@evident-ai/cli 3.1.1-dev.315e29d → 3.1.1-dev.3583438

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
@@ -532,6 +532,343 @@ async function whoami() {
532
532
  blank();
533
533
  }
534
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
+
535
872
  // src/lib/claude-usage.ts
536
873
  import { execFileSync } from "child_process";
537
874
  import { readFileSync } from "fs";
@@ -582,6 +919,10 @@ var ClaudeUsageError = class extends Error {
582
919
  function isLocalCredentialProblem(err) {
583
920
  return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
584
921
  }
922
+ function normalizeResetsAt(value) {
923
+ const ms = Date.parse(value);
924
+ return Number.isNaN(ms) ? null : new Date(ms).toISOString();
925
+ }
585
926
  function toWindow(value) {
586
927
  if (!value || typeof value !== "object") {
587
928
  return null;
@@ -590,7 +931,11 @@ function toWindow(value) {
590
931
  if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
591
932
  return null;
592
933
  }
593
- return { utilization: window.utilization, resetsAt: window.resets_at };
934
+ const resetsAt = normalizeResetsAt(window.resets_at);
935
+ if (resetsAt === null) {
936
+ return null;
937
+ }
938
+ return { utilization: window.utilization, resetsAt };
594
939
  }
595
940
  async function getClaudeUsage() {
596
941
  const credentials2 = readClaudeCliCredentials();
@@ -652,6 +997,9 @@ import { homedir as homedir3 } from "os";
652
997
  import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
653
998
  import chalk6 from "chalk";
654
999
 
1000
+ // ../../packages/types/src/agents/index.ts
1001
+ var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
1002
+
655
1003
  // ../../packages/types/src/telemetry/index.ts
656
1004
  var TelemetryEventTypes = {
657
1005
  // Agent activity events (shown in web UI activity log)
@@ -934,49 +1282,6 @@ function forwardRunnerActivity(entry, context) {
934
1282
  }
935
1283
  }
936
1284
 
937
- // src/lib/auth.ts
938
- async function getAuthCredentials() {
939
- const runnerKey = process.env.EVIDENT_RUNNER_KEY;
940
- const agentKey = process.env.EVIDENT_AGENT_KEY;
941
- if (runnerKey) {
942
- return {
943
- token: runnerKey,
944
- authType: "agent_key",
945
- keySource: "runner_key",
946
- notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
947
- };
948
- }
949
- if (agentKey) {
950
- return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
951
- }
952
- const userToken = process.env.EVIDENT_TOKEN;
953
- if (userToken) {
954
- return { token: userToken, authType: "bearer" };
955
- }
956
- const keychainCreds = await getToken();
957
- if (keychainCreds) {
958
- return {
959
- token: keychainCreds.token,
960
- authType: "bearer",
961
- user: keychainCreds.user
962
- };
963
- }
964
- return null;
965
- }
966
- function getAuthHeader(credentials2) {
967
- if (credentials2.authType === "agent_key") {
968
- return `SandboxKey ${credentials2.token}`;
969
- }
970
- return `Bearer ${credentials2.token}`;
971
- }
972
- function isInteractive(jsonOutput) {
973
- if (jsonOutput) return false;
974
- if (process.env.CI) return false;
975
- if (process.env.GITHUB_ACTIONS) return false;
976
- if (!process.stdin.isTTY) return false;
977
- return true;
978
- }
979
-
980
1285
  // src/lib/opencode/health.ts
981
1286
  async function checkOpenCodeHealth(port) {
982
1287
  try {
@@ -1784,6 +2089,21 @@ function messageError(messages, userMessageId) {
1784
2089
  }
1785
2090
  return "The agent run failed.";
1786
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
+ }
1787
2107
  function messageFailure(messages, userMessageId) {
1788
2108
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1789
2109
  const error2 = errorOf(reply);
@@ -2386,6 +2706,10 @@ function nextReportDelayMs(random = Math.random) {
2386
2706
  return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
2387
2707
  }
2388
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
+ }
2389
2713
 
2390
2714
  // src/lib/channels/driver.ts
2391
2715
  import { homedir as homedir2 } from "os";
@@ -2708,8 +3032,8 @@ async function applyOne(options, file) {
2708
3032
  await ack(options, file, "applied");
2709
3033
  return true;
2710
3034
  }
2711
- function durableDownloadCode(status) {
2712
- return status === 413 ? "file_too_large" : "write_failed";
3035
+ function durableDownloadCode(status2) {
3036
+ return status2 === 413 ? "file_too_large" : "write_failed";
2713
3037
  }
2714
3038
  async function downloadContent(options, file, label) {
2715
3039
  try {
@@ -2742,8 +3066,8 @@ async function downloadContent(options, file, label) {
2742
3066
  return { ok: false, terminal: false };
2743
3067
  }
2744
3068
  }
2745
- async function ack(options, file, status, reason) {
2746
- const outcome = `${status}${reason ? ` (${reason})` : ""}`;
3069
+ async function ack(options, file, status2, reason) {
3070
+ const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
2747
3071
  try {
2748
3072
  const res = await options.fetchImpl(
2749
3073
  `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
@@ -2753,7 +3077,7 @@ async function ack(options, file, status, reason) {
2753
3077
  Authorization: options.getAuthHeader(),
2754
3078
  "Content-Type": "application/json"
2755
3079
  },
2756
- body: JSON.stringify(reason ? { status, reason } : { status })
3080
+ body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
2757
3081
  }
2758
3082
  );
2759
3083
  if (!res.ok) {
@@ -2817,6 +3141,7 @@ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
2817
3141
  var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2818
3142
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
2819
3143
  var MAX_SUPERSEDED_CONVERSATIONS = 256;
3144
+ var MAX_IDENTICAL_REDRIVE_POLL_FAILURES = 5;
2820
3145
  var ChannelAuthError = class extends Error {
2821
3146
  constructor(message) {
2822
3147
  super(message);
@@ -2825,10 +3150,10 @@ var ChannelAuthError = class extends Error {
2825
3150
  };
2826
3151
  var ChannelTerminalError = class extends Error {
2827
3152
  status;
2828
- constructor(message, status) {
3153
+ constructor(message, status2) {
2829
3154
  super(message);
2830
3155
  this.name = "ChannelTerminalError";
2831
- this.status = status;
3156
+ this.status = status2;
2832
3157
  }
2833
3158
  };
2834
3159
  function backoffDelay(attempt, policy) {
@@ -2836,8 +3161,12 @@ function backoffDelay(attempt, policy) {
2836
3161
  const capped = Math.min(policy.maxDelayMs, exp);
2837
3162
  return Math.floor(Math.random() * capped);
2838
3163
  }
2839
- function isRetryableStatus(status) {
2840
- return status === 429 || status >= 500 && status <= 599;
3164
+ function isRetryableStatus(status2) {
3165
+ return status2 === 429 || status2 >= 500 && status2 <= 599;
3166
+ }
3167
+ var VOLATILE_BODY_FIELD_PATTERN = /("(?:ref|requestId|request_id|traceId|trace_id)"\s*:\s*)"[^"]*"/gi;
3168
+ function normalizeRedrivePollFailureBody(body) {
3169
+ return body.replace(VOLATILE_BODY_FIELD_PATTERN, '$1"<redacted>"').replace(/\s+/g, " ").trim().slice(0, 200);
2841
3170
  }
2842
3171
  var ChannelDriver = class _ChannelDriver {
2843
3172
  agentId;
@@ -2952,6 +3281,33 @@ var ChannelDriver = class _ChannelDriver {
2952
3281
  * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
2953
3282
  */
2954
3283
  readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
3284
+ /**
3285
+ * "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
3286
+ * `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
3287
+ * every ~2s drain until opencode's status becomes readable, but the
3288
+ * server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
3289
+ * on any non-`unresolved` outcome so the set cannot grow beyond the currently
3290
+ * unresolvable rows.
3291
+ */
3292
+ redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
3293
+ /**
3294
+ * First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
3295
+ * `pending` row is invisible to every cron arm (all require `status =
3296
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
3297
+ * nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
3298
+ * takes `dispatch` instead of `unresolved` (reusing the existing knob — see
3299
+ * ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
3300
+ */
3301
+ redriveUnresolvedSince = /* @__PURE__ */ new Map();
3302
+ /**
3303
+ * Consecutive-identical-poll-failure streak for the re-drive fence (#1348),
3304
+ * keyed by Evident **message id** (not session) so `clearRedriveUnresolved`
3305
+ * can drop it with the other two trackers and it cannot leak. `sessionId` is
3306
+ * carried inside the entry, not the key: a session change is a different
3307
+ * situation and resets the streak, which gives the `(sessionId, message.id)`
3308
+ * pairing #1348 asks for without a composite map key.
3309
+ */
3310
+ redrivePollFailures = /* @__PURE__ */ new Map();
2955
3311
  /**
2956
3312
  * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
2957
3313
  * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
@@ -3291,6 +3647,12 @@ var ChannelDriver = class _ChannelDriver {
3291
3647
  skippedAlreadyDispatched += 1;
3292
3648
  continue;
3293
3649
  }
3650
+ if (message.opencode_message_id) {
3651
+ const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
3652
+ if (outcome !== "dispatch") {
3653
+ break;
3654
+ }
3655
+ }
3294
3656
  const options = {
3295
3657
  agent: message.opencode_agent ?? void 0,
3296
3658
  model: message.opencode_model ?? void 0
@@ -3349,36 +3711,347 @@ var ChannelDriver = class _ChannelDriver {
3349
3711
  });
3350
3712
  });
3351
3713
  this.log({
3352
- level: "error",
3353
- message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
3714
+ level: "error",
3715
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
3716
+ conversation_id: conv.id,
3717
+ message_id: message.id
3718
+ });
3719
+ break;
3720
+ }
3721
+ if (opencodeMessageId === null) {
3722
+ this.log({
3723
+ level: "warn",
3724
+ 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`,
3725
+ conversation_id: conv.id,
3726
+ message_id: message.id
3727
+ });
3728
+ continue;
3729
+ }
3730
+ this.dispatched.add(message.id);
3731
+ this.registerInFlight(conv, sessionId, message, opencodeMessageId);
3732
+ dispatched += 1;
3733
+ void this.postSignal(conv.id, message.id, "dispatched");
3734
+ }
3735
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3736
+ this.log({
3737
+ level: "warn",
3738
+ message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
3739
+ conversation_id: conv.id
3740
+ });
3741
+ }
3742
+ this.ensureWatcherRunning(sessionId);
3743
+ return dispatched;
3744
+ }
3745
+ /**
3746
+ * Poll a session's message list for the re-drive fence (#965), via the
3747
+ * INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
3748
+ * hits the global `fetch` and would bypass the same override every other
3749
+ * opencode poll in this file respects. Mirrors `readoptProcessing`'s own
3750
+ * snapshot fetch (`:3081-3111`).
3751
+ *
3752
+ * Returns `{ ok: true, messages }` on a readable snapshot, or
3753
+ * `{ ok: false, signature }` on failure — `signature` is a string that
3754
+ * repeats across attempts for the SAME underlying fault (used by the
3755
+ * consecutive-identical-failure bound, #1348), or `null` for a thrown
3756
+ * exception, which is NOT countable toward that bound (a network blip / an
3757
+ * opencode restart also throws identically every tick, and must keep
3758
+ * retrying unbounded rather than ever being treated as permanent).
3759
+ */
3760
+ async pollSessionMessagesForRedrive(conv, message, sessionId) {
3761
+ try {
3762
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
3763
+ if (!res.ok) {
3764
+ const rawBody = await res.text();
3765
+ const normalized = normalizeRedrivePollFailureBody(rawBody);
3766
+ this.log({
3767
+ level: "warn",
3768
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status}${normalized ? `: ${normalized}` : ""} \u2014 treating as unreadable this tick`,
3769
+ conversation_id: conv.id,
3770
+ message_id: message.id
3771
+ });
3772
+ return { ok: false, signature: `HTTP ${res.status}${normalized ? `: ${normalized}` : ""}` };
3773
+ }
3774
+ const body = await res.json();
3775
+ if (!Array.isArray(body)) {
3776
+ this.log({
3777
+ level: "warn",
3778
+ 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`,
3779
+ conversation_id: conv.id,
3780
+ message_id: message.id
3781
+ });
3782
+ return { ok: false, signature: "non-array message body" };
3783
+ }
3784
+ return { ok: true, messages: body };
3785
+ } catch (err) {
3786
+ this.log({
3787
+ level: "warn",
3788
+ 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)}`,
3789
+ conversation_id: conv.id,
3790
+ message_id: message.id
3791
+ });
3792
+ return { ok: false, signature: null };
3793
+ }
3794
+ }
3795
+ /**
3796
+ * The re-drive fence for a `pending` row that already carries a stored
3797
+ * `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
3798
+ * least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
3799
+ * The lifecycle cron can falsely reclaim a `processing` row back to `pending`
3800
+ * mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
3801
+ * without this fence the drain loop would re-`prompt_async` the SAME turn a
3802
+ * second time against live GitHub state. Mirrors `readoptOne`'s job for the
3803
+ * `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
3804
+ * needed here because `refusedSessionId` already handles the one case
3805
+ * (#553 abandoned session) that path exists for.
3806
+ *
3807
+ * Only `ChannelAuthError` propagates. A poll that fails identically
3808
+ * `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row reports the message
3809
+ * failed instead of retrying it (#1348) — SEPARATE from, not a replacement
3810
+ * for, `resolveRedriveUnresolved`'s own `pausedMaxWaitMs` bound below. Every
3811
+ * other failure resolves to `unresolved` and is retried whole on the next
3812
+ * ~2s drain tick.
3813
+ */
3814
+ async resolveRedrive(conv, sessionId, message, refusedSessionId) {
3815
+ const ocId = message.opencode_message_id ?? null;
3816
+ if (refusedSessionId) {
3817
+ this.clearRedriveUnresolved(message.id);
3818
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3819
+ return "dispatch";
3820
+ }
3821
+ const polled = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
3822
+ if (!polled.ok) {
3823
+ const streak = this.recordRedrivePollFailure(message.id, sessionId, polled.signature);
3824
+ if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES && polled.signature !== null) {
3825
+ return this.failRedrivePollPermanent(conv, sessionId, message, polled.signature, streak);
3826
+ }
3827
+ return this.resolveRedriveUnresolved(conv, message);
3828
+ }
3829
+ this.redrivePollFailures.delete(message.id);
3830
+ const messages = polled.messages;
3831
+ if (messages.length === 0) {
3832
+ return this.resolveRedriveUnresolved(conv, message);
3833
+ }
3834
+ const state = messageRunState(messages, ocId ?? "");
3835
+ if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
3836
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3837
+ if (ongoing === false) {
3838
+ this.log({
3839
+ level: "info",
3840
+ 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`,
3841
+ conversation_id: conv.id,
3842
+ message_id: message.id
3843
+ });
3844
+ this.clearRedriveUnresolved(message.id);
3845
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3846
+ return "dispatch";
3847
+ }
3848
+ }
3849
+ if (state === "done" || state === "failed") {
3850
+ return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
3851
+ }
3852
+ if (state === "running" || state === "queued") {
3853
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3854
+ if (ongoing === true) {
3855
+ return this.reattachRedrive(conv, sessionId, message, ocId);
3856
+ }
3857
+ if (ongoing === false) {
3858
+ this.clearRedriveUnresolved(message.id);
3859
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3860
+ return "dispatch";
3861
+ }
3862
+ return this.resolveRedriveUnresolved(conv, message);
3863
+ }
3864
+ this.clearRedriveUnresolved(message.id);
3865
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3866
+ return "dispatch";
3867
+ }
3868
+ /**
3869
+ * The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
3870
+ * opencode's own status map — undo the false reclaim instead of starting a
3871
+ * second turn.
3872
+ */
3873
+ async reattachRedrive(conv, sessionId, message, ocId) {
3874
+ let anchorMs;
3875
+ const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
3876
+ if (!Number.isNaN(parsed)) {
3877
+ anchorMs = parsed;
3878
+ } else {
3879
+ anchorMs = this.now();
3880
+ this.log({
3881
+ level: "error",
3882
+ 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)`,
3883
+ conversation_id: conv.id,
3884
+ message_id: message.id
3885
+ });
3886
+ }
3887
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3888
+ try {
3889
+ await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
3890
+ } catch (err) {
3891
+ if (err instanceof ChannelAuthError) throw err;
3892
+ this.log({
3893
+ level: "warn",
3894
+ 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)}`,
3895
+ conversation_id: conv.id,
3896
+ message_id: message.id
3897
+ });
3898
+ return "unresolved";
3899
+ }
3900
+ this.clearRedriveUnresolved(message.id);
3901
+ this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
3902
+ this.dispatched.add(message.id);
3903
+ this.readopted.add(message.id);
3904
+ this.ensureWatcherRunning(sessionId);
3905
+ const watchedForMs = this.now() - anchorMs;
3906
+ void this.postSignal(conv.id, message.id, "redrive_reattached", {
3907
+ watched_for_ms: watchedForMs
3908
+ });
3909
+ this.log({
3910
+ level: "warn",
3911
+ 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`,
3912
+ conversation_id: conv.id,
3913
+ message_id: message.id
3914
+ });
3915
+ return "reattached";
3916
+ }
3917
+ /**
3918
+ * The `settled` outcome (Task 3.2): the prior turn already finished (or
3919
+ * errored) while nobody was watching — deliver/report it instead of re-running.
3920
+ * Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
3921
+ * (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
3922
+ * drain, same as any other non-auth failure). The restart-abort carve-out that
3923
+ * keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
3924
+ * so a row reaching this `failed` branch is a GENUINE failure.
3925
+ */
3926
+ async settleRedrive(conv, sessionId, message, ocId, messages, state) {
3927
+ try {
3928
+ if (state === "done") {
3929
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3930
+ const usage = messageUsage(messages, ocId ?? "");
3931
+ this.log({
3932
+ level: "info",
3933
+ 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`,
3354
3934
  conversation_id: conv.id,
3355
3935
  message_id: message.id
3356
3936
  });
3357
- break;
3358
- }
3359
- if (opencodeMessageId === null) {
3937
+ await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
3938
+ } else {
3939
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
3940
+ const usage = messageUsage(messages, ocId ?? "");
3941
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3360
3942
  this.log({
3361
- level: "warn",
3362
- 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`,
3943
+ level: "error",
3944
+ 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)"}`,
3363
3945
  conversation_id: conv.id,
3364
3946
  message_id: message.id
3365
3947
  });
3366
- continue;
3948
+ await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
3367
3949
  }
3368
- this.dispatched.add(message.id);
3369
- this.registerInFlight(conv, sessionId, message, opencodeMessageId);
3370
- dispatched += 1;
3371
- void this.postSignal(conv.id, message.id, "dispatched");
3950
+ } catch (err) {
3951
+ if (err instanceof ChannelAuthError) throw err;
3952
+ this.log({
3953
+ level: "warn",
3954
+ message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3955
+ conversation_id: conv.id,
3956
+ message_id: message.id
3957
+ });
3958
+ return "unresolved";
3372
3959
  }
3373
- if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3960
+ this.clearRedriveUnresolved(message.id);
3961
+ void this.postSignal(conv.id, message.id, "redrive_settled");
3962
+ return "settled";
3963
+ }
3964
+ /**
3965
+ * The permanent-failure outcome (#1348): the fence's own poll of this session
3966
+ * failed with the SAME opencode-answered signature
3967
+ * `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row — a transient blip
3968
+ * would have varied or eventually cleared (see `pollSessionMessagesForRedrive`
3969
+ * and `recordRedrivePollFailure`), so this is a durable fault (e.g. #1345's
3970
+ * corrupted opencode session) rather than something worth retrying forever.
3971
+ * Mirrors `settleRedrive`'s error discipline: no `usage`/`failure` args to
3972
+ * `markFailed` (no opencode snapshot to extract them from — this poll never
3973
+ * got a readable one).
3974
+ */
3975
+ async failRedrivePollPermanent(conv, sessionId, message, signature, streak) {
3976
+ this.log({
3977
+ level: "error",
3978
+ message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) failed to poll with the identical signature "${signature}" ${streak} times in a row \u2014 reporting the message failed instead of retrying forever`,
3979
+ conversation_id: conv.id,
3980
+ message_id: message.id
3981
+ });
3982
+ try {
3983
+ await this.markFailed(
3984
+ conv.id,
3985
+ message.id,
3986
+ sessionId,
3987
+ `The runner could not read this conversation's state from OpenCode (${signature}). The same failure repeated ${streak} times in a row, so the message was not retried further.`
3988
+ );
3989
+ } catch (err) {
3990
+ if (err instanceof ChannelAuthError) throw err;
3374
3991
  this.log({
3375
3992
  level: "warn",
3376
- 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).`,
3377
- conversation_id: conv.id
3993
+ message: `Re-drive: failed to report message ${message.id.slice(0, 8)} permanently failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3994
+ conversation_id: conv.id,
3995
+ message_id: message.id
3378
3996
  });
3997
+ return "unresolved";
3379
3998
  }
3380
- this.ensureWatcherRunning(sessionId);
3381
- return dispatched;
3999
+ this.clearRedriveUnresolved(message.id);
4000
+ void this.postSignal(conv.id, message.id, "redrive_poll_failed");
4001
+ return "settled";
4002
+ }
4003
+ /**
4004
+ * The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
4005
+ * observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
4006
+ * A `pending` row is invisible to every cron arm (all require `status =
4007
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
4008
+ * nothing driving it — bound it to the existing `pausedMaxWaitMs` window
4009
+ * (reusing the knob, not a new constant) and take `dispatch` once elapsed.
4010
+ */
4011
+ resolveRedriveUnresolved(conv, message) {
4012
+ const now = this.now();
4013
+ const since = this.redriveUnresolvedSince.get(message.id);
4014
+ if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
4015
+ this.clearRedriveUnresolved(message.id);
4016
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
4017
+ return "dispatch";
4018
+ }
4019
+ if (since === void 0) {
4020
+ this.redriveUnresolvedSince.set(message.id, now);
4021
+ }
4022
+ if (!this.redriveUnresolvedSignalled.has(message.id)) {
4023
+ this.redriveUnresolvedSignalled.add(message.id);
4024
+ void this.postSignal(conv.id, message.id, "redrive_unresolved");
4025
+ }
4026
+ return "unresolved";
4027
+ }
4028
+ /** Clear all three `unresolved`/failure-streak trackers for a row (any non-`unresolved` outcome). */
4029
+ clearRedriveUnresolved(messageId) {
4030
+ this.redriveUnresolvedSince.delete(messageId);
4031
+ this.redriveUnresolvedSignalled.delete(messageId);
4032
+ this.redrivePollFailures.delete(messageId);
4033
+ }
4034
+ /**
4035
+ * Record one poll outcome toward the re-drive fence's consecutive-identical-
4036
+ * failure streak (#1348) and return the resulting count. `signature === null`
4037
+ * (a thrown exception, H1) always clears the streak and returns `0` — it is
4038
+ * never countable. Otherwise the streak continues only when BOTH the session
4039
+ * and the signature match the previous failure; anything else (a different
4040
+ * session, or the same session failing a DIFFERENT way) starts a fresh streak
4041
+ * at `1`.
4042
+ */
4043
+ recordRedrivePollFailure(messageId, sessionId, signature) {
4044
+ if (signature === null) {
4045
+ this.redrivePollFailures.delete(messageId);
4046
+ return 0;
4047
+ }
4048
+ const existing = this.redrivePollFailures.get(messageId);
4049
+ if (existing && existing.sessionId === sessionId && existing.signature === signature) {
4050
+ existing.count += 1;
4051
+ return existing.count;
4052
+ }
4053
+ this.redrivePollFailures.set(messageId, { sessionId, signature, count: 1 });
4054
+ return 1;
3382
4055
  }
3383
4056
  /**
3384
4057
  * Record that `sessionId` is no longer a valid binding for `conversationId`
@@ -3663,9 +4336,10 @@ var ChannelDriver = class _ChannelDriver {
3663
4336
  * opencode reports ACTIVELY `running` is watched to completion (its liveness
3664
4337
  * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
3665
4338
  * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
3666
- * handed to the cron. The old "the `deadline` must settle before the ~15-min
3667
- * cron or they double-drive" reasoning is superseded: liveness now settles the
3668
- * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
4339
+ * handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
4340
+ * runner still holds; a reclaimed row that already ran is never re-dispatched
4341
+ * while opencode reports its turn ongoing (readopt's own gate here, and the
4342
+ * `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
3669
4343
  * (only the appear-guard uses it).
3670
4344
  *
3671
4345
  * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
@@ -4225,7 +4899,10 @@ var ChannelDriver = class _ChannelDriver {
4225
4899
  * re-dispatched (at most once, see `forceReadoptRun`):
4226
4900
  * - `done` → `markDone` now (guarded like the watcher's done branch);
4227
4901
  * - `failed` → `markFailed` with the surfaced error (issue #182), so an
4228
- * errored turn is reported failed on restart, NOT re-dispatched;
4902
+ * errored turn is reported failed on restart, NOT re-dispatched
4903
+ * EXCEPT a restart-ABORTED turn under a not-ongoing session,
4904
+ * which is a restart orphan wearing a terminal error and is
4905
+ * re-dispatched instead (issue #1310, see the branch below);
4229
4906
  * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
4230
4907
  * tracking the stored id so the reply correlates by it;
4231
4908
  * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
@@ -4289,7 +4966,16 @@ var ChannelDriver = class _ChannelDriver {
4289
4966
  void this.postSignal(row.conversation_id, row.id, "readopt_done");
4290
4967
  return;
4291
4968
  }
4292
- if (state === "failed") {
4969
+ const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
4970
+ if (restartAborted) {
4971
+ this.log({
4972
+ level: "info",
4973
+ 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`,
4974
+ conversation_id: row.conversation_id,
4975
+ message_id: row.id
4976
+ });
4977
+ }
4978
+ if (state === "failed" && !restartAborted) {
4293
4979
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
4294
4980
  const usage = messageUsage(messages, ocId ?? "");
4295
4981
  const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
@@ -4567,7 +5253,8 @@ var ChannelDriver = class _ChannelDriver {
4567
5253
  opencode_model: row.opencode_model,
4568
5254
  source_message_id: row.source_message_id,
4569
5255
  slack_user_id: row.slack_user_id,
4570
- attachments: row.attachments ?? null
5256
+ attachments: row.attachments ?? null,
5257
+ opencode_message_id: row.opencode_message_id
4571
5258
  };
4572
5259
  }
4573
5260
  /**
@@ -5159,11 +5846,11 @@ var ChannelDriver = class _ChannelDriver {
5159
5846
  * `ensureSession` guard, not this suppression, is what makes the self-heal
5160
5847
  * stick.
5161
5848
  */
5162
- sessionIdBody(sessionId, conversationId, messageId, status) {
5849
+ sessionIdBody(sessionId, conversationId, messageId, status2) {
5163
5850
  if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
5164
5851
  this.log({
5165
5852
  level: "debug",
5166
- 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)}`,
5853
+ 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)}`,
5167
5854
  conversation_id: conversationId,
5168
5855
  message_id: messageId
5169
5856
  });
@@ -5609,171 +6296,6 @@ Port ${port} is already in use.`));
5609
6296
  return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
5610
6297
  }
5611
6298
 
5612
- // src/commands/agent-lookup.ts
5613
- async function readErrorMessage(response) {
5614
- const text = await response.text().catch(() => "");
5615
- if (!text) return response.statusText || void 0;
5616
- try {
5617
- const data = JSON.parse(text);
5618
- const message = data.message ?? data.error;
5619
- if (typeof message === "string" && message.trim()) {
5620
- return message;
5621
- }
5622
- } catch {
5623
- }
5624
- return text.trim() || response.statusText || void 0;
5625
- }
5626
- function authFailureHint(apiUrl, serverMessage) {
5627
- const reason = serverMessage ? `: ${serverMessage}` : "";
5628
- return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
5629
- }
5630
- async function resolveAgentIdFromKey(authHeader) {
5631
- const apiUrl = getApiUrlConfig();
5632
- try {
5633
- const response = await fetch(`${apiUrl}/me`, {
5634
- headers: { Authorization: authHeader }
5635
- });
5636
- if (response.status === 401) {
5637
- const serverMessage = await readErrorMessage(response);
5638
- return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
5639
- }
5640
- if (!response.ok) {
5641
- const serverMessage = await readErrorMessage(response);
5642
- return {
5643
- error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
5644
- };
5645
- }
5646
- const data = await response.json();
5647
- if (data.auth_type === "agent_key" && data.agent_id) {
5648
- return { agent_id: data.agent_id };
5649
- }
5650
- return {
5651
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
5652
- };
5653
- } catch (error2) {
5654
- const message = error2 instanceof Error ? error2.message : "Unknown error";
5655
- return { error: `Failed to resolve runner from key: ${message}` };
5656
- }
5657
- }
5658
- var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
5659
- async function notifyAgentDisconnected(agentId, authHeader) {
5660
- const apiUrl = getApiUrlConfig();
5661
- try {
5662
- const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
5663
- method: "POST",
5664
- headers: { Authorization: authHeader },
5665
- signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
5666
- });
5667
- if (!response.ok) {
5668
- const serverMessage = await readErrorMessage(response);
5669
- return {
5670
- ok: false,
5671
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
5672
- };
5673
- }
5674
- return { ok: true };
5675
- } catch (error2) {
5676
- return { ok: false, error: describeBestEffortError(error2) };
5677
- }
5678
- }
5679
- function describeBestEffortError(error2) {
5680
- const name = error2?.name;
5681
- if (name === "TimeoutError" || name === "AbortError") {
5682
- return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
5683
- }
5684
- return error2 instanceof Error ? error2.message : String(error2);
5685
- }
5686
- async function reportMicrovmId(agentId, authHeader, microvmId) {
5687
- try {
5688
- const apiUrl = getApiUrlConfig();
5689
- const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
5690
- method: "POST",
5691
- headers: { Authorization: authHeader, "Content-Type": "application/json" },
5692
- body: JSON.stringify({ microvm_id: microvmId }),
5693
- signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
5694
- });
5695
- if (!response.ok) {
5696
- const serverMessage = await readErrorMessage(response);
5697
- return {
5698
- ok: false,
5699
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
5700
- };
5701
- }
5702
- return { ok: true };
5703
- } catch (error2) {
5704
- return { ok: false, error: describeBestEffortError(error2) };
5705
- }
5706
- }
5707
- function toReportedWindow(window) {
5708
- if (!window) return null;
5709
- return { utilization: window.utilization, resets_at: window.resetsAt };
5710
- }
5711
- async function reportClaudeUsage(agentId, authHeader, snapshot) {
5712
- try {
5713
- const apiUrl = getApiUrlConfig();
5714
- const response = await fetch(`${apiUrl}/runners/${agentId}/claude-usage`, {
5715
- method: "POST",
5716
- headers: { Authorization: authHeader, "Content-Type": "application/json" },
5717
- body: JSON.stringify({
5718
- five_hour: toReportedWindow(snapshot.fiveHour),
5719
- seven_day: toReportedWindow(snapshot.sevenDay)
5720
- }),
5721
- signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
5722
- });
5723
- if (!response.ok) {
5724
- const serverMessage = await readErrorMessage(response);
5725
- return {
5726
- ok: false,
5727
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
5728
- };
5729
- }
5730
- return { ok: true };
5731
- } catch (error2) {
5732
- return { ok: false, error: describeBestEffortError(error2) };
5733
- }
5734
- }
5735
- async function getAgentInfo(agentId, authHeader) {
5736
- const apiUrl = getApiUrlConfig();
5737
- try {
5738
- const response = await fetch(`${apiUrl}/runners/${agentId}`, {
5739
- headers: { Authorization: authHeader }
5740
- });
5741
- if (response.status === 401) {
5742
- const serverMessage = await readErrorMessage(response);
5743
- return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
5744
- }
5745
- if (response.status === 403) {
5746
- const serverMessage = await readErrorMessage(response);
5747
- return {
5748
- valid: false,
5749
- error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
5750
- };
5751
- }
5752
- if (response.status === 404) {
5753
- const serverMessage = await readErrorMessage(response);
5754
- return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
5755
- }
5756
- if (!response.ok) {
5757
- const serverMessage = await readErrorMessage(response);
5758
- return {
5759
- valid: false,
5760
- error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
5761
- };
5762
- }
5763
- const agent = await response.json();
5764
- if (agent.agent_type !== "local") {
5765
- return {
5766
- valid: false,
5767
- error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
5768
- };
5769
- }
5770
- return { valid: true, agent };
5771
- } catch (error2) {
5772
- const message = error2 instanceof Error ? error2.message : "Unknown error";
5773
- return { valid: false, error: `Failed to validate runner: ${message}` };
5774
- }
5775
- }
5776
-
5777
6299
  // src/commands/run.ts
5778
6300
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
5779
6301
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
@@ -5984,9 +6506,15 @@ async function handleAuthError(state, error2) {
5984
6506
  }
5985
6507
  async function driveChannels(state, driver) {
5986
6508
  let idlePolls = 0;
6509
+ let idleMs = 0;
6510
+ let consecutiveDrainFailures = 0;
6511
+ let unreachableMs = 0;
5987
6512
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
5988
6513
  let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
5989
6514
  while (state.running) {
6515
+ const cycleStartedAtMs = performance.now();
6516
+ let idleThisCycle = false;
6517
+ let unreachableThisCycle = false;
5990
6518
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
5991
6519
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
5992
6520
  if (state.interactive) displayStatus(state);
@@ -6001,17 +6529,23 @@ async function driveChannels(state, driver) {
6001
6529
  );
6002
6530
  try {
6003
6531
  const processed = await driver.drainPending();
6532
+ consecutiveDrainFailures = 0;
6533
+ unreachableMs = 0;
6004
6534
  state.messageCount += processed;
6005
6535
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
6006
6536
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
6007
6537
  const appliedFiles = driver.fileSyncActivity().appliedFiles;
6008
- const fileActivity = carriedOverFileSync || appliedFiles !== lastSeenAppliedFiles;
6538
+ const filesApplied = appliedFiles !== lastSeenAppliedFiles;
6539
+ const fileActivity = carriedOverFileSync || filesApplied;
6009
6540
  lastSeenAppliedFiles = appliedFiles;
6541
+ if (filesApplied) state.claudeUsageRearm?.();
6010
6542
  if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
6011
6543
  idlePolls = 0;
6544
+ idleMs = 0;
6012
6545
  if (processed > 0 && state.interactive) displayStatus(state);
6013
6546
  } else if (state.idleTimeout !== null) {
6014
6547
  idlePolls++;
6548
+ idleThisCycle = true;
6015
6549
  if (idlePolls === 1) {
6016
6550
  logActivity(state, {
6017
6551
  type: "info",
@@ -6035,15 +6569,38 @@ async function driveChannels(state, driver) {
6035
6569
  const errorMessage = error2 instanceof Error ? error2.message : String(error2);
6036
6570
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
6037
6571
  if (state.interactive) displayStatus(state);
6572
+ if (driver.hasInFlightWatchers()) {
6573
+ consecutiveDrainFailures = 0;
6574
+ unreachableMs = 0;
6575
+ } else if (state.idleTimeout !== null) {
6576
+ consecutiveDrainFailures++;
6577
+ unreachableThisCycle = true;
6578
+ if (consecutiveDrainFailures === 1) {
6579
+ logActivity(state, {
6580
+ type: "info",
6581
+ message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
6582
+ });
6583
+ if (state.interactive) displayStatus(state);
6584
+ }
6585
+ }
6038
6586
  }
6039
6587
  await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
6040
- if (state.idleTimeout !== null && idlePolls >= 2) {
6041
- const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
6042
- if (idleMs > state.idleTimeout * 1e3) {
6043
- logActivity(state, { type: "info", message: "Idle timeout reached" });
6044
- if (state.interactive) displayStatus(state);
6045
- break;
6046
- }
6588
+ const cycleMs = performance.now() - cycleStartedAtMs;
6589
+ if (idleThisCycle) idleMs += cycleMs;
6590
+ if (unreachableThisCycle) unreachableMs += cycleMs;
6591
+ if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
6592
+ logActivity(state, {
6593
+ type: "info",
6594
+ level: "warn",
6595
+ message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
6596
+ });
6597
+ if (state.interactive) displayStatus(state);
6598
+ break;
6599
+ }
6600
+ if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
6601
+ logActivity(state, { type: "info", message: "Idle timeout reached" });
6602
+ if (state.interactive) displayStatus(state);
6603
+ break;
6047
6604
  }
6048
6605
  }
6049
6606
  }
@@ -6122,6 +6679,9 @@ function scheduleSessionCleanup(state, driver, options) {
6122
6679
  );
6123
6680
  state.sessionCleanupTimers.push(interval, firstSweep);
6124
6681
  }
6682
+ function claudeUsageFailureStreakSuffix(consecutiveFailures) {
6683
+ return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
6684
+ }
6125
6685
  function scheduleClaudeUsageReporting(state, options) {
6126
6686
  const { mode, warnings } = resolveClaudeUsageReportingMode(
6127
6687
  options.claudeUsageReporting,
@@ -6140,13 +6700,26 @@ function scheduleClaudeUsageReporting(state, options) {
6140
6700
  level: "debug",
6141
6701
  message: "Claude usage reporting is off (--claude-usage-reporting off)"
6142
6702
  });
6143
- return;
6703
+ return null;
6144
6704
  }
6145
6705
  let consecutiveFailures = 0;
6706
+ let armed = false;
6707
+ let rearmRequested = false;
6146
6708
  const scheduleNextTick = () => {
6709
+ armed = true;
6710
+ rearmRequested = false;
6147
6711
  state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
6148
6712
  };
6149
- const tick = async (isFirst) => {
6713
+ const rearm = () => {
6714
+ if (armed) {
6715
+ rearmRequested = true;
6716
+ return;
6717
+ }
6718
+ rearmRequested = false;
6719
+ armed = true;
6720
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6721
+ };
6722
+ const tick = async (isProbe) => {
6150
6723
  try {
6151
6724
  const usage = await getClaudeUsage();
6152
6725
  const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
@@ -6168,8 +6741,8 @@ function scheduleClaudeUsageReporting(state, options) {
6168
6741
  consecutiveFailures++;
6169
6742
  logActivity(state, {
6170
6743
  type: "info",
6171
- level: consecutiveFailures === 1 ? "warn" : "debug",
6172
- message: `Failed to report Claude usage: ${result.error}`
6744
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
6745
+ message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
6173
6746
  });
6174
6747
  }
6175
6748
  scheduleNextTick();
@@ -6182,12 +6755,14 @@ function scheduleClaudeUsageReporting(state, options) {
6182
6755
  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"
6183
6756
  });
6184
6757
  scheduleNextTick();
6185
- } else if (isFirst) {
6758
+ } else if (isProbe) {
6186
6759
  logActivity(state, {
6187
6760
  type: "info",
6188
6761
  level: "debug",
6189
6762
  message: `Claude usage reporting: ${error2.message}`
6190
6763
  });
6764
+ armed = false;
6765
+ if (rearmRequested) rearm();
6191
6766
  } else {
6192
6767
  logActivity(state, {
6193
6768
  type: "info",
@@ -6201,14 +6776,16 @@ function scheduleClaudeUsageReporting(state, options) {
6201
6776
  const message = error2 instanceof Error ? error2.message : String(error2);
6202
6777
  logActivity(state, {
6203
6778
  type: "info",
6204
- level: consecutiveFailures === 1 ? "warn" : "debug",
6205
- message: `Claude usage reporting failed: ${message}`
6779
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
6780
+ message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
6206
6781
  });
6207
6782
  scheduleNextTick();
6208
6783
  }
6209
6784
  }
6210
6785
  };
6786
+ armed = true;
6211
6787
  state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6788
+ return rearm;
6212
6789
  }
6213
6790
  async function notifyOffline(state) {
6214
6791
  if (!state.agentId || !state.authHeader) return;
@@ -6249,6 +6826,7 @@ async function cleanup(state, opts = {}) {
6249
6826
  clearTimeout(state.claudeUsageTimer);
6250
6827
  state.claudeUsageTimer = null;
6251
6828
  }
6829
+ state.claudeUsageRearm = null;
6252
6830
  if (opts.graceful && state.channelDriver) {
6253
6831
  state.channelDriver.stop();
6254
6832
  log2(state, "Draining in-flight channel work before shutdown...");
@@ -6330,6 +6908,7 @@ async function run(options) {
6330
6908
  lastProxiedActivityAt: null,
6331
6909
  sessionCleanupTimers: [],
6332
6910
  claudeUsageTimer: null,
6911
+ claudeUsageRearm: null,
6333
6912
  authHeader: ""
6334
6913
  };
6335
6914
  setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
@@ -6409,6 +6988,7 @@ async function run(options) {
6409
6988
  console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
6410
6989
  blank();
6411
6990
  process.exit(1);
6991
+ return;
6412
6992
  }
6413
6993
  blank();
6414
6994
  console.log(chalk6.yellow("You are not logged in to Evident."));
@@ -6453,6 +7033,7 @@ async function run(options) {
6453
7033
  } else {
6454
7034
  printError(resolved.error || "Failed to resolve runner ID from key");
6455
7035
  process.exit(1);
7036
+ return;
6456
7037
  }
6457
7038
  } else {
6458
7039
  printError(
@@ -6466,6 +7047,7 @@ async function run(options) {
6466
7047
  );
6467
7048
  blank();
6468
7049
  process.exit(1);
7050
+ return;
6469
7051
  }
6470
7052
  }
6471
7053
  telemetry.info(
@@ -6713,7 +7295,7 @@ async function run(options) {
6713
7295
  throw error2;
6714
7296
  }
6715
7297
  scheduleSessionCleanup(state, channelDriver, options);
6716
- scheduleClaudeUsageReporting(state, options);
7298
+ state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
6717
7299
  if (!interactive || state.json) {
6718
7300
  log2(state, "Driving channel messages...");
6719
7301
  }
@@ -6768,6 +7350,7 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
6768
7350
  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);
6769
7351
  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 }));
6770
7352
  program.command("whoami").description("Show the currently logged in user").action(whoami);
7353
+ program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
6771
7354
  program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
6772
7355
  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(
6773
7356
  "-a, --agent [id]",
@@ -6797,7 +7380,7 @@ program.command("run").description("Connect to Evident and process messages").op
6797
7380
  []
6798
7381
  ).option(
6799
7382
  "--tunnel-ready-file <path>",
6800
- "Path to write once the tunnel is connected (set by the MicroVM hooks; unused on a developer machine)"
7383
+ "Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
6801
7384
  ).action(
6802
7385
  (options) => {
6803
7386
  run({