || state === "queued") {
3818
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3819
+ if (ongoing === true) {
3820
+ return this.reattachRedrive(conv, sessionId, message, ocId);
3821
+ }
3822
+ if (ongoing === false) {
3823
+ this.clearRedriveUnresolved(message.id);
3824
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3825
+ return "dispatch";
3826
+ }
3827
+ return this.resolveRedriveUnresolved(conv, message);
3828
+ }
3829
+ this.clearRedriveUnresolved(message.id);
3830
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3831
+ return "dispatch";
3832
+ }
3833
+ /**
3834
+ * The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
3835
+ * opencode's own status map — undo the false reclaim instead of starting a
3836
+ * second turn.
3837
+ */
3838
+ async reattachRedrive(conv, sessionId, message, ocId) {
3839
+ let anchorMs;
3840
+ const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
3841
+ if (!Number.isNaN(parsed)) {
3842
+ anchorMs = parsed;
3843
+ } else {
3844
+ anchorMs = this.now();
3845
+ this.log({
3846
+ level: "error",
3847
+ message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
3848
+ conversation_id: conv.id,
3849
+ message_id: message.id
3850
+ });
3851
+ }
3852
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3853
+ try {
3854
+ await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
3855
+ } catch (err) {
3856
+ if (err instanceof ChannelAuthError) throw err;
3857
+ this.log({
3858
+ level: "warn",
3859
+ message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3860
+ conversation_id: conv.id,
3861
+ message_id: message.id
3862
+ });
3863
+ return "unresolved";
3864
+ }
3865
+ this.clearRedriveUnresolved(message.id);
3866
+ this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
3867
+ this.dispatched.add(message.id);
3868
+ this.readopted.add(message.id);
3869
+ this.ensureWatcherRunning(sessionId);
3870
+ const watchedForMs = this.now() - anchorMs;
3871
+ void this.postSignal(conv.id, message.id, "redrive_reattached", {
3872
+ watched_for_ms: watchedForMs
3873
+ });
3874
+ this.log({
3875
+ level: "warn",
3876
+ message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
3877
+ conversation_id: conv.id,
3878
+ message_id: message.id
3879
+ });
3880
+ return "reattached";
3881
+ }
3882
+ /**
3883
+ * The `settled` outcome (Task 3.2): the prior turn already finished (or
3884
+ * errored) while nobody was watching — deliver/report it instead of re-running.
3885
+ * Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
3886
+ * (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
3887
+ * drain, same as any other non-auth failure). The restart-abort carve-out that
3888
+ * keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
3889
+ * so a row reaching this `failed` branch is a GENUINE failure.
3890
+ */
3891
+ async settleRedrive(conv, sessionId, message, ocId, messages, state) {
3892
+ try {
3893
+ if (state === "done") {
3894
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3895
+ const usage = messageUsage(messages, ocId ?? "");
3044
3896
  this.log({
3045
- level: "warn",
3046
- message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
3897
+ level: "info",
3898
+ message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
3047
3899
  conversation_id: conv.id,
3048
3900
  message_id: message.id
3049
3901
  });
3050
- continue;
3902
+ await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
3903
+ } else {
3904
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
3905
+ const usage = messageUsage(messages, ocId ?? "");
3906
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3907
+ this.log({
3908
+ level: "error",
3909
+ message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
3910
+ conversation_id: conv.id,
3911
+ message_id: message.id
3912
+ });
3913
+ await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
3051
3914
  }
3052
- this.dispatched.add(message.id);
3053
- this.registerInFlight(conv, sessionId, message, opencodeMessageId);
3054
- dispatched += 1;
3055
- void this.postSignal(conv.id, message.id, "dispatched");
3056
- }
3057
- if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3915
+ } catch (err) {
3916
+ if (err instanceof ChannelAuthError) throw err;
3058
3917
  this.log({
3059
3918
  level: "warn",
3060
- message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
3061
- conversation_id: conv.id
3919
+ message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3920
+ conversation_id: conv.id,
3921
+ message_id: message.id
3062
3922
  });
3923
+ return "unresolved";
3063
3924
  }
3064
- this.ensureWatcherRunning(sessionId);
3065
- return dispatched;
3925
+ this.clearRedriveUnresolved(message.id);
3926
+ void this.postSignal(conv.id, message.id, "redrive_settled");
3927
+ return "settled";
3928
+ }
3929
+ /**
3930
+ * The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
3931
+ * observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
3932
+ * A `pending` row is invisible to every cron arm (all require `status =
3933
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
3934
+ * nothing driving it — bound it to the existing `pausedMaxWaitMs` window
3935
+ * (reusing the knob, not a new constant) and take `dispatch` once elapsed.
3936
+ */
3937
+ resolveRedriveUnresolved(conv, message) {
3938
+ const now = this.now();
3939
+ const since = this.redriveUnresolvedSince.get(message.id);
3940
+ if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
3941
+ this.clearRedriveUnresolved(message.id);
3942
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3943
+ return "dispatch";
3944
+ }
3945
+ if (since === void 0) {
3946
+ this.redriveUnresolvedSince.set(message.id, now);
3947
+ }
3948
+ if (!this.redriveUnresolvedSignalled.has(message.id)) {
3949
+ this.redriveUnresolvedSignalled.add(message.id);
3950
+ void this.postSignal(conv.id, message.id, "redrive_unresolved");
3951
+ }
3952
+ return "unresolved";
3953
+ }
3954
+ /** Clear both `unresolved`-bound trackers for a row (any non-`unresolved` outcome). */
3955
+ clearRedriveUnresolved(messageId) {
3956
+ this.redriveUnresolvedSince.delete(messageId);
3957
+ this.redriveUnresolvedSignalled.delete(messageId);
3066
3958
  }
3067
3959
  /**
3068
3960
  * Record that `sessionId` is no longer a valid binding for `conversationId`
@@ -3125,7 +4017,12 @@ var ChannelDriver = class _ChannelDriver {
3125
4017
  const directory = await this.resolveOpenCodeDirectory();
3126
4018
  const sessionId = await createOpenCodeSession(this.port, directory);
3127
4019
  this.sessions.set(conversationId, sessionId);
3128
- await this.persistSession(conversationId, sessionId).catch(() => {
4020
+ await this.persistSession(conversationId, sessionId).catch((err) => {
4021
+ this.log({
4022
+ level: "warn",
4023
+ message: `Persisting the OpenCode session binding ${sessionId.slice(0, 8)} for conversation ${conversationId.slice(0, 8)} failed (best-effort, not retried) \u2014 the completion PATCH also carries opencode_session_id, so the binding is repaired when the turn finishes: ${err instanceof Error ? err.message : String(err)}`,
4024
+ conversation_id: conversationId
4025
+ });
3129
4026
  });
3130
4027
  return sessionId;
3131
4028
  }
@@ -3317,12 +4214,17 @@ var ChannelDriver = class _ChannelDriver {
3317
4214
  stuckReported: false,
3318
4215
  lastAliveAt: 0,
3319
4216
  aliveInFlight: false,
4217
+ titleSynced: false,
4218
+ titleSyncInFlight: false,
3320
4219
  awaitingHumanLatched: false,
3321
4220
  pausedOnQuestion: false,
3322
4221
  pausedOnPermission: false,
3323
4222
  pausedClearConfirmed: false,
3324
4223
  pausedInFlight: false,
3325
- deliveryDeadlineAnchored: false
4224
+ deliveryDeadlineAnchored: false,
4225
+ b2PinnedSinceMs: 0,
4226
+ b2LastDescendantCheckMs: 0,
4227
+ b2AbandonedSignalled: false
3326
4228
  });
3327
4229
  }
3328
4230
  /**
@@ -3337,9 +4239,10 @@ var ChannelDriver = class _ChannelDriver {
3337
4239
  * opencode reports ACTIVELY `running` is watched to completion (its liveness
3338
4240
  * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
3339
4241
  * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
3340
- * handed to the cron. The old "the `deadline` must settle before the ~15-min
3341
- * cron or they double-drive" reasoning is superseded: liveness now settles the
3342
- * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
4242
+ * handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
4243
+ * runner still holds; a reclaimed row that already ran is never re-dispatched
4244
+ * while opencode reports its turn ongoing (readopt's own gate here, and the
4245
+ * `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
3343
4246
  * (only the appear-guard uses it).
3344
4247
  *
3345
4248
  * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
@@ -3390,12 +4293,17 @@ var ChannelDriver = class _ChannelDriver {
3390
4293
  // with no extra `re_adopted` signal needed (folds old WI-6).
3391
4294
  lastAliveAt: 0,
3392
4295
  aliveInFlight: false,
4296
+ titleSynced: false,
4297
+ titleSyncInFlight: false,
3393
4298
  awaitingHumanLatched: false,
3394
4299
  pausedOnQuestion: false,
3395
4300
  pausedOnPermission: false,
3396
4301
  pausedClearConfirmed: false,
3397
4302
  pausedInFlight: false,
3398
- deliveryDeadlineAnchored: false
4303
+ deliveryDeadlineAnchored: false,
4304
+ b2PinnedSinceMs: 0,
4305
+ b2LastDescendantCheckMs: 0,
4306
+ b2AbandonedSignalled: false
3399
4307
  });
3400
4308
  }
3401
4309
  /**
@@ -3557,58 +4465,7 @@ var ChannelDriver = class _ChannelDriver {
3557
4465
  }
3558
4466
  }
3559
4467
  if (state === "done") {
3560
- this.anchorDeliveryDeadline(inFlight);
3561
- if (!inFlight.done) {
3562
- this.log({
3563
- level: "info",
3564
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
3565
- conversation_id: conv.id,
3566
- message_id: inFlight.evidentMessageId
3567
- });
3568
- const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
3569
- const usage = messageUsage(messages, inFlight.opencodeMessageId);
3570
- try {
3571
- await this.markDone(
3572
- conv.id,
3573
- inFlight.evidentMessageId,
3574
- sessionId,
3575
- inFlight.opencodeMessageId,
3576
- title,
3577
- usage
3578
- );
3579
- } catch (err) {
3580
- if (err instanceof ChannelAuthError) throw err;
3581
- if (err instanceof ChannelTerminalError) {
3582
- this.log({
3583
- level: "warn",
3584
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
3585
- conversation_id: conv.id,
3586
- message_id: inFlight.evidentMessageId
3587
- });
3588
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3589
- return;
3590
- }
3591
- if (this.now() >= inFlight.deadline) {
3592
- this.log({
3593
- level: "warn",
3594
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
3595
- conversation_id: conv.id,
3596
- message_id: inFlight.evidentMessageId
3597
- });
3598
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3599
- return;
3600
- }
3601
- this.log({
3602
- level: "warn",
3603
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
3604
- conversation_id: conv.id,
3605
- message_id: inFlight.evidentMessageId
3606
- });
3607
- return;
3608
- }
3609
- inFlight.done = true;
3610
- }
3611
- this.removeInFlight(watcher, inFlight.evidentMessageId);
4468
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3612
4469
  return;
3613
4470
  }
3614
4471
  if (state === "failed") {
@@ -3622,8 +4479,16 @@ var ChannelDriver = class _ChannelDriver {
3622
4479
  message_id: inFlight.evidentMessageId
3623
4480
  });
3624
4481
  const usage = messageUsage(messages, inFlight.opencodeMessageId);
4482
+ const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
3625
4483
  try {
3626
- await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2, usage);
4484
+ await this.markFailed(
4485
+ conv.id,
4486
+ inFlight.evidentMessageId,
4487
+ sessionId,
4488
+ error2,
4489
+ usage,
4490
+ failure
4491
+ );
3627
4492
  } catch (err) {
3628
4493
  if (err instanceof ChannelAuthError) throw err;
3629
4494
  if (err instanceof ChannelTerminalError) {
@@ -3668,6 +4533,44 @@ var ChannelDriver = class _ChannelDriver {
3668
4533
  });
3669
4534
  }
3670
4535
  const activelyRunning = state === "running" && !awaitingHuman;
4536
+ const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
4537
+ const snapshotReadable = messages != null && messages.length > 0;
4538
+ if (!pinnedNow) {
4539
+ if (snapshotReadable) {
4540
+ inFlight.b2PinnedSinceMs = 0;
4541
+ inFlight.b2LastDescendantCheckMs = 0;
4542
+ inFlight.b2AbandonedSignalled = false;
4543
+ }
4544
+ } else {
4545
+ if (inFlight.b2AbandonedSignalled) {
4546
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
4547
+ return;
4548
+ }
4549
+ if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
4550
+ const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
4551
+ if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
4552
+ inFlight.b2LastDescendantCheckMs = this.now();
4553
+ const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
4554
+ if (isB2AbandonmentConfirmed({
4555
+ pinnedForMs,
4556
+ minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
4557
+ descendantOngoing
4558
+ })) {
4559
+ inFlight.b2AbandonedSignalled = true;
4560
+ this.log({
4561
+ level: "warn",
4562
+ message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
4563
+ conversation_id: conv.id,
4564
+ message_id: id
4565
+ });
4566
+ void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
4567
+ watched_for_ms: pinnedForMs
4568
+ });
4569
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
4570
+ return;
4571
+ }
4572
+ }
4573
+ }
3671
4574
  if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
3672
4575
  this.log({
3673
4576
  level: "warn",
@@ -3687,6 +4590,18 @@ var ChannelDriver = class _ChannelDriver {
3687
4590
  inFlight.aliveInFlight = false;
3688
4591
  if (ok) inFlight.lastAliveAt = this.now();
3689
4592
  });
4593
+ if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
4594
+ inFlight.titleSyncInFlight = true;
4595
+ void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
4596
+ if (!title) {
4597
+ inFlight.titleSyncInFlight = false;
4598
+ return;
4599
+ }
4600
+ const ok = await this.patchConversationTitle(conv.id, title);
4601
+ inFlight.titleSyncInFlight = false;
4602
+ if (ok) inFlight.titleSynced = true;
4603
+ });
4604
+ }
3690
4605
  }
3691
4606
  if (awaitingHuman) {
3692
4607
  if (!inFlight.awaitingHumanLatched) {
@@ -3724,6 +4639,70 @@ var ChannelDriver = class _ChannelDriver {
3724
4639
  this.removeInFlight(watcher, inFlight.evidentMessageId);
3725
4640
  }
3726
4641
  }
4642
+ /**
4643
+ * Settle a message whose run-state has resolved `'done'` — extracted verbatim
4644
+ * (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
4645
+ * inline `state === 'done'` branch body, so a SECOND caller (the #721
4646
+ * b2-abandonment resolution) can reach the exact same completion behavior
4647
+ * (delivery-deadline anchoring, title resolution, usage extraction, and
4648
+ * `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
4649
+ * and risking the two copies silently drifting apart.
4650
+ */
4651
+ async settleMessageDone(sessionId, watcher, inFlight, messages) {
4652
+ const conv = watcher.conv;
4653
+ this.anchorDeliveryDeadline(inFlight);
4654
+ if (!inFlight.done) {
4655
+ this.log({
4656
+ level: "info",
4657
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
4658
+ conversation_id: conv.id,
4659
+ message_id: inFlight.evidentMessageId
4660
+ });
4661
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
4662
+ const usage = messageUsage(messages, inFlight.opencodeMessageId);
4663
+ try {
4664
+ await this.markDone(
4665
+ conv.id,
4666
+ inFlight.evidentMessageId,
4667
+ sessionId,
4668
+ inFlight.opencodeMessageId,
4669
+ title,
4670
+ usage
4671
+ );
4672
+ } catch (err) {
4673
+ if (err instanceof ChannelAuthError) throw err;
4674
+ if (err instanceof ChannelTerminalError) {
4675
+ this.log({
4676
+ level: "warn",
4677
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
4678
+ conversation_id: conv.id,
4679
+ message_id: inFlight.evidentMessageId
4680
+ });
4681
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4682
+ return;
4683
+ }
4684
+ if (this.now() >= inFlight.deadline) {
4685
+ this.log({
4686
+ level: "warn",
4687
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
4688
+ conversation_id: conv.id,
4689
+ message_id: inFlight.evidentMessageId
4690
+ });
4691
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4692
+ return;
4693
+ }
4694
+ this.log({
4695
+ level: "warn",
4696
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
4697
+ conversation_id: conv.id,
4698
+ message_id: inFlight.evidentMessageId
4699
+ });
4700
+ return;
4701
+ }
4702
+ inFlight.done = true;
4703
+ }
4704
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4705
+ }
3727
4706
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
3728
4707
  /**
3729
4708
  * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
@@ -3823,7 +4802,10 @@ var ChannelDriver = class _ChannelDriver {
3823
4802
  * re-dispatched (at most once, see `forceReadoptRun`):
3824
4803
  * - `done` → `markDone` now (guarded like the watcher's done branch);
3825
4804
  * - `failed` → `markFailed` with the surfaced error (issue #182), so an
3826
- * errored turn is reported failed on restart, NOT re-dispatched;
4805
+ * errored turn is reported failed on restart, NOT re-dispatched
4806
+ * EXCEPT a restart-ABORTED turn under a not-ongoing session,
4807
+ * which is a restart orphan wearing a terminal error and is
4808
+ * re-dispatched instead (issue #1310, see the branch below);
3827
4809
  * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
3828
4810
  * tracking the stored id so the reply correlates by it;
3829
4811
  * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
@@ -3887,9 +4869,19 @@ var ChannelDriver = class _ChannelDriver {
3887
4869
  void this.postSignal(row.conversation_id, row.id, "readopt_done");
3888
4870
  return;
3889
4871
  }
3890
- if (state === "failed") {
4872
+ const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
4873
+ if (restartAborted) {
4874
+ this.log({
4875
+ level: "info",
4876
+ message: `Re-adopt: message ${row.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
4877
+ conversation_id: row.conversation_id,
4878
+ message_id: row.id
4879
+ });
4880
+ }
4881
+ if (state === "failed" && !restartAborted) {
3891
4882
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
3892
4883
  const usage = messageUsage(messages, ocId ?? "");
4884
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3893
4885
  this.log({
3894
4886
  level: "error",
3895
4887
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -3897,7 +4889,7 @@ var ChannelDriver = class _ChannelDriver {
3897
4889
  message_id: row.id
3898
4890
  });
3899
4891
  try {
3900
- await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
4892
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
3901
4893
  } catch (err) {
3902
4894
  if (err instanceof ChannelAuthError) throw err;
3903
4895
  if (err instanceof ChannelTerminalError) {
@@ -4164,7 +5156,8 @@ var ChannelDriver = class _ChannelDriver {
4164
5156
  opencode_model: row.opencode_model,
4165
5157
  source_message_id: row.source_message_id,
4166
5158
  slack_user_id: row.slack_user_id,
4167
- attachments: row.attachments ?? null
5159
+ attachments: row.attachments ?? null,
5160
+ opencode_message_id: row.opencode_message_id
4168
5161
  };
4169
5162
  }
4170
5163
  /**
@@ -4303,6 +5296,47 @@ var ChannelDriver = class _ChannelDriver {
4303
5296
  }
4304
5297
  return false;
4305
5298
  }
5299
+ /**
5300
+ * Tri-state variant of the upward parentID membership walk (#721), used ONLY
5301
+ * by `isAnyDescendantSessionOngoing`. Walks the SAME cached
5302
+ * `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
5303
+ * `sessionBelongsTo`, which deliberately collapses "confirmed not a
5304
+ * descendant" and "the walk's fetch failed" into the same `false` (safe for
5305
+ * its OTHER callers: interaction attribution and the recovery-path
5306
+ * `isAnyDescendantSessionAlive`, both of which just retry next tick with no
5307
+ * safety consequence either way) — this variant keeps those two outcomes
5308
+ * SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
5309
+ * (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
5310
+ * not ongoing".
5311
+ *
5312
+ * Return contract:
5313
+ * - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
5314
+ * - `false` → the walk reached a definitive, parent-less root session
5315
+ * WITHOUT ever matching `rootSessionId` — `sessionId` is
5316
+ * CONFIRMED NOT a descendant of it.
5317
+ * - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
5318
+ * through the walk (`resolveSessionParent` returned `undefined`),
5319
+ * or the depth cap (32) was hit without a definitive answer (a
5320
+ * pathological/cyclic chain proves nothing either way). NEVER
5321
+ * treat this the same as `false` — see `sessionBelongsTo`'s own
5322
+ * doc comment above for why that collapse is safe THERE but not
5323
+ * here.
5324
+ *
5325
+ * `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
5326
+ * to the live-path descendant check, not a modification of shared code used
5327
+ * by interaction attribution or the recovery path.
5328
+ */
5329
+ async resolveSessionMembership(sessionId, rootSessionId) {
5330
+ let current = sessionId;
5331
+ for (let depth = 0; current && depth < 32; depth++) {
5332
+ if (current === rootSessionId) return true;
5333
+ const parent = await this.resolveSessionParent(current);
5334
+ if (parent === void 0) return null;
5335
+ if (parent === null) return false;
5336
+ current = parent;
5337
+ }
5338
+ return null;
5339
+ }
4306
5340
  /**
4307
5341
  * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
4308
5342
  * `null` for a root session (no parent) and `undefined` when opencode is
@@ -4387,6 +5421,54 @@ var ChannelDriver = class _ChannelDriver {
4387
5421
  }
4388
5422
  return null;
4389
5423
  }
5424
+ /**
5425
+ * Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
5426
+ * session title onto the conversation via the PLAIN conversation-update
5427
+ * endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
5428
+ * message-status endpoint `markProcessing`/`markDone` use. Deliberately a
5429
+ * separate, lighter call: it carries no `status`, so it cannot re-trigger the
5430
+ * `processing`/`done` transition side effects (Slack notices, activity-log
5431
+ * rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
5432
+ * ever touches `conversations.title`. That route (`routes/conversations.ts`)
5433
+ * skips a title write matching the stored value, so a redundant call with the
5434
+ * same title is a real no-op — it does not bump `updated_at`, which the
5435
+ * conversation list sorts and paginates on. (Note this is a DIFFERENT guard
5436
+ * from `threads.ts`'s "non-empty AND changed" one, which only covers the
5437
+ * message-status PATCH; the non-empty half is enforced here instead, by
5438
+ * `resolveSessionTitle` never returning an empty/placeholder title.)
5439
+ *
5440
+ * Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
5441
+ * is logged and the title is simply retried on the next heartbeat tick (the
5442
+ * caller only latches `titleSynced` on `true`).
5443
+ */
5444
+ async patchConversationTitle(conversationId, title) {
5445
+ try {
5446
+ const res = await this.fetchImpl(
5447
+ `${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
5448
+ {
5449
+ method: "PATCH",
5450
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
5451
+ body: JSON.stringify({ title })
5452
+ }
5453
+ );
5454
+ if (!res.ok) {
5455
+ this.log({
5456
+ level: "debug",
5457
+ message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
5458
+ conversation_id: conversationId
5459
+ });
5460
+ return false;
5461
+ }
5462
+ return true;
5463
+ } catch (err) {
5464
+ this.log({
5465
+ level: "debug",
5466
+ message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
5467
+ conversation_id: conversationId
5468
+ });
5469
+ return false;
5470
+ }
5471
+ }
4390
5472
  /**
4391
5473
  * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
4392
5474
  * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
@@ -4419,31 +5501,109 @@ var ChannelDriver = class _ChannelDriver {
4419
5501
  * longer exists". The inversion lives in the caller; this method just reports
4420
5502
  * true/false/null faithfully.
4421
5503
  *
4422
- * VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
4423
- * (already proven by the existing child-session interaction tests, via
4424
- * `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
4425
- * terminal state. We do NOT depend on any session-level `busy`/`idle` field —
4426
- * there is none on `GET /session/:id`; OpenCode's busy state is in-memory
4427
- * `SessionStatus` only.
5504
+ * VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
5505
+ * (already proven by the existing child-session interaction tests, via
5506
+ * `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
5507
+ * terminal state. We do NOT depend on any session-level `busy`/`idle` field —
5508
+ * there is none on `GET /session/:id`; OpenCode's busy state is in-memory
5509
+ * `SessionStatus` only.
5510
+ */
5511
+ async isAnyDescendantSessionAlive(rootSessionId) {
5512
+ const sessions = await listSessions(this.port);
5513
+ if (!sessions) {
5514
+ this.log({
5515
+ level: "warn",
5516
+ message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
5517
+ });
5518
+ return null;
5519
+ }
5520
+ for (const candidate of sessions) {
5521
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
5522
+ if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
5523
+ const childMsgs = await getSessionMessages(this.port, candidate.id);
5524
+ if (isSessionActivelyGenerating(childMsgs)) {
5525
+ return true;
5526
+ }
5527
+ }
5528
+ return false;
5529
+ }
5530
+ /**
5531
+ * LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
5532
+ * sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
5533
+ * in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
5534
+ *
5535
+ * Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
5536
+ * cross-check above): that method judges liveness from the child's OWN
5537
+ * TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
5538
+ * on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
5539
+ * path the local opencode server IS running, so its in-memory status map is
5540
+ * live and authoritative — and per ADR-0047 §4a ("the child has its own entry
5541
+ * [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
5542
+ * ENTIRE turn (including any tool call it is itself executing), not a
5543
+ * per-message transcript snapshot. This sidesteps the "child's own tool is
5544
+ * executing, between its step's completion and the next generation step"
5545
+ * transcript gap that a transcript-based check would need a second,
5546
+ * sustained-window bound to guard against — it is simply not derived from
5547
+ * message timestamps at all.
5548
+ *
5549
+ * Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
5550
+ * status, as the recovery path does per §4a)? Because on the LIVE path the
5551
+ * root session can be shared: a SECOND, unrelated user message can land on the
5552
+ * SAME session (issue #721's own root cause) and keep the root `busy` for a
5553
+ * reason that has nothing to do with THIS message's delegation. A `task`
5554
+ * descendant session is spawned for exactly one delegated turn and never
5555
+ * reused, so its OWN status-map entry is unambiguous evidence about that one
5556
+ * delegation — which the root's status is not.
5557
+ *
5558
+ * Why membership is checked via `resolveSessionMembership`, NOT
5559
+ * `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
5560
+ * `GET /session/:id` fetch failure into "not a descendant", which would
5561
+ * silently drop a genuinely-live candidate from consideration on the one
5562
+ * unlucky tick its membership-walk fetch hiccups (#721).
5563
+ * `resolveSessionMembership` keeps that failure mode as a distinct `null`
5564
+ * (indeterminate) so it is folded into THIS method's own `indeterminate` flag
5565
+ * instead.
5566
+ *
5567
+ * Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
5568
+ * - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
5569
+ * - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
5570
+ * confirmed either way (`resolveSessionMembership` never
5571
+ * returned `null`), and every CONFIRMED descendant's status read
5572
+ * succeeded and is not ongoing (includes "no descendant session
5573
+ * exists at all" — e.g. a plain, non-`task` tool call).
5574
+ * - `null` → INDETERMINATE: `listSessions` failed, OR at least one
5575
+ * candidate's MEMBERSHIP could not be confirmed
5576
+ * (`resolveSessionMembership` returned `null` — a fetch failure
5577
+ * or pathological chain partway through the parent walk), OR at
5578
+ * least one CONFIRMED descendant's `isSessionOngoing` read
5579
+ * failed — and no OTHER candidate was already confirmed `true`.
5580
+ * The caller MUST NOT treat `null` the same as `false` here
5581
+ * (unlike the recovery cross-check's contract) — see
5582
+ * `isB2AbandonmentConfirmed`.
4428
5583
  */
4429
- async isAnyDescendantSessionAlive(rootSessionId) {
5584
+ async isAnyDescendantSessionOngoing(rootSessionId) {
4430
5585
  const sessions = await listSessions(this.port);
4431
5586
  if (!sessions) {
4432
5587
  this.log({
4433
5588
  level: "warn",
4434
- message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
5589
+ message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
4435
5590
  });
4436
5591
  return null;
4437
5592
  }
5593
+ let indeterminate = false;
4438
5594
  for (const candidate of sessions) {
4439
5595
  if (!candidate?.id || candidate.id === rootSessionId) continue;
4440
- if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
4441
- const childMsgs = await getSessionMessages(this.port, candidate.id);
4442
- if (isSessionActivelyGenerating(childMsgs)) {
4443
- return true;
5596
+ const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
5597
+ if (membership === null) {
5598
+ indeterminate = true;
5599
+ continue;
4444
5600
  }
5601
+ if (membership === false) continue;
5602
+ const ongoing = await isSessionOngoing(this.port, candidate.id);
5603
+ if (ongoing === true) return true;
5604
+ if (ongoing === null) indeterminate = true;
4445
5605
  }
4446
- return false;
5606
+ return indeterminate ? null : false;
4447
5607
  }
4448
5608
  /**
4449
5609
  * Cheap decision-telemetry label for a running row's LAST correlated reply
@@ -4589,11 +5749,11 @@ var ChannelDriver = class _ChannelDriver {
4589
5749
  * `ensureSession` guard, not this suppression, is what makes the self-heal
4590
5750
  * stick.
4591
5751
  */
4592
- sessionIdBody(sessionId, conversationId, messageId, status) {
5752
+ sessionIdBody(sessionId, conversationId, messageId, status2) {
4593
5753
  if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
4594
5754
  this.log({
4595
5755
  level: "debug",
4596
- message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
5756
+ message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
4597
5757
  conversation_id: conversationId,
4598
5758
  message_id: messageId
4599
5759
  });
@@ -4708,7 +5868,7 @@ var ChannelDriver = class _ChannelDriver {
4708
5868
  * exists but is wedged, so the next attempt must get a fresh one
4709
5869
  * instead of reusing it — see WI-1's server-side null-clearing PATCH).
4710
5870
  */
4711
- async markFailed(conversationId, messageId, sessionId, error2, usage) {
5871
+ async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
4712
5872
  const body = { status: "failed" };
4713
5873
  if (sessionId === null) {
4714
5874
  body.opencode_session_id = null;
@@ -4717,6 +5877,12 @@ var ChannelDriver = class _ChannelDriver {
4717
5877
  }
4718
5878
  if (error2 !== void 0) body.error = error2;
4719
5879
  if (usage) Object.assign(body, usage);
5880
+ if (failure) {
5881
+ body.failure_kind = failure.kind;
5882
+ body.failure_provider_id = failure.providerId;
5883
+ body.failure_model_id = failure.modelId;
5884
+ body.failure_reason = failure.reason;
5885
+ }
4720
5886
  await this.callWithRetry(
4721
5887
  "marking message as failed",
4722
5888
  () => this.fetchImpl(
@@ -4729,6 +5895,29 @@ var ChannelDriver = class _ChannelDriver {
4729
5895
  )
4730
5896
  );
4731
5897
  }
5898
+ /**
5899
+ * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
5900
+ *
5901
+ * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
5902
+ * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
5903
+ * to the P1-2b zero-provider check — one extra loopback call to
5904
+ * `hasAnyConfiguredProvider`, only reached when the structured classifier
5905
+ * couldn't place it. Fails open (never throws): a fallback probe failure
5906
+ * (`null`/indeterminate) leaves the classification `null`, which produces
5907
+ * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
5908
+ */
5909
+ async classifyModelAuthFailure(messages, userMessageId) {
5910
+ const classified = messageFailure(messages, userMessageId);
5911
+ if (classified != null) return classified;
5912
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
5913
+ const hasProvider = await hasAnyConfiguredProvider(this.port);
5914
+ return applyZeroProviderFallback(
5915
+ classified,
5916
+ hasProvider,
5917
+ reply?.info?.providerID ?? null,
5918
+ reply?.info?.modelID ?? null
5919
+ );
5920
+ }
4732
5921
  /**
4733
5922
  * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
4734
5923
  * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
@@ -4881,10 +6070,16 @@ var ChannelDriver = class _ChannelDriver {
4881
6070
  import chalk5 from "chalk";
4882
6071
  import ora2 from "ora";
4883
6072
  import { select as select2 } from "@inquirer/prompts";
6073
+ var INTERACTIVE_START_TIMEOUT_MS = 3e4;
4884
6074
  async function ensureOpenCodeRunning(ctx) {
4885
6075
  const healthCheck = await checkOpenCodeHealth(ctx.port);
4886
6076
  if (healthCheck.healthy) {
4887
- return { port: ctx.port, process: null, version: healthCheck.version ?? null };
6077
+ return {
6078
+ port: ctx.port,
6079
+ process: null,
6080
+ version: healthCheck.version ?? null,
6081
+ notReadyReason: null
6082
+ };
4888
6083
  }
4889
6084
  const runningInstances = await findHealthyOpenCodeInstances();
4890
6085
  if (runningInstances.length > 0) {
@@ -4905,7 +6100,7 @@ async function ensureOpenCodeRunning(ctx) {
4905
6100
  console.log(chalk5.yellow("Tip: Run with the correct port:"));
4906
6101
  console.log(
4907
6102
  chalk5.dim(
4908
- ` ${getCliName()} run --agent ${ctx.agentId} --port ${runningInstances[0].port}`
6103
+ ` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
4909
6104
  )
4910
6105
  );
4911
6106
  }
@@ -4925,14 +6120,22 @@ async function ensureOpenCodeRunning(ctx) {
4925
6120
  if (!ctx.interactive) {
4926
6121
  ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
4927
6122
  const proc = await startOpenCode(ctx.port);
4928
- const health = await waitForOpenCodeHealth(ctx.port, 3e4);
6123
+ const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
4929
6124
  if (!health.healthy) {
4930
- throw new Error(
4931
- `OpenCode failed to start on port ${ctx.port}. Install with: npm install -g opencode-ai`
4932
- );
6125
+ return {
6126
+ port: ctx.port,
6127
+ process: proc,
6128
+ version: null,
6129
+ notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
6130
+ };
4933
6131
  }
4934
6132
  ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
4935
- return { port: ctx.port, process: proc, version: health.version ?? null };
6133
+ return {
6134
+ port: ctx.port,
6135
+ process: proc,
6136
+ version: health.version ?? null,
6137
+ notReadyReason: null
6138
+ };
4936
6139
  }
4937
6140
  let port = ctx.port;
4938
6141
  if (isPortInUse(port)) {
@@ -4985,122 +6188,15 @@ Port ${port} is already in use.`));
4985
6188
  if (action === "start") {
4986
6189
  const spinner = ora2("Starting OpenCode...").start();
4987
6190
  const proc = await startOpenCode(port);
4988
- const health = await waitForOpenCodeHealth(port, 3e4);
6191
+ const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
4989
6192
  if (!health.healthy) {
4990
6193
  spinner.fail("Failed to start OpenCode");
4991
6194
  throw new Error("OpenCode failed to start");
4992
6195
  }
4993
6196
  spinner.stop();
4994
- return { port, process: proc, version: health.version ?? null };
4995
- }
4996
- return { port, process: null, version: null };
4997
- }
4998
-
4999
- // src/commands/agent-lookup.ts
5000
- async function readErrorMessage(response) {
5001
- const text = await response.text().catch(() => "");
5002
- if (!text) return response.statusText || void 0;
5003
- try {
5004
- const data = JSON.parse(text);
5005
- const message = data.message ?? data.error;
5006
- if (typeof message === "string" && message.trim()) {
5007
- return message;
5008
- }
5009
- } catch {
5010
- }
5011
- return text.trim() || response.statusText || void 0;
5012
- }
5013
- function authFailureHint(apiUrl, serverMessage) {
5014
- const reason = serverMessage ? `: ${serverMessage}` : "";
5015
- return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
5016
- }
5017
- async function resolveAgentIdFromKey(authHeader) {
5018
- const apiUrl = getApiUrlConfig();
5019
- try {
5020
- const response = await fetch(`${apiUrl}/me`, {
5021
- headers: { Authorization: authHeader }
5022
- });
5023
- if (response.status === 401) {
5024
- const serverMessage = await readErrorMessage(response);
5025
- return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
5026
- }
5027
- if (!response.ok) {
5028
- const serverMessage = await readErrorMessage(response);
5029
- return {
5030
- error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
5031
- };
5032
- }
5033
- const data = await response.json();
5034
- if (data.auth_type === "agent_key" && data.agent_id) {
5035
- return { agent_id: data.agent_id };
5036
- }
5037
- return {
5038
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
5039
- };
5040
- } catch (error2) {
5041
- const message = error2 instanceof Error ? error2.message : "Unknown error";
5042
- return { error: `Failed to resolve runner from key: ${message}` };
5043
- }
5044
- }
5045
- async function notifyAgentDisconnected(agentId, authHeader) {
5046
- const apiUrl = getApiUrlConfig();
5047
- try {
5048
- const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
5049
- method: "POST",
5050
- headers: { Authorization: authHeader }
5051
- });
5052
- if (!response.ok) {
5053
- const serverMessage = await readErrorMessage(response);
5054
- return {
5055
- ok: false,
5056
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
5057
- };
5058
- }
5059
- return { ok: true };
5060
- } catch (error2) {
5061
- return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
5062
- }
5063
- }
5064
- async function getAgentInfo(agentId, authHeader) {
5065
- const apiUrl = getApiUrlConfig();
5066
- try {
5067
- const response = await fetch(`${apiUrl}/runners/${agentId}`, {
5068
- headers: { Authorization: authHeader }
5069
- });
5070
- if (response.status === 401) {
5071
- const serverMessage = await readErrorMessage(response);
5072
- return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
5073
- }
5074
- if (response.status === 403) {
5075
- const serverMessage = await readErrorMessage(response);
5076
- return {
5077
- valid: false,
5078
- error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
5079
- };
5080
- }
5081
- if (response.status === 404) {
5082
- const serverMessage = await readErrorMessage(response);
5083
- return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
5084
- }
5085
- if (!response.ok) {
5086
- const serverMessage = await readErrorMessage(response);
5087
- return {
5088
- valid: false,
5089
- error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
5090
- };
5091
- }
5092
- const agent = await response.json();
5093
- if (agent.agent_type !== "local") {
5094
- return {
5095
- valid: false,
5096
- error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
5097
- };
5098
- }
5099
- return { valid: true, agent };
5100
- } catch (error2) {
5101
- const message = error2 instanceof Error ? error2.message : "Unknown error";
5102
- return { valid: false, error: `Failed to validate runner: ${message}` };
6197
+ return { port, process: proc, version: health.version ?? null, notReadyReason: null };
5103
6198
  }
6199
+ return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
5104
6200
  }
5105
6201
 
5106
6202
  // src/commands/run.ts
@@ -5108,6 +6204,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
5108
6204
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
5109
6205
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
5110
6206
  var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
6207
+ var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
5111
6208
  function resolveLogLevel(options) {
5112
6209
  const accepted = Object.keys(LOG_LEVELS);
5113
6210
  const validate = (value, source) => {
@@ -5138,7 +6235,7 @@ function resolveFileSyncDirectories(raw, homeDir) {
5138
6235
  if (trimmed === "") {
5139
6236
  throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
5140
6237
  }
5141
- const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join2(homeDir, trimmed.slice(2)) : trimmed;
6238
+ const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
5142
6239
  if (!isAbsolute2(expanded)) {
5143
6240
  throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
5144
6241
  }
@@ -5159,6 +6256,35 @@ function resolveFileSyncDirectories(raw, homeDir) {
5159
6256
  }
5160
6257
  return directories;
5161
6258
  }
6259
+ var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
6260
+ var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
6261
+ var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
6262
+ function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
6263
+ const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
6264
+ let raw;
6265
+ let source;
6266
+ if (options.opencodeStartTimeout !== void 0) {
6267
+ raw = options.opencodeStartTimeout;
6268
+ source = "--opencode-start-timeout";
6269
+ } else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
6270
+ raw = env[OPENCODE_START_TIMEOUT_ENV];
6271
+ source = OPENCODE_START_TIMEOUT_ENV;
6272
+ } else {
6273
+ return { timeoutMs: defaultMs, warnings: [] };
6274
+ }
6275
+ const trimmed = raw.trim();
6276
+ const seconds = Number(trimmed);
6277
+ const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
6278
+ if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
6279
+ return {
6280
+ timeoutMs: defaultMs,
6281
+ warnings: [
6282
+ `Ignoring invalid ${source} "${raw}": expected a positive integer number of seconds (at most ${MAX_OPENCODE_START_TIMEOUT_SECONDS}); using the default ${DEFAULT_OPENCODE_START_TIMEOUT_SECONDS}s`
6283
+ ]
6284
+ };
6285
+ }
6286
+ return { timeoutMs: seconds * 1e3, warnings: [] };
6287
+ }
5162
6288
  function meetsThreshold(state, level) {
5163
6289
  return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
5164
6290
  }
@@ -5180,6 +6306,10 @@ function log2(state, message, level = "info") {
5180
6306
  function logActivity(state, entry) {
5181
6307
  const level = entry.level ?? (entry.type === "error" ? "error" : "info");
5182
6308
  if (!meetsThreshold(state, level)) return;
6309
+ forwardRunnerActivity(
6310
+ { level, message: entry.message, error: entry.error },
6311
+ { agentId: state.agentId, authHeader: state.authHeader }
6312
+ );
5183
6313
  const fullEntry = {
5184
6314
  ...entry,
5185
6315
  level,
@@ -5279,9 +6409,15 @@ async function handleAuthError(state, error2) {
5279
6409
  }
5280
6410
  async function driveChannels(state, driver) {
5281
6411
  let idlePolls = 0;
6412
+ let idleMs = 0;
6413
+ let consecutiveDrainFailures = 0;
6414
+ let unreachableMs = 0;
5282
6415
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
5283
6416
  let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
5284
6417
  while (state.running) {
6418
+ const cycleStartedAtMs = performance.now();
6419
+ let idleThisCycle = false;
6420
+ let unreachableThisCycle = false;
5285
6421
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
5286
6422
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
5287
6423
  if (state.interactive) displayStatus(state);
@@ -5296,17 +6432,23 @@ async function driveChannels(state, driver) {
5296
6432
  );
5297
6433
  try {
5298
6434
  const processed = await driver.drainPending();
6435
+ consecutiveDrainFailures = 0;
6436
+ unreachableMs = 0;
5299
6437
  state.messageCount += processed;
5300
6438
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
5301
6439
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
5302
6440
  const appliedFiles = driver.fileSyncActivity().appliedFiles;
5303
- const fileActivity = carriedOverFileSync || appliedFiles !== lastSeenAppliedFiles;
6441
+ const filesApplied = appliedFiles !== lastSeenAppliedFiles;
6442
+ const fileActivity = carriedOverFileSync || filesApplied;
5304
6443
  lastSeenAppliedFiles = appliedFiles;
6444
+ if (filesApplied) state.claudeUsageRearm?.();
5305
6445
  if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
5306
6446
  idlePolls = 0;
6447
+ idleMs = 0;
5307
6448
  if (processed > 0 && state.interactive) displayStatus(state);
5308
6449
  } else if (state.idleTimeout !== null) {
5309
6450
  idlePolls++;
6451
+ idleThisCycle = true;
5310
6452
  if (idlePolls === 1) {
5311
6453
  logActivity(state, {
5312
6454
  type: "info",
@@ -5330,21 +6472,44 @@ async function driveChannels(state, driver) {
5330
6472
  const errorMessage = error2 instanceof Error ? error2.message : String(error2);
5331
6473
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
5332
6474
  if (state.interactive) displayStatus(state);
6475
+ if (driver.hasInFlightWatchers()) {
6476
+ consecutiveDrainFailures = 0;
6477
+ unreachableMs = 0;
6478
+ } else if (state.idleTimeout !== null) {
6479
+ consecutiveDrainFailures++;
6480
+ unreachableThisCycle = true;
6481
+ if (consecutiveDrainFailures === 1) {
6482
+ logActivity(state, {
6483
+ type: "info",
6484
+ message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
6485
+ });
6486
+ if (state.interactive) displayStatus(state);
6487
+ }
6488
+ }
5333
6489
  }
5334
6490
  await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
5335
- if (state.idleTimeout !== null && idlePolls >= 2) {
5336
- const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
5337
- if (idleMs > state.idleTimeout * 1e3) {
5338
- logActivity(state, { type: "info", message: "Idle timeout reached" });
5339
- if (state.interactive) displayStatus(state);
5340
- break;
5341
- }
6491
+ const cycleMs = performance.now() - cycleStartedAtMs;
6492
+ if (idleThisCycle) idleMs += cycleMs;
6493
+ if (unreachableThisCycle) unreachableMs += cycleMs;
6494
+ if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
6495
+ logActivity(state, {
6496
+ type: "info",
6497
+ level: "warn",
6498
+ message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
6499
+ });
6500
+ if (state.interactive) displayStatus(state);
6501
+ break;
6502
+ }
6503
+ if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
6504
+ logActivity(state, { type: "info", message: "Idle timeout reached" });
6505
+ if (state.interactive) displayStatus(state);
6506
+ break;
5342
6507
  }
5343
6508
  }
5344
6509
  }
5345
6510
  var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
5346
- async function runSweep(state, driver, config2) {
5347
- const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
6511
+ async function runSweep(state, driver, config) {
6512
+ const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
5348
6513
  try {
5349
6514
  const sessions = await listSessions(state.port);
5350
6515
  if (sessions === null) {
@@ -5357,8 +6522,8 @@ async function runSweep(state, driver, config2) {
5357
6522
  const toDelete = selectSessionsToDelete(
5358
6523
  sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
5359
6524
  {
5360
- maxAgeMs: config2.maxAgeMs,
5361
- maxCount: config2.maxCount,
6525
+ maxAgeMs: config.maxAgeMs,
6526
+ maxCount: config.maxCount,
5362
6527
  nowMs: Date.now(),
5363
6528
  protectedIds: driver.protectedSessionIds()
5364
6529
  }
@@ -5394,7 +6559,7 @@ async function runSweep(state, driver, config2) {
5394
6559
  }
5395
6560
  }
5396
6561
  function scheduleSessionCleanup(state, driver, options) {
5397
- const config2 = resolveSessionCleanupConfig(
6562
+ const config = resolveSessionCleanupConfig(
5398
6563
  {
5399
6564
  maxAge: options.sessionCleanupMaxAge,
5400
6565
  maxCount: options.sessionCleanupMaxCount,
@@ -5402,21 +6567,129 @@ function scheduleSessionCleanup(state, driver, options) {
5402
6567
  },
5403
6568
  process.env
5404
6569
  );
5405
- for (const warning2 of config2.warnings) {
6570
+ for (const warning2 of config.warnings) {
5406
6571
  logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
5407
6572
  }
5408
- if (!config2.enabled) return;
6573
+ if (!config.enabled) return;
5409
6574
  logActivity(state, {
5410
6575
  type: "info",
5411
- message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
6576
+ message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
5412
6577
  });
5413
- const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
6578
+ const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
5414
6579
  const firstSweep = setTimeout(
5415
- () => void runSweep(state, driver, config2),
6580
+ () => void runSweep(state, driver, config),
5416
6581
  SESSION_CLEANUP_FIRST_SWEEP_MS
5417
6582
  );
5418
6583
  state.sessionCleanupTimers.push(interval, firstSweep);
5419
6584
  }
6585
+ function claudeUsageFailureStreakSuffix(consecutiveFailures) {
6586
+ return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
6587
+ }
6588
+ function scheduleClaudeUsageReporting(state, options) {
6589
+ const { mode, warnings } = resolveClaudeUsageReportingMode(
6590
+ options.claudeUsageReporting,
6591
+ process.env
6592
+ );
6593
+ for (const warning2 of warnings) {
6594
+ logActivity(state, {
6595
+ type: "info",
6596
+ level: "warn",
6597
+ message: `Claude usage reporting: ${warning2}`
6598
+ });
6599
+ }
6600
+ if (mode === "off") {
6601
+ logActivity(state, {
6602
+ type: "info",
6603
+ level: "debug",
6604
+ message: "Claude usage reporting is off (--claude-usage-reporting off)"
6605
+ });
6606
+ return null;
6607
+ }
6608
+ let consecutiveFailures = 0;
6609
+ let armed = false;
6610
+ let rearmRequested = false;
6611
+ const scheduleNextTick = () => {
6612
+ armed = true;
6613
+ rearmRequested = false;
6614
+ state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
6615
+ };
6616
+ const rearm = () => {
6617
+ if (armed) {
6618
+ rearmRequested = true;
6619
+ return;
6620
+ }
6621
+ rearmRequested = false;
6622
+ armed = true;
6623
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6624
+ };
6625
+ const tick = async (isProbe) => {
6626
+ try {
6627
+ const usage = await getClaudeUsage();
6628
+ const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
6629
+ if (result.ok) {
6630
+ if (consecutiveFailures > 0) {
6631
+ logActivity(state, {
6632
+ type: "info",
6633
+ level: "info",
6634
+ message: "Claude usage reporting recovered"
6635
+ });
6636
+ }
6637
+ consecutiveFailures = 0;
6638
+ logActivity(state, {
6639
+ type: "info",
6640
+ level: "debug",
6641
+ message: "Reported Claude usage to Evident"
6642
+ });
6643
+ } else {
6644
+ consecutiveFailures++;
6645
+ logActivity(state, {
6646
+ type: "info",
6647
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
6648
+ message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
6649
+ });
6650
+ }
6651
+ scheduleNextTick();
6652
+ } catch (error2) {
6653
+ if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
6654
+ if (mode === "on") {
6655
+ logActivity(state, {
6656
+ type: "info",
6657
+ level: "warn",
6658
+ message: "Claude usage reporting is forced on but no usable Claude Code login was found \u2014 run `claude` to sign in; reporting will keep retrying"
6659
+ });
6660
+ scheduleNextTick();
6661
+ } else if (isProbe) {
6662
+ logActivity(state, {
6663
+ type: "info",
6664
+ level: "debug",
6665
+ message: `Claude usage reporting: ${error2.message}`
6666
+ });
6667
+ armed = false;
6668
+ if (rearmRequested) rearm();
6669
+ } else {
6670
+ logActivity(state, {
6671
+ type: "info",
6672
+ level: "debug",
6673
+ message: `Claude usage reporting: ${error2.message}`
6674
+ });
6675
+ scheduleNextTick();
6676
+ }
6677
+ } else {
6678
+ consecutiveFailures++;
6679
+ const message = error2 instanceof Error ? error2.message : String(error2);
6680
+ logActivity(state, {
6681
+ type: "info",
6682
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
6683
+ message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
6684
+ });
6685
+ scheduleNextTick();
6686
+ }
6687
+ }
6688
+ };
6689
+ armed = true;
6690
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6691
+ return rearm;
6692
+ }
5420
6693
  async function notifyOffline(state) {
5421
6694
  if (!state.agentId || !state.authHeader) return;
5422
6695
  if (!state.connected) {
@@ -5434,13 +6707,29 @@ async function notifyOffline(state) {
5434
6707
  if (state.interactive) displayStatus(state);
5435
6708
  }
5436
6709
  }
6710
+ async function timeShutdownPhase(state, durations, name, run2) {
6711
+ const startedAt = Date.now();
6712
+ try {
6713
+ return await run2();
6714
+ } finally {
6715
+ const elapsedMs = Date.now() - startedAt;
6716
+ durations[name] = elapsedMs;
6717
+ log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
6718
+ }
6719
+ }
5437
6720
  async function cleanup(state, opts = {}) {
6721
+ const durations = {};
5438
6722
  state.running = false;
5439
6723
  for (const timer of state.sessionCleanupTimers) {
5440
6724
  clearInterval(timer);
5441
6725
  clearTimeout(timer);
5442
6726
  }
5443
6727
  state.sessionCleanupTimers = [];
6728
+ if (state.claudeUsageTimer) {
6729
+ clearTimeout(state.claudeUsageTimer);
6730
+ state.claudeUsageTimer = null;
6731
+ }
6732
+ state.claudeUsageRearm = null;
5444
6733
  if (opts.graceful && state.channelDriver) {
5445
6734
  state.channelDriver.stop();
5446
6735
  log2(state, "Draining in-flight channel work before shutdown...");
@@ -5448,7 +6737,13 @@ async function cleanup(state, opts = {}) {
5448
6737
  logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
5449
6738
  displayStatus(state);
5450
6739
  }
5451
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
6740
+ const driver = state.channelDriver;
6741
+ const settled = await timeShutdownPhase(
6742
+ state,
6743
+ durations,
6744
+ "drain",
6745
+ () => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
6746
+ );
5452
6747
  if (!settled) {
5453
6748
  logActivity(state, {
5454
6749
  type: "info",
@@ -5457,13 +6752,15 @@ async function cleanup(state, opts = {}) {
5457
6752
  if (state.interactive) displayStatus(state);
5458
6753
  }
5459
6754
  }
5460
- await notifyOffline(state);
6755
+ await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
5461
6756
  if (state.connection) {
5462
- state.connection.close();
6757
+ const connection = state.connection;
6758
+ await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
5463
6759
  state.connection = null;
5464
6760
  }
5465
6761
  if (state.opencodeProcess) {
5466
- stopOpenCode(state.opencodeProcess);
6762
+ const opencodeProcess = state.opencodeProcess;
6763
+ await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
5467
6764
  if (state.interactive) {
5468
6765
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
5469
6766
  displayStatus(state);
@@ -5472,6 +6769,7 @@ async function cleanup(state, opts = {}) {
5472
6769
  }
5473
6770
  state.opencodeProcess = null;
5474
6771
  }
6772
+ return durations;
5475
6773
  }
5476
6774
  async function run(options) {
5477
6775
  const interactive = isInteractive(options.json);
@@ -5479,7 +6777,7 @@ async function run(options) {
5479
6777
  let fileSyncDirectories;
5480
6778
  try {
5481
6779
  logLevel = resolveLogLevel(options);
5482
- fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir2());
6780
+ fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
5483
6781
  } catch (error2) {
5484
6782
  const message = error2 instanceof Error ? error2.message : String(error2);
5485
6783
  if (options.json) {
@@ -5512,8 +6810,11 @@ async function run(options) {
5512
6810
  messageCount: 0,
5513
6811
  lastProxiedActivityAt: null,
5514
6812
  sessionCleanupTimers: [],
6813
+ claudeUsageTimer: null,
6814
+ claudeUsageRearm: null,
5515
6815
  authHeader: ""
5516
6816
  };
6817
+ setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
5517
6818
  if (fileSyncDirectories.length > 0) {
5518
6819
  log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
5519
6820
  } else {
@@ -5542,14 +6843,38 @@ async function run(options) {
5542
6843
  const handleSignal = async () => {
5543
6844
  if (state.shuttingDown) return;
5544
6845
  state.shuttingDown = true;
6846
+ const shutdownStartedAt = Date.now();
5545
6847
  if (state.interactive) {
5546
6848
  logActivity(state, { type: "info", message: "Shutting down..." });
5547
6849
  displayStatus(state);
5548
6850
  } else {
5549
6851
  log2(state, "Shutting down...");
5550
6852
  }
5551
- await cleanup(state, { graceful: true });
5552
- await shutdownTelemetry();
6853
+ const durations = await cleanup(state, { graceful: true });
6854
+ const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
6855
+ await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
6856
+ let timer;
6857
+ const flushed = shutdownTelemetry().then(
6858
+ () => true,
6859
+ (error2) => {
6860
+ log2(
6861
+ state,
6862
+ `Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
6863
+ "warn"
6864
+ );
6865
+ return true;
6866
+ }
6867
+ );
6868
+ const timedOut = new Promise((resolve3) => {
6869
+ timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
6870
+ });
6871
+ if (!await Promise.race([flushed, timedOut])) {
6872
+ log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
6873
+ }
6874
+ clearTimeout(timer);
6875
+ });
6876
+ const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
6877
+ log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
5553
6878
  process.exit(0);
5554
6879
  };
5555
6880
  process.on("SIGINT", handleSignal);
@@ -5566,6 +6891,7 @@ async function run(options) {
5566
6891
  console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
5567
6892
  blank();
5568
6893
  process.exit(1);
6894
+ return;
5569
6895
  }
5570
6896
  blank();
5571
6897
  console.log(chalk6.yellow("You are not logged in to Evident."));
@@ -5610,6 +6936,7 @@ async function run(options) {
5610
6936
  } else {
5611
6937
  printError(resolved.error || "Failed to resolve runner ID from key");
5612
6938
  process.exit(1);
6939
+ return;
5613
6940
  }
5614
6941
  } else {
5615
6942
  printError(
@@ -5623,6 +6950,7 @@ async function run(options) {
5623
6950
  );
5624
6951
  blank();
5625
6952
  process.exit(1);
6953
+ return;
5626
6954
  }
5627
6955
  }
5628
6956
  telemetry.info(
@@ -5663,40 +6991,67 @@ async function run(options) {
5663
6991
  }
5664
6992
  spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
5665
6993
  state.agentName = validation.agent.name;
6994
+ const microvmId = process.env.MICROVM_ID?.trim();
6995
+ if (microvmId) {
6996
+ const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
6997
+ if (reported.ok) {
6998
+ log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
6999
+ } else {
7000
+ const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
7001
+ log2(state, message, "warn");
7002
+ if (state.interactive && !state.json) {
7003
+ logActivity(state, { type: "info", level: "warn", message });
7004
+ }
7005
+ }
7006
+ } else {
7007
+ log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
7008
+ }
7009
+ const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
7010
+ for (const warning2 of opencodeStartTimeoutWarnings) {
7011
+ logActivity(state, { type: "info", level: "warn", message: warning2 });
7012
+ }
5666
7013
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
5667
7014
  try {
5668
7015
  const oc = await ensureOpenCodeRunning({
5669
7016
  port: state.port,
5670
7017
  interactive: state.interactive,
5671
7018
  agentId: state.agentId,
5672
- log: (message) => log2(state, message)
7019
+ log: (message) => log2(state, message),
7020
+ startTimeoutMs: opencodeStartTimeoutMs
5673
7021
  });
5674
7022
  state.port = oc.port;
5675
7023
  state.opencodeProcess = oc.process;
5676
7024
  state.opencodeVersion = oc.version;
5677
- state.opencodeConnected = oc.process !== null || oc.version !== null;
7025
+ state.opencodeConnected = oc.notReadyReason === null;
5678
7026
  const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
5679
7027
  ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
5680
- const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
5681
- if (versionWarning) {
5682
- log2(state, versionWarning, "warn");
5683
- if (state.interactive && !state.json) {
5684
- logActivity(state, { type: "info", level: "warn", message: versionWarning });
7028
+ if (!state.interactive && oc.notReadyReason !== null) {
7029
+ const message = `OpenCode is not ready on port ${state.port}: ${oc.notReadyReason}. The runner will still come online, but messages will fail until opencode answers \u2014 raise the wait with --opencode-start-timeout <seconds> (env ${OPENCODE_START_TIMEOUT_ENV}).`;
7030
+ logActivity(state, { type: "info", level: "warn", message });
7031
+ } else {
7032
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
7033
+ if (versionWarning) {
7034
+ log2(state, versionWarning, "warn");
7035
+ if (state.interactive && !state.json) {
7036
+ logActivity(state, { type: "info", level: "warn", message: versionWarning });
7037
+ }
5685
7038
  }
5686
- }
5687
- const noProviderWarning = buildNoProviderWarning(await hasAnyConfiguredProvider(state.port));
5688
- if (noProviderWarning) {
5689
- log2(state, noProviderWarning, "warn");
5690
- if (state.interactive && !state.json) {
5691
- logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
5692
- blank();
5693
- console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
5694
- console.log(
5695
- chalk6.dim(
5696
- `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
5697
- )
5698
- );
5699
- blank();
7039
+ const noProviderWarning = buildNoProviderWarning(
7040
+ await hasAnyConfiguredProvider(state.port)
7041
+ );
7042
+ if (noProviderWarning) {
7043
+ log2(state, noProviderWarning, "warn");
7044
+ if (state.interactive && !state.json) {
7045
+ logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
7046
+ blank();
7047
+ console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
7048
+ console.log(
7049
+ chalk6.dim(
7050
+ `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
7051
+ )
7052
+ );
7053
+ blank();
7054
+ }
5700
7055
  }
5701
7056
  }
5702
7057
  } catch (error2) {
@@ -5714,7 +7069,7 @@ async function run(options) {
5714
7069
  // #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
5715
7070
  // REJECTED with `file_sync_disabled` on the ack, not silently ignored.
5716
7071
  fileSyncDirectories,
5717
- homeDir: homedir2(),
7072
+ homeDir: homedir3(),
5718
7073
  log: (entry) => (
5719
7074
  // Thread the driver's real level straight through so `debug`/`warn`
5720
7075
  // survive the sink filter (they no longer collapse to info). `type`
@@ -5741,6 +7096,18 @@ async function run(options) {
5741
7096
  type: "info",
5742
7097
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
5743
7098
  });
7099
+ if (options.tunnelReadyFile) {
7100
+ const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
7101
+ if (marker.ok) {
7102
+ log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
7103
+ } else {
7104
+ log2(
7105
+ state,
7106
+ `Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
7107
+ "error"
7108
+ );
7109
+ }
7110
+ }
5744
7111
  emitAgentConnected(state.agentId, {
5745
7112
  port: state.port,
5746
7113
  cli_version: getCliVersion(),
@@ -5831,6 +7198,7 @@ async function run(options) {
5831
7198
  throw error2;
5832
7199
  }
5833
7200
  scheduleSessionCleanup(state, channelDriver, options);
7201
+ state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
5834
7202
  if (!interactive || state.json) {
5835
7203
  log2(state, "Driving channel messages...");
5836
7204
  }
@@ -5885,13 +7253,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
5885
7253
  program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
5886
7254
  program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
5887
7255
  program.command("whoami").description("Show the currently logged in user").action(whoami);
7256
+ program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
7257
+ program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
5888
7258
  program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
5889
7259
  "-a, --agent [id]",
5890
7260
  "Deprecated alias for --runner (still supported; --runner wins if both are given)"
5891
7261
  ).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
5892
7262
  "--log-level <level>",
5893
7263
  "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
5894
- ).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").option(
7264
+ ).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
7265
+ "--opencode-start-timeout <seconds>",
7266
+ "Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
7267
+ ).option("--json", "Output in JSON format").option(
5895
7268
  "--session-cleanup-max-age <duration>",
5896
7269
  "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
5897
7270
  ).option(
@@ -5900,11 +7273,17 @@ program.command("run").description("Connect to Evident and process messages").op
5900
7273
  ).option(
5901
7274
  "--session-cleanup-interval <duration>",
5902
7275
  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
7276
+ ).option(
7277
+ "--claude-usage-reporting <mode>",
7278
+ "Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
5903
7279
  ).option(
5904
7280
  "--enable-file-sync-to <dir>",
5905
7281
  "Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
5906
7282
  (value, previous) => previous.concat([value]),
5907
7283
  []
7284
+ ).option(
7285
+ "--tunnel-ready-file <path>",
7286
+ "Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
5908
7287
  ).action(
5909
7288
  (options) => {
5910
7289
  run({
@@ -5917,14 +7296,21 @@ program.command("run").description("Connect to Evident and process messages").op
5917
7296
  verbose: options.verbose,
5918
7297
  conversation: options.conversation,
5919
7298
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
7299
+ // Raw string — validation/precedence is single-sourced in run.ts's
7300
+ // resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
7301
+ opencodeStartTimeout: options.opencodeStartTimeout,
5920
7302
  json: options.json,
5921
7303
  // Raw strings — the resolver in run.ts single-sources parsing (M1).
5922
7304
  sessionCleanupMaxAge: options.sessionCleanupMaxAge,
5923
7305
  sessionCleanupMaxCount: options.sessionCleanupMaxCount,
5924
7306
  sessionCleanupInterval: options.sessionCleanupInterval,
7307
+ // Raw string — the resolver in run.ts single-sources parsing
7308
+ // (resolveClaudeUsageReportingMode).
7309
+ claudeUsageReporting: options.claudeUsageReporting,
5925
7310
  // Raw values — expansion/validation is single-sourced in run.ts's
5926
7311
  // resolveFileSyncDirectories.
5927
- enableFileSyncTo: options.enableFileSyncTo
7312
+ enableFileSyncTo: options.enableFileSyncTo,
7313
+ tunnelReadyFile: options.tunnelReadyFile
5928
7314
  });
5929
7315
  }
5930
7316
  );