@evident-ai/cli 3.1.1-dev.2f78b33 → 3.1.1-dev.315e29d

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
@@ -11,8 +11,8 @@ import chalk2 from "chalk";
11
11
 
12
12
  // src/lib/config.ts
13
13
  import Conf from "conf";
14
- import { homedir } from "os";
15
- import { join } from "path";
14
+ import { chmodSync, existsSync, statSync } from "fs";
15
+ import { dirname } from "path";
16
16
  var PRODUCTION_API_URL = "https://api.production.evident.run/v1";
17
17
  var PRODUCTION_TUNNEL_URL = "wss://tunnel.production.evident.run";
18
18
  var defaults = {
@@ -38,17 +38,39 @@ 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: "",
49
44
  configName: "credentials",
50
- defaults: {}
45
+ defaults: {},
46
+ configFileMode: 384
51
47
  });
48
+ var CREDENTIALS_FILE_MODE = 384;
49
+ var CREDENTIALS_DIR_MODE = 448;
50
+ var permissionWarningEmitted = false;
51
+ function hardenCredentialsPermissions() {
52
+ if (process.platform === "win32") {
53
+ return;
54
+ }
55
+ const file = credentials.path;
56
+ for (const [path, mode] of [
57
+ [file, CREDENTIALS_FILE_MODE],
58
+ [dirname(file), CREDENTIALS_DIR_MODE]
59
+ ]) {
60
+ try {
61
+ if (existsSync(path) && (statSync(path).mode & 511) !== mode) {
62
+ chmodSync(path, mode);
63
+ }
64
+ } catch (err) {
65
+ if (!permissionWarningEmitted) {
66
+ permissionWarningEmitted = true;
67
+ console.error(
68
+ `[config] could not restrict permissions on ${path}; the credentials file may be readable by other users on this machine: ${err instanceof Error ? err.message : String(err)}`
69
+ );
70
+ }
71
+ }
72
+ }
73
+ }
52
74
  function getApiUrlConfig() {
53
75
  return getApiUrl();
54
76
  }
@@ -59,6 +81,7 @@ function credentialsKey() {
59
81
  return getApiUrl();
60
82
  }
61
83
  function getCredentials() {
84
+ hardenCredentialsPermissions();
62
85
  const byEndpoint = credentials.get("byEndpoint") ?? {};
63
86
  return byEndpoint[credentialsKey()] ?? {};
64
87
  }
@@ -70,14 +93,17 @@ function setCredentials(creds) {
70
93
  expiresAt: creds.expiresAt
71
94
  };
72
95
  credentials.set("byEndpoint", byEndpoint);
96
+ hardenCredentialsPermissions();
73
97
  }
74
98
  function clearCredentials() {
75
99
  const byEndpoint = credentials.get("byEndpoint") ?? {};
76
100
  delete byEndpoint[credentialsKey()];
77
101
  credentials.set("byEndpoint", byEndpoint);
102
+ hardenCredentialsPermissions();
78
103
  }
79
104
  function clearAllCredentials() {
80
105
  credentials.clear();
106
+ hardenCredentialsPermissions();
81
107
  }
82
108
  function getCliName() {
83
109
  const argv1 = process.argv[1] || "";
@@ -236,16 +262,28 @@ async function getToken() {
236
262
  }
237
263
  return null;
238
264
  }
265
+ function toError(err) {
266
+ return err instanceof Error ? err : new Error(String(err));
267
+ }
239
268
  async function deleteToken(options = {}) {
240
269
  const keytar = await getKeytar();
270
+ const failures = [];
241
271
  if (keytar) {
242
272
  if (options.all) {
243
- 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
+ }
244
279
  await Promise.all(
245
- all.map(
246
- (entry) => keytar.deletePassword(SERVICE_NAME, entry.account).catch(() => {
247
- })
248
- )
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
+ })
249
287
  );
250
288
  } else {
251
289
  await keytar.deletePassword(SERVICE_NAME, keychainAccount());
@@ -256,6 +294,7 @@ async function deleteToken(options = {}) {
256
294
  } else {
257
295
  clearCredentials();
258
296
  }
297
+ return { failures };
259
298
  }
260
299
 
261
300
  // src/utils/ui.ts
@@ -285,14 +324,14 @@ function blank() {
285
324
  console.log();
286
325
  }
287
326
  function waitForEnter(prompt = "Press Enter to continue...") {
288
- return new Promise((resolve2) => {
327
+ return new Promise((resolve3) => {
289
328
  process.stdout.write(chalk.dim(prompt));
290
329
  const handler = () => {
291
330
  process.stdin.removeListener("data", handler);
292
331
  process.stdin.setRawMode?.(false);
293
332
  process.stdin.pause();
294
333
  console.log();
295
- resolve2();
334
+ resolve3();
296
335
  };
297
336
  if (process.stdin.isTTY) {
298
337
  process.stdin.setRawMode?.(true);
@@ -302,7 +341,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
302
341
  });
303
342
  }
304
343
  function sleep(ms) {
305
- return new Promise((resolve2) => setTimeout(resolve2, ms));
344
+ return new Promise((resolve3) => setTimeout(resolve3, ms));
306
345
  }
307
346
 
308
347
  // src/commands/login.ts
@@ -373,22 +412,25 @@ async function deviceFlowLogin(options) {
373
412
  }
374
413
  async function tokenLogin() {
375
414
  console.log("Token login mode.");
376
- console.log("Visit your Evident dashboard to generate a CLI token.");
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
+ );
377
419
  blank();
378
420
  process.stdout.write("Paste token: ");
379
- const token = await new Promise((resolve2) => {
421
+ const token = await new Promise((resolve3) => {
380
422
  let data = "";
381
423
  process.stdin.setEncoding("utf8");
382
424
  process.stdin.on("data", (chunk) => {
383
425
  data += chunk;
384
426
  });
385
427
  process.stdin.on("end", () => {
386
- resolve2(data.trim());
428
+ resolve3(data.trim());
387
429
  });
388
430
  if (process.stdin.isTTY) {
389
431
  process.stdin.once("data", (chunk) => {
390
432
  process.stdin.pause();
391
- resolve2(chunk.toString().trim());
433
+ resolve3(chunk.toString().trim());
392
434
  });
393
435
  process.stdin.resume();
394
436
  }
@@ -397,13 +439,22 @@ async function tokenLogin() {
397
439
  printError("No token provided.");
398
440
  process.exit(1);
399
441
  }
442
+ await validateAndStoreToken(token);
443
+ }
444
+ async function validateAndStoreToken(token) {
400
445
  const spinner = ora("Validating token...").start();
401
446
  try {
402
- 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
+ }
403
455
  await storeToken({
404
456
  token,
405
- user: result.user,
406
- expiresAt: result.expires_at
457
+ user: { email: result.user.email }
407
458
  });
408
459
  spinner.stop();
409
460
  printSuccess(`Logged in as ${chalk2.bold(result.user.email)}`);
@@ -423,9 +474,22 @@ async function login(options) {
423
474
  }
424
475
 
425
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
+ }
426
483
  async function logout(options = {}) {
427
484
  if (options.all) {
428
- 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
+ }
429
493
  printSuccess("Logged out of all endpoints.");
430
494
  return;
431
495
  }
@@ -450,7 +514,9 @@ async function whoami() {
450
514
  blank();
451
515
  console.log(keyValue("Endpoint", apiUrl));
452
516
  console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
453
- console.log(keyValue("User ID", credentials2.user.id));
517
+ if (credentials2.user.id) {
518
+ console.log(keyValue("User ID", credentials2.user.id));
519
+ }
454
520
  if (credentials2.expiresAt) {
455
521
  const expiresAt = new Date(credentials2.expiresAt);
456
522
  const now = /* @__PURE__ */ new Date();
@@ -466,10 +532,125 @@ async function whoami() {
466
532
  blank();
467
533
  }
468
534
 
535
+ // src/lib/claude-usage.ts
536
+ import { execFileSync } from "child_process";
537
+ import { readFileSync } from "fs";
538
+ import { homedir } from "os";
539
+ import { join } from "path";
540
+ var CLAUDE_USAGE_URL = "https://api.anthropic.com/api/oauth/usage";
541
+ var KEYCHAIN_SERVICE = "Claude Code-credentials";
542
+ function parseClaudeCliCredentials(raw) {
543
+ let parsed;
544
+ try {
545
+ parsed = JSON.parse(raw);
546
+ } catch {
547
+ return null;
548
+ }
549
+ const data = parsed.claudeAiOauth ?? parsed;
550
+ const creds = data;
551
+ if (typeof creds.accessToken !== "string" || typeof creds.expiresAt !== "number") {
552
+ return null;
553
+ }
554
+ return { accessToken: creds.accessToken, expiresAt: creds.expiresAt };
555
+ }
556
+ function readClaudeCliCredentials() {
557
+ if (process.platform === "darwin") {
558
+ try {
559
+ const raw = execFileSync(
560
+ "/usr/bin/security",
561
+ ["find-generic-password", "-s", KEYCHAIN_SERVICE, "-w"],
562
+ { encoding: "utf-8", timeout: 2e3, stdio: ["pipe", "pipe", "ignore"] }
563
+ );
564
+ return parseClaudeCliCredentials(raw);
565
+ } catch {
566
+ return null;
567
+ }
568
+ }
569
+ try {
570
+ const raw = readFileSync(join(homedir(), ".claude", ".credentials.json"), "utf-8");
571
+ return parseClaudeCliCredentials(raw);
572
+ } catch {
573
+ return null;
574
+ }
575
+ }
576
+ var ClaudeUsageError = class extends Error {
577
+ constructor(message, reason) {
578
+ super(message);
579
+ this.reason = reason;
580
+ }
581
+ };
582
+ function isLocalCredentialProblem(err) {
583
+ return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
584
+ }
585
+ function toWindow(value) {
586
+ if (!value || typeof value !== "object") {
587
+ return null;
588
+ }
589
+ const window = value;
590
+ if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
591
+ return null;
592
+ }
593
+ return { utilization: window.utilization, resetsAt: window.resets_at };
594
+ }
595
+ async function getClaudeUsage() {
596
+ const credentials2 = readClaudeCliCredentials();
597
+ if (!credentials2) {
598
+ throw new ClaudeUsageError(
599
+ "No local Claude Code login found. Run `claude` once to sign in with your Claude subscription.",
600
+ "no_credentials"
601
+ );
602
+ }
603
+ if (credentials2.expiresAt < Date.now()) {
604
+ throw new ClaudeUsageError(
605
+ "Claude Code credentials have expired. Run `claude` to refresh them.",
606
+ "credentials_expired"
607
+ );
608
+ }
609
+ const res = await fetch(CLAUDE_USAGE_URL, {
610
+ headers: {
611
+ Authorization: `Bearer ${credentials2.accessToken}`,
612
+ "Content-Type": "application/json",
613
+ "anthropic-version": "2023-06-01"
614
+ }
615
+ });
616
+ if (!res.ok) {
617
+ throw new ClaudeUsageError(`Claude usage request failed: HTTP ${res.status}`, "request_failed");
618
+ }
619
+ const body = await res.json();
620
+ return {
621
+ fiveHour: toWindow(body.five_hour),
622
+ sevenDay: toWindow(body.seven_day)
623
+ };
624
+ }
625
+
626
+ // src/commands/claude-usage.ts
627
+ function formatWindow(label, window) {
628
+ if (!window) {
629
+ return keyValue(label, "not available for this plan");
630
+ }
631
+ const resetsAt = new Date(window.resetsAt);
632
+ return keyValue(label, `${window.utilization}% used, resets ${resetsAt.toLocaleString()}`);
633
+ }
634
+ async function claudeUsage() {
635
+ try {
636
+ const usage = await getClaudeUsage();
637
+ blank();
638
+ console.log(formatWindow("5-hour session", usage.fiveHour));
639
+ console.log(formatWindow("7-day", usage.sevenDay));
640
+ blank();
641
+ } catch (err) {
642
+ if (err instanceof ClaudeUsageError) {
643
+ printError(err.message);
644
+ process.exit(1);
645
+ }
646
+ throw err;
647
+ }
648
+ }
649
+
469
650
  // src/commands/run.ts
651
+ import { homedir as homedir3 } from "os";
652
+ import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
470
653
  import chalk6 from "chalk";
471
- import ora3 from "ora";
472
- import { select as select3 } from "@inquirer/prompts";
473
654
 
474
655
  // ../../packages/types/src/telemetry/index.ts
475
656
  var TelemetryEventTypes = {
@@ -478,13 +659,20 @@ var TelemetryEventTypes = {
478
659
  AGENT_DISCONNECTED: "agent.disconnected",
479
660
  AGENT_MESSAGE_PROCESSING: "agent.message_processing",
480
661
  AGENT_MESSAGE_DONE: "agent.message_done",
481
- AGENT_MESSAGE_FAILED: "agent.message_failed"
662
+ AGENT_MESSAGE_FAILED: "agent.message_failed",
663
+ // A `warn`/`error` runner-side log line forwarded server-side for
664
+ // observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
665
+ RUNNER_ACTIVITY: "runner.activity"
482
666
  };
483
667
 
484
668
  // ../../packages/types/src/tunnel/index.ts
485
669
  var MAX_FRAME_BYTES = 256 * 1024;
486
670
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
487
671
 
672
+ // ../../packages/types/src/runner-files.ts
673
+ var MAX_FILE_PUSH_BYTES = 64 * 1024;
674
+ var MAX_FILE_SYNC_DIRECTORIES = 16;
675
+
488
676
  // ../../packages/types/src/logging/index.ts
489
677
  var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
678
  function log(level, event, fields) {
@@ -514,6 +702,10 @@ function stripQuery(url) {
514
702
  }
515
703
  }
516
704
 
705
+ // src/commands/run.ts
706
+ import ora3 from "ora";
707
+ import { select as select3 } from "@inquirer/prompts";
708
+
517
709
  // src/lib/telemetry.ts
518
710
  var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
519
711
  function getCliVersion() {
@@ -525,6 +717,13 @@ var isShuttingDown = false;
525
717
  var FLUSH_INTERVAL_MS = 5e3;
526
718
  var MAX_BUFFER_SIZE = 50;
527
719
  var FLUSH_TIMEOUT_MS = 3e3;
720
+ var authProvider = null;
721
+ function setTelemetryAuthProvider(provider) {
722
+ authProvider = provider;
723
+ }
724
+ var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
725
+ var lastFlushFailureLoggedAt = 0;
726
+ var suppressedFlushFailureCount = 0;
528
727
  function logEvent(eventType, options = {}) {
529
728
  const event = {
530
729
  event_type: eventType,
@@ -559,9 +758,16 @@ async function flushEvents() {
559
758
  flushTimeout = null;
560
759
  }
561
760
  try {
562
- const credentials2 = await getToken();
563
- if (!credentials2) {
564
- return;
761
+ const providerContext = authProvider?.();
762
+ let authHeader;
763
+ if (providerContext?.authHeader) {
764
+ authHeader = providerContext.authHeader;
765
+ } else {
766
+ const credentials2 = await getToken();
767
+ if (!credentials2) {
768
+ return;
769
+ }
770
+ authHeader = `Bearer ${credentials2.token}`;
565
771
  }
566
772
  const apiUrl = getApiUrlConfig();
567
773
  const controller = new AbortController();
@@ -576,7 +782,7 @@ async function flushEvents() {
576
782
  method: "POST",
577
783
  headers: {
578
784
  "Content-Type": "application/json",
579
- Authorization: `Bearer ${credentials2.token}`
785
+ Authorization: authHeader
580
786
  },
581
787
  body: JSON.stringify(request),
582
788
  signal: controller.signal
@@ -588,8 +794,15 @@ async function flushEvents() {
588
794
  clearTimeout(timeout);
589
795
  }
590
796
  } catch (error2) {
591
- if (process.env.DEBUG) {
592
- console.error("Telemetry error:", error2);
797
+ const now = Date.now();
798
+ if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
799
+ const message = error2 instanceof Error ? error2.message : String(error2);
800
+ const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
801
+ console.error(`Telemetry flush error: ${message}${suffix}`);
802
+ lastFlushFailureLoggedAt = now;
803
+ suppressedFlushFailureCount = 0;
804
+ } else {
805
+ suppressedFlushFailureCount++;
593
806
  }
594
807
  }
595
808
  }
@@ -658,6 +871,69 @@ var EventTypes = {
658
871
  DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
659
872
  };
660
873
 
874
+ // src/lib/runner-activity-telemetry.ts
875
+ var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
876
+ var SEVERITY_BY_LEVEL = {
877
+ warn: "warning",
878
+ error: "error"
879
+ };
880
+ var MAX_MESSAGE_LENGTH = 500;
881
+ var TRUNCATION_MARKER = "\u2026";
882
+ function redact(message) {
883
+ return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
884
+ }
885
+ function truncate(message) {
886
+ if (message.length <= MAX_MESSAGE_LENGTH) return message;
887
+ return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
888
+ }
889
+ var RATE_LIMIT_WINDOW_MS = 6e4;
890
+ var RATE_LIMIT_MAX_EVENTS = 30;
891
+ var windowStartedAt = 0;
892
+ var windowCount = 0;
893
+ var windowDroppedCount = 0;
894
+ function admitUnderRateLimit(now) {
895
+ if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
896
+ if (windowDroppedCount > 0) {
897
+ console.error(
898
+ `[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
899
+ );
900
+ }
901
+ windowStartedAt = now;
902
+ windowCount = 0;
903
+ windowDroppedCount = 0;
904
+ }
905
+ if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
906
+ windowDroppedCount++;
907
+ if (windowDroppedCount === 1) {
908
+ console.error(
909
+ `[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
910
+ );
911
+ }
912
+ return false;
913
+ }
914
+ windowCount++;
915
+ return true;
916
+ }
917
+ function forwardRunnerActivity(entry, context) {
918
+ try {
919
+ if (!FORWARDED_LEVELS.has(entry.level)) return;
920
+ if (!context.agentId || !context.authHeader) return;
921
+ if (!admitUnderRateLimit(Date.now())) return;
922
+ const rawMessage = entry.error ?? entry.message ?? "";
923
+ const message = truncate(redact(rawMessage));
924
+ logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
925
+ severity: SEVERITY_BY_LEVEL[entry.level],
926
+ message,
927
+ metadata: { source: "cli.run" },
928
+ agentId: context.agentId
929
+ });
930
+ } catch (err) {
931
+ console.error(
932
+ `[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
933
+ );
934
+ }
935
+ }
936
+
661
937
  // src/lib/auth.ts
662
938
  async function getAuthCredentials() {
663
939
  const runnerKey = process.env.EVIDENT_RUNNER_KEY;
@@ -725,7 +1001,7 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
725
1001
  if (health.healthy) {
726
1002
  return health;
727
1003
  }
728
- await new Promise((resolve2) => setTimeout(resolve2, 1e3));
1004
+ await new Promise((resolve3) => setTimeout(resolve3, 1e3));
729
1005
  }
730
1006
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
731
1007
  }
@@ -1364,7 +1640,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
1364
1640
  }
1365
1641
  }
1366
1642
  if (attempt < READ_BACK_ATTEMPTS - 1) {
1367
- await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1643
+ await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
1368
1644
  }
1369
1645
  }
1370
1646
  return null;
@@ -1492,6 +1768,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
1492
1768
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1493
1769
  return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1494
1770
  }
1771
+ function isB2AbandonmentConfirmed(params) {
1772
+ return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
1773
+ }
1495
1774
  function messageError(messages, userMessageId) {
1496
1775
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1497
1776
  const error2 = errorOf(reply);
@@ -1505,6 +1784,42 @@ function messageError(messages, userMessageId) {
1505
1784
  }
1506
1785
  return "The agent run failed.";
1507
1786
  }
1787
+ function messageFailure(messages, userMessageId) {
1788
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1789
+ const error2 = errorOf(reply);
1790
+ if (error2 == null || typeof error2 !== "object") return null;
1791
+ const e = error2;
1792
+ const replyProviderId = reply?.info?.providerID ?? null;
1793
+ const replyModelId = reply?.info?.modelID ?? null;
1794
+ if (e.name === "ProviderAuthError") {
1795
+ const data = e.data;
1796
+ const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
1797
+ return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
1798
+ }
1799
+ if (e.name === "APIError") {
1800
+ const data = e.data;
1801
+ const statusCode = data?.statusCode;
1802
+ if (statusCode === 401 || statusCode === 403) {
1803
+ return {
1804
+ kind: "model_auth",
1805
+ providerId: replyProviderId,
1806
+ modelId: replyModelId,
1807
+ reason: "rejected"
1808
+ };
1809
+ }
1810
+ }
1811
+ return null;
1812
+ }
1813
+ function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
1814
+ if (classified != null) return classified;
1815
+ if (hasConfiguredProvider !== false) return null;
1816
+ return {
1817
+ kind: "model_auth",
1818
+ providerId: replyProviderId,
1819
+ modelId: replyModelId,
1820
+ reason: "missing"
1821
+ };
1822
+ }
1508
1823
  function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1509
1824
  if (!messages || messages.length === 0) return false;
1510
1825
  return messages.some(
@@ -1738,12 +2053,12 @@ var StreamForwarder = class {
1738
2053
  let endBody;
1739
2054
  if (has_body) {
1740
2055
  const chunks = [];
1741
- bodyPromise = new Promise((resolve2) => {
2056
+ bodyPromise = new Promise((resolve3) => {
1742
2057
  pushBody = (buf) => {
1743
2058
  chunks.push(buf);
1744
2059
  };
1745
2060
  endBody = () => {
1746
- resolve2(Buffer.concat(chunks));
2061
+ resolve3(Buffer.concat(chunks));
1747
2062
  };
1748
2063
  });
1749
2064
  }
@@ -1860,7 +2175,7 @@ function connectTunnel(options) {
1860
2175
  } = options;
1861
2176
  const tunnelUrl = getTunnelUrlConfig();
1862
2177
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1863
- return new Promise((resolve2, reject) => {
2178
+ return new Promise((resolve3, reject) => {
1864
2179
  const ws = new WebSocket2(url, {
1865
2180
  headers: {
1866
2181
  Authorization: authHeader
@@ -1915,7 +2230,7 @@ function connectTunnel(options) {
1915
2230
  clearTimeout(connectionTimeout);
1916
2231
  const connectedAgentId = message.agent_id ?? agentId;
1917
2232
  onConnected?.(connectedAgentId);
1918
- resolve2({
2233
+ resolve3({
1919
2234
  ws,
1920
2235
  close: () => ws.close(1e3, "CLI shutdown")
1921
2236
  });
@@ -2033,6 +2348,443 @@ var RunnerConnection = class {
2033
2348
  }
2034
2349
  };
2035
2350
 
2351
+ // src/lib/tunnel/ready-marker.ts
2352
+ import { writeFileSync } from "fs";
2353
+ function writeTunnelReadyMarker(path, agentId) {
2354
+ try {
2355
+ writeFileSync(path, `${agentId}
2356
+ `);
2357
+ return { ok: true };
2358
+ } catch (error2) {
2359
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2360
+ }
2361
+ }
2362
+
2363
+ // src/lib/claude-usage-reporting.ts
2364
+ var VALID_MODES = ["auto", "on", "off"];
2365
+ function resolveClaudeUsageReportingMode(flagValue, env) {
2366
+ const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
2367
+ if (raw === void 0 || raw === "") {
2368
+ return { mode: "auto", warnings: [] };
2369
+ }
2370
+ const normalized = raw.trim().toLowerCase();
2371
+ if (VALID_MODES.includes(normalized)) {
2372
+ return { mode: normalized, warnings: [] };
2373
+ }
2374
+ const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
2375
+ return {
2376
+ mode: "auto",
2377
+ warnings: [
2378
+ `Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
2379
+ ]
2380
+ };
2381
+ }
2382
+ var BASE_REPORT_DELAY_MS = 10 * 6e4;
2383
+ var REPORT_DELAY_JITTER_FRACTION = 0.2;
2384
+ function nextReportDelayMs(random = Math.random) {
2385
+ const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
2386
+ return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
2387
+ }
2388
+ var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
2389
+
2390
+ // src/lib/channels/driver.ts
2391
+ import { homedir as homedir2 } from "os";
2392
+
2393
+ // src/lib/file-push.ts
2394
+ import { randomUUID } from "crypto";
2395
+ import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
2396
+ import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
2397
+ var FILE_MODE = 384;
2398
+ var DIRECTORY_MODE = 448;
2399
+ async function writePushedFile(request) {
2400
+ const { requestedPath, content, allowedDirectories, homeDir } = request;
2401
+ const bytes = content.byteLength;
2402
+ if (allowedDirectories.length === 0) {
2403
+ return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
2404
+ path: requestedPath,
2405
+ bytes
2406
+ });
2407
+ }
2408
+ if (bytes > MAX_FILE_PUSH_BYTES) {
2409
+ return refuse(
2410
+ "file_too_large",
2411
+ `File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
2412
+ {
2413
+ path: requestedPath,
2414
+ bytes
2415
+ }
2416
+ );
2417
+ }
2418
+ const candidate = expandAndValidate(requestedPath, homeDir);
2419
+ if (candidate === null) {
2420
+ return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
2421
+ path: requestedPath,
2422
+ bytes
2423
+ });
2424
+ }
2425
+ try {
2426
+ const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
2427
+ dirname2(candidate)
2428
+ );
2429
+ const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
2430
+ const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
2431
+ if (allowedDirectory === null) {
2432
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2433
+ path: realTarget,
2434
+ bytes
2435
+ });
2436
+ }
2437
+ if (missingSegments.length > 0) {
2438
+ await createMissingDirectories(existingAncestor, missingSegments);
2439
+ const realParent = await realpath(dirname2(realTarget));
2440
+ if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
2441
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2442
+ path: realTarget,
2443
+ bytes,
2444
+ reason: "parent_changed_after_create"
2445
+ });
2446
+ }
2447
+ }
2448
+ await writeAtomically(realTarget, content);
2449
+ log("info", "file_push_written", { path: realTarget, bytes });
2450
+ return { ok: true, path: realTarget };
2451
+ } catch (err) {
2452
+ const errno = err.code ?? "UNKNOWN";
2453
+ return refuse("write_failed", `The runner could not write the file (${errno}).`, {
2454
+ path: candidate,
2455
+ bytes,
2456
+ errno,
2457
+ ...errorFields(err)
2458
+ });
2459
+ }
2460
+ }
2461
+ function expandAndValidate(requestedPath, homeDir) {
2462
+ if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
2463
+ return null;
2464
+ }
2465
+ const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
2466
+ if (expanded.split(/[/\\]/).includes("..")) {
2467
+ return null;
2468
+ }
2469
+ if (!isAbsolute(expanded)) {
2470
+ return null;
2471
+ }
2472
+ const candidate = resolve2(expanded);
2473
+ const name = basename(candidate);
2474
+ return name === "" || name === "." || name === ".." ? null : candidate;
2475
+ }
2476
+ async function resolveNearestExistingAncestor(directory) {
2477
+ const missingSegments = [];
2478
+ let current = directory;
2479
+ for (; ; ) {
2480
+ try {
2481
+ return { existingAncestor: await realpath(current), missingSegments };
2482
+ } catch (err) {
2483
+ const parent = dirname2(current);
2484
+ if (err.code !== "ENOENT" || parent === current) {
2485
+ throw err;
2486
+ }
2487
+ missingSegments.unshift(basename(current));
2488
+ current = parent;
2489
+ }
2490
+ }
2491
+ }
2492
+ async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
2493
+ for (const directory of allowedDirectories) {
2494
+ if (!isAbsolute(directory)) {
2495
+ log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
2496
+ continue;
2497
+ }
2498
+ const realDirectory = await realpathCreatingIfMissing(directory);
2499
+ if (realDirectory !== null && contains(realDirectory, realTarget)) {
2500
+ return realDirectory;
2501
+ }
2502
+ }
2503
+ return null;
2504
+ }
2505
+ async function realpathCreatingIfMissing(directory) {
2506
+ try {
2507
+ return await realpath(directory);
2508
+ } catch (err) {
2509
+ if (err.code !== "ENOENT") {
2510
+ log("warn", "file_push_allowed_directory_skipped", {
2511
+ directory,
2512
+ reason: "unresolvable",
2513
+ ...errorFields(err)
2514
+ });
2515
+ return null;
2516
+ }
2517
+ }
2518
+ try {
2519
+ await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
2520
+ await chmod(directory, DIRECTORY_MODE);
2521
+ return await realpath(directory);
2522
+ } catch (err) {
2523
+ log("warn", "file_push_allowed_directory_skipped", {
2524
+ directory,
2525
+ reason: "create_failed",
2526
+ ...errorFields(err)
2527
+ });
2528
+ return null;
2529
+ }
2530
+ }
2531
+ function contains(realDirectory, realTarget) {
2532
+ const rel = relative(realDirectory, realTarget);
2533
+ return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
2534
+ }
2535
+ async function createMissingDirectories(existingAncestor, missingSegments) {
2536
+ let current = existingAncestor;
2537
+ for (const segment of missingSegments) {
2538
+ current = join2(current, segment);
2539
+ await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
2540
+ await chmod(current, DIRECTORY_MODE);
2541
+ }
2542
+ }
2543
+ async function writeAtomically(realTarget, content) {
2544
+ const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2545
+ let handle;
2546
+ try {
2547
+ handle = await open2(temporaryPath, "wx", FILE_MODE);
2548
+ await handle.writeFile(content);
2549
+ await handle.chmod(FILE_MODE);
2550
+ await handle.close();
2551
+ handle = void 0;
2552
+ await rename(temporaryPath, realTarget);
2553
+ } catch (err) {
2554
+ await discardTemporaryFile(temporaryPath, handle);
2555
+ throw err;
2556
+ }
2557
+ }
2558
+ async function discardTemporaryFile(temporaryPath, handle) {
2559
+ try {
2560
+ await handle?.close();
2561
+ } catch (err) {
2562
+ log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
2563
+ }
2564
+ try {
2565
+ await unlink(temporaryPath);
2566
+ } catch (err) {
2567
+ const errno = err.code;
2568
+ if (errno !== "ENOENT" && errno !== "ENOTDIR") {
2569
+ log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
2570
+ }
2571
+ }
2572
+ }
2573
+ function refuse(code, message, fields) {
2574
+ log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
2575
+ return { ok: false, code, message };
2576
+ }
2577
+
2578
+ // src/lib/runner-file-sync.ts
2579
+ var MAX_ACK_ATTEMPTS = 5;
2580
+ async function syncPendingRunnerFiles(options) {
2581
+ const pending = await listPendingFiles(options);
2582
+ const pendingIds = new Set(pending.map((file) => file.id));
2583
+ for (const id of options.ackFailures.keys()) {
2584
+ if (!pendingIds.has(id)) options.ackFailures.delete(id);
2585
+ }
2586
+ if (pending.length === 0) return 0;
2587
+ options.log({
2588
+ level: "info",
2589
+ message: `Runner file sync: ${pending.length} file(s) queued for this runner`
2590
+ });
2591
+ let applied = 0;
2592
+ for (const file of pending) {
2593
+ if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
2594
+ if (await applyOne(options, file)) applied += 1;
2595
+ }
2596
+ return applied;
2597
+ }
2598
+ async function listPendingFiles(options) {
2599
+ let res;
2600
+ try {
2601
+ res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
2602
+ headers: { Authorization: options.getAuthHeader() }
2603
+ });
2604
+ } catch (err) {
2605
+ options.log({
2606
+ level: "warn",
2607
+ message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
2608
+ });
2609
+ return [];
2610
+ }
2611
+ if (!res.ok) {
2612
+ options.log({
2613
+ level: res.status === 404 ? "debug" : "warn",
2614
+ message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
2615
+ });
2616
+ return [];
2617
+ }
2618
+ let body;
2619
+ try {
2620
+ body = await res.json();
2621
+ } catch (err) {
2622
+ options.log({
2623
+ level: "warn",
2624
+ message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
2625
+ });
2626
+ return [];
2627
+ }
2628
+ if (!Array.isArray(body)) {
2629
+ options.log({
2630
+ level: "warn",
2631
+ message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
2632
+ });
2633
+ return [];
2634
+ }
2635
+ const files = [];
2636
+ for (const entry of body) {
2637
+ const file = asPendingFile(entry);
2638
+ if (file === null) {
2639
+ options.log({
2640
+ level: "warn",
2641
+ message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
2642
+ });
2643
+ continue;
2644
+ }
2645
+ files.push(file);
2646
+ }
2647
+ return files;
2648
+ }
2649
+ function asPendingFile(entry) {
2650
+ if (entry === null || typeof entry !== "object") return null;
2651
+ const { id, path, size } = entry;
2652
+ if (typeof id !== "string" || id === "") return null;
2653
+ if (typeof path !== "string" || path === "") return null;
2654
+ if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
2655
+ return { id, path, size };
2656
+ }
2657
+ async function applyOne(options, file) {
2658
+ const label = `${file.id.slice(0, 8)} (${file.path})`;
2659
+ if (options.allowedDirectories.length === 0) {
2660
+ options.log({
2661
+ level: "warn",
2662
+ message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
2663
+ });
2664
+ await ack(options, file, "rejected", "file_sync_disabled");
2665
+ return false;
2666
+ }
2667
+ if (file.size > MAX_FILE_PUSH_BYTES) {
2668
+ options.log({
2669
+ level: "warn",
2670
+ message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
2671
+ });
2672
+ await ack(options, file, "rejected", "file_too_large");
2673
+ return false;
2674
+ }
2675
+ const download = await downloadContent(options, file, label);
2676
+ if (!download.ok) {
2677
+ if (download.terminal) await ack(options, file, "rejected", download.code);
2678
+ return false;
2679
+ }
2680
+ let outcome;
2681
+ try {
2682
+ outcome = await writePushedFile({
2683
+ requestedPath: file.path,
2684
+ content: download.content,
2685
+ allowedDirectories: options.allowedDirectories,
2686
+ homeDir: options.homeDir
2687
+ });
2688
+ } catch (err) {
2689
+ options.log({
2690
+ level: "error",
2691
+ message: `Runner file ${label} could not be written: ${describe(err)}`
2692
+ });
2693
+ await ack(options, file, "rejected", "write_failed");
2694
+ return false;
2695
+ }
2696
+ if (!outcome.ok) {
2697
+ options.log({
2698
+ level: "warn",
2699
+ message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
2700
+ });
2701
+ await ack(options, file, "rejected", outcome.code);
2702
+ return false;
2703
+ }
2704
+ options.log({
2705
+ level: "info",
2706
+ message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
2707
+ });
2708
+ await ack(options, file, "applied");
2709
+ return true;
2710
+ }
2711
+ function durableDownloadCode(status) {
2712
+ return status === 413 ? "file_too_large" : "write_failed";
2713
+ }
2714
+ async function downloadContent(options, file, label) {
2715
+ try {
2716
+ const res = await options.fetchImpl(
2717
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
2718
+ { headers: { Authorization: options.getAuthHeader() } }
2719
+ );
2720
+ if (!res.ok) {
2721
+ const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
2722
+ if (!terminal) {
2723
+ options.log({
2724
+ level: "warn",
2725
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
2726
+ });
2727
+ return { ok: false, terminal: false };
2728
+ }
2729
+ const code = durableDownloadCode(res.status);
2730
+ options.log({
2731
+ level: "error",
2732
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
2733
+ });
2734
+ return { ok: false, terminal: true, code };
2735
+ }
2736
+ return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
2737
+ } catch (err) {
2738
+ options.log({
2739
+ level: "warn",
2740
+ message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
2741
+ });
2742
+ return { ok: false, terminal: false };
2743
+ }
2744
+ }
2745
+ async function ack(options, file, status, reason) {
2746
+ const outcome = `${status}${reason ? ` (${reason})` : ""}`;
2747
+ try {
2748
+ const res = await options.fetchImpl(
2749
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
2750
+ {
2751
+ method: "POST",
2752
+ headers: {
2753
+ Authorization: options.getAuthHeader(),
2754
+ "Content-Type": "application/json"
2755
+ },
2756
+ body: JSON.stringify(reason ? { status, reason } : { status })
2757
+ }
2758
+ );
2759
+ if (!res.ok) {
2760
+ recordAckFailure(
2761
+ options,
2762
+ file,
2763
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
2764
+ );
2765
+ return;
2766
+ }
2767
+ options.ackFailures.delete(file.id);
2768
+ } catch (err) {
2769
+ recordAckFailure(
2770
+ options,
2771
+ file,
2772
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
2773
+ );
2774
+ }
2775
+ }
2776
+ function recordAckFailure(options, file, what) {
2777
+ const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
2778
+ options.ackFailures.set(file.id, attempts);
2779
+ options.log({
2780
+ level: "error",
2781
+ message: attempts >= MAX_ACK_ATTEMPTS ? `${what} \u2014 giving up after ${attempts} attempts. It stays pending until the server expires it; restart the runner to retry.` : `${what} \u2014 it stays pending until a later drain re-acks it (attempt ${attempts} of ${MAX_ACK_ATTEMPTS})`
2782
+ });
2783
+ }
2784
+ function describe(err) {
2785
+ return err instanceof Error ? err.message : String(err);
2786
+ }
2787
+
2036
2788
  // src/lib/channels/driver.ts
2037
2789
  function messageIdOf(m) {
2038
2790
  if (!m || typeof m !== "object") return void 0;
@@ -2061,7 +2813,10 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
2061
2813
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
2062
2814
  var HEARTBEAT_MS = 6e4;
2063
2815
  var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
2816
+ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
2817
+ var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2064
2818
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
2819
+ var MAX_SUPERSEDED_CONVERSATIONS = 256;
2065
2820
  var ChannelAuthError = class extends Error {
2066
2821
  constructor(message) {
2067
2822
  super(message);
@@ -2098,8 +2853,38 @@ var ChannelDriver = class _ChannelDriver {
2098
2853
  pausedMaxWaitMs;
2099
2854
  stuckQueuedMs;
2100
2855
  now;
2856
+ fileSyncDirectories;
2857
+ homeDir;
2101
2858
  /** Cache of conversationId → opencode sessionId. */
2102
2859
  sessions = /* @__PURE__ */ new Map();
2860
+ /**
2861
+ * conversationId → the opencode session this runner has ABANDONED as that
2862
+ * conversation's binding (#553), after a genuine (`sessionExists === true`)
2863
+ * dispatch failure: the session still exists but is wedged, so #485's self-heal
2864
+ * must bind a fresh one.
2865
+ *
2866
+ * Dropping the local binding + clearing the server row is not enough on its own:
2867
+ * a SIBLING message dispatched earlier in the same drain is still in-flight under
2868
+ * the same session, and its watcher's routine status writes carry
2869
+ * `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
2870
+ * and `ensureSession`'s persisted-id fallback then reuses it, defeating the
2871
+ * self-heal. This map makes the runner authoritative instead of racing those
2872
+ * writes: *`ensureSession` never reuses an abandoned id for that conversation,
2873
+ * whatever the server row says* — which holds even when the resurrecting write
2874
+ * is one we deliberately keep (see `markDone`).
2875
+ *
2876
+ * Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
2877
+ * one conversation hold ONE entry (the newest abandonment replaces the older), and
2878
+ * hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
2879
+ * NEWEST abandoned id per conversation is guarded: after a second abandonment a
2880
+ * late sibling of the FIRST session can write that id back and `ensureSession`
2881
+ * will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
2882
+ * NOT dropped when the session's watcher tears down: `markDone` still writes the
2883
+ * abandoned id back (it must, or the reply is lost), so the guard has to outlive
2884
+ * the turn that resurrects it. In-memory only — a restart forgets it, at the same
2885
+ * bounded cost.
2886
+ */
2887
+ supersededSessions = /* @__PURE__ */ new Map();
2103
2888
  /**
2104
2889
  * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
2105
2890
  * longer idempotent (no caller-supplied `messageID`), and its read-back picks
@@ -2222,6 +3007,24 @@ var ChannelDriver = class _ChannelDriver {
2222
3007
  sessionTitles = /* @__PURE__ */ new Map();
2223
3008
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
2224
3009
  draining = false;
3010
+ /**
3011
+ * Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
3012
+ * drain ping don't download, write and ack the same file twice.
3013
+ */
3014
+ syncingFiles = false;
3015
+ /**
3016
+ * Consecutive failed acks per pending file (#559). Lives on the driver so it
3017
+ * survives across drains — without it, a file whose ack keeps failing is
3018
+ * re-downloaded and re-written every ~2s until the server expires it.
3019
+ */
3020
+ fileAckFailures = /* @__PURE__ */ new Map();
3021
+ /**
3022
+ * Monotonic count of files this runner has pulled and written (#559). Only
3023
+ * ever increases, so `run.ts` detects work by comparing it against the value
3024
+ * it saw on the previous cycle — including work that landed mid-sleep, the
3025
+ * same trick `lastProxiedActivityAt` uses.
3026
+ */
3027
+ appliedFileCount = 0;
2225
3028
  /**
2226
3029
  * The currently-executing `drainPending()` promise, or null when idle. Lets a
2227
3030
  * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
@@ -2237,21 +3040,23 @@ var ChannelDriver = class _ChannelDriver {
2237
3040
  * and stops opencode.
2238
3041
  */
2239
3042
  stopped = false;
2240
- constructor(config2) {
2241
- this.agentId = config2.agentId;
2242
- this.port = config2.port;
2243
- this.apiUrl = config2.apiUrl.replace(/\/$/, "");
2244
- this.getAuthHeader = config2.getAuthHeader;
2245
- this.conversationFilter = config2.conversationFilter ?? null;
2246
- this.retry = { ...DEFAULT_RETRY_POLICY, ...config2.retry };
2247
- this.log = config2.log ?? (() => {
3043
+ constructor(config) {
3044
+ this.agentId = config.agentId;
3045
+ this.port = config.port;
3046
+ this.apiUrl = config.apiUrl.replace(/\/$/, "");
3047
+ this.getAuthHeader = config.getAuthHeader;
3048
+ this.conversationFilter = config.conversationFilter ?? null;
3049
+ this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
3050
+ this.log = config.log ?? (() => {
2248
3051
  });
2249
- this.fetchImpl = config2.fetchImpl ?? fetch;
2250
- this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
2251
- this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
2252
- this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2253
- this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2254
- this.now = config2.now ?? (() => Date.now());
3052
+ this.fetchImpl = config.fetchImpl ?? fetch;
3053
+ this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
3054
+ this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
3055
+ this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
3056
+ this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
3057
+ this.now = config.now ?? (() => Date.now());
3058
+ this.fileSyncDirectories = config.fileSyncDirectories ?? [];
3059
+ this.homeDir = config.homeDir ?? homedir2();
2255
3060
  }
2256
3061
  /** The IPv4-loopback base URL for the local `opencode serve`. */
2257
3062
  get opencodeBase() {
@@ -2279,6 +3084,47 @@ var ChannelDriver = class _ChannelDriver {
2279
3084
  );
2280
3085
  return run2;
2281
3086
  }
3087
+ /**
3088
+ * Pull-and-apply any files Evident has queued for this runner (#559), riding
3089
+ * the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
3090
+ * and drain ping that call `drainPending()`. There is deliberately no channel,
3091
+ * control frame or poll loop of its own: worst-case latency is one poll tick.
3092
+ *
3093
+ * NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
3094
+ * cost a conversation turn. Failures are logged and either acked as a terminal
3095
+ * outcome or left pending for the next drain (see `runner-file-sync.ts`).
3096
+ *
3097
+ * Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
3098
+ *
3099
+ * @returns the number of files written to disk.
3100
+ */
3101
+ async syncPendingFiles() {
3102
+ if (this.stopped) return 0;
3103
+ if (this.syncingFiles) return 0;
3104
+ this.syncingFiles = true;
3105
+ try {
3106
+ const applied = await syncPendingRunnerFiles({
3107
+ agentId: this.agentId,
3108
+ apiUrl: this.apiUrl,
3109
+ getAuthHeader: this.getAuthHeader,
3110
+ fetchImpl: this.fetchImpl,
3111
+ allowedDirectories: this.fileSyncDirectories,
3112
+ homeDir: this.homeDir,
3113
+ ackFailures: this.fileAckFailures,
3114
+ log: this.log
3115
+ });
3116
+ this.appliedFileCount += applied;
3117
+ return applied;
3118
+ } catch (err) {
3119
+ this.log({
3120
+ level: "error",
3121
+ message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
3122
+ });
3123
+ return 0;
3124
+ } finally {
3125
+ this.syncingFiles = false;
3126
+ }
3127
+ }
2282
3128
  async runDrain() {
2283
3129
  let dispatched = 0;
2284
3130
  try {
@@ -2312,6 +3158,28 @@ var ChannelDriver = class _ChannelDriver {
2312
3158
  }
2313
3159
  return false;
2314
3160
  }
3161
+ /**
3162
+ * File-pull work, for `run.ts`'s idle accounting (#559).
3163
+ *
3164
+ * Pulling a file is real work that `drainPending()` knows nothing about, so
3165
+ * without this a near-idle runner counts a credential pull as an empty tick
3166
+ * and `--idle-timeout` can `process.exit` mid-pull — leaving a
3167
+ * `.evident-push-*.tmp` behind — or immediately after the write, before the
3168
+ * browser has run the authorize/callback that activates it (the user then sees
3169
+ * `saved_not_activated` for a runner that was fine).
3170
+ *
3171
+ * Two signals because one cannot cover both cases: `inFlight` is the pull
3172
+ * happening RIGHT NOW (it may outlive the tick that started it), and
3173
+ * `appliedFiles` is monotonic so a pull that started AND finished between two
3174
+ * idle checks still shows up as an advance.
3175
+ *
3176
+ * CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
3177
+ * the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
3178
+ * samples afterwards reads `true` every single cycle and can never idle out.
3179
+ */
3180
+ fileSyncActivity() {
3181
+ return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
3182
+ }
2315
3183
  /**
2316
3184
  * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2317
3185
  * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
@@ -2373,7 +3241,7 @@ var ChannelDriver = class _ChannelDriver {
2373
3241
  await this.sleep(step);
2374
3242
  }
2375
3243
  }
2376
- while (this.hasInFlightWatchers()) {
3244
+ while (this.hasInFlightWatchers() || this.syncingFiles) {
2377
3245
  if (this.now() >= deadline) return false;
2378
3246
  await this.sleep(step);
2379
3247
  }
@@ -2408,10 +3276,15 @@ var ChannelDriver = class _ChannelDriver {
2408
3276
  * @returns the count of messages NEWLY dispatched (not already in-flight).
2409
3277
  */
2410
3278
  async processConversation(conv) {
2411
- const sessionId = await this.ensureSession(conv);
3279
+ const { sessionId, refusedSessionId } = await this.ensureSession(conv);
2412
3280
  const messages = await this.getPendingMessages(conv.id);
2413
3281
  let dispatched = 0;
2414
3282
  let skippedAlreadyDispatched = 0;
3283
+ if (refusedSessionId && messages.length > 0) {
3284
+ void this.postSignal(conv.id, messages[0].id, "session_superseded", {
3285
+ superseded_session_id: refusedSessionId
3286
+ });
3287
+ }
2415
3288
  for (const message of messages) {
2416
3289
  if (this.stopped) break;
2417
3290
  if (this.dispatched.has(message.id)) {
@@ -2460,6 +3333,13 @@ var ChannelDriver = class _ChannelDriver {
2460
3333
  }
2461
3334
  const errorMessage = err instanceof Error ? err.message : String(err);
2462
3335
  this.sessions.delete(conv.id);
3336
+ this.supersede(conv.id, sessionId);
3337
+ this.log({
3338
+ level: "warn",
3339
+ message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
3340
+ conversation_id: conv.id,
3341
+ message_id: message.id
3342
+ });
2463
3343
  await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
2464
3344
  this.log({
2465
3345
  level: "warn",
@@ -2500,8 +3380,42 @@ var ChannelDriver = class _ChannelDriver {
2500
3380
  this.ensureWatcherRunning(sessionId);
2501
3381
  return dispatched;
2502
3382
  }
3383
+ /**
3384
+ * Record that `sessionId` is no longer a valid binding for `conversationId`
3385
+ * (#553). Keyed by conversation and hard-capped, so it cannot grow with the
3386
+ * number of failures — see the `supersededSessions` field doc.
3387
+ */
3388
+ supersede(conversationId, sessionId) {
3389
+ this.supersededSessions.delete(conversationId);
3390
+ this.supersededSessions.set(conversationId, sessionId);
3391
+ while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
3392
+ const oldest = this.supersededSessions.keys().next().value;
3393
+ if (oldest === void 0) return;
3394
+ this.supersededSessions.delete(oldest);
3395
+ }
3396
+ }
3397
+ /** Whether `sessionId` is the session this conversation has abandoned (#553). */
3398
+ isSuperseded(conversationId, sessionId) {
3399
+ return this.supersededSessions.get(conversationId) === sessionId;
3400
+ }
3401
+ /**
3402
+ * Resolve the opencode session to run this conversation's turns in.
3403
+ *
3404
+ * `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
3405
+ * binding was an id this runner had abandoned, so a resurrection genuinely
3406
+ * happened and a fresh session was bound instead. The caller reports it.
3407
+ */
2503
3408
  async ensureSession(conv) {
2504
3409
  const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
3410
+ if (bound && this.isSuperseded(conv.id, bound)) {
3411
+ this.log({
3412
+ level: "warn",
3413
+ message: `OpenCode session ${bound.slice(0, 8)} was abandoned for conversation ${conv.id.slice(0, 8)} after a failed dispatch but is still bound to it (the persisted id was written back by a turn already in flight) \u2014 ignoring it and binding a fresh session.`,
3414
+ conversation_id: conv.id
3415
+ });
3416
+ this.sessions.delete(conv.id);
3417
+ return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
3418
+ }
2505
3419
  if (bound) {
2506
3420
  const exists = await sessionExists(this.port, bound);
2507
3421
  if (exists === false) {
@@ -2511,12 +3425,12 @@ var ChannelDriver = class _ChannelDriver {
2511
3425
  conversation_id: conv.id
2512
3426
  });
2513
3427
  this.sessions.delete(conv.id);
2514
- return this.createAndBindSession(conv.id);
3428
+ return { sessionId: await this.createAndBindSession(conv.id) };
2515
3429
  }
2516
3430
  this.sessions.set(conv.id, bound);
2517
- return bound;
3431
+ return { sessionId: bound };
2518
3432
  }
2519
- return this.createAndBindSession(conv.id);
3433
+ return { sessionId: await this.createAndBindSession(conv.id) };
2520
3434
  }
2521
3435
  /**
2522
3436
  * Create a fresh OpenCode session for a conversation, cache the binding, and
@@ -2527,7 +3441,12 @@ var ChannelDriver = class _ChannelDriver {
2527
3441
  const directory = await this.resolveOpenCodeDirectory();
2528
3442
  const sessionId = await createOpenCodeSession(this.port, directory);
2529
3443
  this.sessions.set(conversationId, sessionId);
2530
- await this.persistSession(conversationId, sessionId).catch(() => {
3444
+ await this.persistSession(conversationId, sessionId).catch((err) => {
3445
+ this.log({
3446
+ level: "warn",
3447
+ 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)}`,
3448
+ conversation_id: conversationId
3449
+ });
2531
3450
  });
2532
3451
  return sessionId;
2533
3452
  }
@@ -2719,12 +3638,17 @@ var ChannelDriver = class _ChannelDriver {
2719
3638
  stuckReported: false,
2720
3639
  lastAliveAt: 0,
2721
3640
  aliveInFlight: false,
3641
+ titleSynced: false,
3642
+ titleSyncInFlight: false,
2722
3643
  awaitingHumanLatched: false,
2723
3644
  pausedOnQuestion: false,
2724
3645
  pausedOnPermission: false,
2725
3646
  pausedClearConfirmed: false,
2726
3647
  pausedInFlight: false,
2727
- deliveryDeadlineAnchored: false
3648
+ deliveryDeadlineAnchored: false,
3649
+ b2PinnedSinceMs: 0,
3650
+ b2LastDescendantCheckMs: 0,
3651
+ b2AbandonedSignalled: false
2728
3652
  });
2729
3653
  }
2730
3654
  /**
@@ -2792,12 +3716,17 @@ var ChannelDriver = class _ChannelDriver {
2792
3716
  // with no extra `re_adopted` signal needed (folds old WI-6).
2793
3717
  lastAliveAt: 0,
2794
3718
  aliveInFlight: false,
3719
+ titleSynced: false,
3720
+ titleSyncInFlight: false,
2795
3721
  awaitingHumanLatched: false,
2796
3722
  pausedOnQuestion: false,
2797
3723
  pausedOnPermission: false,
2798
3724
  pausedClearConfirmed: false,
2799
3725
  pausedInFlight: false,
2800
- deliveryDeadlineAnchored: false
3726
+ deliveryDeadlineAnchored: false,
3727
+ b2PinnedSinceMs: 0,
3728
+ b2LastDescendantCheckMs: 0,
3729
+ b2AbandonedSignalled: false
2801
3730
  });
2802
3731
  }
2803
3732
  /**
@@ -2959,58 +3888,7 @@ var ChannelDriver = class _ChannelDriver {
2959
3888
  }
2960
3889
  }
2961
3890
  if (state === "done") {
2962
- this.anchorDeliveryDeadline(inFlight);
2963
- if (!inFlight.done) {
2964
- this.log({
2965
- level: "info",
2966
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2967
- conversation_id: conv.id,
2968
- message_id: inFlight.evidentMessageId
2969
- });
2970
- const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2971
- const usage = messageUsage(messages, inFlight.opencodeMessageId);
2972
- try {
2973
- await this.markDone(
2974
- conv.id,
2975
- inFlight.evidentMessageId,
2976
- sessionId,
2977
- inFlight.opencodeMessageId,
2978
- title,
2979
- usage
2980
- );
2981
- } catch (err) {
2982
- if (err instanceof ChannelAuthError) throw err;
2983
- if (err instanceof ChannelTerminalError) {
2984
- this.log({
2985
- level: "warn",
2986
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2987
- conversation_id: conv.id,
2988
- message_id: inFlight.evidentMessageId
2989
- });
2990
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2991
- return;
2992
- }
2993
- if (this.now() >= inFlight.deadline) {
2994
- this.log({
2995
- level: "warn",
2996
- 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)}`,
2997
- conversation_id: conv.id,
2998
- message_id: inFlight.evidentMessageId
2999
- });
3000
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3001
- return;
3002
- }
3003
- this.log({
3004
- level: "warn",
3005
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
3006
- conversation_id: conv.id,
3007
- message_id: inFlight.evidentMessageId
3008
- });
3009
- return;
3010
- }
3011
- inFlight.done = true;
3012
- }
3013
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3891
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3014
3892
  return;
3015
3893
  }
3016
3894
  if (state === "failed") {
@@ -3024,8 +3902,16 @@ var ChannelDriver = class _ChannelDriver {
3024
3902
  message_id: inFlight.evidentMessageId
3025
3903
  });
3026
3904
  const usage = messageUsage(messages, inFlight.opencodeMessageId);
3905
+ const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
3027
3906
  try {
3028
- await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2, usage);
3907
+ await this.markFailed(
3908
+ conv.id,
3909
+ inFlight.evidentMessageId,
3910
+ sessionId,
3911
+ error2,
3912
+ usage,
3913
+ failure
3914
+ );
3029
3915
  } catch (err) {
3030
3916
  if (err instanceof ChannelAuthError) throw err;
3031
3917
  if (err instanceof ChannelTerminalError) {
@@ -3070,6 +3956,44 @@ var ChannelDriver = class _ChannelDriver {
3070
3956
  });
3071
3957
  }
3072
3958
  const activelyRunning = state === "running" && !awaitingHuman;
3959
+ const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
3960
+ const snapshotReadable = messages != null && messages.length > 0;
3961
+ if (!pinnedNow) {
3962
+ if (snapshotReadable) {
3963
+ inFlight.b2PinnedSinceMs = 0;
3964
+ inFlight.b2LastDescendantCheckMs = 0;
3965
+ inFlight.b2AbandonedSignalled = false;
3966
+ }
3967
+ } else {
3968
+ if (inFlight.b2AbandonedSignalled) {
3969
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3970
+ return;
3971
+ }
3972
+ if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
3973
+ const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
3974
+ if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
3975
+ inFlight.b2LastDescendantCheckMs = this.now();
3976
+ const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
3977
+ if (isB2AbandonmentConfirmed({
3978
+ pinnedForMs,
3979
+ minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
3980
+ descendantOngoing
3981
+ })) {
3982
+ inFlight.b2AbandonedSignalled = true;
3983
+ this.log({
3984
+ level: "warn",
3985
+ 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`,
3986
+ conversation_id: conv.id,
3987
+ message_id: id
3988
+ });
3989
+ void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
3990
+ watched_for_ms: pinnedForMs
3991
+ });
3992
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3993
+ return;
3994
+ }
3995
+ }
3996
+ }
3073
3997
  if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
3074
3998
  this.log({
3075
3999
  level: "warn",
@@ -3089,6 +4013,18 @@ var ChannelDriver = class _ChannelDriver {
3089
4013
  inFlight.aliveInFlight = false;
3090
4014
  if (ok) inFlight.lastAliveAt = this.now();
3091
4015
  });
4016
+ if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
4017
+ inFlight.titleSyncInFlight = true;
4018
+ void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
4019
+ if (!title) {
4020
+ inFlight.titleSyncInFlight = false;
4021
+ return;
4022
+ }
4023
+ const ok = await this.patchConversationTitle(conv.id, title);
4024
+ inFlight.titleSyncInFlight = false;
4025
+ if (ok) inFlight.titleSynced = true;
4026
+ });
4027
+ }
3092
4028
  }
3093
4029
  if (awaitingHuman) {
3094
4030
  if (!inFlight.awaitingHumanLatched) {
@@ -3102,29 +4038,93 @@ var ChannelDriver = class _ChannelDriver {
3102
4038
  if (ok && inFlight.awaitingHumanLatched) inFlight.pausedClearConfirmed = true;
3103
4039
  });
3104
4040
  }
3105
- } else if (inFlight.awaitingHumanLatched) {
3106
- inFlight.awaitingHumanLatched = false;
3107
- inFlight.pausedOnQuestion = false;
3108
- inFlight.pausedOnPermission = false;
3109
- inFlight.pausedClearConfirmed = false;
3110
- }
3111
- const siblingPaused = (sib) => openQuestions.has(sib.evidentMessageId) || openPermissions.has(sib.evidentMessageId) || sib.awaitingHumanLatched || sib.pausedOnQuestion || sib.pausedOnPermission;
3112
- const hasActivelyRunningSibling = [...watcher.inFlight.values()].some(
3113
- (sib) => sib.evidentMessageId !== inFlight.evidentMessageId && messageRunState(messages, sib.opencodeMessageId) === "running" && !siblingPaused(sib)
3114
- );
3115
- const queuedBehindRunningSibling = state === "queued" && hasActivelyRunningSibling;
3116
- if (!activelyRunning && !queuedBehindRunningSibling && this.now() >= inFlight.deadline) {
3117
- this.log({
3118
- level: "debug",
3119
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
3120
- conversation_id: conv.id,
3121
- message_id: inFlight.evidentMessageId
3122
- });
3123
- void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
3124
- watched_for_ms: this.now() - inFlight.dispatchedAt
3125
- });
3126
- this.removeInFlight(watcher, inFlight.evidentMessageId);
4041
+ } else if (inFlight.awaitingHumanLatched) {
4042
+ inFlight.awaitingHumanLatched = false;
4043
+ inFlight.pausedOnQuestion = false;
4044
+ inFlight.pausedOnPermission = false;
4045
+ inFlight.pausedClearConfirmed = false;
4046
+ }
4047
+ const siblingPaused = (sib) => openQuestions.has(sib.evidentMessageId) || openPermissions.has(sib.evidentMessageId) || sib.awaitingHumanLatched || sib.pausedOnQuestion || sib.pausedOnPermission;
4048
+ const hasActivelyRunningSibling = [...watcher.inFlight.values()].some(
4049
+ (sib) => sib.evidentMessageId !== inFlight.evidentMessageId && messageRunState(messages, sib.opencodeMessageId) === "running" && !siblingPaused(sib)
4050
+ );
4051
+ const queuedBehindRunningSibling = state === "queued" && hasActivelyRunningSibling;
4052
+ if (!activelyRunning && !queuedBehindRunningSibling && this.now() >= inFlight.deadline) {
4053
+ this.log({
4054
+ level: "debug",
4055
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
4056
+ conversation_id: conv.id,
4057
+ message_id: inFlight.evidentMessageId
4058
+ });
4059
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
4060
+ watched_for_ms: this.now() - inFlight.dispatchedAt
4061
+ });
4062
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4063
+ }
4064
+ }
4065
+ /**
4066
+ * Settle a message whose run-state has resolved `'done'` — extracted verbatim
4067
+ * (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
4068
+ * inline `state === 'done'` branch body, so a SECOND caller (the #721
4069
+ * b2-abandonment resolution) can reach the exact same completion behavior
4070
+ * (delivery-deadline anchoring, title resolution, usage extraction, and
4071
+ * `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
4072
+ * and risking the two copies silently drifting apart.
4073
+ */
4074
+ async settleMessageDone(sessionId, watcher, inFlight, messages) {
4075
+ const conv = watcher.conv;
4076
+ this.anchorDeliveryDeadline(inFlight);
4077
+ if (!inFlight.done) {
4078
+ this.log({
4079
+ level: "info",
4080
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
4081
+ conversation_id: conv.id,
4082
+ message_id: inFlight.evidentMessageId
4083
+ });
4084
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
4085
+ const usage = messageUsage(messages, inFlight.opencodeMessageId);
4086
+ try {
4087
+ await this.markDone(
4088
+ conv.id,
4089
+ inFlight.evidentMessageId,
4090
+ sessionId,
4091
+ inFlight.opencodeMessageId,
4092
+ title,
4093
+ usage
4094
+ );
4095
+ } catch (err) {
4096
+ if (err instanceof ChannelAuthError) throw err;
4097
+ if (err instanceof ChannelTerminalError) {
4098
+ this.log({
4099
+ level: "warn",
4100
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
4101
+ conversation_id: conv.id,
4102
+ message_id: inFlight.evidentMessageId
4103
+ });
4104
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4105
+ return;
4106
+ }
4107
+ if (this.now() >= inFlight.deadline) {
4108
+ this.log({
4109
+ level: "warn",
4110
+ 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)}`,
4111
+ conversation_id: conv.id,
4112
+ message_id: inFlight.evidentMessageId
4113
+ });
4114
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4115
+ return;
4116
+ }
4117
+ this.log({
4118
+ level: "warn",
4119
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
4120
+ conversation_id: conv.id,
4121
+ message_id: inFlight.evidentMessageId
4122
+ });
4123
+ return;
4124
+ }
4125
+ inFlight.done = true;
3127
4126
  }
4127
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3128
4128
  }
3129
4129
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
3130
4130
  /**
@@ -3292,6 +4292,7 @@ var ChannelDriver = class _ChannelDriver {
3292
4292
  if (state === "failed") {
3293
4293
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
3294
4294
  const usage = messageUsage(messages, ocId ?? "");
4295
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3295
4296
  this.log({
3296
4297
  level: "error",
3297
4298
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -3299,7 +4300,7 @@ var ChannelDriver = class _ChannelDriver {
3299
4300
  message_id: row.id
3300
4301
  });
3301
4302
  try {
3302
- await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
4303
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
3303
4304
  } catch (err) {
3304
4305
  if (err instanceof ChannelAuthError) throw err;
3305
4306
  if (err instanceof ChannelTerminalError) {
@@ -3705,6 +4706,47 @@ var ChannelDriver = class _ChannelDriver {
3705
4706
  }
3706
4707
  return false;
3707
4708
  }
4709
+ /**
4710
+ * Tri-state variant of the upward parentID membership walk (#721), used ONLY
4711
+ * by `isAnyDescendantSessionOngoing`. Walks the SAME cached
4712
+ * `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
4713
+ * `sessionBelongsTo`, which deliberately collapses "confirmed not a
4714
+ * descendant" and "the walk's fetch failed" into the same `false` (safe for
4715
+ * its OTHER callers: interaction attribution and the recovery-path
4716
+ * `isAnyDescendantSessionAlive`, both of which just retry next tick with no
4717
+ * safety consequence either way) — this variant keeps those two outcomes
4718
+ * SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
4719
+ * (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
4720
+ * not ongoing".
4721
+ *
4722
+ * Return contract:
4723
+ * - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
4724
+ * - `false` → the walk reached a definitive, parent-less root session
4725
+ * WITHOUT ever matching `rootSessionId` — `sessionId` is
4726
+ * CONFIRMED NOT a descendant of it.
4727
+ * - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
4728
+ * through the walk (`resolveSessionParent` returned `undefined`),
4729
+ * or the depth cap (32) was hit without a definitive answer (a
4730
+ * pathological/cyclic chain proves nothing either way). NEVER
4731
+ * treat this the same as `false` — see `sessionBelongsTo`'s own
4732
+ * doc comment above for why that collapse is safe THERE but not
4733
+ * here.
4734
+ *
4735
+ * `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
4736
+ * to the live-path descendant check, not a modification of shared code used
4737
+ * by interaction attribution or the recovery path.
4738
+ */
4739
+ async resolveSessionMembership(sessionId, rootSessionId) {
4740
+ let current = sessionId;
4741
+ for (let depth = 0; current && depth < 32; depth++) {
4742
+ if (current === rootSessionId) return true;
4743
+ const parent = await this.resolveSessionParent(current);
4744
+ if (parent === void 0) return null;
4745
+ if (parent === null) return false;
4746
+ current = parent;
4747
+ }
4748
+ return null;
4749
+ }
3708
4750
  /**
3709
4751
  * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3710
4752
  * `null` for a root session (no parent) and `undefined` when opencode is
@@ -3789,6 +4831,54 @@ var ChannelDriver = class _ChannelDriver {
3789
4831
  }
3790
4832
  return null;
3791
4833
  }
4834
+ /**
4835
+ * Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
4836
+ * session title onto the conversation via the PLAIN conversation-update
4837
+ * endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
4838
+ * message-status endpoint `markProcessing`/`markDone` use. Deliberately a
4839
+ * separate, lighter call: it carries no `status`, so it cannot re-trigger the
4840
+ * `processing`/`done` transition side effects (Slack notices, activity-log
4841
+ * rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
4842
+ * ever touches `conversations.title`. That route (`routes/conversations.ts`)
4843
+ * skips a title write matching the stored value, so a redundant call with the
4844
+ * same title is a real no-op — it does not bump `updated_at`, which the
4845
+ * conversation list sorts and paginates on. (Note this is a DIFFERENT guard
4846
+ * from `threads.ts`'s "non-empty AND changed" one, which only covers the
4847
+ * message-status PATCH; the non-empty half is enforced here instead, by
4848
+ * `resolveSessionTitle` never returning an empty/placeholder title.)
4849
+ *
4850
+ * Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
4851
+ * is logged and the title is simply retried on the next heartbeat tick (the
4852
+ * caller only latches `titleSynced` on `true`).
4853
+ */
4854
+ async patchConversationTitle(conversationId, title) {
4855
+ try {
4856
+ const res = await this.fetchImpl(
4857
+ `${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
4858
+ {
4859
+ method: "PATCH",
4860
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4861
+ body: JSON.stringify({ title })
4862
+ }
4863
+ );
4864
+ if (!res.ok) {
4865
+ this.log({
4866
+ level: "debug",
4867
+ message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
4868
+ conversation_id: conversationId
4869
+ });
4870
+ return false;
4871
+ }
4872
+ return true;
4873
+ } catch (err) {
4874
+ this.log({
4875
+ level: "debug",
4876
+ 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)}`,
4877
+ conversation_id: conversationId
4878
+ });
4879
+ return false;
4880
+ }
4881
+ }
3792
4882
  /**
3793
4883
  * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3794
4884
  * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
@@ -3847,6 +4937,84 @@ var ChannelDriver = class _ChannelDriver {
3847
4937
  }
3848
4938
  return false;
3849
4939
  }
4940
+ /**
4941
+ * LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
4942
+ * sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
4943
+ * in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
4944
+ *
4945
+ * Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
4946
+ * cross-check above): that method judges liveness from the child's OWN
4947
+ * TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
4948
+ * on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
4949
+ * path the local opencode server IS running, so its in-memory status map is
4950
+ * live and authoritative — and per ADR-0047 §4a ("the child has its own entry
4951
+ * [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
4952
+ * ENTIRE turn (including any tool call it is itself executing), not a
4953
+ * per-message transcript snapshot. This sidesteps the "child's own tool is
4954
+ * executing, between its step's completion and the next generation step"
4955
+ * transcript gap that a transcript-based check would need a second,
4956
+ * sustained-window bound to guard against — it is simply not derived from
4957
+ * message timestamps at all.
4958
+ *
4959
+ * Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
4960
+ * status, as the recovery path does per §4a)? Because on the LIVE path the
4961
+ * root session can be shared: a SECOND, unrelated user message can land on the
4962
+ * SAME session (issue #721's own root cause) and keep the root `busy` for a
4963
+ * reason that has nothing to do with THIS message's delegation. A `task`
4964
+ * descendant session is spawned for exactly one delegated turn and never
4965
+ * reused, so its OWN status-map entry is unambiguous evidence about that one
4966
+ * delegation — which the root's status is not.
4967
+ *
4968
+ * Why membership is checked via `resolveSessionMembership`, NOT
4969
+ * `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
4970
+ * `GET /session/:id` fetch failure into "not a descendant", which would
4971
+ * silently drop a genuinely-live candidate from consideration on the one
4972
+ * unlucky tick its membership-walk fetch hiccups (#721).
4973
+ * `resolveSessionMembership` keeps that failure mode as a distinct `null`
4974
+ * (indeterminate) so it is folded into THIS method's own `indeterminate` flag
4975
+ * instead.
4976
+ *
4977
+ * Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
4978
+ * - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
4979
+ * - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
4980
+ * confirmed either way (`resolveSessionMembership` never
4981
+ * returned `null`), and every CONFIRMED descendant's status read
4982
+ * succeeded and is not ongoing (includes "no descendant session
4983
+ * exists at all" — e.g. a plain, non-`task` tool call).
4984
+ * - `null` → INDETERMINATE: `listSessions` failed, OR at least one
4985
+ * candidate's MEMBERSHIP could not be confirmed
4986
+ * (`resolveSessionMembership` returned `null` — a fetch failure
4987
+ * or pathological chain partway through the parent walk), OR at
4988
+ * least one CONFIRMED descendant's `isSessionOngoing` read
4989
+ * failed — and no OTHER candidate was already confirmed `true`.
4990
+ * The caller MUST NOT treat `null` the same as `false` here
4991
+ * (unlike the recovery cross-check's contract) — see
4992
+ * `isB2AbandonmentConfirmed`.
4993
+ */
4994
+ async isAnyDescendantSessionOngoing(rootSessionId) {
4995
+ const sessions = await listSessions(this.port);
4996
+ if (!sessions) {
4997
+ this.log({
4998
+ level: "warn",
4999
+ message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
5000
+ });
5001
+ return null;
5002
+ }
5003
+ let indeterminate = false;
5004
+ for (const candidate of sessions) {
5005
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
5006
+ const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
5007
+ if (membership === null) {
5008
+ indeterminate = true;
5009
+ continue;
5010
+ }
5011
+ if (membership === false) continue;
5012
+ const ongoing = await isSessionOngoing(this.port, candidate.id);
5013
+ if (ongoing === true) return true;
5014
+ if (ongoing === null) indeterminate = true;
5015
+ }
5016
+ return indeterminate ? null : false;
5017
+ }
3850
5018
  /**
3851
5019
  * Cheap decision-telemetry label for a running row's LAST correlated reply
3852
5020
  * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
@@ -3975,6 +5143,32 @@ var ChannelDriver = class _ChannelDriver {
3975
5143
  }
3976
5144
  return messages;
3977
5145
  }
5146
+ /**
5147
+ * The `opencode_session_id` fragment of a status PATCH body — `{}` when this
5148
+ * conversation has ABANDONED that session (#553). The field is optional
5149
+ * server-side and an absent one leaves the persisted binding untouched, so
5150
+ * omitting it is how a routine status write stops resurrecting it.
5151
+ *
5152
+ * ONLY for writes whose sole cost is a lost deep link. The `processing` notice
5153
+ * degrades to no "View in Evident" link (the reaction swap still fires) and the
5154
+ * turn-failure notice is built from the PATCH's own `error` text with a link off
5155
+ * the persisted row — neither loses content the user came for. `markDone`
5156
+ * deliberately does NOT use this helper: the server fetches the reply text
5157
+ * THROUGH the session id it is given, so suppressing there would replace the
5158
+ * agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
5159
+ * `ensureSession` guard, not this suppression, is what makes the self-heal
5160
+ * stick.
5161
+ */
5162
+ sessionIdBody(sessionId, conversationId, messageId, status) {
5163
+ if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
5164
+ this.log({
5165
+ 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)}`,
5167
+ conversation_id: conversationId,
5168
+ message_id: messageId
5169
+ });
5170
+ return {};
5171
+ }
3978
5172
  /**
3979
5173
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
3980
5174
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -4004,7 +5198,7 @@ var ChannelDriver = class _ChannelDriver {
4004
5198
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4005
5199
  body: JSON.stringify({
4006
5200
  status: "processing",
4007
- opencode_session_id: sessionId,
5201
+ ...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
4008
5202
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
4009
5203
  ...title ? { title } : {}
4010
5204
  })
@@ -4053,6 +5247,11 @@ var ChannelDriver = class _ChannelDriver {
4053
5247
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4054
5248
  body: JSON.stringify({
4055
5249
  status: "done",
5250
+ // ALWAYS sent, even for a session this conversation has abandoned
5251
+ // (#553): the server reads the reply text back out of THIS session id
5252
+ // to deliver it. Omitting it would leave the user with "✅ Done!"
5253
+ // instead of the answer — a worse regression than the resurrection it
5254
+ // would prevent, which `ensureSession`'s guard handles anyway.
4056
5255
  opencode_session_id: sessionId,
4057
5256
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
4058
5257
  ...title ? { title } : {},
@@ -4079,11 +5278,21 @@ var ChannelDriver = class _ChannelDriver {
4079
5278
  * exists but is wedged, so the next attempt must get a fresh one
4080
5279
  * instead of reusing it — see WI-1's server-side null-clearing PATCH).
4081
5280
  */
4082
- async markFailed(conversationId, messageId, sessionId, error2, usage) {
5281
+ async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
4083
5282
  const body = { status: "failed" };
4084
- if (sessionId !== void 0) body.opencode_session_id = sessionId;
5283
+ if (sessionId === null) {
5284
+ body.opencode_session_id = null;
5285
+ } else if (sessionId !== void 0) {
5286
+ Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
5287
+ }
4085
5288
  if (error2 !== void 0) body.error = error2;
4086
5289
  if (usage) Object.assign(body, usage);
5290
+ if (failure) {
5291
+ body.failure_kind = failure.kind;
5292
+ body.failure_provider_id = failure.providerId;
5293
+ body.failure_model_id = failure.modelId;
5294
+ body.failure_reason = failure.reason;
5295
+ }
4087
5296
  await this.callWithRetry(
4088
5297
  "marking message as failed",
4089
5298
  () => this.fetchImpl(
@@ -4096,6 +5305,29 @@ var ChannelDriver = class _ChannelDriver {
4096
5305
  )
4097
5306
  );
4098
5307
  }
5308
+ /**
5309
+ * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
5310
+ *
5311
+ * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
5312
+ * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
5313
+ * to the P1-2b zero-provider check — one extra loopback call to
5314
+ * `hasAnyConfiguredProvider`, only reached when the structured classifier
5315
+ * couldn't place it. Fails open (never throws): a fallback probe failure
5316
+ * (`null`/indeterminate) leaves the classification `null`, which produces
5317
+ * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
5318
+ */
5319
+ async classifyModelAuthFailure(messages, userMessageId) {
5320
+ const classified = messageFailure(messages, userMessageId);
5321
+ if (classified != null) return classified;
5322
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
5323
+ const hasProvider = await hasAnyConfiguredProvider(this.port);
5324
+ return applyZeroProviderFallback(
5325
+ classified,
5326
+ hasProvider,
5327
+ reply?.info?.providerID ?? null,
5328
+ reply?.info?.modelID ?? null
5329
+ );
5330
+ }
4099
5331
  /**
4100
5332
  * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
4101
5333
  * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
@@ -4248,10 +5480,16 @@ var ChannelDriver = class _ChannelDriver {
4248
5480
  import chalk5 from "chalk";
4249
5481
  import ora2 from "ora";
4250
5482
  import { select as select2 } from "@inquirer/prompts";
5483
+ var INTERACTIVE_START_TIMEOUT_MS = 3e4;
4251
5484
  async function ensureOpenCodeRunning(ctx) {
4252
5485
  const healthCheck = await checkOpenCodeHealth(ctx.port);
4253
5486
  if (healthCheck.healthy) {
4254
- return { port: ctx.port, process: null, version: healthCheck.version ?? null };
5487
+ return {
5488
+ port: ctx.port,
5489
+ process: null,
5490
+ version: healthCheck.version ?? null,
5491
+ notReadyReason: null
5492
+ };
4255
5493
  }
4256
5494
  const runningInstances = await findHealthyOpenCodeInstances();
4257
5495
  if (runningInstances.length > 0) {
@@ -4272,7 +5510,7 @@ async function ensureOpenCodeRunning(ctx) {
4272
5510
  console.log(chalk5.yellow("Tip: Run with the correct port:"));
4273
5511
  console.log(
4274
5512
  chalk5.dim(
4275
- ` ${getCliName()} run --agent ${ctx.agentId} --port ${runningInstances[0].port}`
5513
+ ` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
4276
5514
  )
4277
5515
  );
4278
5516
  }
@@ -4292,14 +5530,22 @@ async function ensureOpenCodeRunning(ctx) {
4292
5530
  if (!ctx.interactive) {
4293
5531
  ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
4294
5532
  const proc = await startOpenCode(ctx.port);
4295
- const health = await waitForOpenCodeHealth(ctx.port, 3e4);
5533
+ const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
4296
5534
  if (!health.healthy) {
4297
- throw new Error(
4298
- `OpenCode failed to start on port ${ctx.port}. Install with: npm install -g opencode-ai`
4299
- );
5535
+ return {
5536
+ port: ctx.port,
5537
+ process: proc,
5538
+ version: null,
5539
+ notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
5540
+ };
4300
5541
  }
4301
5542
  ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
4302
- return { port: ctx.port, process: proc, version: health.version ?? null };
5543
+ return {
5544
+ port: ctx.port,
5545
+ process: proc,
5546
+ version: health.version ?? null,
5547
+ notReadyReason: null
5548
+ };
4303
5549
  }
4304
5550
  let port = ctx.port;
4305
5551
  if (isPortInUse(port)) {
@@ -4352,15 +5598,15 @@ Port ${port} is already in use.`));
4352
5598
  if (action === "start") {
4353
5599
  const spinner = ora2("Starting OpenCode...").start();
4354
5600
  const proc = await startOpenCode(port);
4355
- const health = await waitForOpenCodeHealth(port, 3e4);
5601
+ const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
4356
5602
  if (!health.healthy) {
4357
5603
  spinner.fail("Failed to start OpenCode");
4358
5604
  throw new Error("OpenCode failed to start");
4359
5605
  }
4360
5606
  spinner.stop();
4361
- return { port, process: proc, version: health.version ?? null };
5607
+ return { port, process: proc, version: health.version ?? null, notReadyReason: null };
4362
5608
  }
4363
- return { port, process: null, version: null };
5609
+ return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
4364
5610
  }
4365
5611
 
4366
5612
  // src/commands/agent-lookup.ts
@@ -4402,19 +5648,21 @@ async function resolveAgentIdFromKey(authHeader) {
4402
5648
  return { agent_id: data.agent_id };
4403
5649
  }
4404
5650
  return {
4405
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
5651
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
4406
5652
  };
4407
5653
  } catch (error2) {
4408
5654
  const message = error2 instanceof Error ? error2.message : "Unknown error";
4409
5655
  return { error: `Failed to resolve runner from key: ${message}` };
4410
5656
  }
4411
5657
  }
5658
+ var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
4412
5659
  async function notifyAgentDisconnected(agentId, authHeader) {
4413
5660
  const apiUrl = getApiUrlConfig();
4414
5661
  try {
4415
5662
  const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
4416
5663
  method: "POST",
4417
- headers: { Authorization: authHeader }
5664
+ headers: { Authorization: authHeader },
5665
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
4418
5666
  });
4419
5667
  if (!response.ok) {
4420
5668
  const serverMessage = await readErrorMessage(response);
@@ -4425,7 +5673,63 @@ async function notifyAgentDisconnected(agentId, authHeader) {
4425
5673
  }
4426
5674
  return { ok: true };
4427
5675
  } catch (error2) {
4428
- return { ok: false, error: error2 instanceof Error ? error2.message : String(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 ? `: ${serverMessage}` : ""}`
5728
+ };
5729
+ }
5730
+ return { ok: true };
5731
+ } catch (error2) {
5732
+ return { ok: false, error: describeBestEffortError(error2) };
4429
5733
  }
4430
5734
  }
4431
5735
  async function getAgentInfo(agentId, authHeader) {
@@ -4475,6 +5779,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
4475
5779
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
4476
5780
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4477
5781
  var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
5782
+ var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
4478
5783
  function resolveLogLevel(options) {
4479
5784
  const accepted = Object.keys(LOG_LEVELS);
4480
5785
  const validate = (value, source) => {
@@ -4498,6 +5803,63 @@ function resolveLogLevel(options) {
4498
5803
  }
4499
5804
  return "info";
4500
5805
  }
5806
+ function resolveFileSyncDirectories(raw, homeDir) {
5807
+ const directories = [];
5808
+ for (const entry of raw ?? []) {
5809
+ const trimmed = entry.trim();
5810
+ if (trimmed === "") {
5811
+ throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
5812
+ }
5813
+ const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
5814
+ if (!isAbsolute2(expanded)) {
5815
+ throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
5816
+ }
5817
+ const normalized = resolvePath(expanded);
5818
+ if (parse(normalized).root === normalized) {
5819
+ throw new Error(
5820
+ `--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
5821
+ );
5822
+ }
5823
+ if (!directories.includes(normalized)) {
5824
+ directories.push(normalized);
5825
+ }
5826
+ }
5827
+ if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
5828
+ throw new Error(
5829
+ `--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
5830
+ );
5831
+ }
5832
+ return directories;
5833
+ }
5834
+ var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
5835
+ var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
5836
+ var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
5837
+ function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
5838
+ const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
5839
+ let raw;
5840
+ let source;
5841
+ if (options.opencodeStartTimeout !== void 0) {
5842
+ raw = options.opencodeStartTimeout;
5843
+ source = "--opencode-start-timeout";
5844
+ } else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
5845
+ raw = env[OPENCODE_START_TIMEOUT_ENV];
5846
+ source = OPENCODE_START_TIMEOUT_ENV;
5847
+ } else {
5848
+ return { timeoutMs: defaultMs, warnings: [] };
5849
+ }
5850
+ const trimmed = raw.trim();
5851
+ const seconds = Number(trimmed);
5852
+ const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
5853
+ if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
5854
+ return {
5855
+ timeoutMs: defaultMs,
5856
+ warnings: [
5857
+ `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`
5858
+ ]
5859
+ };
5860
+ }
5861
+ return { timeoutMs: seconds * 1e3, warnings: [] };
5862
+ }
4501
5863
  function meetsThreshold(state, level) {
4502
5864
  return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
4503
5865
  }
@@ -4519,6 +5881,10 @@ function log2(state, message, level = "info") {
4519
5881
  function logActivity(state, entry) {
4520
5882
  const level = entry.level ?? (entry.type === "error" ? "error" : "info");
4521
5883
  if (!meetsThreshold(state, level)) return;
5884
+ forwardRunnerActivity(
5885
+ { level, message: entry.message, error: entry.error },
5886
+ { agentId: state.agentId, authHeader: state.authHeader }
5887
+ );
4522
5888
  const fullEntry = {
4523
5889
  ...entry,
4524
5890
  level,
@@ -4619,18 +5985,29 @@ async function handleAuthError(state, error2) {
4619
5985
  async function driveChannels(state, driver) {
4620
5986
  let idlePolls = 0;
4621
5987
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
5988
+ let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
4622
5989
  while (state.running) {
4623
5990
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
4624
5991
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
4625
5992
  if (state.interactive) displayStatus(state);
4626
5993
  await state.connection.reconnectPromise;
4627
5994
  }
5995
+ const carriedOverFileSync = driver.fileSyncActivity().inFlight;
5996
+ void driver.syncPendingFiles().catch(
5997
+ (error2) => logActivity(state, {
5998
+ type: "error",
5999
+ error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
6000
+ })
6001
+ );
4628
6002
  try {
4629
6003
  const processed = await driver.drainPending();
4630
6004
  state.messageCount += processed;
4631
6005
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4632
6006
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4633
- if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
6007
+ const appliedFiles = driver.fileSyncActivity().appliedFiles;
6008
+ const fileActivity = carriedOverFileSync || appliedFiles !== lastSeenAppliedFiles;
6009
+ lastSeenAppliedFiles = appliedFiles;
6010
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
4634
6011
  idlePolls = 0;
4635
6012
  if (processed > 0 && state.interactive) displayStatus(state);
4636
6013
  } else if (state.idleTimeout !== null) {
@@ -4659,7 +6036,7 @@ async function driveChannels(state, driver) {
4659
6036
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
4660
6037
  if (state.interactive) displayStatus(state);
4661
6038
  }
4662
- await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
6039
+ await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
4663
6040
  if (state.idleTimeout !== null && idlePolls >= 2) {
4664
6041
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
4665
6042
  if (idleMs > state.idleTimeout * 1e3) {
@@ -4671,8 +6048,8 @@ async function driveChannels(state, driver) {
4671
6048
  }
4672
6049
  }
4673
6050
  var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
4674
- async function runSweep(state, driver, config2) {
4675
- const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
6051
+ async function runSweep(state, driver, config) {
6052
+ const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
4676
6053
  try {
4677
6054
  const sessions = await listSessions(state.port);
4678
6055
  if (sessions === null) {
@@ -4685,8 +6062,8 @@ async function runSweep(state, driver, config2) {
4685
6062
  const toDelete = selectSessionsToDelete(
4686
6063
  sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
4687
6064
  {
4688
- maxAgeMs: config2.maxAgeMs,
4689
- maxCount: config2.maxCount,
6065
+ maxAgeMs: config.maxAgeMs,
6066
+ maxCount: config.maxCount,
4690
6067
  nowMs: Date.now(),
4691
6068
  protectedIds: driver.protectedSessionIds()
4692
6069
  }
@@ -4722,7 +6099,7 @@ async function runSweep(state, driver, config2) {
4722
6099
  }
4723
6100
  }
4724
6101
  function scheduleSessionCleanup(state, driver, options) {
4725
- const config2 = resolveSessionCleanupConfig(
6102
+ const config = resolveSessionCleanupConfig(
4726
6103
  {
4727
6104
  maxAge: options.sessionCleanupMaxAge,
4728
6105
  maxCount: options.sessionCleanupMaxCount,
@@ -4730,21 +6107,109 @@ function scheduleSessionCleanup(state, driver, options) {
4730
6107
  },
4731
6108
  process.env
4732
6109
  );
4733
- for (const warning2 of config2.warnings) {
6110
+ for (const warning2 of config.warnings) {
4734
6111
  logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
4735
6112
  }
4736
- if (!config2.enabled) return;
6113
+ if (!config.enabled) return;
4737
6114
  logActivity(state, {
4738
6115
  type: "info",
4739
- message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
6116
+ message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
4740
6117
  });
4741
- const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
6118
+ const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
4742
6119
  const firstSweep = setTimeout(
4743
- () => void runSweep(state, driver, config2),
6120
+ () => void runSweep(state, driver, config),
4744
6121
  SESSION_CLEANUP_FIRST_SWEEP_MS
4745
6122
  );
4746
6123
  state.sessionCleanupTimers.push(interval, firstSweep);
4747
6124
  }
6125
+ function scheduleClaudeUsageReporting(state, options) {
6126
+ const { mode, warnings } = resolveClaudeUsageReportingMode(
6127
+ options.claudeUsageReporting,
6128
+ process.env
6129
+ );
6130
+ for (const warning2 of warnings) {
6131
+ logActivity(state, {
6132
+ type: "info",
6133
+ level: "warn",
6134
+ message: `Claude usage reporting: ${warning2}`
6135
+ });
6136
+ }
6137
+ if (mode === "off") {
6138
+ logActivity(state, {
6139
+ type: "info",
6140
+ level: "debug",
6141
+ message: "Claude usage reporting is off (--claude-usage-reporting off)"
6142
+ });
6143
+ return;
6144
+ }
6145
+ let consecutiveFailures = 0;
6146
+ const scheduleNextTick = () => {
6147
+ state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
6148
+ };
6149
+ const tick = async (isFirst) => {
6150
+ try {
6151
+ const usage = await getClaudeUsage();
6152
+ const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
6153
+ if (result.ok) {
6154
+ if (consecutiveFailures > 0) {
6155
+ logActivity(state, {
6156
+ type: "info",
6157
+ level: "info",
6158
+ message: "Claude usage reporting recovered"
6159
+ });
6160
+ }
6161
+ consecutiveFailures = 0;
6162
+ logActivity(state, {
6163
+ type: "info",
6164
+ level: "debug",
6165
+ message: "Reported Claude usage to Evident"
6166
+ });
6167
+ } else {
6168
+ consecutiveFailures++;
6169
+ logActivity(state, {
6170
+ type: "info",
6171
+ level: consecutiveFailures === 1 ? "warn" : "debug",
6172
+ message: `Failed to report Claude usage: ${result.error}`
6173
+ });
6174
+ }
6175
+ scheduleNextTick();
6176
+ } catch (error2) {
6177
+ if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
6178
+ if (mode === "on") {
6179
+ logActivity(state, {
6180
+ type: "info",
6181
+ level: "warn",
6182
+ 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
+ });
6184
+ scheduleNextTick();
6185
+ } else if (isFirst) {
6186
+ logActivity(state, {
6187
+ type: "info",
6188
+ level: "debug",
6189
+ message: `Claude usage reporting: ${error2.message}`
6190
+ });
6191
+ } else {
6192
+ logActivity(state, {
6193
+ type: "info",
6194
+ level: "debug",
6195
+ message: `Claude usage reporting: ${error2.message}`
6196
+ });
6197
+ scheduleNextTick();
6198
+ }
6199
+ } else {
6200
+ consecutiveFailures++;
6201
+ const message = error2 instanceof Error ? error2.message : String(error2);
6202
+ logActivity(state, {
6203
+ type: "info",
6204
+ level: consecutiveFailures === 1 ? "warn" : "debug",
6205
+ message: `Claude usage reporting failed: ${message}`
6206
+ });
6207
+ scheduleNextTick();
6208
+ }
6209
+ }
6210
+ };
6211
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6212
+ }
4748
6213
  async function notifyOffline(state) {
4749
6214
  if (!state.agentId || !state.authHeader) return;
4750
6215
  if (!state.connected) {
@@ -4762,13 +6227,28 @@ async function notifyOffline(state) {
4762
6227
  if (state.interactive) displayStatus(state);
4763
6228
  }
4764
6229
  }
6230
+ async function timeShutdownPhase(state, durations, name, run2) {
6231
+ const startedAt = Date.now();
6232
+ try {
6233
+ return await run2();
6234
+ } finally {
6235
+ const elapsedMs = Date.now() - startedAt;
6236
+ durations[name] = elapsedMs;
6237
+ log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
6238
+ }
6239
+ }
4765
6240
  async function cleanup(state, opts = {}) {
6241
+ const durations = {};
4766
6242
  state.running = false;
4767
6243
  for (const timer of state.sessionCleanupTimers) {
4768
6244
  clearInterval(timer);
4769
6245
  clearTimeout(timer);
4770
6246
  }
4771
6247
  state.sessionCleanupTimers = [];
6248
+ if (state.claudeUsageTimer) {
6249
+ clearTimeout(state.claudeUsageTimer);
6250
+ state.claudeUsageTimer = null;
6251
+ }
4772
6252
  if (opts.graceful && state.channelDriver) {
4773
6253
  state.channelDriver.stop();
4774
6254
  log2(state, "Draining in-flight channel work before shutdown...");
@@ -4776,7 +6256,13 @@ async function cleanup(state, opts = {}) {
4776
6256
  logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4777
6257
  displayStatus(state);
4778
6258
  }
4779
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
6259
+ const driver = state.channelDriver;
6260
+ const settled = await timeShutdownPhase(
6261
+ state,
6262
+ durations,
6263
+ "drain",
6264
+ () => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
6265
+ );
4780
6266
  if (!settled) {
4781
6267
  logActivity(state, {
4782
6268
  type: "info",
@@ -4785,13 +6271,15 @@ async function cleanup(state, opts = {}) {
4785
6271
  if (state.interactive) displayStatus(state);
4786
6272
  }
4787
6273
  }
4788
- await notifyOffline(state);
6274
+ await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
4789
6275
  if (state.connection) {
4790
- state.connection.close();
6276
+ const connection = state.connection;
6277
+ await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
4791
6278
  state.connection = null;
4792
6279
  }
4793
6280
  if (state.opencodeProcess) {
4794
- stopOpenCode(state.opencodeProcess);
6281
+ const opencodeProcess = state.opencodeProcess;
6282
+ await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
4795
6283
  if (state.interactive) {
4796
6284
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
4797
6285
  displayStatus(state);
@@ -4800,12 +6288,15 @@ async function cleanup(state, opts = {}) {
4800
6288
  }
4801
6289
  state.opencodeProcess = null;
4802
6290
  }
6291
+ return durations;
4803
6292
  }
4804
6293
  async function run(options) {
4805
6294
  const interactive = isInteractive(options.json);
4806
6295
  let logLevel;
6296
+ let fileSyncDirectories;
4807
6297
  try {
4808
6298
  logLevel = resolveLogLevel(options);
6299
+ fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
4809
6300
  } catch (error2) {
4810
6301
  const message = error2 instanceof Error ? error2.message : String(error2);
4811
6302
  if (options.json) {
@@ -4838,8 +6329,15 @@ async function run(options) {
4838
6329
  messageCount: 0,
4839
6330
  lastProxiedActivityAt: null,
4840
6331
  sessionCleanupTimers: [],
6332
+ claudeUsageTimer: null,
4841
6333
  authHeader: ""
4842
6334
  };
6335
+ setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
6336
+ if (fileSyncDirectories.length > 0) {
6337
+ log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
6338
+ } else {
6339
+ log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
6340
+ }
4843
6341
  if (!options.runner && options.agent) {
4844
6342
  telemetry.info(
4845
6343
  EventTypes.DEPRECATED_AGENT_FLAG_USED,
@@ -4863,14 +6361,38 @@ async function run(options) {
4863
6361
  const handleSignal = async () => {
4864
6362
  if (state.shuttingDown) return;
4865
6363
  state.shuttingDown = true;
6364
+ const shutdownStartedAt = Date.now();
4866
6365
  if (state.interactive) {
4867
6366
  logActivity(state, { type: "info", message: "Shutting down..." });
4868
6367
  displayStatus(state);
4869
6368
  } else {
4870
6369
  log2(state, "Shutting down...");
4871
6370
  }
4872
- await cleanup(state, { graceful: true });
4873
- await shutdownTelemetry();
6371
+ const durations = await cleanup(state, { graceful: true });
6372
+ const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
6373
+ await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
6374
+ let timer;
6375
+ const flushed = shutdownTelemetry().then(
6376
+ () => true,
6377
+ (error2) => {
6378
+ log2(
6379
+ state,
6380
+ `Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
6381
+ "warn"
6382
+ );
6383
+ return true;
6384
+ }
6385
+ );
6386
+ const timedOut = new Promise((resolve3) => {
6387
+ timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
6388
+ });
6389
+ if (!await Promise.race([flushed, timedOut])) {
6390
+ log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
6391
+ }
6392
+ clearTimeout(timer);
6393
+ });
6394
+ const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
6395
+ log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
4874
6396
  process.exit(0);
4875
6397
  };
4876
6398
  process.on("SIGINT", handleSignal);
@@ -4984,40 +6506,67 @@ async function run(options) {
4984
6506
  }
4985
6507
  spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
4986
6508
  state.agentName = validation.agent.name;
6509
+ const microvmId = process.env.MICROVM_ID?.trim();
6510
+ if (microvmId) {
6511
+ const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
6512
+ if (reported.ok) {
6513
+ log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
6514
+ } else {
6515
+ const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
6516
+ log2(state, message, "warn");
6517
+ if (state.interactive && !state.json) {
6518
+ logActivity(state, { type: "info", level: "warn", message });
6519
+ }
6520
+ }
6521
+ } else {
6522
+ log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
6523
+ }
6524
+ const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
6525
+ for (const warning2 of opencodeStartTimeoutWarnings) {
6526
+ logActivity(state, { type: "info", level: "warn", message: warning2 });
6527
+ }
4987
6528
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
4988
6529
  try {
4989
6530
  const oc = await ensureOpenCodeRunning({
4990
6531
  port: state.port,
4991
6532
  interactive: state.interactive,
4992
6533
  agentId: state.agentId,
4993
- log: (message) => log2(state, message)
6534
+ log: (message) => log2(state, message),
6535
+ startTimeoutMs: opencodeStartTimeoutMs
4994
6536
  });
4995
6537
  state.port = oc.port;
4996
6538
  state.opencodeProcess = oc.process;
4997
6539
  state.opencodeVersion = oc.version;
4998
- state.opencodeConnected = oc.process !== null || oc.version !== null;
6540
+ state.opencodeConnected = oc.notReadyReason === null;
4999
6541
  const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
5000
6542
  ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
5001
- const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
5002
- if (versionWarning) {
5003
- log2(state, versionWarning, "warn");
5004
- if (state.interactive && !state.json) {
5005
- logActivity(state, { type: "info", level: "warn", message: versionWarning });
6543
+ if (!state.interactive && oc.notReadyReason !== null) {
6544
+ 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}).`;
6545
+ logActivity(state, { type: "info", level: "warn", message });
6546
+ } else {
6547
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
6548
+ if (versionWarning) {
6549
+ log2(state, versionWarning, "warn");
6550
+ if (state.interactive && !state.json) {
6551
+ logActivity(state, { type: "info", level: "warn", message: versionWarning });
6552
+ }
5006
6553
  }
5007
- }
5008
- const noProviderWarning = buildNoProviderWarning(await hasAnyConfiguredProvider(state.port));
5009
- if (noProviderWarning) {
5010
- log2(state, noProviderWarning, "warn");
5011
- if (state.interactive && !state.json) {
5012
- logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
5013
- blank();
5014
- console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
5015
- console.log(
5016
- chalk6.dim(
5017
- `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
5018
- )
5019
- );
5020
- blank();
6554
+ const noProviderWarning = buildNoProviderWarning(
6555
+ await hasAnyConfiguredProvider(state.port)
6556
+ );
6557
+ if (noProviderWarning) {
6558
+ log2(state, noProviderWarning, "warn");
6559
+ if (state.interactive && !state.json) {
6560
+ logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
6561
+ blank();
6562
+ console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
6563
+ console.log(
6564
+ chalk6.dim(
6565
+ `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
6566
+ )
6567
+ );
6568
+ blank();
6569
+ }
5021
6570
  }
5022
6571
  }
5023
6572
  } catch (error2) {
@@ -5032,6 +6581,10 @@ async function run(options) {
5032
6581
  getAuthHeader: () => state.authHeader,
5033
6582
  conversationFilter: state.conversationFilter,
5034
6583
  stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
6584
+ // #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
6585
+ // REJECTED with `file_sync_disabled` on the ack, not silently ignored.
6586
+ fileSyncDirectories,
6587
+ homeDir: homedir3(),
5035
6588
  log: (entry) => (
5036
6589
  // Thread the driver's real level straight through so `debug`/`warn`
5037
6590
  // survive the sink filter (they no longer collapse to info). `type`
@@ -5058,6 +6611,18 @@ async function run(options) {
5058
6611
  type: "info",
5059
6612
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
5060
6613
  });
6614
+ if (options.tunnelReadyFile) {
6615
+ const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
6616
+ if (marker.ok) {
6617
+ log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
6618
+ } else {
6619
+ log2(
6620
+ state,
6621
+ `Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
6622
+ "error"
6623
+ );
6624
+ }
6625
+ }
5061
6626
  emitAgentConnected(state.agentId, {
5062
6627
  port: state.port,
5063
6628
  cli_version: getCliVersion(),
@@ -5113,6 +6678,12 @@ async function run(options) {
5113
6678
  onDrainPing: () => {
5114
6679
  if (!state.running) return;
5115
6680
  logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
6681
+ void channelDriver.syncPendingFiles().catch(
6682
+ (error2) => logActivity(state, {
6683
+ type: "error",
6684
+ error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
6685
+ })
6686
+ );
5116
6687
  channelDriver.drainPending().then((processed) => {
5117
6688
  if (processed > 0) {
5118
6689
  state.messageCount += processed;
@@ -5142,6 +6713,7 @@ async function run(options) {
5142
6713
  throw error2;
5143
6714
  }
5144
6715
  scheduleSessionCleanup(state, channelDriver, options);
6716
+ scheduleClaudeUsageReporting(state, options);
5145
6717
  if (!interactive || state.json) {
5146
6718
  log2(state, "Driving channel messages...");
5147
6719
  }
@@ -5196,10 +6768,17 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
5196
6768
  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);
5197
6769
  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 }));
5198
6770
  program.command("whoami").description("Show the currently logged in user").action(whoami);
5199
- program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Runner ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("--runner [id]", "Alias for --agent (preferred name; wins if both are given)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
6771
+ program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
6772
+ 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
+ "-a, --agent [id]",
6774
+ "Deprecated alias for --runner (still supported; --runner wins if both are given)"
6775
+ ).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
5200
6776
  "--log-level <level>",
5201
6777
  "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
5202
- ).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(
6778
+ ).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(
6779
+ "--opencode-start-timeout <seconds>",
6780
+ "Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
6781
+ ).option("--json", "Output in JSON format").option(
5203
6782
  "--session-cleanup-max-age <duration>",
5204
6783
  "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
5205
6784
  ).option(
@@ -5208,6 +6787,17 @@ program.command("run").description("Connect to Evident and process messages").op
5208
6787
  ).option(
5209
6788
  "--session-cleanup-interval <duration>",
5210
6789
  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
6790
+ ).option(
6791
+ "--claude-usage-reporting <mode>",
6792
+ "Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
6793
+ ).option(
6794
+ "--enable-file-sync-to <dir>",
6795
+ "Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
6796
+ (value, previous) => previous.concat([value]),
6797
+ []
6798
+ ).option(
6799
+ "--tunnel-ready-file <path>",
6800
+ "Path to write once the tunnel is connected (set by the MicroVM hooks; unused on a developer machine)"
5211
6801
  ).action(
5212
6802
  (options) => {
5213
6803
  run({
@@ -5220,11 +6810,21 @@ program.command("run").description("Connect to Evident and process messages").op
5220
6810
  verbose: options.verbose,
5221
6811
  conversation: options.conversation,
5222
6812
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
6813
+ // Raw string — validation/precedence is single-sourced in run.ts's
6814
+ // resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
6815
+ opencodeStartTimeout: options.opencodeStartTimeout,
5223
6816
  json: options.json,
5224
6817
  // Raw strings — the resolver in run.ts single-sources parsing (M1).
5225
6818
  sessionCleanupMaxAge: options.sessionCleanupMaxAge,
5226
6819
  sessionCleanupMaxCount: options.sessionCleanupMaxCount,
5227
- sessionCleanupInterval: options.sessionCleanupInterval
6820
+ sessionCleanupInterval: options.sessionCleanupInterval,
6821
+ // Raw string — the resolver in run.ts single-sources parsing
6822
+ // (resolveClaudeUsageReportingMode).
6823
+ claudeUsageReporting: options.claudeUsageReporting,
6824
+ // Raw values — expansion/validation is single-sourced in run.ts's
6825
+ // resolveFileSyncDirectories.
6826
+ enableFileSyncTo: options.enableFileSyncTo,
6827
+ tunnelReadyFile: options.tunnelReadyFile
5228
6828
  });
5229
6829
  }
5230
6830
  );