@evident-ai/cli 3.1.1-dev.e69750c → 3.1.1-dev.eec1cca

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();
@@ -466,188 +532,6 @@ async function whoami() {
466
532
  blank();
467
533
  }
468
534
 
469
- // src/commands/run.ts
470
- import chalk6 from "chalk";
471
- import ora3 from "ora";
472
- import { select as select3 } from "@inquirer/prompts";
473
-
474
- // ../../packages/types/src/telemetry/index.ts
475
- var TelemetryEventTypes = {
476
- // Agent activity events (shown in web UI activity log)
477
- AGENT_CONNECTED: "agent.connected",
478
- AGENT_DISCONNECTED: "agent.disconnected",
479
- AGENT_MESSAGE_PROCESSING: "agent.message_processing",
480
- AGENT_MESSAGE_DONE: "agent.message_done",
481
- AGENT_MESSAGE_FAILED: "agent.message_failed"
482
- };
483
-
484
- // ../../packages/types/src/tunnel/index.ts
485
- var MAX_FRAME_BYTES = 256 * 1024;
486
- var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
487
-
488
- // ../../packages/types/src/logging/index.ts
489
- var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
- function log(level, event, fields) {
491
- const method = level === "debug" ? "log" : level;
492
- try {
493
- console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
494
- } catch (err) {
495
- console.error(
496
- "[evident] log_serialize_failed",
497
- event,
498
- err instanceof Error ? err.message : String(err)
499
- );
500
- }
501
- }
502
- function stripQuery(url) {
503
- try {
504
- return new URL(url).pathname;
505
- } catch {
506
- const q = url.indexOf("?");
507
- return q === -1 ? url : url.slice(0, q);
508
- }
509
- }
510
-
511
- // src/lib/telemetry.ts
512
- var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
513
- function getCliVersion() {
514
- return CLI_VERSION;
515
- }
516
- var eventBuffer = [];
517
- var flushTimeout = null;
518
- var isShuttingDown = false;
519
- var FLUSH_INTERVAL_MS = 5e3;
520
- var MAX_BUFFER_SIZE = 50;
521
- var FLUSH_TIMEOUT_MS = 3e3;
522
- function logEvent(eventType, options = {}) {
523
- const event = {
524
- event_type: eventType,
525
- severity: options.severity || "info",
526
- message: options.message,
527
- metadata: options.metadata,
528
- agent_id: options.agentId,
529
- timestamp: (/* @__PURE__ */ new Date()).toISOString()
530
- };
531
- eventBuffer.push(event);
532
- if (options.severity === "error" || eventBuffer.length >= MAX_BUFFER_SIZE) {
533
- void flushEvents();
534
- } else if (!flushTimeout && !isShuttingDown) {
535
- flushTimeout = setTimeout(() => {
536
- flushTimeout = null;
537
- void flushEvents();
538
- }, FLUSH_INTERVAL_MS);
539
- }
540
- }
541
- var telemetry = {
542
- debug: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "debug", message, metadata, agentId }),
543
- info: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "info", message, metadata, agentId }),
544
- warn: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "warning", message, metadata, agentId }),
545
- error: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "error", message, metadata, agentId })
546
- };
547
- async function flushEvents() {
548
- if (eventBuffer.length === 0) return;
549
- const events = eventBuffer;
550
- eventBuffer = [];
551
- if (flushTimeout) {
552
- clearTimeout(flushTimeout);
553
- flushTimeout = null;
554
- }
555
- try {
556
- const credentials2 = await getToken();
557
- if (!credentials2) {
558
- return;
559
- }
560
- const apiUrl = getApiUrlConfig();
561
- const controller = new AbortController();
562
- const timeout = setTimeout(() => controller.abort(), FLUSH_TIMEOUT_MS);
563
- try {
564
- const request = {
565
- events,
566
- client_type: "cli",
567
- client_version: CLI_VERSION
568
- };
569
- const response = await fetch(`${apiUrl}/telemetry/events`, {
570
- method: "POST",
571
- headers: {
572
- "Content-Type": "application/json",
573
- Authorization: `Bearer ${credentials2.token}`
574
- },
575
- body: JSON.stringify(request),
576
- signal: controller.signal
577
- });
578
- if (!response.ok) {
579
- console.error(`Telemetry flush failed: ${response.status}`);
580
- }
581
- } finally {
582
- clearTimeout(timeout);
583
- }
584
- } catch (error2) {
585
- if (process.env.DEBUG) {
586
- console.error("Telemetry error:", error2);
587
- }
588
- }
589
- }
590
- async function shutdownTelemetry() {
591
- isShuttingDown = true;
592
- if (flushTimeout) {
593
- clearTimeout(flushTimeout);
594
- flushTimeout = null;
595
- }
596
- await flushEvents();
597
- }
598
- function emitEvent(event) {
599
- logEvent(event.event_type, {
600
- severity: event.severity,
601
- message: event.message,
602
- metadata: event.metadata,
603
- agentId: event.agent_id
604
- });
605
- }
606
- function emitAgentConnected(agentId, metadata) {
607
- emitEvent({
608
- event_type: TelemetryEventTypes.AGENT_CONNECTED,
609
- severity: "info",
610
- message: "Agent CLI connected",
611
- metadata,
612
- agent_id: agentId
613
- });
614
- }
615
- function emitAgentDisconnected(agentId, metadata) {
616
- emitEvent({
617
- event_type: TelemetryEventTypes.AGENT_DISCONNECTED,
618
- severity: "info",
619
- message: `Agent CLI disconnected (code: ${metadata.code})`,
620
- metadata,
621
- agent_id: agentId
622
- });
623
- }
624
- var EventTypes = {
625
- // Tunnel lifecycle
626
- TUNNEL_STARTING: "tunnel.starting",
627
- TUNNEL_CONNECTED: "tunnel.connected",
628
- TUNNEL_DISCONNECTED: "tunnel.disconnected",
629
- TUNNEL_RECONNECTING: "tunnel.reconnecting",
630
- TUNNEL_ERROR: "tunnel.error",
631
- // OpenCode communication
632
- OPENCODE_HEALTH_CHECK: "opencode.health_check",
633
- OPENCODE_HEALTH_OK: "opencode.health_ok",
634
- OPENCODE_HEALTH_FAILED: "opencode.health_failed",
635
- OPENCODE_REQUEST_RECEIVED: "opencode.request_received",
636
- OPENCODE_REQUEST_FORWARDED: "opencode.request_forwarded",
637
- OPENCODE_RESPONSE_SENT: "opencode.response_sent",
638
- OPENCODE_UNREACHABLE: "opencode.unreachable",
639
- OPENCODE_ERROR: "opencode.error",
640
- // Authentication
641
- AUTH_LOGIN_STARTED: "auth.login_started",
642
- AUTH_LOGIN_SUCCESS: "auth.login_success",
643
- AUTH_LOGIN_FAILED: "auth.login_failed",
644
- AUTH_LOGOUT: "auth.logout",
645
- // CLI lifecycle
646
- CLI_STARTED: "cli.started",
647
- CLI_COMMAND: "cli.command",
648
- CLI_ERROR: "cli.error"
649
- };
650
-
651
535
  // src/lib/auth.ts
652
536
  async function getAuthCredentials() {
653
537
  const runnerKey = process.env.EVIDENT_RUNNER_KEY;
@@ -656,11 +540,12 @@ async function getAuthCredentials() {
656
540
  return {
657
541
  token: runnerKey,
658
542
  authType: "agent_key",
543
+ keySource: "runner_key",
659
544
  notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
660
545
  };
661
546
  }
662
547
  if (agentKey) {
663
- return { token: agentKey, authType: "agent_key" };
548
+ return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
664
549
  }
665
550
  const userToken = process.env.EVIDENT_TOKEN;
666
551
  if (userToken) {
@@ -690,18 +575,735 @@ function isInteractive(jsonOutput) {
690
575
  return true;
691
576
  }
692
577
 
693
- // src/lib/opencode/health.ts
694
- async function checkOpenCodeHealth(port) {
578
+ // src/commands/agent-lookup.ts
579
+ async function readErrorMessage(response) {
580
+ const text = await response.text().catch(() => "");
581
+ if (!text) return response.statusText || void 0;
695
582
  try {
696
- const response = await fetch(`http://127.0.0.1:${port}/global/health`, {
697
- signal: AbortSignal.timeout(2e3)
698
- // 2 second timeout
583
+ const data = JSON.parse(text);
584
+ const message = data.message ?? data.error;
585
+ if (typeof message === "string" && message.trim()) {
586
+ return message;
587
+ }
588
+ } catch {
589
+ }
590
+ return text.trim() || response.statusText || void 0;
591
+ }
592
+ function authFailureHint(apiUrl, serverMessage) {
593
+ const reason = serverMessage ? `: ${serverMessage}` : "";
594
+ return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
595
+ }
596
+ async function resolveAgentIdFromKey(authHeader) {
597
+ const apiUrl = getApiUrlConfig();
598
+ try {
599
+ const response = await fetch(`${apiUrl}/me`, {
600
+ headers: { Authorization: authHeader }
699
601
  });
602
+ if (response.status === 401) {
603
+ const serverMessage = await readErrorMessage(response);
604
+ return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
605
+ }
700
606
  if (!response.ok) {
701
- return { healthy: false, error: `HTTP ${response.status}` };
607
+ const serverMessage = await readErrorMessage(response);
608
+ return {
609
+ error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
610
+ };
702
611
  }
703
- const data = await response.json().catch(() => ({}));
704
- return { healthy: true, version: data.version };
612
+ const data = await response.json();
613
+ if (data.auth_type === "agent_key" && data.agent_id) {
614
+ return { agent_id: data.agent_id };
615
+ }
616
+ return {
617
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
618
+ };
619
+ } catch (error2) {
620
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
621
+ return { error: `Failed to resolve runner from key: ${message}` };
622
+ }
623
+ }
624
+ var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
625
+ async function notifyAgentDisconnected(agentId, authHeader) {
626
+ const apiUrl = getApiUrlConfig();
627
+ try {
628
+ const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
629
+ method: "POST",
630
+ headers: { Authorization: authHeader },
631
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
632
+ });
633
+ if (!response.ok) {
634
+ const serverMessage = await readErrorMessage(response);
635
+ return {
636
+ ok: false,
637
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
638
+ };
639
+ }
640
+ return { ok: true };
641
+ } catch (error2) {
642
+ return { ok: false, error: describeBestEffortError(error2) };
643
+ }
644
+ }
645
+ function describeBestEffortError(error2) {
646
+ const name = error2?.name;
647
+ if (name === "TimeoutError" || name === "AbortError") {
648
+ return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
649
+ }
650
+ return error2 instanceof Error ? error2.message : String(error2);
651
+ }
652
+ async function reportMicrovmId(agentId, authHeader, microvmId) {
653
+ try {
654
+ const apiUrl = getApiUrlConfig();
655
+ const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
656
+ method: "POST",
657
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
658
+ body: JSON.stringify({ microvm_id: microvmId }),
659
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
660
+ });
661
+ if (!response.ok) {
662
+ const serverMessage = await readErrorMessage(response);
663
+ return {
664
+ ok: false,
665
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
666
+ };
667
+ }
668
+ return { ok: true };
669
+ } catch (error2) {
670
+ return { ok: false, error: describeBestEffortError(error2) };
671
+ }
672
+ }
673
+ function toReportedWindow(window) {
674
+ if (!window) return null;
675
+ return { utilization: window.utilization, resets_at: window.resetsAt };
676
+ }
677
+ async function reportClaudeUsage(agentId, authHeader, snapshot) {
678
+ try {
679
+ const apiUrl = getApiUrlConfig();
680
+ const response = await fetch(`${apiUrl}/runners/${agentId}/claude-usage`, {
681
+ method: "POST",
682
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
683
+ body: JSON.stringify({
684
+ five_hour: toReportedWindow(snapshot.fiveHour),
685
+ seven_day: toReportedWindow(snapshot.sevenDay)
686
+ }),
687
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
688
+ });
689
+ if (!response.ok) {
690
+ const serverMessage = await readErrorMessage(response);
691
+ return {
692
+ ok: false,
693
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
694
+ };
695
+ }
696
+ return { ok: true };
697
+ } catch (error2) {
698
+ return { ok: false, error: describeBestEffortError(error2) };
699
+ }
700
+ }
701
+ async function getAgentInfo(agentId, authHeader) {
702
+ const apiUrl = getApiUrlConfig();
703
+ try {
704
+ const response = await fetch(`${apiUrl}/runners/${agentId}`, {
705
+ headers: { Authorization: authHeader }
706
+ });
707
+ if (response.status === 401) {
708
+ const serverMessage = await readErrorMessage(response);
709
+ return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
710
+ }
711
+ if (response.status === 403) {
712
+ const serverMessage = await readErrorMessage(response);
713
+ return {
714
+ valid: false,
715
+ error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
716
+ };
717
+ }
718
+ if (response.status === 404) {
719
+ const serverMessage = await readErrorMessage(response);
720
+ return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
721
+ }
722
+ if (!response.ok) {
723
+ const serverMessage = await readErrorMessage(response);
724
+ return {
725
+ valid: false,
726
+ error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
727
+ };
728
+ }
729
+ const agent = await response.json();
730
+ if (agent.agent_type !== "local") {
731
+ return {
732
+ valid: false,
733
+ error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
734
+ };
735
+ }
736
+ return { valid: true, agent };
737
+ } catch (error2) {
738
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
739
+ return { valid: false, error: `Failed to validate runner: ${message}` };
740
+ }
741
+ }
742
+
743
+ // src/commands/status.ts
744
+ var STATUS_TIMEOUT_MS = 1e4;
745
+ function authLabelFor(credentials2) {
746
+ if (credentials2.authType === "agent_key") {
747
+ return credentials2.keySource === "agent_key" ? "runner key (EVIDENT_AGENT_KEY)" : "runner key (EVIDENT_RUNNER_KEY)";
748
+ }
749
+ return "user token";
750
+ }
751
+ function describeFetchError(error2) {
752
+ const name = error2?.name;
753
+ if (name === "TimeoutError" || name === "AbortError") {
754
+ return `timed out after ${STATUS_TIMEOUT_MS}ms waiting for a response`;
755
+ }
756
+ return error2 instanceof Error ? error2.message : String(error2);
757
+ }
758
+ async function checkStatus(jsonMode) {
759
+ const apiUrl = getApiUrlConfig();
760
+ const credentials2 = await getAuthCredentials();
761
+ if (!credentials2) {
762
+ return {
763
+ ok: false,
764
+ endpoint: apiUrl,
765
+ reason: "no_credentials",
766
+ error: "No credentials configured. Set EVIDENT_RUNNER_KEY (or EVIDENT_AGENT_KEY), or EVIDENT_TOKEN, or run `evident login`.",
767
+ exitCode: 1
768
+ };
769
+ }
770
+ if (credentials2.notice && !jsonMode) {
771
+ printWarning(credentials2.notice);
772
+ }
773
+ let response;
774
+ try {
775
+ response = await fetch(`${apiUrl}/me`, {
776
+ headers: { Authorization: getAuthHeader(credentials2) },
777
+ signal: AbortSignal.timeout(STATUS_TIMEOUT_MS)
778
+ });
779
+ } catch (error2) {
780
+ return {
781
+ ok: false,
782
+ endpoint: apiUrl,
783
+ authLabel: authLabelFor(credentials2),
784
+ reason: "unreachable",
785
+ error: `Could not reach ${apiUrl}: ${describeFetchError(error2)}. The credentials were NOT validated.`,
786
+ exitCode: 75
787
+ };
788
+ }
789
+ if (response.status === 401) {
790
+ const serverMessage = await readErrorMessage(response);
791
+ return {
792
+ ok: false,
793
+ endpoint: apiUrl,
794
+ authLabel: authLabelFor(credentials2),
795
+ reason: "unauthorized",
796
+ error: authFailureHint(apiUrl, serverMessage),
797
+ exitCode: 1
798
+ };
799
+ }
800
+ if (response.status === 404) {
801
+ return {
802
+ ok: false,
803
+ endpoint: apiUrl,
804
+ authLabel: authLabelFor(credentials2),
805
+ reason: "endpoint_not_found",
806
+ error: `${apiUrl}/me returned HTTP 404 \u2014 that endpoint has no /me route, so it is probably missing the /v1 prefix. The credentials were NOT validated.`,
807
+ exitCode: 75
808
+ };
809
+ }
810
+ if (response.status >= 500) {
811
+ const serverMessage = await readErrorMessage(response);
812
+ return {
813
+ ok: false,
814
+ endpoint: apiUrl,
815
+ authLabel: authLabelFor(credentials2),
816
+ reason: "unreachable",
817
+ error: `${apiUrl} returned HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}. The credentials were NOT validated.`,
818
+ exitCode: 75
819
+ };
820
+ }
821
+ if (!response.ok) {
822
+ const serverMessage = await readErrorMessage(response);
823
+ return {
824
+ ok: false,
825
+ endpoint: apiUrl,
826
+ authLabel: authLabelFor(credentials2),
827
+ reason: "http_error",
828
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`,
829
+ exitCode: 1
830
+ };
831
+ }
832
+ const data = await response.json();
833
+ return {
834
+ ok: true,
835
+ endpoint: apiUrl,
836
+ authType: data.auth_type,
837
+ authLabel: authLabelFor(credentials2),
838
+ runnerId: data.auth_type === "agent_key" ? data.agent_id : void 0,
839
+ reason: "ok",
840
+ exitCode: 0
841
+ };
842
+ }
843
+ function printJson(result) {
844
+ const payload = {
845
+ ok: result.ok,
846
+ endpoint: result.endpoint
847
+ };
848
+ if (result.authType) payload.auth_type = result.authType;
849
+ if (result.runnerId) payload.runner_id = result.runnerId;
850
+ if (result.reason) payload.reason = result.reason;
851
+ if (result.error) payload.error = result.error;
852
+ console.log(JSON.stringify(payload));
853
+ }
854
+ function printHuman(result) {
855
+ blank();
856
+ console.log(keyValue("Endpoint", result.endpoint));
857
+ if (result.ok) {
858
+ console.log(keyValue("Auth", result.authLabel ?? "\u2014"));
859
+ if (result.runnerId) {
860
+ console.log(keyValue("Runner", result.runnerId));
861
+ }
862
+ console.log(keyValue("Status", "OK \u2014 credentials accepted"));
863
+ blank();
864
+ return;
865
+ }
866
+ if (result.authLabel) {
867
+ console.log(keyValue("Auth", result.authLabel));
868
+ }
869
+ blank();
870
+ printError(result.error ?? "Unknown error");
871
+ }
872
+ async function status(options = {}) {
873
+ const result = await checkStatus(Boolean(options.json));
874
+ if (options.json) {
875
+ printJson(result);
876
+ } else {
877
+ printHuman(result);
878
+ }
879
+ process.exit(result.exitCode);
880
+ }
881
+
882
+ // src/lib/claude-usage.ts
883
+ import { execFileSync } from "child_process";
884
+ import { readFileSync } from "fs";
885
+ import { homedir } from "os";
886
+ import { join } from "path";
887
+ var CLAUDE_USAGE_URL = "https://api.anthropic.com/api/oauth/usage";
888
+ var KEYCHAIN_SERVICE = "Claude Code-credentials";
889
+ function parseClaudeCliCredentials(raw) {
890
+ let parsed;
891
+ try {
892
+ parsed = JSON.parse(raw);
893
+ } catch {
894
+ return null;
895
+ }
896
+ const data = parsed.claudeAiOauth ?? parsed;
897
+ const creds = data;
898
+ if (typeof creds.accessToken !== "string" || typeof creds.expiresAt !== "number") {
899
+ return null;
900
+ }
901
+ return { accessToken: creds.accessToken, expiresAt: creds.expiresAt };
902
+ }
903
+ function readClaudeCliCredentials() {
904
+ if (process.platform === "darwin") {
905
+ try {
906
+ const raw = execFileSync(
907
+ "/usr/bin/security",
908
+ ["find-generic-password", "-s", KEYCHAIN_SERVICE, "-w"],
909
+ { encoding: "utf-8", timeout: 2e3, stdio: ["pipe", "pipe", "ignore"] }
910
+ );
911
+ return parseClaudeCliCredentials(raw);
912
+ } catch {
913
+ return null;
914
+ }
915
+ }
916
+ try {
917
+ const raw = readFileSync(join(homedir(), ".claude", ".credentials.json"), "utf-8");
918
+ return parseClaudeCliCredentials(raw);
919
+ } catch {
920
+ return null;
921
+ }
922
+ }
923
+ var ClaudeUsageError = class extends Error {
924
+ constructor(message, reason) {
925
+ super(message);
926
+ this.reason = reason;
927
+ }
928
+ };
929
+ function isLocalCredentialProblem(err) {
930
+ return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
931
+ }
932
+ function normalizeResetsAt(value) {
933
+ const ms = Date.parse(value);
934
+ return Number.isNaN(ms) ? null : new Date(ms).toISOString();
935
+ }
936
+ function toWindow(value) {
937
+ if (!value || typeof value !== "object") {
938
+ return null;
939
+ }
940
+ const window = value;
941
+ if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
942
+ return null;
943
+ }
944
+ const resetsAt = normalizeResetsAt(window.resets_at);
945
+ if (resetsAt === null) {
946
+ return null;
947
+ }
948
+ return { utilization: window.utilization, resetsAt };
949
+ }
950
+ async function getClaudeUsage() {
951
+ const credentials2 = readClaudeCliCredentials();
952
+ if (!credentials2) {
953
+ throw new ClaudeUsageError(
954
+ "No local Claude Code login found. Run `claude` once to sign in with your Claude subscription.",
955
+ "no_credentials"
956
+ );
957
+ }
958
+ if (credentials2.expiresAt < Date.now()) {
959
+ throw new ClaudeUsageError(
960
+ "Claude Code credentials have expired. Run `claude` to refresh them.",
961
+ "credentials_expired"
962
+ );
963
+ }
964
+ const res = await fetch(CLAUDE_USAGE_URL, {
965
+ headers: {
966
+ Authorization: `Bearer ${credentials2.accessToken}`,
967
+ "Content-Type": "application/json",
968
+ "anthropic-version": "2023-06-01"
969
+ }
970
+ });
971
+ if (!res.ok) {
972
+ throw new ClaudeUsageError(`Claude usage request failed: HTTP ${res.status}`, "request_failed");
973
+ }
974
+ const body = await res.json();
975
+ return {
976
+ fiveHour: toWindow(body.five_hour),
977
+ sevenDay: toWindow(body.seven_day)
978
+ };
979
+ }
980
+
981
+ // src/commands/claude-usage.ts
982
+ function formatWindow(label, window) {
983
+ if (!window) {
984
+ return keyValue(label, "not available for this plan");
985
+ }
986
+ const resetsAt = new Date(window.resetsAt);
987
+ return keyValue(label, `${window.utilization}% used, resets ${resetsAt.toLocaleString()}`);
988
+ }
989
+ async function claudeUsage() {
990
+ try {
991
+ const usage = await getClaudeUsage();
992
+ blank();
993
+ console.log(formatWindow("5-hour session", usage.fiveHour));
994
+ console.log(formatWindow("7-day", usage.sevenDay));
995
+ blank();
996
+ } catch (err) {
997
+ if (err instanceof ClaudeUsageError) {
998
+ printError(err.message);
999
+ process.exit(1);
1000
+ }
1001
+ throw err;
1002
+ }
1003
+ }
1004
+
1005
+ // src/commands/run.ts
1006
+ import { homedir as homedir3 } from "os";
1007
+ import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
1008
+ import chalk6 from "chalk";
1009
+
1010
+ // ../../packages/types/src/agents/index.ts
1011
+ var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
1012
+
1013
+ // ../../packages/types/src/telemetry/index.ts
1014
+ var TelemetryEventTypes = {
1015
+ // Agent activity events (shown in web UI activity log)
1016
+ AGENT_CONNECTED: "agent.connected",
1017
+ AGENT_DISCONNECTED: "agent.disconnected",
1018
+ AGENT_MESSAGE_PROCESSING: "agent.message_processing",
1019
+ AGENT_MESSAGE_DONE: "agent.message_done",
1020
+ AGENT_MESSAGE_FAILED: "agent.message_failed",
1021
+ // A `warn`/`error` runner-side log line forwarded server-side for
1022
+ // observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
1023
+ RUNNER_ACTIVITY: "runner.activity"
1024
+ };
1025
+
1026
+ // ../../packages/types/src/tunnel/index.ts
1027
+ var MAX_FRAME_BYTES = 256 * 1024;
1028
+ var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
1029
+
1030
+ // ../../packages/types/src/runner-files.ts
1031
+ var MAX_FILE_PUSH_BYTES = 64 * 1024;
1032
+ var MAX_FILE_SYNC_DIRECTORIES = 16;
1033
+
1034
+ // ../../packages/types/src/logging/index.ts
1035
+ var CORRELATION_ID_HEADER = "x-evident-correlation-id";
1036
+ function log(level, event, fields) {
1037
+ const method = level === "debug" ? "log" : level;
1038
+ try {
1039
+ console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
1040
+ } catch (err) {
1041
+ console.error(
1042
+ "[evident] log_serialize_failed",
1043
+ event,
1044
+ err instanceof Error ? err.message : String(err)
1045
+ );
1046
+ }
1047
+ }
1048
+ function errorFields(err) {
1049
+ if (err instanceof Error) {
1050
+ return { error: err.message, error_name: err.name };
1051
+ }
1052
+ return { error: String(err) };
1053
+ }
1054
+ function stripQuery(url) {
1055
+ try {
1056
+ return new URL(url).pathname;
1057
+ } catch {
1058
+ const q = url.indexOf("?");
1059
+ return q === -1 ? url : url.slice(0, q);
1060
+ }
1061
+ }
1062
+
1063
+ // src/commands/run.ts
1064
+ import ora3 from "ora";
1065
+ import { select as select3 } from "@inquirer/prompts";
1066
+
1067
+ // src/lib/telemetry.ts
1068
+ var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
1069
+ function getCliVersion() {
1070
+ return CLI_VERSION;
1071
+ }
1072
+ var eventBuffer = [];
1073
+ var flushTimeout = null;
1074
+ var isShuttingDown = false;
1075
+ var FLUSH_INTERVAL_MS = 5e3;
1076
+ var MAX_BUFFER_SIZE = 50;
1077
+ var FLUSH_TIMEOUT_MS = 3e3;
1078
+ var authProvider = null;
1079
+ function setTelemetryAuthProvider(provider) {
1080
+ authProvider = provider;
1081
+ }
1082
+ var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
1083
+ var lastFlushFailureLoggedAt = 0;
1084
+ var suppressedFlushFailureCount = 0;
1085
+ function logEvent(eventType, options = {}) {
1086
+ const event = {
1087
+ event_type: eventType,
1088
+ severity: options.severity || "info",
1089
+ message: options.message,
1090
+ metadata: options.metadata,
1091
+ agent_id: options.agentId,
1092
+ timestamp: (/* @__PURE__ */ new Date()).toISOString()
1093
+ };
1094
+ eventBuffer.push(event);
1095
+ if (options.severity === "error" || eventBuffer.length >= MAX_BUFFER_SIZE) {
1096
+ void flushEvents();
1097
+ } else if (!flushTimeout && !isShuttingDown) {
1098
+ flushTimeout = setTimeout(() => {
1099
+ flushTimeout = null;
1100
+ void flushEvents();
1101
+ }, FLUSH_INTERVAL_MS);
1102
+ }
1103
+ }
1104
+ var telemetry = {
1105
+ debug: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "debug", message, metadata, agentId }),
1106
+ info: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "info", message, metadata, agentId }),
1107
+ warn: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "warning", message, metadata, agentId }),
1108
+ error: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "error", message, metadata, agentId })
1109
+ };
1110
+ async function flushEvents() {
1111
+ if (eventBuffer.length === 0) return;
1112
+ const events = eventBuffer;
1113
+ eventBuffer = [];
1114
+ if (flushTimeout) {
1115
+ clearTimeout(flushTimeout);
1116
+ flushTimeout = null;
1117
+ }
1118
+ try {
1119
+ const providerContext = authProvider?.();
1120
+ let authHeader;
1121
+ if (providerContext?.authHeader) {
1122
+ authHeader = providerContext.authHeader;
1123
+ } else {
1124
+ const credentials2 = await getToken();
1125
+ if (!credentials2) {
1126
+ return;
1127
+ }
1128
+ authHeader = `Bearer ${credentials2.token}`;
1129
+ }
1130
+ const apiUrl = getApiUrlConfig();
1131
+ const controller = new AbortController();
1132
+ const timeout = setTimeout(() => controller.abort(), FLUSH_TIMEOUT_MS);
1133
+ try {
1134
+ const request = {
1135
+ events,
1136
+ client_type: "cli",
1137
+ client_version: CLI_VERSION
1138
+ };
1139
+ const response = await fetch(`${apiUrl}/telemetry/events`, {
1140
+ method: "POST",
1141
+ headers: {
1142
+ "Content-Type": "application/json",
1143
+ Authorization: authHeader
1144
+ },
1145
+ body: JSON.stringify(request),
1146
+ signal: controller.signal
1147
+ });
1148
+ if (!response.ok) {
1149
+ console.error(`Telemetry flush failed: ${response.status}`);
1150
+ }
1151
+ } finally {
1152
+ clearTimeout(timeout);
1153
+ }
1154
+ } catch (error2) {
1155
+ const now = Date.now();
1156
+ if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
1157
+ const message = error2 instanceof Error ? error2.message : String(error2);
1158
+ const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
1159
+ console.error(`Telemetry flush error: ${message}${suffix}`);
1160
+ lastFlushFailureLoggedAt = now;
1161
+ suppressedFlushFailureCount = 0;
1162
+ } else {
1163
+ suppressedFlushFailureCount++;
1164
+ }
1165
+ }
1166
+ }
1167
+ async function shutdownTelemetry() {
1168
+ isShuttingDown = true;
1169
+ if (flushTimeout) {
1170
+ clearTimeout(flushTimeout);
1171
+ flushTimeout = null;
1172
+ }
1173
+ await flushEvents();
1174
+ }
1175
+ function emitEvent(event) {
1176
+ logEvent(event.event_type, {
1177
+ severity: event.severity,
1178
+ message: event.message,
1179
+ metadata: event.metadata,
1180
+ agentId: event.agent_id
1181
+ });
1182
+ }
1183
+ function emitAgentConnected(agentId, metadata) {
1184
+ emitEvent({
1185
+ event_type: TelemetryEventTypes.AGENT_CONNECTED,
1186
+ severity: "info",
1187
+ message: "Agent CLI connected",
1188
+ metadata,
1189
+ agent_id: agentId
1190
+ });
1191
+ }
1192
+ function emitAgentDisconnected(agentId, metadata) {
1193
+ emitEvent({
1194
+ event_type: TelemetryEventTypes.AGENT_DISCONNECTED,
1195
+ severity: "info",
1196
+ message: `Agent CLI disconnected (code: ${metadata.code})`,
1197
+ metadata,
1198
+ agent_id: agentId
1199
+ });
1200
+ }
1201
+ var EventTypes = {
1202
+ // Tunnel lifecycle
1203
+ TUNNEL_STARTING: "tunnel.starting",
1204
+ TUNNEL_CONNECTED: "tunnel.connected",
1205
+ TUNNEL_DISCONNECTED: "tunnel.disconnected",
1206
+ TUNNEL_RECONNECTING: "tunnel.reconnecting",
1207
+ TUNNEL_ERROR: "tunnel.error",
1208
+ // OpenCode communication
1209
+ OPENCODE_HEALTH_CHECK: "opencode.health_check",
1210
+ OPENCODE_HEALTH_OK: "opencode.health_ok",
1211
+ OPENCODE_HEALTH_FAILED: "opencode.health_failed",
1212
+ OPENCODE_REQUEST_RECEIVED: "opencode.request_received",
1213
+ OPENCODE_REQUEST_FORWARDED: "opencode.request_forwarded",
1214
+ OPENCODE_RESPONSE_SENT: "opencode.response_sent",
1215
+ OPENCODE_UNREACHABLE: "opencode.unreachable",
1216
+ OPENCODE_ERROR: "opencode.error",
1217
+ // Authentication
1218
+ AUTH_LOGIN_STARTED: "auth.login_started",
1219
+ AUTH_LOGIN_SUCCESS: "auth.login_success",
1220
+ AUTH_LOGIN_FAILED: "auth.login_failed",
1221
+ AUTH_LOGOUT: "auth.logout",
1222
+ // CLI lifecycle
1223
+ CLI_STARTED: "cli.started",
1224
+ CLI_COMMAND: "cli.command",
1225
+ CLI_ERROR: "cli.error",
1226
+ // Deprecation telemetry (#412) — usage of the old `--agent`/`EVIDENT_AGENT_KEY`
1227
+ // names instead of the preferred `--runner`/`EVIDENT_RUNNER_KEY` (#409).
1228
+ DEPRECATED_AGENT_FLAG_USED: "cli.deprecated_agent_flag_used",
1229
+ DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
1230
+ };
1231
+
1232
+ // src/lib/runner-activity-telemetry.ts
1233
+ var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
1234
+ var SEVERITY_BY_LEVEL = {
1235
+ warn: "warning",
1236
+ error: "error"
1237
+ };
1238
+ var MAX_MESSAGE_LENGTH = 500;
1239
+ var TRUNCATION_MARKER = "\u2026";
1240
+ function redact(message) {
1241
+ return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
1242
+ }
1243
+ function truncate(message) {
1244
+ if (message.length <= MAX_MESSAGE_LENGTH) return message;
1245
+ return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
1246
+ }
1247
+ var RATE_LIMIT_WINDOW_MS = 6e4;
1248
+ var RATE_LIMIT_MAX_EVENTS = 30;
1249
+ var windowStartedAt = 0;
1250
+ var windowCount = 0;
1251
+ var windowDroppedCount = 0;
1252
+ function admitUnderRateLimit(now) {
1253
+ if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
1254
+ if (windowDroppedCount > 0) {
1255
+ console.error(
1256
+ `[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
1257
+ );
1258
+ }
1259
+ windowStartedAt = now;
1260
+ windowCount = 0;
1261
+ windowDroppedCount = 0;
1262
+ }
1263
+ if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
1264
+ windowDroppedCount++;
1265
+ if (windowDroppedCount === 1) {
1266
+ console.error(
1267
+ `[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
1268
+ );
1269
+ }
1270
+ return false;
1271
+ }
1272
+ windowCount++;
1273
+ return true;
1274
+ }
1275
+ function forwardRunnerActivity(entry, context) {
1276
+ try {
1277
+ if (!FORWARDED_LEVELS.has(entry.level)) return;
1278
+ if (!context.agentId || !context.authHeader) return;
1279
+ if (!admitUnderRateLimit(Date.now())) return;
1280
+ const rawMessage = entry.error ?? entry.message ?? "";
1281
+ const message = truncate(redact(rawMessage));
1282
+ logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
1283
+ severity: SEVERITY_BY_LEVEL[entry.level],
1284
+ message,
1285
+ metadata: { source: "cli.run" },
1286
+ agentId: context.agentId
1287
+ });
1288
+ } catch (err) {
1289
+ console.error(
1290
+ `[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
1291
+ );
1292
+ }
1293
+ }
1294
+
1295
+ // src/lib/opencode/health.ts
1296
+ async function checkOpenCodeHealth(port) {
1297
+ try {
1298
+ const response = await fetch(`http://127.0.0.1:${port}/global/health`, {
1299
+ signal: AbortSignal.timeout(2e3)
1300
+ // 2 second timeout
1301
+ });
1302
+ if (!response.ok) {
1303
+ return { healthy: false, error: `HTTP ${response.status}` };
1304
+ }
1305
+ const data = await response.json().catch(() => ({}));
1306
+ return { healthy: true, version: data.version };
705
1307
  } catch (error2) {
706
1308
  const message = error2 instanceof Error ? error2.message : "Unknown error";
707
1309
  return { healthy: false, error: message };
@@ -714,7 +1316,7 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
714
1316
  if (health.healthy) {
715
1317
  return health;
716
1318
  }
717
- await new Promise((resolve2) => setTimeout(resolve2, 1e3));
1319
+ await new Promise((resolve3) => setTimeout(resolve3, 1e3));
718
1320
  }
719
1321
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
720
1322
  }
@@ -729,7 +1331,7 @@ function buildOpenCodeVersionWarning(version2) {
729
1331
  if (isQueueValidatedVersion(version2)) return null;
730
1332
  const detected = version2 ? `v${version2}` : "unknown";
731
1333
  const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
732
- return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
1334
+ return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
733
1335
  }
734
1336
 
735
1337
  // src/lib/opencode/process.ts
@@ -1021,6 +1623,12 @@ async function promptOpenCodeInstall(interactive) {
1021
1623
  return action;
1022
1624
  }
1023
1625
 
1626
+ // src/lib/opencode/provider-check.ts
1627
+ function buildNoProviderWarning(hasProvider) {
1628
+ if (hasProvider !== false) return null;
1629
+ return "Warning: opencode has no authenticated model provider configured, so it won't be able to answer prompts. Run `opencode auth login` to set one up (see https://opencode.ai for details).";
1630
+ }
1631
+
1024
1632
  // src/lib/opencode/session.ts
1025
1633
  function opencodeBase(port) {
1026
1634
  return `http://127.0.0.1:${port}`;
@@ -1223,6 +1831,11 @@ async function getModelAttachmentCapability(port, model) {
1223
1831
  }
1224
1832
  const entry = provider.models[modelId];
1225
1833
  if (!entry || typeof entry !== "object") return null;
1834
+ if (entry.capabilities && typeof entry.capabilities === "object") {
1835
+ if (typeof entry.capabilities.attachment === "boolean") {
1836
+ return entry.capabilities.attachment;
1837
+ }
1838
+ }
1226
1839
  return typeof entry.attachment === "boolean" ? entry.attachment : null;
1227
1840
  } catch (err) {
1228
1841
  console.error(
@@ -1251,6 +1864,16 @@ async function buildFileParts(attachments, capable) {
1251
1864
  );
1252
1865
  dataUrl = null;
1253
1866
  }
1867
+ if (dataUrl !== null && typeof dataUrl === "object") {
1868
+ outcomes.push({
1869
+ index: a.index,
1870
+ mime: a.mime,
1871
+ filename: a.filename,
1872
+ status: "failed",
1873
+ reason: "needs_reauth"
1874
+ });
1875
+ continue;
1876
+ }
1254
1877
  if (dataUrl == null) {
1255
1878
  outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
1256
1879
  continue;
@@ -1332,7 +1955,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
1332
1955
  }
1333
1956
  }
1334
1957
  if (attempt < READ_BACK_ATTEMPTS - 1) {
1335
- await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1958
+ await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
1336
1959
  }
1337
1960
  }
1338
1961
  return null;
@@ -1460,6 +2083,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
1460
2083
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1461
2084
  return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1462
2085
  }
2086
+ function isB2AbandonmentConfirmed(params) {
2087
+ return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
2088
+ }
1463
2089
  function messageError(messages, userMessageId) {
1464
2090
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1465
2091
  const error2 = errorOf(reply);
@@ -1473,12 +2099,94 @@ function messageError(messages, userMessageId) {
1473
2099
  }
1474
2100
  return "The agent run failed.";
1475
2101
  }
2102
+ function isAbortedTerminalReply(messages, userMessageId) {
2103
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
2104
+ const error2 = errorOf(reply);
2105
+ if (error2 == null) return false;
2106
+ if (typeof error2 === "string") return error2.trim() === "Aborted";
2107
+ if (typeof error2 === "object") {
2108
+ const e = error2;
2109
+ if (e.name === "MessageAbortedError") return true;
2110
+ if (e.name === "AbortError") return true;
2111
+ const dataMessage = e.data?.message;
2112
+ const rendered = typeof dataMessage === "string" ? dataMessage : typeof e.message === "string" ? e.message : null;
2113
+ return rendered != null && rendered.trim() === "Aborted";
2114
+ }
2115
+ return false;
2116
+ }
2117
+ function messageFailure(messages, userMessageId) {
2118
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
2119
+ const error2 = errorOf(reply);
2120
+ if (error2 == null || typeof error2 !== "object") return null;
2121
+ const e = error2;
2122
+ const replyProviderId = reply?.info?.providerID ?? null;
2123
+ const replyModelId = reply?.info?.modelID ?? null;
2124
+ if (e.name === "ProviderAuthError") {
2125
+ const data = e.data;
2126
+ const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
2127
+ return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
2128
+ }
2129
+ if (e.name === "APIError") {
2130
+ const data = e.data;
2131
+ const statusCode = data?.statusCode;
2132
+ if (statusCode === 401 || statusCode === 403) {
2133
+ return {
2134
+ kind: "model_auth",
2135
+ providerId: replyProviderId,
2136
+ modelId: replyModelId,
2137
+ reason: "rejected"
2138
+ };
2139
+ }
2140
+ }
2141
+ return null;
2142
+ }
2143
+ function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
2144
+ if (classified != null) return classified;
2145
+ if (hasConfiguredProvider !== false) return null;
2146
+ return {
2147
+ kind: "model_auth",
2148
+ providerId: replyProviderId,
2149
+ modelId: replyModelId,
2150
+ reason: "missing"
2151
+ };
2152
+ }
1476
2153
  function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1477
2154
  if (!messages || messages.length === 0) return false;
1478
2155
  return messages.some(
1479
2156
  (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1480
2157
  );
1481
2158
  }
2159
+ async function hasAnyConfiguredProvider(port) {
2160
+ try {
2161
+ const res = await fetch(`${opencodeBase(port)}/config/providers`);
2162
+ if (!res.ok) {
2163
+ console.error(
2164
+ `[hasAnyConfiguredProvider] GET /config/providers returned HTTP ${res.status} (port ${port})`
2165
+ );
2166
+ return null;
2167
+ }
2168
+ const body = await res.json();
2169
+ if (!body || typeof body !== "object" || Array.isArray(body)) {
2170
+ console.error(
2171
+ `[hasAnyConfiguredProvider] GET /config/providers body was not a plain object (port ${port})`
2172
+ );
2173
+ return null;
2174
+ }
2175
+ const defaults2 = body.default;
2176
+ if (!defaults2 || typeof defaults2 !== "object" || Array.isArray(defaults2)) {
2177
+ console.error(
2178
+ `[hasAnyConfiguredProvider] GET /config/providers body had no \`default\` object (port ${port})`
2179
+ );
2180
+ return null;
2181
+ }
2182
+ return Object.keys(defaults2).length > 0;
2183
+ } catch (err) {
2184
+ console.error(
2185
+ `[hasAnyConfiguredProvider] GET /config/providers failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
2186
+ );
2187
+ return null;
2188
+ }
2189
+ }
1482
2190
 
1483
2191
  // src/lib/opencode/session-cleanup.ts
1484
2192
  var DURATION_UNIT_MS = {
@@ -1637,10 +2345,11 @@ var StreamForwarder = class {
1637
2345
  * Abort every in-flight stream (e.g. on WebSocket close).
1638
2346
  */
1639
2347
  abortAll() {
1640
- for (const stream of this.inflight.values()) {
2348
+ for (const [sid, stream] of this.inflight.entries()) {
1641
2349
  try {
1642
2350
  stream.abort();
1643
- } catch {
2351
+ } catch (err) {
2352
+ log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
1644
2353
  }
1645
2354
  }
1646
2355
  this.inflight.clear();
@@ -1674,12 +2383,12 @@ var StreamForwarder = class {
1674
2383
  let endBody;
1675
2384
  if (has_body) {
1676
2385
  const chunks = [];
1677
- bodyPromise = new Promise((resolve2) => {
2386
+ bodyPromise = new Promise((resolve3) => {
1678
2387
  pushBody = (buf) => {
1679
2388
  chunks.push(buf);
1680
2389
  };
1681
2390
  endBody = () => {
1682
- resolve2(Buffer.concat(chunks));
2391
+ resolve3(Buffer.concat(chunks));
1683
2392
  };
1684
2393
  });
1685
2394
  }
@@ -1790,31 +2499,20 @@ function connectTunnel(options) {
1790
2499
  onConnected,
1791
2500
  onDisconnected,
1792
2501
  onError,
1793
- onRequest,
1794
2502
  onResponse,
1795
2503
  onInfo,
1796
2504
  onDrainPing
1797
2505
  } = options;
1798
2506
  const tunnelUrl = getTunnelUrlConfig();
1799
2507
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1800
- return new Promise((resolve2, reject) => {
2508
+ return new Promise((resolve3, reject) => {
1801
2509
  const ws = new WebSocket2(url, {
1802
2510
  headers: {
1803
2511
  Authorization: authHeader
1804
2512
  }
1805
2513
  });
1806
- const streamStartTimes = /* @__PURE__ */ new Map();
1807
2514
  const forwarder = new StreamForwarder(ws, port, {
1808
- onOpen: (sid, method, path) => {
1809
- if (path === TUNNEL_DRAIN_PING_PATH) return;
1810
- streamStartTimes.set(sid, Date.now());
1811
- onRequest?.(method, path, sid);
1812
- },
1813
- onHead: (sid, status) => {
1814
- const startedAt = streamStartTimes.get(sid);
1815
- streamStartTimes.delete(sid);
1816
- onResponse?.(status, startedAt ? Date.now() - startedAt : 0, sid);
1817
- },
2515
+ onHead: () => onResponse?.(),
1818
2516
  onDrainPing: () => onDrainPing?.()
1819
2517
  });
1820
2518
  const connectionTimeout = setTimeout(() => {
@@ -1862,7 +2560,7 @@ function connectTunnel(options) {
1862
2560
  clearTimeout(connectionTimeout);
1863
2561
  const connectedAgentId = message.agent_id ?? agentId;
1864
2562
  onConnected?.(connectedAgentId);
1865
- resolve2({
2563
+ resolve3({
1866
2564
  ws,
1867
2565
  close: () => ws.close(1e3, "CLI shutdown")
1868
2566
  });
@@ -1890,7 +2588,6 @@ function connectTunnel(options) {
1890
2588
  ws.on("close", (code, reason) => {
1891
2589
  const reasonStr = reason.toString() || upgradeRejection || (code === 1006 ? "abnormal closure" : "No reason provided");
1892
2590
  forwarder.abortAll();
1893
- streamStartTimes.clear();
1894
2591
  onDisconnected?.(code, reasonStr);
1895
2592
  });
1896
2593
  });
@@ -1925,7 +2622,11 @@ var RunnerConnection = class {
1925
2622
  if (this.connection) {
1926
2623
  try {
1927
2624
  this.connection.close();
1928
- } catch {
2625
+ } catch (err) {
2626
+ log("error", "runner_connection_close_failed", {
2627
+ agent_id: this.resolvedAgentId,
2628
+ ...errorFields(err)
2629
+ });
1929
2630
  }
1930
2631
  this.connection = null;
1931
2632
  }
@@ -1975,7 +2676,448 @@ var RunnerConnection = class {
1975
2676
  }
1976
2677
  this.reconnecting = false;
1977
2678
  }
1978
- };
2679
+ };
2680
+
2681
+ // src/lib/tunnel/ready-marker.ts
2682
+ import { writeFileSync } from "fs";
2683
+ function writeTunnelReadyMarker(path, agentId) {
2684
+ try {
2685
+ writeFileSync(path, `${agentId}
2686
+ `);
2687
+ return { ok: true };
2688
+ } catch (error2) {
2689
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2690
+ }
2691
+ }
2692
+
2693
+ // src/lib/claude-usage-reporting.ts
2694
+ var VALID_MODES = ["auto", "on", "off"];
2695
+ function resolveClaudeUsageReportingMode(flagValue, env) {
2696
+ const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
2697
+ if (raw === void 0 || raw === "") {
2698
+ return { mode: "auto", warnings: [] };
2699
+ }
2700
+ const normalized = raw.trim().toLowerCase();
2701
+ if (VALID_MODES.includes(normalized)) {
2702
+ return { mode: normalized, warnings: [] };
2703
+ }
2704
+ const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
2705
+ return {
2706
+ mode: "auto",
2707
+ warnings: [
2708
+ `Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
2709
+ ]
2710
+ };
2711
+ }
2712
+ var BASE_REPORT_DELAY_MS = 10 * 6e4;
2713
+ var REPORT_DELAY_JITTER_FRACTION = 0.2;
2714
+ function nextReportDelayMs(random = Math.random) {
2715
+ const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
2716
+ return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
2717
+ }
2718
+ var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
2719
+ var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
2720
+ function claudeUsageFailureLogLevel(consecutiveFailures) {
2721
+ return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
2722
+ }
2723
+
2724
+ // src/lib/channels/driver.ts
2725
+ import { homedir as homedir2 } from "os";
2726
+
2727
+ // src/lib/file-push.ts
2728
+ import { randomUUID } from "crypto";
2729
+ import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
2730
+ import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
2731
+ var FILE_MODE = 384;
2732
+ var DIRECTORY_MODE = 448;
2733
+ async function writePushedFile(request) {
2734
+ const { requestedPath, content, allowedDirectories, homeDir } = request;
2735
+ const bytes = content.byteLength;
2736
+ if (allowedDirectories.length === 0) {
2737
+ return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
2738
+ path: requestedPath,
2739
+ bytes
2740
+ });
2741
+ }
2742
+ if (bytes > MAX_FILE_PUSH_BYTES) {
2743
+ return refuse(
2744
+ "file_too_large",
2745
+ `File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
2746
+ {
2747
+ path: requestedPath,
2748
+ bytes
2749
+ }
2750
+ );
2751
+ }
2752
+ const candidate = expandAndValidate(requestedPath, homeDir);
2753
+ if (candidate === null) {
2754
+ return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
2755
+ path: requestedPath,
2756
+ bytes
2757
+ });
2758
+ }
2759
+ try {
2760
+ const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
2761
+ dirname2(candidate)
2762
+ );
2763
+ const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
2764
+ const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
2765
+ if (allowedDirectory === null) {
2766
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2767
+ path: realTarget,
2768
+ bytes
2769
+ });
2770
+ }
2771
+ if (missingSegments.length > 0) {
2772
+ await createMissingDirectories(existingAncestor, missingSegments);
2773
+ const realParent = await realpath(dirname2(realTarget));
2774
+ if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
2775
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2776
+ path: realTarget,
2777
+ bytes,
2778
+ reason: "parent_changed_after_create"
2779
+ });
2780
+ }
2781
+ }
2782
+ await writeAtomically(realTarget, content);
2783
+ log("info", "file_push_written", { path: realTarget, bytes });
2784
+ return { ok: true, path: realTarget };
2785
+ } catch (err) {
2786
+ const errno = err.code ?? "UNKNOWN";
2787
+ return refuse("write_failed", `The runner could not write the file (${errno}).`, {
2788
+ path: candidate,
2789
+ bytes,
2790
+ errno,
2791
+ ...errorFields(err)
2792
+ });
2793
+ }
2794
+ }
2795
+ function expandAndValidate(requestedPath, homeDir) {
2796
+ if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
2797
+ return null;
2798
+ }
2799
+ const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
2800
+ if (expanded.split(/[/\\]/).includes("..")) {
2801
+ return null;
2802
+ }
2803
+ if (!isAbsolute(expanded)) {
2804
+ return null;
2805
+ }
2806
+ const candidate = resolve2(expanded);
2807
+ const name = basename(candidate);
2808
+ return name === "" || name === "." || name === ".." ? null : candidate;
2809
+ }
2810
+ async function resolveNearestExistingAncestor(directory) {
2811
+ const missingSegments = [];
2812
+ let current = directory;
2813
+ for (; ; ) {
2814
+ try {
2815
+ return { existingAncestor: await realpath(current), missingSegments };
2816
+ } catch (err) {
2817
+ const parent = dirname2(current);
2818
+ if (err.code !== "ENOENT" || parent === current) {
2819
+ throw err;
2820
+ }
2821
+ missingSegments.unshift(basename(current));
2822
+ current = parent;
2823
+ }
2824
+ }
2825
+ }
2826
+ async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
2827
+ for (const directory of allowedDirectories) {
2828
+ if (!isAbsolute(directory)) {
2829
+ log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
2830
+ continue;
2831
+ }
2832
+ const realDirectory = await realpathCreatingIfMissing(directory);
2833
+ if (realDirectory !== null && contains(realDirectory, realTarget)) {
2834
+ return realDirectory;
2835
+ }
2836
+ }
2837
+ return null;
2838
+ }
2839
+ async function realpathCreatingIfMissing(directory) {
2840
+ try {
2841
+ return await realpath(directory);
2842
+ } catch (err) {
2843
+ if (err.code !== "ENOENT") {
2844
+ log("warn", "file_push_allowed_directory_skipped", {
2845
+ directory,
2846
+ reason: "unresolvable",
2847
+ ...errorFields(err)
2848
+ });
2849
+ return null;
2850
+ }
2851
+ }
2852
+ try {
2853
+ await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
2854
+ await chmod(directory, DIRECTORY_MODE);
2855
+ return await realpath(directory);
2856
+ } catch (err) {
2857
+ log("warn", "file_push_allowed_directory_skipped", {
2858
+ directory,
2859
+ reason: "create_failed",
2860
+ ...errorFields(err)
2861
+ });
2862
+ return null;
2863
+ }
2864
+ }
2865
+ function contains(realDirectory, realTarget) {
2866
+ const rel = relative(realDirectory, realTarget);
2867
+ return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
2868
+ }
2869
+ async function createMissingDirectories(existingAncestor, missingSegments) {
2870
+ let current = existingAncestor;
2871
+ for (const segment of missingSegments) {
2872
+ current = join2(current, segment);
2873
+ await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
2874
+ await chmod(current, DIRECTORY_MODE);
2875
+ }
2876
+ }
2877
+ async function writeAtomically(realTarget, content) {
2878
+ const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2879
+ let handle;
2880
+ try {
2881
+ handle = await open2(temporaryPath, "wx", FILE_MODE);
2882
+ await handle.writeFile(content);
2883
+ await handle.chmod(FILE_MODE);
2884
+ await handle.close();
2885
+ handle = void 0;
2886
+ await rename(temporaryPath, realTarget);
2887
+ } catch (err) {
2888
+ await discardTemporaryFile(temporaryPath, handle);
2889
+ throw err;
2890
+ }
2891
+ }
2892
+ async function discardTemporaryFile(temporaryPath, handle) {
2893
+ try {
2894
+ await handle?.close();
2895
+ } catch (err) {
2896
+ log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
2897
+ }
2898
+ try {
2899
+ await unlink(temporaryPath);
2900
+ } catch (err) {
2901
+ const errno = err.code;
2902
+ if (errno !== "ENOENT" && errno !== "ENOTDIR") {
2903
+ log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
2904
+ }
2905
+ }
2906
+ }
2907
+ function refuse(code, message, fields) {
2908
+ log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
2909
+ return { ok: false, code, message };
2910
+ }
2911
+
2912
+ // src/lib/runner-file-sync.ts
2913
+ var MAX_ACK_ATTEMPTS = 5;
2914
+ async function syncPendingRunnerFiles(options) {
2915
+ const pending = await listPendingFiles(options);
2916
+ const pendingIds = new Set(pending.map((file) => file.id));
2917
+ for (const id of options.ackFailures.keys()) {
2918
+ if (!pendingIds.has(id)) options.ackFailures.delete(id);
2919
+ }
2920
+ if (pending.length === 0) return 0;
2921
+ options.log({
2922
+ level: "info",
2923
+ message: `Runner file sync: ${pending.length} file(s) queued for this runner`
2924
+ });
2925
+ let applied = 0;
2926
+ for (const file of pending) {
2927
+ if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
2928
+ if (await applyOne(options, file)) applied += 1;
2929
+ }
2930
+ return applied;
2931
+ }
2932
+ async function listPendingFiles(options) {
2933
+ let res;
2934
+ try {
2935
+ res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
2936
+ headers: { Authorization: options.getAuthHeader() }
2937
+ });
2938
+ } catch (err) {
2939
+ options.log({
2940
+ level: "warn",
2941
+ message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
2942
+ });
2943
+ return [];
2944
+ }
2945
+ if (!res.ok) {
2946
+ options.log({
2947
+ level: res.status === 404 ? "debug" : "warn",
2948
+ message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
2949
+ });
2950
+ return [];
2951
+ }
2952
+ let body;
2953
+ try {
2954
+ body = await res.json();
2955
+ } catch (err) {
2956
+ options.log({
2957
+ level: "warn",
2958
+ message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
2959
+ });
2960
+ return [];
2961
+ }
2962
+ if (!Array.isArray(body)) {
2963
+ options.log({
2964
+ level: "warn",
2965
+ message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
2966
+ });
2967
+ return [];
2968
+ }
2969
+ const files = [];
2970
+ for (const entry of body) {
2971
+ const file = asPendingFile(entry);
2972
+ if (file === null) {
2973
+ options.log({
2974
+ level: "warn",
2975
+ message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
2976
+ });
2977
+ continue;
2978
+ }
2979
+ files.push(file);
2980
+ }
2981
+ return files;
2982
+ }
2983
+ function asPendingFile(entry) {
2984
+ if (entry === null || typeof entry !== "object") return null;
2985
+ const { id, path, size } = entry;
2986
+ if (typeof id !== "string" || id === "") return null;
2987
+ if (typeof path !== "string" || path === "") return null;
2988
+ if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
2989
+ return { id, path, size };
2990
+ }
2991
+ async function applyOne(options, file) {
2992
+ const label = `${file.id.slice(0, 8)} (${file.path})`;
2993
+ if (options.allowedDirectories.length === 0) {
2994
+ options.log({
2995
+ level: "warn",
2996
+ message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
2997
+ });
2998
+ await ack(options, file, "rejected", "file_sync_disabled");
2999
+ return false;
3000
+ }
3001
+ if (file.size > MAX_FILE_PUSH_BYTES) {
3002
+ options.log({
3003
+ level: "warn",
3004
+ message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
3005
+ });
3006
+ await ack(options, file, "rejected", "file_too_large");
3007
+ return false;
3008
+ }
3009
+ const download = await downloadContent(options, file, label);
3010
+ if (!download.ok) {
3011
+ if (download.terminal) await ack(options, file, "rejected", download.code);
3012
+ return false;
3013
+ }
3014
+ let outcome;
3015
+ try {
3016
+ outcome = await writePushedFile({
3017
+ requestedPath: file.path,
3018
+ content: download.content,
3019
+ allowedDirectories: options.allowedDirectories,
3020
+ homeDir: options.homeDir
3021
+ });
3022
+ } catch (err) {
3023
+ options.log({
3024
+ level: "error",
3025
+ message: `Runner file ${label} could not be written: ${describe(err)}`
3026
+ });
3027
+ await ack(options, file, "rejected", "write_failed");
3028
+ return false;
3029
+ }
3030
+ if (!outcome.ok) {
3031
+ options.log({
3032
+ level: "warn",
3033
+ message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
3034
+ });
3035
+ await ack(options, file, "rejected", outcome.code);
3036
+ return false;
3037
+ }
3038
+ options.log({
3039
+ level: "info",
3040
+ message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
3041
+ });
3042
+ await ack(options, file, "applied");
3043
+ return true;
3044
+ }
3045
+ function durableDownloadCode(status2) {
3046
+ return status2 === 413 ? "file_too_large" : "write_failed";
3047
+ }
3048
+ async function downloadContent(options, file, label) {
3049
+ try {
3050
+ const res = await options.fetchImpl(
3051
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
3052
+ { headers: { Authorization: options.getAuthHeader() } }
3053
+ );
3054
+ if (!res.ok) {
3055
+ const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
3056
+ if (!terminal) {
3057
+ options.log({
3058
+ level: "warn",
3059
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
3060
+ });
3061
+ return { ok: false, terminal: false };
3062
+ }
3063
+ const code = durableDownloadCode(res.status);
3064
+ options.log({
3065
+ level: "error",
3066
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
3067
+ });
3068
+ return { ok: false, terminal: true, code };
3069
+ }
3070
+ return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
3071
+ } catch (err) {
3072
+ options.log({
3073
+ level: "warn",
3074
+ message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
3075
+ });
3076
+ return { ok: false, terminal: false };
3077
+ }
3078
+ }
3079
+ async function ack(options, file, status2, reason) {
3080
+ const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
3081
+ try {
3082
+ const res = await options.fetchImpl(
3083
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
3084
+ {
3085
+ method: "POST",
3086
+ headers: {
3087
+ Authorization: options.getAuthHeader(),
3088
+ "Content-Type": "application/json"
3089
+ },
3090
+ body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
3091
+ }
3092
+ );
3093
+ if (!res.ok) {
3094
+ recordAckFailure(
3095
+ options,
3096
+ file,
3097
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
3098
+ );
3099
+ return;
3100
+ }
3101
+ options.ackFailures.delete(file.id);
3102
+ } catch (err) {
3103
+ recordAckFailure(
3104
+ options,
3105
+ file,
3106
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
3107
+ );
3108
+ }
3109
+ }
3110
+ function recordAckFailure(options, file, what) {
3111
+ const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
3112
+ options.ackFailures.set(file.id, attempts);
3113
+ options.log({
3114
+ level: "error",
3115
+ message: attempts >= MAX_ACK_ATTEMPTS ? `${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})`
3116
+ });
3117
+ }
3118
+ function describe(err) {
3119
+ return err instanceof Error ? err.message : String(err);
3120
+ }
1979
3121
 
1980
3122
  // src/lib/channels/driver.ts
1981
3123
  function messageIdOf(m) {
@@ -2005,7 +3147,11 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
2005
3147
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
2006
3148
  var HEARTBEAT_MS = 6e4;
2007
3149
  var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
3150
+ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
3151
+ var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2008
3152
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
3153
+ var MAX_SUPERSEDED_CONVERSATIONS = 256;
3154
+ var MAX_IDENTICAL_REDRIVE_POLL_FAILURES = 5;
2009
3155
  var ChannelAuthError = class extends Error {
2010
3156
  constructor(message) {
2011
3157
  super(message);
@@ -2014,10 +3160,10 @@ var ChannelAuthError = class extends Error {
2014
3160
  };
2015
3161
  var ChannelTerminalError = class extends Error {
2016
3162
  status;
2017
- constructor(message, status) {
3163
+ constructor(message, status2) {
2018
3164
  super(message);
2019
3165
  this.name = "ChannelTerminalError";
2020
- this.status = status;
3166
+ this.status = status2;
2021
3167
  }
2022
3168
  };
2023
3169
  function backoffDelay(attempt, policy) {
@@ -2025,10 +3171,14 @@ function backoffDelay(attempt, policy) {
2025
3171
  const capped = Math.min(policy.maxDelayMs, exp);
2026
3172
  return Math.floor(Math.random() * capped);
2027
3173
  }
2028
- function isRetryableStatus(status) {
2029
- return status === 429 || status >= 500 && status <= 599;
3174
+ function isRetryableStatus(status2) {
3175
+ return status2 === 429 || status2 >= 500 && status2 <= 599;
3176
+ }
3177
+ var VOLATILE_BODY_FIELD_PATTERN = /("(?:ref|requestId|request_id|traceId|trace_id)"\s*:\s*)"[^"]*"/gi;
3178
+ function normalizeRedrivePollFailureBody(body) {
3179
+ return body.replace(VOLATILE_BODY_FIELD_PATTERN, '$1"<redacted>"').replace(/\s+/g, " ").trim().slice(0, 200);
2030
3180
  }
2031
- var ChannelDriver = class {
3181
+ var ChannelDriver = class _ChannelDriver {
2032
3182
  agentId;
2033
3183
  port;
2034
3184
  apiUrl;
@@ -2042,8 +3192,38 @@ var ChannelDriver = class {
2042
3192
  pausedMaxWaitMs;
2043
3193
  stuckQueuedMs;
2044
3194
  now;
3195
+ fileSyncDirectories;
3196
+ homeDir;
2045
3197
  /** Cache of conversationId → opencode sessionId. */
2046
3198
  sessions = /* @__PURE__ */ new Map();
3199
+ /**
3200
+ * conversationId → the opencode session this runner has ABANDONED as that
3201
+ * conversation's binding (#553), after a genuine (`sessionExists === true`)
3202
+ * dispatch failure: the session still exists but is wedged, so #485's self-heal
3203
+ * must bind a fresh one.
3204
+ *
3205
+ * Dropping the local binding + clearing the server row is not enough on its own:
3206
+ * a SIBLING message dispatched earlier in the same drain is still in-flight under
3207
+ * the same session, and its watcher's routine status writes carry
3208
+ * `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
3209
+ * and `ensureSession`'s persisted-id fallback then reuses it, defeating the
3210
+ * self-heal. This map makes the runner authoritative instead of racing those
3211
+ * writes: *`ensureSession` never reuses an abandoned id for that conversation,
3212
+ * whatever the server row says* — which holds even when the resurrecting write
3213
+ * is one we deliberately keep (see `markDone`).
3214
+ *
3215
+ * Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
3216
+ * one conversation hold ONE entry (the newest abandonment replaces the older), and
3217
+ * hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
3218
+ * NEWEST abandoned id per conversation is guarded: after a second abandonment a
3219
+ * late sibling of the FIRST session can write that id back and `ensureSession`
3220
+ * will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
3221
+ * NOT dropped when the session's watcher tears down: `markDone` still writes the
3222
+ * abandoned id back (it must, or the reply is lost), so the guard has to outlive
3223
+ * the turn that resurrects it. In-memory only — a restart forgets it, at the same
3224
+ * bounded cost.
3225
+ */
3226
+ supersededSessions = /* @__PURE__ */ new Map();
2047
3227
  /**
2048
3228
  * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
2049
3229
  * longer idempotent (no caller-supplied `messageID`), and its read-back picks
@@ -2111,6 +3291,84 @@ var ChannelDriver = class {
2111
3291
  * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
2112
3292
  */
2113
3293
  readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
3294
+ /**
3295
+ * "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
3296
+ * `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
3297
+ * every ~2s drain until opencode's status becomes readable, but the
3298
+ * server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
3299
+ * on any non-`unresolved` outcome so the set cannot grow beyond the currently
3300
+ * unresolvable rows.
3301
+ */
3302
+ redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
3303
+ /**
3304
+ * First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
3305
+ * `pending` row is invisible to every cron arm (all require `status =
3306
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
3307
+ * nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
3308
+ * takes `dispatch` instead of `unresolved` (reusing the existing knob — see
3309
+ * ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
3310
+ */
3311
+ redriveUnresolvedSince = /* @__PURE__ */ new Map();
3312
+ /**
3313
+ * Consecutive-identical-poll-failure streak for the re-drive fence (#1348),
3314
+ * keyed by Evident **message id** (not session) so `clearRedriveUnresolved`
3315
+ * can drop it with the other two trackers and it cannot leak. `sessionId` is
3316
+ * carried inside the entry, not the key: a session change is a different
3317
+ * situation and resets the streak, which gives the `(sessionId, message.id)`
3318
+ * pairing #1348 asks for without a composite map key.
3319
+ */
3320
+ redrivePollFailures = /* @__PURE__ */ new Map();
3321
+ /**
3322
+ * "Already emitted `redrive_outcome_unreported` for THIS (message, outcome)
3323
+ * streak" (Class B, #1340: the runner DECIDED reattach/settle/fail_permanent
3324
+ * but its own PATCH to record it failed — distinct from Class A's
3325
+ * `redrive_poll_failed`, where opencode itself can't be observed). Keyed by
3326
+ * message id, valued by the outcome currently failing to report, so a
3327
+ * change of outcome starts a fresh signal. Cleared by
3328
+ * `clearRedriveUnresolved` the instant either PATCH succeeds.
3329
+ */
3330
+ redriveOutcomeUnreportedSignalled = /* @__PURE__ */ new Map();
3331
+ /**
3332
+ * First `now()` a Class B outcome PATCH (reattach/settle/fail_permanent) was
3333
+ * observed to fail for this message (#1366's failure-window trip arm,
3334
+ * `boundRedriveOutcome`). Duration, not a tick count — bounded by the
3335
+ * existing `pausedMaxWaitMs` window (reusing the knob, not a new constant).
3336
+ * Cleared by `clearRedriveUnresolved` the instant the original PATCH
3337
+ * succeeds.
3338
+ */
3339
+ redriveOutcomeFailingSince = /* @__PURE__ */ new Map();
3340
+ /**
3341
+ * "Already posted `redrive_outcome_abandoned` with `reported: false` for this
3342
+ * row" (#1366) — the bound tripped but the terminal `markFailed` fallback ALSO
3343
+ * failed (the route-level fault of G2), so every following tick re-attempts
3344
+ * the same terminal PATCH. Guards that quiet retry from re-signalling on
3345
+ * every tick. Cleared by `clearRedriveUnresolved`.
3346
+ */
3347
+ redriveOutcomeAbandonedSignalled = /* @__PURE__ */ new Set();
3348
+ /**
3349
+ * "Already emitted `dispatch_not_started` for THIS (message, branch) streak"
3350
+ * (#1340). Valued by the branch currently firing, so a row that moves between
3351
+ * exits re-signals — the move IS the finding. Cleared only on a CONFIRMED
3352
+ * dispatch, never on the fence's decision to dispatch: `clearRedriveUnresolved`
3353
+ * runs on that decision (`resolveRedriveUnresolved`), so clearing there would
3354
+ * re-signal on every one of the 15h of re-dispatch attempts #1110 made.
3355
+ */
3356
+ dispatchNotStartedSignalled = /* @__PURE__ */ new Map();
3357
+ /**
3358
+ * Consecutive-UNCONFIRMED-dispatch streak for a `pending` row with NO stored
3359
+ * `opencode_message_id` yet — i.e. one that has never even reached the
3360
+ * re-drive fence above. `sendPromptAsync`'s POST may 2xx, but its own
3361
+ * read-back retries can never confirm the assigned id when the session's
3362
+ * message list is PERMANENTLY unreadable (e.g. a corrupted local opencode
3363
+ * SQLite DB, #1345/#1348's exact fault, just hit BEFORE the row is ever
3364
+ * dispatched instead of after). Unlike an already-dispatched row, THIS row has
3365
+ * no other safety net at all: the lifecycle cron only reclaims `status =
3366
+ * 'processing'` rows, and a row stuck here never reaches `processing`. Keyed
3367
+ * by message id, carrying `sessionId` so a session change (a fresh one bound
3368
+ * after abandonment) starts a new streak rather than inheriting the old
3369
+ * session's count — same shape as `redrivePollFailures` above.
3370
+ */
3371
+ unconfirmedDispatchFailures = /* @__PURE__ */ new Map();
2114
3372
  /**
2115
3373
  * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
2116
3374
  * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
@@ -2153,9 +3411,12 @@ var ChannelDriver = class {
2153
3411
  sessionParents = /* @__PURE__ */ new Map();
2154
3412
  /**
2155
3413
  * Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
2156
- * NON-EMPTY name is stored (terminal — a real session name won't later un-name),
2157
- * so we do NOT re-GET `/session/:id` every tick. A missing entry = not yet
2158
- * resolved OR resolved-but-still-empty re-fetch on next need, since OpenCode
3414
+ * NON-EMPTY, non-placeholder name is stored (terminal — a real session name
3415
+ * won't later un-name), so we do NOT re-GET `/session/:id` every tick. "Non-empty"
3416
+ * excludes OpenCode's synchronous default title (see
3417
+ * `OPENCODE_DEFAULT_TITLE_PREFIX`, #549) — that placeholder is treated the same
3418
+ * as an empty title so it never latches. A missing entry = not yet resolved OR
3419
+ * resolved-but-still-empty/placeholder → re-fetch on next need, since OpenCode
2159
3420
  * names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
2160
3421
  * the watcher completion path AND the restart-recovery re-adopt path (which has
2161
3422
  * no watcher) can resolve the title.
@@ -2163,6 +3424,24 @@ var ChannelDriver = class {
2163
3424
  sessionTitles = /* @__PURE__ */ new Map();
2164
3425
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
2165
3426
  draining = false;
3427
+ /**
3428
+ * Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
3429
+ * drain ping don't download, write and ack the same file twice.
3430
+ */
3431
+ syncingFiles = false;
3432
+ /**
3433
+ * Consecutive failed acks per pending file (#559). Lives on the driver so it
3434
+ * survives across drains — without it, a file whose ack keeps failing is
3435
+ * re-downloaded and re-written every ~2s until the server expires it.
3436
+ */
3437
+ fileAckFailures = /* @__PURE__ */ new Map();
3438
+ /**
3439
+ * Monotonic count of files this runner has pulled and written (#559). Only
3440
+ * ever increases, so `run.ts` detects work by comparing it against the value
3441
+ * it saw on the previous cycle — including work that landed mid-sleep, the
3442
+ * same trick `lastProxiedActivityAt` uses.
3443
+ */
3444
+ appliedFileCount = 0;
2166
3445
  /**
2167
3446
  * The currently-executing `drainPending()` promise, or null when idle. Lets a
2168
3447
  * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
@@ -2178,21 +3457,23 @@ var ChannelDriver = class {
2178
3457
  * and stops opencode.
2179
3458
  */
2180
3459
  stopped = false;
2181
- constructor(config2) {
2182
- this.agentId = config2.agentId;
2183
- this.port = config2.port;
2184
- this.apiUrl = config2.apiUrl.replace(/\/$/, "");
2185
- this.getAuthHeader = config2.getAuthHeader;
2186
- this.conversationFilter = config2.conversationFilter ?? null;
2187
- this.retry = { ...DEFAULT_RETRY_POLICY, ...config2.retry };
2188
- this.log = config2.log ?? (() => {
3460
+ constructor(config) {
3461
+ this.agentId = config.agentId;
3462
+ this.port = config.port;
3463
+ this.apiUrl = config.apiUrl.replace(/\/$/, "");
3464
+ this.getAuthHeader = config.getAuthHeader;
3465
+ this.conversationFilter = config.conversationFilter ?? null;
3466
+ this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
3467
+ this.log = config.log ?? (() => {
2189
3468
  });
2190
- this.fetchImpl = config2.fetchImpl ?? fetch;
2191
- this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
2192
- this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
2193
- this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2194
- this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2195
- this.now = config2.now ?? (() => Date.now());
3469
+ this.fetchImpl = config.fetchImpl ?? fetch;
3470
+ this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
3471
+ this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
3472
+ this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
3473
+ this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
3474
+ this.now = config.now ?? (() => Date.now());
3475
+ this.fileSyncDirectories = config.fileSyncDirectories ?? [];
3476
+ this.homeDir = config.homeDir ?? homedir2();
2196
3477
  }
2197
3478
  /** The IPv4-loopback base URL for the local `opencode serve`. */
2198
3479
  get opencodeBase() {
@@ -2220,6 +3501,47 @@ var ChannelDriver = class {
2220
3501
  );
2221
3502
  return run2;
2222
3503
  }
3504
+ /**
3505
+ * Pull-and-apply any files Evident has queued for this runner (#559), riding
3506
+ * the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
3507
+ * and drain ping that call `drainPending()`. There is deliberately no channel,
3508
+ * control frame or poll loop of its own: worst-case latency is one poll tick.
3509
+ *
3510
+ * NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
3511
+ * cost a conversation turn. Failures are logged and either acked as a terminal
3512
+ * outcome or left pending for the next drain (see `runner-file-sync.ts`).
3513
+ *
3514
+ * Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
3515
+ *
3516
+ * @returns the number of files written to disk.
3517
+ */
3518
+ async syncPendingFiles() {
3519
+ if (this.stopped) return 0;
3520
+ if (this.syncingFiles) return 0;
3521
+ this.syncingFiles = true;
3522
+ try {
3523
+ const applied = await syncPendingRunnerFiles({
3524
+ agentId: this.agentId,
3525
+ apiUrl: this.apiUrl,
3526
+ getAuthHeader: this.getAuthHeader,
3527
+ fetchImpl: this.fetchImpl,
3528
+ allowedDirectories: this.fileSyncDirectories,
3529
+ homeDir: this.homeDir,
3530
+ ackFailures: this.fileAckFailures,
3531
+ log: this.log
3532
+ });
3533
+ this.appliedFileCount += applied;
3534
+ return applied;
3535
+ } catch (err) {
3536
+ this.log({
3537
+ level: "error",
3538
+ message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
3539
+ });
3540
+ return 0;
3541
+ } finally {
3542
+ this.syncingFiles = false;
3543
+ }
3544
+ }
2223
3545
  async runDrain() {
2224
3546
  let dispatched = 0;
2225
3547
  try {
@@ -2253,6 +3575,28 @@ var ChannelDriver = class {
2253
3575
  }
2254
3576
  return false;
2255
3577
  }
3578
+ /**
3579
+ * File-pull work, for `run.ts`'s idle accounting (#559).
3580
+ *
3581
+ * Pulling a file is real work that `drainPending()` knows nothing about, so
3582
+ * without this a near-idle runner counts a credential pull as an empty tick
3583
+ * and `--idle-timeout` can `process.exit` mid-pull — leaving a
3584
+ * `.evident-push-*.tmp` behind — or immediately after the write, before the
3585
+ * browser has run the authorize/callback that activates it (the user then sees
3586
+ * `saved_not_activated` for a runner that was fine).
3587
+ *
3588
+ * Two signals because one cannot cover both cases: `inFlight` is the pull
3589
+ * happening RIGHT NOW (it may outlive the tick that started it), and
3590
+ * `appliedFiles` is monotonic so a pull that started AND finished between two
3591
+ * idle checks still shows up as an advance.
3592
+ *
3593
+ * CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
3594
+ * the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
3595
+ * samples afterwards reads `true` every single cycle and can never idle out.
3596
+ */
3597
+ fileSyncActivity() {
3598
+ return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
3599
+ }
2256
3600
  /**
2257
3601
  * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2258
3602
  * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
@@ -2314,7 +3658,7 @@ var ChannelDriver = class {
2314
3658
  await this.sleep(step);
2315
3659
  }
2316
3660
  }
2317
- while (this.hasInFlightWatchers()) {
3661
+ while (this.hasInFlightWatchers() || this.syncingFiles) {
2318
3662
  if (this.now() >= deadline) return false;
2319
3663
  await this.sleep(step);
2320
3664
  }
@@ -2349,16 +3693,30 @@ var ChannelDriver = class {
2349
3693
  * @returns the count of messages NEWLY dispatched (not already in-flight).
2350
3694
  */
2351
3695
  async processConversation(conv) {
2352
- const sessionId = await this.ensureSession(conv);
3696
+ const { sessionId, refusedSessionId, created: sessionCreated } = await this.ensureSession(conv);
2353
3697
  const messages = await this.getPendingMessages(conv.id);
2354
3698
  let dispatched = 0;
2355
3699
  let skippedAlreadyDispatched = 0;
3700
+ if (refusedSessionId && messages.length > 0) {
3701
+ void this.postSignal(conv.id, messages[0].id, "session_superseded", {
3702
+ superseded_session_id: refusedSessionId
3703
+ });
3704
+ }
2356
3705
  for (const message of messages) {
2357
3706
  if (this.stopped) break;
2358
3707
  if (this.dispatched.has(message.id)) {
2359
3708
  skippedAlreadyDispatched += 1;
2360
3709
  continue;
2361
3710
  }
3711
+ if (message.opencode_message_id) {
3712
+ const outcome = await this.resolveRedrive(conv, sessionId, message, sessionCreated);
3713
+ if (outcome === "abandoned") {
3714
+ continue;
3715
+ }
3716
+ if (outcome !== "dispatch") {
3717
+ break;
3718
+ }
3719
+ }
2362
3720
  const options = {
2363
3721
  agent: message.opencode_agent ?? void 0,
2364
3722
  model: message.opencode_model ?? void 0
@@ -2379,7 +3737,8 @@ var ChannelDriver = class {
2379
3737
  } catch (err) {
2380
3738
  if (err instanceof ChannelAuthError) throw err;
2381
3739
  this.dispatched.delete(message.id);
2382
- if (await sessionExists(this.port, sessionId) === false) {
3740
+ const exists = await sessionExists(this.port, sessionId);
3741
+ if (exists === false) {
2383
3742
  this.sessions.delete(conv.id);
2384
3743
  this.log({
2385
3744
  level: "warn",
@@ -2387,44 +3746,587 @@ var ChannelDriver = class {
2387
3746
  conversation_id: conv.id,
2388
3747
  message_id: message.id
2389
3748
  });
3749
+ this.signalDispatchNotStarted(conv, message, "session_deleted_race");
3750
+ break;
3751
+ }
3752
+ if (exists === null) {
3753
+ this.log({
3754
+ level: "warn",
3755
+ message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
3756
+ conversation_id: conv.id,
3757
+ message_id: message.id
3758
+ });
3759
+ this.signalDispatchNotStarted(conv, message, "session_existence_unknown");
2390
3760
  break;
2391
3761
  }
2392
- await this.markFailed(conv.id, message.id).catch(() => {
3762
+ const errorMessage = err instanceof Error ? err.message : String(err);
3763
+ this.sessions.delete(conv.id);
3764
+ this.supersede(conv.id, sessionId);
3765
+ this.log({
3766
+ level: "warn",
3767
+ 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.`,
3768
+ conversation_id: conv.id,
3769
+ message_id: message.id
3770
+ });
3771
+ await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
3772
+ this.log({
3773
+ level: "warn",
3774
+ message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
3775
+ conversation_id: conv.id,
3776
+ message_id: message.id
3777
+ });
3778
+ this.signalDispatchNotStarted(conv, message, "failure_unreported");
2393
3779
  });
2394
3780
  this.log({
2395
3781
  level: "error",
2396
- message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
3782
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
2397
3783
  conversation_id: conv.id,
2398
3784
  message_id: message.id
2399
3785
  });
2400
- continue;
3786
+ break;
2401
3787
  }
2402
3788
  if (opencodeMessageId === null) {
3789
+ const streak = this.recordUnconfirmedDispatch(message.id, sessionId);
3790
+ if (streak < MAX_IDENTICAL_REDRIVE_POLL_FAILURES) {
3791
+ this.log({
3792
+ level: "warn",
3793
+ message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back (${streak}/${MAX_IDENTICAL_REDRIVE_POLL_FAILURES}) \u2014 leaving un-tracked to retry next tick`,
3794
+ conversation_id: conv.id,
3795
+ message_id: message.id
3796
+ });
3797
+ this.signalDispatchNotStarted(conv, message, "readback_unconfirmed");
3798
+ continue;
3799
+ }
3800
+ this.unconfirmedDispatchFailures.delete(message.id);
3801
+ this.sessions.delete(conv.id);
3802
+ this.supersede(conv.id, sessionId);
3803
+ const errorMessage = `OpenCode accepted this message ${streak} times in a row but never confirmed its assigned id (session ${sessionId.slice(0, 8)}'s message list could not be read back) \u2014 the session was abandoned; a fresh one is used for further messages.`;
2403
3804
  this.log({
2404
- level: "warn",
2405
- message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
3805
+ level: "error",
3806
+ message: errorMessage,
2406
3807
  conversation_id: conv.id,
2407
3808
  message_id: message.id
2408
3809
  });
2409
- continue;
3810
+ await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
3811
+ this.log({
3812
+ level: "warn",
3813
+ message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
3814
+ conversation_id: conv.id,
3815
+ message_id: message.id
3816
+ });
3817
+ this.signalDispatchNotStarted(conv, message, "abandon_unreported");
3818
+ });
3819
+ break;
2410
3820
  }
3821
+ this.unconfirmedDispatchFailures.delete(message.id);
3822
+ this.dispatchNotStartedSignalled.delete(message.id);
2411
3823
  this.dispatched.add(message.id);
2412
3824
  this.registerInFlight(conv, sessionId, message, opencodeMessageId);
2413
3825
  dispatched += 1;
2414
3826
  void this.postSignal(conv.id, message.id, "dispatched");
2415
3827
  }
2416
- if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3828
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3829
+ this.log({
3830
+ level: "warn",
3831
+ message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
3832
+ conversation_id: conv.id
3833
+ });
3834
+ }
3835
+ this.ensureWatcherRunning(sessionId);
3836
+ return dispatched;
3837
+ }
3838
+ /**
3839
+ * Poll a session's message list for the re-drive fence (#965), via the
3840
+ * INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
3841
+ * hits the global `fetch` and would bypass the same override every other
3842
+ * opencode poll in this file respects. Mirrors `readoptProcessing`'s own
3843
+ * snapshot fetch (`:3081-3111`).
3844
+ *
3845
+ * Returns `{ ok: true, messages }` on a readable snapshot, or
3846
+ * `{ ok: false, signature }` on failure — `signature` is a string that
3847
+ * repeats across attempts for the SAME underlying fault (used by the
3848
+ * consecutive-identical-failure bound, #1348), or `null` for a thrown
3849
+ * exception, which is NOT countable toward that bound (a network blip / an
3850
+ * opencode restart also throws identically every tick, and must keep
3851
+ * retrying unbounded rather than ever being treated as permanent).
3852
+ */
3853
+ async pollSessionMessagesForRedrive(conv, message, sessionId) {
3854
+ try {
3855
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
3856
+ if (!res.ok) {
3857
+ const rawBody = await res.text();
3858
+ const normalized = normalizeRedrivePollFailureBody(rawBody);
3859
+ this.log({
3860
+ level: "warn",
3861
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status}${normalized ? `: ${normalized}` : ""} \u2014 treating as unreadable this tick`,
3862
+ conversation_id: conv.id,
3863
+ message_id: message.id
3864
+ });
3865
+ return { ok: false, signature: `HTTP ${res.status}${normalized ? `: ${normalized}` : ""}` };
3866
+ }
3867
+ const body = await res.json();
3868
+ if (!Array.isArray(body)) {
3869
+ this.log({
3870
+ level: "warn",
3871
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
3872
+ conversation_id: conv.id,
3873
+ message_id: message.id
3874
+ });
3875
+ return { ok: false, signature: "non-array message body" };
3876
+ }
3877
+ return { ok: true, messages: body };
3878
+ } catch (err) {
3879
+ this.log({
3880
+ level: "warn",
3881
+ message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3882
+ conversation_id: conv.id,
3883
+ message_id: message.id
3884
+ });
3885
+ return { ok: false, signature: null };
3886
+ }
3887
+ }
3888
+ /**
3889
+ * The re-drive fence for a `pending` row that already carries a stored
3890
+ * `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
3891
+ * least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
3892
+ * The lifecycle cron can falsely reclaim a `processing` row back to `pending`
3893
+ * mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
3894
+ * without this fence the drain loop would re-`prompt_async` the SAME turn a
3895
+ * second time against live GitHub state. Mirrors `readoptOne`'s job for the
3896
+ * `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
3897
+ * needed here because `sessionCreated` already handles the cases (a #553
3898
+ * abandoned session, a #190 vanished one) that path exists for.
3899
+ *
3900
+ * Only `ChannelAuthError` propagates. A poll that fails identically
3901
+ * `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row reports the message
3902
+ * failed instead of retrying it (#1348) — SEPARATE from, not a replacement
3903
+ * for, `resolveRedriveUnresolved`'s own `pausedMaxWaitMs` bound below. Every
3904
+ * other failure resolves to `unresolved` and is retried whole on the next
3905
+ * ~2s drain tick.
3906
+ */
3907
+ async resolveRedrive(conv, sessionId, message, sessionCreated) {
3908
+ const ocId = message.opencode_message_id ?? null;
3909
+ if (sessionCreated) {
3910
+ this.clearRedriveUnresolved(message.id);
3911
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3912
+ return "dispatch";
3913
+ }
3914
+ const polled = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
3915
+ if (!polled.ok) {
3916
+ const streak = this.recordRedrivePollFailure(message.id, sessionId, polled.signature);
3917
+ if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES && polled.signature !== null) {
3918
+ return this.failRedrivePollPermanent(conv, sessionId, message, polled.signature, streak);
3919
+ }
3920
+ return this.resolveRedriveUnresolved(conv, message);
3921
+ }
3922
+ this.redrivePollFailures.delete(message.id);
3923
+ const messages = polled.messages;
3924
+ if (messages.length === 0) {
3925
+ return this.resolveRedriveUnresolved(conv, message);
3926
+ }
3927
+ const state = messageRunState(messages, ocId ?? "");
3928
+ if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
3929
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3930
+ if (ongoing === false) {
3931
+ this.log({
3932
+ level: "info",
3933
+ message: `Re-drive: message ${message.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
3934
+ conversation_id: conv.id,
3935
+ message_id: message.id
3936
+ });
3937
+ this.clearRedriveUnresolved(message.id);
3938
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3939
+ return "dispatch";
3940
+ }
3941
+ }
3942
+ if (state === "done" || state === "failed") {
3943
+ return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
3944
+ }
3945
+ if (state === "running" || state === "queued") {
3946
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3947
+ if (ongoing === true) {
3948
+ return this.reattachRedrive(conv, sessionId, message, ocId);
3949
+ }
3950
+ if (ongoing === false) {
3951
+ this.clearRedriveUnresolved(message.id);
3952
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3953
+ return "dispatch";
3954
+ }
3955
+ return this.resolveRedriveUnresolved(conv, message);
3956
+ }
3957
+ this.clearRedriveUnresolved(message.id);
3958
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3959
+ return "dispatch";
3960
+ }
3961
+ /**
3962
+ * The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
3963
+ * opencode's own status map — undo the false reclaim instead of starting a
3964
+ * second turn.
3965
+ */
3966
+ async reattachRedrive(conv, sessionId, message, ocId) {
3967
+ let anchorMs;
3968
+ const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
3969
+ if (!Number.isNaN(parsed)) {
3970
+ anchorMs = parsed;
3971
+ } else {
3972
+ anchorMs = this.now();
3973
+ this.log({
3974
+ level: "error",
3975
+ message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
3976
+ conversation_id: conv.id,
3977
+ message_id: message.id
3978
+ });
3979
+ }
3980
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3981
+ try {
3982
+ await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
3983
+ } catch (err) {
3984
+ if (err instanceof ChannelAuthError) throw err;
3985
+ if (err instanceof ChannelTerminalError) {
3986
+ this.log({
3987
+ level: "error",
3988
+ message: `Re-drive: the server definitively refused to restore message ${message.id.slice(0, 8)} to processing (terminal HTTP ${err.status} \u2014 the row is gone or the update was rejected); NOT reporting a re-attach`,
3989
+ conversation_id: conv.id,
3990
+ message_id: message.id
3991
+ });
3992
+ } else {
3993
+ this.log({
3994
+ level: "warn",
3995
+ message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3996
+ conversation_id: conv.id,
3997
+ message_id: message.id
3998
+ });
3999
+ }
4000
+ const bound = await this.boundRedriveOutcome(conv, message, "reattach");
4001
+ return bound === "abandoned" ? "abandoned" : "unresolved";
4002
+ }
4003
+ this.clearRedriveUnresolved(message.id);
4004
+ this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
4005
+ this.dispatched.add(message.id);
4006
+ this.readopted.add(message.id);
4007
+ this.ensureWatcherRunning(sessionId);
4008
+ const watchedForMs = this.now() - anchorMs;
4009
+ void this.postSignal(conv.id, message.id, "redrive_reattached", {
4010
+ watched_for_ms: watchedForMs
4011
+ });
4012
+ this.log({
4013
+ level: "warn",
4014
+ message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
4015
+ conversation_id: conv.id,
4016
+ message_id: message.id
4017
+ });
4018
+ return "reattached";
4019
+ }
4020
+ /**
4021
+ * The `settled` outcome (Task 3.2): the prior turn already finished (or
4022
+ * errored) while nobody was watching — deliver/report it instead of re-running.
4023
+ * Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
4024
+ * (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
4025
+ * drain, same as any other non-auth failure). The restart-abort carve-out that
4026
+ * keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
4027
+ * so a row reaching this `failed` branch is a GENUINE failure.
4028
+ */
4029
+ async settleRedrive(conv, sessionId, message, ocId, messages, state) {
4030
+ try {
4031
+ if (state === "done") {
4032
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
4033
+ const usage = messageUsage(messages, ocId ?? "");
4034
+ this.log({
4035
+ level: "info",
4036
+ message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
4037
+ conversation_id: conv.id,
4038
+ message_id: message.id
4039
+ });
4040
+ await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
4041
+ } else {
4042
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
4043
+ const usage = messageUsage(messages, ocId ?? "");
4044
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
4045
+ this.log({
4046
+ level: "error",
4047
+ message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
4048
+ conversation_id: conv.id,
4049
+ message_id: message.id
4050
+ });
4051
+ await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
4052
+ }
4053
+ } catch (err) {
4054
+ if (err instanceof ChannelAuthError) throw err;
4055
+ this.log({
4056
+ level: "warn",
4057
+ message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
4058
+ conversation_id: conv.id,
4059
+ message_id: message.id
4060
+ });
4061
+ const bound = await this.boundRedriveOutcome(conv, message, "settle");
4062
+ return bound === "abandoned" ? "abandoned" : "unresolved";
4063
+ }
4064
+ this.clearRedriveUnresolved(message.id);
4065
+ void this.postSignal(conv.id, message.id, "redrive_settled");
4066
+ return "settled";
4067
+ }
4068
+ /**
4069
+ * The permanent-failure outcome (#1348): the fence's own poll of this session
4070
+ * failed with the SAME opencode-answered signature
4071
+ * `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row — a transient blip
4072
+ * would have varied or eventually cleared (see `pollSessionMessagesForRedrive`
4073
+ * and `recordRedrivePollFailure`), so this is a durable fault (e.g. #1345's
4074
+ * corrupted opencode session) rather than something worth retrying forever.
4075
+ * Mirrors `settleRedrive`'s error discipline: no `usage`/`failure` args to
4076
+ * `markFailed` (no opencode snapshot to extract them from — this poll never
4077
+ * got a readable one).
4078
+ */
4079
+ async failRedrivePollPermanent(conv, sessionId, message, signature, streak) {
4080
+ this.log({
4081
+ level: "error",
4082
+ message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) failed to poll with the identical signature "${signature}" ${streak} times in a row \u2014 reporting the message failed instead of retrying forever`,
4083
+ conversation_id: conv.id,
4084
+ message_id: message.id
4085
+ });
4086
+ try {
4087
+ await this.markFailed(
4088
+ conv.id,
4089
+ message.id,
4090
+ sessionId,
4091
+ `The runner could not read this conversation's state from OpenCode (${signature}). The same failure repeated ${streak} times in a row, so the message was not retried further.`
4092
+ );
4093
+ } catch (err) {
4094
+ if (err instanceof ChannelAuthError) throw err;
4095
+ this.log({
4096
+ level: "warn",
4097
+ message: `Re-drive: failed to report message ${message.id.slice(0, 8)} permanently failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
4098
+ conversation_id: conv.id,
4099
+ message_id: message.id
4100
+ });
4101
+ const bound = await this.boundRedriveOutcome(conv, message, "fail_permanent");
4102
+ return bound === "abandoned" ? "abandoned" : "unresolved";
4103
+ }
4104
+ this.clearRedriveUnresolved(message.id);
4105
+ void this.postSignal(conv.id, message.id, "redrive_poll_failed");
4106
+ return "settled";
4107
+ }
4108
+ /**
4109
+ * The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
4110
+ * observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
4111
+ * A `pending` row is swept by the server's own `PENDING_MAX_AGE_MS` (24h,
4112
+ * #1368) cron arm, but that is a day-scale backstop — this local bound acts
4113
+ * in minutes so the row (and the conversation it starves, per the ordering
4114
+ * invariant below) isn't left stranded for that long. Bound to the existing
4115
+ * `pausedMaxWaitMs` window (reusing the knob, not a new constant); takes
4116
+ * `dispatch` once elapsed.
4117
+ */
4118
+ resolveRedriveUnresolved(conv, message) {
4119
+ const now = this.now();
4120
+ const since = this.redriveUnresolvedSince.get(message.id);
4121
+ if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
4122
+ this.clearRedriveUnresolved(message.id);
4123
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
4124
+ return "dispatch";
4125
+ }
4126
+ if (since === void 0) {
4127
+ this.redriveUnresolvedSince.set(message.id, now);
4128
+ }
4129
+ if (!this.redriveUnresolvedSignalled.has(message.id)) {
4130
+ this.redriveUnresolvedSignalled.add(message.id);
4131
+ void this.postSignal(conv.id, message.id, "redrive_unresolved");
4132
+ }
4133
+ return "unresolved";
4134
+ }
4135
+ /** Clear all `unresolved`/failure-streak trackers for a row (any non-`unresolved` outcome). */
4136
+ clearRedriveUnresolved(messageId) {
4137
+ this.redriveUnresolvedSince.delete(messageId);
4138
+ this.redriveUnresolvedSignalled.delete(messageId);
4139
+ this.redrivePollFailures.delete(messageId);
4140
+ this.redriveOutcomeUnreportedSignalled.delete(messageId);
4141
+ this.redriveOutcomeFailingSince.delete(messageId);
4142
+ this.redriveOutcomeAbandonedSignalled.delete(messageId);
4143
+ }
4144
+ /**
4145
+ * #1340: the dispatch loop reached a message and did NOT start a turn. Fires at
4146
+ * most once per (message, branch) streak — a wedged row is re-tried every tick,
4147
+ * and the per-tick count is already carried by the co-occurring
4148
+ * `redrive_unresolved`/`redrive_redispatched` signals.
4149
+ */
4150
+ signalDispatchNotStarted(conv, message, branch) {
4151
+ if (this.dispatchNotStartedSignalled.get(message.id) === branch) return;
4152
+ this.dispatchNotStartedSignalled.set(message.id, branch);
4153
+ void this.postSignal(conv.id, message.id, "dispatch_not_started", { branch });
4154
+ }
4155
+ /**
4156
+ * Class B (#1340): the runner DECIDED an outcome (reattach/settle/fail_permanent)
4157
+ * but its own PATCH to record it failed. Fires at most once per (message,
4158
+ * outcome) streak, and only while `boundRedriveOutcome` has not yet tripped —
4159
+ * once it trips, `redrive_outcome_abandoned` takes over reporting for the row
4160
+ * (#1366).
4161
+ */
4162
+ signalRedriveOutcomeUnreported(conv, message, outcome) {
4163
+ if (this.redriveOutcomeUnreportedSignalled.get(message.id) === outcome) return;
4164
+ this.redriveOutcomeUnreportedSignalled.set(message.id, outcome);
4165
+ void this.postSignal(conv.id, message.id, "redrive_outcome_unreported", {
4166
+ attempted_outcome: outcome
4167
+ });
4168
+ }
4169
+ /**
4170
+ * The runner-authored, honest error text for the terminal fallback a tripped
4171
+ * `boundRedriveOutcome` sends. Distinguishable per outcome and truthful about
4172
+ * what actually happened — the `settle`/done case must say the turn finished
4173
+ * but its result could not be recorded, never that the runner stopped
4174
+ * responding (that would be a lie for this shape, see #1366's "why this ships").
4175
+ */
4176
+ static REDRIVE_ABANDON_ERROR = {
4177
+ reattach: "your runner could not record that this message had started, so it was given up on",
4178
+ settle: "your runner finished this message but could not record the result, so the reply could not be delivered",
4179
+ fail_permanent: "the runner could not read this conversation's state from OpenCode, and could not record that failure either, so the message was given up on"
4180
+ };
4181
+ /**
4182
+ * Bound for Class B (#1340, #1366): the runner DECIDED an outcome but its own
4183
+ * PATCH to record it failed. Two independent trip arms (either sufficient):
4184
+ * (1) this failure streak has lasted `pausedMaxWaitMs` — DURATION, not a tick
4185
+ * count, reusing the knob `resolveRedriveUnresolved` already established; (2)
4186
+ * the turn's `processing_started_at` age has crossed
4187
+ * `ABSOLUTE_MAX_PROCESSING_MS` — durable and restart-surviving, since arm (1)'s
4188
+ * in-memory streak resets on a scale-to-zero restart.
4189
+ *
4190
+ * INVARIANT — a tripped bound never suppresses the original outcome attempt;
4191
+ * it only adds a fallback after that attempt has failed again. This is only
4192
+ * ever reached from inside the catch of the ORIGINAL outcome PATCH, which is
4193
+ * attempted first on every tick whether or not this bound tripped before —
4194
+ * there is no give-up latch that would short-circuit it. That is what lets a
4195
+ * route-level fault that heals later still deliver the turn's real
4196
+ * `done`/`failed` payload: once the original PATCH succeeds again, this
4197
+ * helper is never entered and the row settles with its real result.
4198
+ */
4199
+ async boundRedriveOutcome(conv, message, outcome) {
4200
+ const now = this.now();
4201
+ const since = this.redriveOutcomeFailingSince.get(message.id);
4202
+ if (since === void 0) this.redriveOutcomeFailingSince.set(message.id, now);
4203
+ const durationTripped = now - (since ?? now) >= this.pausedMaxWaitMs;
4204
+ const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
4205
+ const absoluteAgeTripped = !Number.isNaN(parsed) && now - parsed >= ABSOLUTE_MAX_PROCESSING_MS;
4206
+ if (!durationTripped && !absoluteAgeTripped) {
4207
+ this.signalRedriveOutcomeUnreported(conv, message, outcome);
4208
+ return "retry";
4209
+ }
4210
+ const arm = durationTripped ? "failure_window" : "absolute_age";
4211
+ try {
4212
+ await this.markFailed(
4213
+ conv.id,
4214
+ message.id,
4215
+ void 0,
4216
+ _ChannelDriver.REDRIVE_ABANDON_ERROR[outcome]
4217
+ );
4218
+ } catch (err) {
4219
+ if (err instanceof ChannelAuthError) throw err;
4220
+ this.log({
4221
+ level: "warn",
4222
+ message: `Re-drive bound: fallback markFailed for message ${message.id.slice(0, 8)} also failed (arm ${arm}, will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
4223
+ conversation_id: conv.id,
4224
+ message_id: message.id
4225
+ });
4226
+ if (!this.redriveOutcomeAbandonedSignalled.has(message.id)) {
4227
+ this.redriveOutcomeAbandonedSignalled.add(message.id);
4228
+ void this.postSignal(conv.id, message.id, "redrive_outcome_abandoned", {
4229
+ attempted_outcome: outcome,
4230
+ reported: false,
4231
+ arm
4232
+ });
4233
+ }
4234
+ return "retry";
4235
+ }
4236
+ this.clearRedriveUnresolved(message.id);
4237
+ void this.postSignal(conv.id, message.id, "redrive_outcome_abandoned", {
4238
+ attempted_outcome: outcome,
4239
+ reported: true,
4240
+ arm
4241
+ });
4242
+ return "abandoned";
4243
+ }
4244
+ /**
4245
+ * Record one poll outcome toward the re-drive fence's consecutive-identical-
4246
+ * failure streak (#1348) and return the resulting count. `signature === null`
4247
+ * (a thrown exception, H1) always clears the streak and returns `0` — it is
4248
+ * never countable. Otherwise the streak continues only when BOTH the session
4249
+ * and the signature match the previous failure; anything else (a different
4250
+ * session, or the same session failing a DIFFERENT way) starts a fresh streak
4251
+ * at `1`.
4252
+ */
4253
+ recordRedrivePollFailure(messageId, sessionId, signature) {
4254
+ if (signature === null) {
4255
+ this.redrivePollFailures.delete(messageId);
4256
+ return 0;
4257
+ }
4258
+ const existing = this.redrivePollFailures.get(messageId);
4259
+ if (existing && existing.sessionId === sessionId && existing.signature === signature) {
4260
+ existing.count += 1;
4261
+ return existing.count;
4262
+ }
4263
+ this.redrivePollFailures.set(messageId, { sessionId, signature, count: 1 });
4264
+ return 1;
4265
+ }
4266
+ /**
4267
+ * Record one UNCONFIRMED-dispatch outcome (a `pending` row with no stored
4268
+ * `opencode_message_id` whose `sendPromptAsync` returned `null`) toward the
4269
+ * bound in `processConversation`'s dispatch loop, and return the resulting
4270
+ * count. Mirrors `recordRedrivePollFailure`'s session-scoping: a session
4271
+ * change starts a fresh streak at `1` rather than inheriting the old one's
4272
+ * count, since a new session is a genuinely different attempt.
4273
+ */
4274
+ recordUnconfirmedDispatch(messageId, sessionId) {
4275
+ const existing = this.unconfirmedDispatchFailures.get(messageId);
4276
+ if (existing && existing.sessionId === sessionId) {
4277
+ existing.count += 1;
4278
+ return existing.count;
4279
+ }
4280
+ this.unconfirmedDispatchFailures.set(messageId, { sessionId, count: 1 });
4281
+ return 1;
4282
+ }
4283
+ /**
4284
+ * Record that `sessionId` is no longer a valid binding for `conversationId`
4285
+ * (#553). Keyed by conversation and hard-capped, so it cannot grow with the
4286
+ * number of failures — see the `supersededSessions` field doc.
4287
+ */
4288
+ supersede(conversationId, sessionId) {
4289
+ this.supersededSessions.delete(conversationId);
4290
+ this.supersededSessions.set(conversationId, sessionId);
4291
+ while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
4292
+ const oldest = this.supersededSessions.keys().next().value;
4293
+ if (oldest === void 0) return;
4294
+ this.supersededSessions.delete(oldest);
4295
+ }
4296
+ }
4297
+ /** Whether `sessionId` is the session this conversation has abandoned (#553). */
4298
+ isSuperseded(conversationId, sessionId) {
4299
+ return this.supersededSessions.get(conversationId) === sessionId;
4300
+ }
4301
+ /**
4302
+ * Resolve the opencode session to run this conversation's turns in.
4303
+ *
4304
+ * `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
4305
+ * binding was an id this runner had abandoned, so a resurrection genuinely
4306
+ * happened and a fresh session was bound instead. The caller reports it.
4307
+ *
4308
+ * `created` says the returned session was made JUST NOW, so it provably holds
4309
+ * no prior turn. The re-drive fence needs that as CONTRARY evidence ("nothing
4310
+ * to reconcile against") — distinct from the ambiguous "I polled and saw an
4311
+ * empty transcript", which stays a deferral. Keep it separate from
4312
+ * `refusedSessionId`: only the latter means a #553 resurrection happened, and
4313
+ * only it may drive the `session_superseded` signal.
4314
+ */
4315
+ async ensureSession(conv) {
4316
+ const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
4317
+ if (bound && this.isSuperseded(conv.id, bound)) {
2417
4318
  this.log({
2418
4319
  level: "warn",
2419
- message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
4320
+ 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.`,
2420
4321
  conversation_id: conv.id
2421
4322
  });
4323
+ this.sessions.delete(conv.id);
4324
+ return {
4325
+ sessionId: await this.createAndBindSession(conv.id),
4326
+ refusedSessionId: bound,
4327
+ created: true
4328
+ };
2422
4329
  }
2423
- this.ensureWatcherRunning(sessionId);
2424
- return dispatched;
2425
- }
2426
- async ensureSession(conv) {
2427
- const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
2428
4330
  if (bound) {
2429
4331
  const exists = await sessionExists(this.port, bound);
2430
4332
  if (exists === false) {
@@ -2434,12 +4336,12 @@ var ChannelDriver = class {
2434
4336
  conversation_id: conv.id
2435
4337
  });
2436
4338
  this.sessions.delete(conv.id);
2437
- return this.createAndBindSession(conv.id);
4339
+ return { sessionId: await this.createAndBindSession(conv.id), created: true };
2438
4340
  }
2439
4341
  this.sessions.set(conv.id, bound);
2440
- return bound;
4342
+ return { sessionId: bound, created: false };
2441
4343
  }
2442
- return this.createAndBindSession(conv.id);
4344
+ return { sessionId: await this.createAndBindSession(conv.id), created: true };
2443
4345
  }
2444
4346
  /**
2445
4347
  * Create a fresh OpenCode session for a conversation, cache the binding, and
@@ -2450,7 +4352,12 @@ var ChannelDriver = class {
2450
4352
  const directory = await this.resolveOpenCodeDirectory();
2451
4353
  const sessionId = await createOpenCodeSession(this.port, directory);
2452
4354
  this.sessions.set(conversationId, sessionId);
2453
- await this.persistSession(conversationId, sessionId).catch(() => {
4355
+ await this.persistSession(conversationId, sessionId).catch((err) => {
4356
+ this.log({
4357
+ level: "warn",
4358
+ 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)}`,
4359
+ conversation_id: conversationId
4360
+ });
2454
4361
  });
2455
4362
  return sessionId;
2456
4363
  }
@@ -2527,7 +4434,11 @@ var ChannelDriver = class {
2527
4434
  * (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
2528
4435
  * (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
2529
4436
  * OMITS that one image and the text turn still sends — NEVER throws the turn.
2530
- * Failures are logged with context (no silent swallow).
4437
+ * A 404 body carrying `{ reason: 'needs_reauth' }` (#547 the server CONFIRMED
4438
+ * a Slack `files:read` scope problem via `files.info`) instead resolves the
4439
+ * `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
4440
+ * user to reconnect Slack instead of a generic "unavailable". Failures are
4441
+ * logged with context (no silent swallow).
2531
4442
  */
2532
4443
  async fetchAttachmentDataUrl(messageId, index, mime) {
2533
4444
  try {
@@ -2536,6 +4447,25 @@ var ChannelDriver = class {
2536
4447
  { headers: { Authorization: this.getAuthHeader() } }
2537
4448
  );
2538
4449
  if (!res.ok) {
4450
+ let reason;
4451
+ try {
4452
+ const body = await res.json();
4453
+ if (body && typeof body.reason === "string") reason = body.reason;
4454
+ } catch (parseErr) {
4455
+ this.log({
4456
+ level: "debug",
4457
+ message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index}: error body was not JSON (${parseErr instanceof Error ? parseErr.message : String(parseErr)}) \u2014 treating as a plain failure`,
4458
+ message_id: messageId
4459
+ });
4460
+ }
4461
+ if (reason === "needs_reauth") {
4462
+ this.log({
4463
+ level: "error",
4464
+ message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 server confirmed a Slack reauth/scope problem \u2014 omitting this image (text turn proceeds)`,
4465
+ message_id: messageId
4466
+ });
4467
+ return { needsReauth: true };
4468
+ }
2539
4469
  this.log({
2540
4470
  level: "error",
2541
4471
  message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
@@ -2575,6 +4505,9 @@ var ChannelDriver = class {
2575
4505
  if (this.attachmentsSkippedSignalled.has(messageId)) return;
2576
4506
  this.attachmentsSkippedSignalled.add(messageId);
2577
4507
  const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
4508
+ const failedReason = outcomes.some(
4509
+ (o) => o.status === "failed" && o.reason === "needs_reauth"
4510
+ ) ? "needs_reauth" : void 0;
2578
4511
  this.log({
2579
4512
  level: "info",
2580
4513
  message: `Message ${messageId.slice(0, 8)}: ${skipped} image(s) skipped (${capabilityUnknown ? "capability was unreadable \u2014 failed open to text-only" : "model not attachment-capable"}), ${failed} image(s) unavailable (deleted-at-source or fetch failure) \u2014 noting to Evident`,
@@ -2584,7 +4517,8 @@ var ChannelDriver = class {
2584
4517
  void this.postSignal(conversationId, messageId, "attachments_skipped", {
2585
4518
  skipped,
2586
4519
  failed,
2587
- ...skipped > 0 ? { skipped_reason: skippedReason } : {}
4520
+ ...skipped > 0 ? { skipped_reason: skippedReason } : {},
4521
+ ...failedReason ? { failed_reason: failedReason } : {}
2588
4522
  });
2589
4523
  }
2590
4524
  /** Register a freshly-dispatched message with its session's watcher state. */
@@ -2615,12 +4549,17 @@ var ChannelDriver = class {
2615
4549
  stuckReported: false,
2616
4550
  lastAliveAt: 0,
2617
4551
  aliveInFlight: false,
4552
+ titleSynced: false,
4553
+ titleSyncInFlight: false,
2618
4554
  awaitingHumanLatched: false,
2619
4555
  pausedOnQuestion: false,
2620
4556
  pausedOnPermission: false,
2621
4557
  pausedClearConfirmed: false,
2622
4558
  pausedInFlight: false,
2623
- deliveryDeadlineAnchored: false
4559
+ deliveryDeadlineAnchored: false,
4560
+ b2PinnedSinceMs: 0,
4561
+ b2LastDescendantCheckMs: 0,
4562
+ b2AbandonedSignalled: false
2624
4563
  });
2625
4564
  }
2626
4565
  /**
@@ -2635,9 +4574,10 @@ var ChannelDriver = class {
2635
4574
  * opencode reports ACTIVELY `running` is watched to completion (its liveness
2636
4575
  * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
2637
4576
  * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
2638
- * handed to the cron. The old "the `deadline` must settle before the ~15-min
2639
- * cron or they double-drive" reasoning is superseded: liveness now settles the
2640
- * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
4577
+ * handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
4578
+ * runner still holds; a reclaimed row that already ran is never re-dispatched
4579
+ * while opencode reports its turn ongoing (readopt's own gate here, and the
4580
+ * `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
2641
4581
  * (only the appear-guard uses it).
2642
4582
  *
2643
4583
  * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
@@ -2688,12 +4628,17 @@ var ChannelDriver = class {
2688
4628
  // with no extra `re_adopted` signal needed (folds old WI-6).
2689
4629
  lastAliveAt: 0,
2690
4630
  aliveInFlight: false,
4631
+ titleSynced: false,
4632
+ titleSyncInFlight: false,
2691
4633
  awaitingHumanLatched: false,
2692
4634
  pausedOnQuestion: false,
2693
4635
  pausedOnPermission: false,
2694
4636
  pausedClearConfirmed: false,
2695
4637
  pausedInFlight: false,
2696
- deliveryDeadlineAnchored: false
4638
+ deliveryDeadlineAnchored: false,
4639
+ b2PinnedSinceMs: 0,
4640
+ b2LastDescendantCheckMs: 0,
4641
+ b2AbandonedSignalled: false
2697
4642
  });
2698
4643
  }
2699
4644
  /**
@@ -2825,9 +4770,8 @@ var ChannelDriver = class {
2825
4770
  const awaitingHuman = observedOpen || latchedPaused;
2826
4771
  if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
2827
4772
  const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2828
- let claimed;
2829
4773
  try {
2830
- claimed = await this.markProcessing(
4774
+ await this.markProcessing(
2831
4775
  conv.id,
2832
4776
  inFlight.evidentMessageId,
2833
4777
  sessionId,
@@ -2836,77 +4780,27 @@ var ChannelDriver = class {
2836
4780
  );
2837
4781
  } catch (err) {
2838
4782
  if (err instanceof ChannelAuthError) throw err;
2839
- this.log({
2840
- level: "warn",
2841
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2842
- conversation_id: conv.id,
2843
- message_id: inFlight.evidentMessageId
2844
- });
2845
- return;
2846
- }
2847
- inFlight.started = true;
2848
- if (!claimed) {
2849
- this.log({
2850
- level: "debug",
2851
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} already marked processing \u2014 continuing`,
2852
- conversation_id: conv.id,
2853
- message_id: inFlight.evidentMessageId
2854
- });
2855
- }
2856
- }
2857
- if (state === "done") {
2858
- this.anchorDeliveryDeadline(inFlight);
2859
- if (!inFlight.done) {
2860
- this.log({
2861
- level: "info",
2862
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2863
- conversation_id: conv.id,
2864
- message_id: inFlight.evidentMessageId
2865
- });
2866
- const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2867
- const usage = messageUsage(messages, inFlight.opencodeMessageId);
2868
- try {
2869
- await this.markDone(
2870
- conv.id,
2871
- inFlight.evidentMessageId,
2872
- sessionId,
2873
- inFlight.opencodeMessageId,
2874
- title,
2875
- usage
2876
- );
2877
- } catch (err) {
2878
- if (err instanceof ChannelAuthError) throw err;
2879
- if (err instanceof ChannelTerminalError) {
2880
- this.log({
2881
- level: "warn",
2882
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2883
- conversation_id: conv.id,
2884
- message_id: inFlight.evidentMessageId
2885
- });
2886
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2887
- return;
2888
- }
2889
- if (this.now() >= inFlight.deadline) {
2890
- this.log({
2891
- level: "warn",
2892
- 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)}`,
2893
- conversation_id: conv.id,
2894
- message_id: inFlight.evidentMessageId
2895
- });
2896
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2897
- return;
2898
- }
4783
+ if (err instanceof ChannelTerminalError) {
4784
+ this.log({
4785
+ level: "error",
4786
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (terminal HTTP ${err.status}) \u2014 the server definitively refused the swap`,
4787
+ conversation_id: conv.id,
4788
+ message_id: inFlight.evidentMessageId
4789
+ });
4790
+ } else {
2899
4791
  this.log({
2900
4792
  level: "warn",
2901
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
4793
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2902
4794
  conversation_id: conv.id,
2903
4795
  message_id: inFlight.evidentMessageId
2904
4796
  });
2905
4797
  return;
2906
4798
  }
2907
- inFlight.done = true;
2908
4799
  }
2909
- this.removeInFlight(watcher, inFlight.evidentMessageId);
4800
+ inFlight.started = true;
4801
+ }
4802
+ if (state === "done") {
4803
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
2910
4804
  return;
2911
4805
  }
2912
4806
  if (state === "failed") {
@@ -2920,8 +4814,16 @@ var ChannelDriver = class {
2920
4814
  message_id: inFlight.evidentMessageId
2921
4815
  });
2922
4816
  const usage = messageUsage(messages, inFlight.opencodeMessageId);
4817
+ const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
2923
4818
  try {
2924
- await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2, usage);
4819
+ await this.markFailed(
4820
+ conv.id,
4821
+ inFlight.evidentMessageId,
4822
+ sessionId,
4823
+ error2,
4824
+ usage,
4825
+ failure
4826
+ );
2925
4827
  } catch (err) {
2926
4828
  if (err instanceof ChannelAuthError) throw err;
2927
4829
  if (err instanceof ChannelTerminalError) {
@@ -2966,6 +4868,44 @@ var ChannelDriver = class {
2966
4868
  });
2967
4869
  }
2968
4870
  const activelyRunning = state === "running" && !awaitingHuman;
4871
+ const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
4872
+ const snapshotReadable = messages != null && messages.length > 0;
4873
+ if (!pinnedNow) {
4874
+ if (snapshotReadable) {
4875
+ inFlight.b2PinnedSinceMs = 0;
4876
+ inFlight.b2LastDescendantCheckMs = 0;
4877
+ inFlight.b2AbandonedSignalled = false;
4878
+ }
4879
+ } else {
4880
+ if (inFlight.b2AbandonedSignalled) {
4881
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
4882
+ return;
4883
+ }
4884
+ if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
4885
+ const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
4886
+ if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
4887
+ inFlight.b2LastDescendantCheckMs = this.now();
4888
+ const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
4889
+ if (isB2AbandonmentConfirmed({
4890
+ pinnedForMs,
4891
+ minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
4892
+ descendantOngoing
4893
+ })) {
4894
+ inFlight.b2AbandonedSignalled = true;
4895
+ this.log({
4896
+ level: "warn",
4897
+ 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`,
4898
+ conversation_id: conv.id,
4899
+ message_id: id
4900
+ });
4901
+ void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
4902
+ watched_for_ms: pinnedForMs
4903
+ });
4904
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
4905
+ return;
4906
+ }
4907
+ }
4908
+ }
2969
4909
  if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
2970
4910
  this.log({
2971
4911
  level: "warn",
@@ -2985,6 +4925,18 @@ var ChannelDriver = class {
2985
4925
  inFlight.aliveInFlight = false;
2986
4926
  if (ok) inFlight.lastAliveAt = this.now();
2987
4927
  });
4928
+ if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
4929
+ inFlight.titleSyncInFlight = true;
4930
+ void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
4931
+ if (!title) {
4932
+ inFlight.titleSyncInFlight = false;
4933
+ return;
4934
+ }
4935
+ const ok = await this.patchConversationTitle(conv.id, title);
4936
+ inFlight.titleSyncInFlight = false;
4937
+ if (ok) inFlight.titleSynced = true;
4938
+ });
4939
+ }
2988
4940
  }
2989
4941
  if (awaitingHuman) {
2990
4942
  if (!inFlight.awaitingHumanLatched) {
@@ -3022,6 +4974,70 @@ var ChannelDriver = class {
3022
4974
  this.removeInFlight(watcher, inFlight.evidentMessageId);
3023
4975
  }
3024
4976
  }
4977
+ /**
4978
+ * Settle a message whose run-state has resolved `'done'` — extracted verbatim
4979
+ * (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
4980
+ * inline `state === 'done'` branch body, so a SECOND caller (the #721
4981
+ * b2-abandonment resolution) can reach the exact same completion behavior
4982
+ * (delivery-deadline anchoring, title resolution, usage extraction, and
4983
+ * `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
4984
+ * and risking the two copies silently drifting apart.
4985
+ */
4986
+ async settleMessageDone(sessionId, watcher, inFlight, messages) {
4987
+ const conv = watcher.conv;
4988
+ this.anchorDeliveryDeadline(inFlight);
4989
+ if (!inFlight.done) {
4990
+ this.log({
4991
+ level: "info",
4992
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
4993
+ conversation_id: conv.id,
4994
+ message_id: inFlight.evidentMessageId
4995
+ });
4996
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
4997
+ const usage = messageUsage(messages, inFlight.opencodeMessageId);
4998
+ try {
4999
+ await this.markDone(
5000
+ conv.id,
5001
+ inFlight.evidentMessageId,
5002
+ sessionId,
5003
+ inFlight.opencodeMessageId,
5004
+ title,
5005
+ usage
5006
+ );
5007
+ } catch (err) {
5008
+ if (err instanceof ChannelAuthError) throw err;
5009
+ if (err instanceof ChannelTerminalError) {
5010
+ this.log({
5011
+ level: "warn",
5012
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
5013
+ conversation_id: conv.id,
5014
+ message_id: inFlight.evidentMessageId
5015
+ });
5016
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
5017
+ return;
5018
+ }
5019
+ if (this.now() >= inFlight.deadline) {
5020
+ this.log({
5021
+ level: "warn",
5022
+ 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)}`,
5023
+ conversation_id: conv.id,
5024
+ message_id: inFlight.evidentMessageId
5025
+ });
5026
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
5027
+ return;
5028
+ }
5029
+ this.log({
5030
+ level: "warn",
5031
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
5032
+ conversation_id: conv.id,
5033
+ message_id: inFlight.evidentMessageId
5034
+ });
5035
+ return;
5036
+ }
5037
+ inFlight.done = true;
5038
+ }
5039
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
5040
+ }
3025
5041
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
3026
5042
  /**
3027
5043
  * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
@@ -3121,7 +5137,10 @@ var ChannelDriver = class {
3121
5137
  * re-dispatched (at most once, see `forceReadoptRun`):
3122
5138
  * - `done` → `markDone` now (guarded like the watcher's done branch);
3123
5139
  * - `failed` → `markFailed` with the surfaced error (issue #182), so an
3124
- * errored turn is reported failed on restart, NOT re-dispatched;
5140
+ * errored turn is reported failed on restart, NOT re-dispatched
5141
+ * EXCEPT a restart-ABORTED turn under a not-ongoing session,
5142
+ * which is a restart orphan wearing a terminal error and is
5143
+ * re-dispatched instead (issue #1310, see the branch below);
3125
5144
  * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
3126
5145
  * tracking the stored id so the reply correlates by it;
3127
5146
  * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
@@ -3185,9 +5204,19 @@ var ChannelDriver = class {
3185
5204
  void this.postSignal(row.conversation_id, row.id, "readopt_done");
3186
5205
  return;
3187
5206
  }
3188
- if (state === "failed") {
5207
+ const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
5208
+ if (restartAborted) {
5209
+ this.log({
5210
+ level: "info",
5211
+ message: `Re-adopt: message ${row.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
5212
+ conversation_id: row.conversation_id,
5213
+ message_id: row.id
5214
+ });
5215
+ }
5216
+ if (state === "failed" && !restartAborted) {
3189
5217
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
3190
5218
  const usage = messageUsage(messages, ocId ?? "");
5219
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3191
5220
  this.log({
3192
5221
  level: "error",
3193
5222
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -3195,7 +5224,7 @@ var ChannelDriver = class {
3195
5224
  message_id: row.id
3196
5225
  });
3197
5226
  try {
3198
- await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
5227
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
3199
5228
  } catch (err) {
3200
5229
  if (err instanceof ChannelAuthError) throw err;
3201
5230
  if (err instanceof ChannelTerminalError) {
@@ -3398,15 +5427,39 @@ var ChannelDriver = class {
3398
5427
  }
3399
5428
  if (ocId === null) {
3400
5429
  this.awaitingReadopt.delete(row.id);
5430
+ const streak = this.recordUnconfirmedDispatch(row.id, sessionId);
5431
+ if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES) {
5432
+ this.unconfirmedDispatchFailures.delete(row.id);
5433
+ this.sessions.delete(readoptConv.id);
5434
+ this.supersede(readoptConv.id, sessionId);
5435
+ const errorMessage = `OpenCode accepted this re-dispatched message ${streak} times in a row but never confirmed its assigned id (session ${sessionId.slice(0, 8)}'s message list could not be read back) \u2014 the session was abandoned; a fresh one is used for further messages.`;
5436
+ this.log({
5437
+ level: "error",
5438
+ message: errorMessage,
5439
+ conversation_id: row.conversation_id,
5440
+ message_id: row.id
5441
+ });
5442
+ await this.markFailed(row.conversation_id, row.id, null, errorMessage).catch((markErr) => {
5443
+ this.log({
5444
+ level: "warn",
5445
+ message: `markFailed PATCH for message ${row.id.slice(0, 8)} (conversation ${row.conversation_id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
5446
+ conversation_id: row.conversation_id,
5447
+ message_id: row.id
5448
+ });
5449
+ });
5450
+ void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
5451
+ return;
5452
+ }
3401
5453
  this.log({
3402
5454
  level: "warn",
3403
- message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next drain`,
5455
+ message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back (${streak}/${MAX_IDENTICAL_REDRIVE_POLL_FAILURES}) \u2014 leaving un-tracked to retry next drain`,
3404
5456
  conversation_id: row.conversation_id,
3405
5457
  message_id: row.id
3406
5458
  });
3407
5459
  void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3408
5460
  return;
3409
5461
  }
5462
+ this.unconfirmedDispatchFailures.delete(row.id);
3410
5463
  this.registerReadopted(readoptConv, sessionId, readoptMessage, ocId, this.processedAtMs(row));
3411
5464
  this.dispatched.add(row.id);
3412
5465
  this.readopted.add(row.id);
@@ -3462,7 +5515,8 @@ var ChannelDriver = class {
3462
5515
  opencode_model: row.opencode_model,
3463
5516
  source_message_id: row.source_message_id,
3464
5517
  slack_user_id: row.slack_user_id,
3465
- attachments: row.attachments ?? null
5518
+ attachments: row.attachments ?? null,
5519
+ opencode_message_id: row.opencode_message_id
3466
5520
  };
3467
5521
  }
3468
5522
  /**
@@ -3601,6 +5655,47 @@ var ChannelDriver = class {
3601
5655
  }
3602
5656
  return false;
3603
5657
  }
5658
+ /**
5659
+ * Tri-state variant of the upward parentID membership walk (#721), used ONLY
5660
+ * by `isAnyDescendantSessionOngoing`. Walks the SAME cached
5661
+ * `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
5662
+ * `sessionBelongsTo`, which deliberately collapses "confirmed not a
5663
+ * descendant" and "the walk's fetch failed" into the same `false` (safe for
5664
+ * its OTHER callers: interaction attribution and the recovery-path
5665
+ * `isAnyDescendantSessionAlive`, both of which just retry next tick with no
5666
+ * safety consequence either way) — this variant keeps those two outcomes
5667
+ * SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
5668
+ * (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
5669
+ * not ongoing".
5670
+ *
5671
+ * Return contract:
5672
+ * - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
5673
+ * - `false` → the walk reached a definitive, parent-less root session
5674
+ * WITHOUT ever matching `rootSessionId` — `sessionId` is
5675
+ * CONFIRMED NOT a descendant of it.
5676
+ * - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
5677
+ * through the walk (`resolveSessionParent` returned `undefined`),
5678
+ * or the depth cap (32) was hit without a definitive answer (a
5679
+ * pathological/cyclic chain proves nothing either way). NEVER
5680
+ * treat this the same as `false` — see `sessionBelongsTo`'s own
5681
+ * doc comment above for why that collapse is safe THERE but not
5682
+ * here.
5683
+ *
5684
+ * `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
5685
+ * to the live-path descendant check, not a modification of shared code used
5686
+ * by interaction attribution or the recovery path.
5687
+ */
5688
+ async resolveSessionMembership(sessionId, rootSessionId) {
5689
+ let current = sessionId;
5690
+ for (let depth = 0; current && depth < 32; depth++) {
5691
+ if (current === rootSessionId) return true;
5692
+ const parent = await this.resolveSessionParent(current);
5693
+ if (parent === void 0) return null;
5694
+ if (parent === null) return false;
5695
+ current = parent;
5696
+ }
5697
+ return null;
5698
+ }
3604
5699
  /**
3605
5700
  * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3606
5701
  * `null` for a root session (no parent) and `undefined` when opencode is
@@ -3623,19 +5718,36 @@ var ChannelDriver = class {
3623
5718
  if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3624
5719
  return parent;
3625
5720
  }
5721
+ /**
5722
+ * OpenCode's synchronous default session title (e.g.
5723
+ * `"New session - 1737800000000"`), assigned immediately when a session is
5724
+ * created — before OpenCode's async LLM-based auto-titling later renames it
5725
+ * mid-turn (#549). Matched by this literal, case-sensitive prefix only; the
5726
+ * timestamp suffix's exact format is deliberately NOT matched, since the prefix
5727
+ * alone is the stable, cheap signal and over-anchoring on the timestamp
5728
+ * representation risks silently breaking if OpenCode ever changes it. Accepted
5729
+ * trade-off: a genuine LLM-assigned title that happens to literally start with
5730
+ * this prefix would also fail to latch (see `resolveSessionTitle`) —
5731
+ * vanishingly unlikely in practice, and deliberately not engineered around.
5732
+ */
5733
+ static OPENCODE_DEFAULT_TITLE_PREFIX = /^New session - /;
3626
5734
  /**
3627
5735
  * Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
3628
5736
  * status PATCH can carry it into the "Live sessions" list. Driver-level cache so
3629
5737
  * BOTH the watcher completion path and the restart-recovery re-adopt path (which
3630
5738
  * has no watcher) can use it. `conversationId` is passed only for log context.
3631
5739
  * Best-effort:
3632
- * - a resolved NON-EMPTY title is cached and terminal (a real session name
5740
+ * - a resolved NON-EMPTY title that does NOT match
5741
+ * `OPENCODE_DEFAULT_TITLE_PREFIX` is cached and terminal (a real session name
3633
5742
  * won't later un-name), so we do NOT re-GET `/session/:id` every tick;
3634
- * - while the title is still absent/empty we do NOT latch it — OpenCode names
3635
- * sessions asynchronously mid-turn, so an early call (e.g. at `processing`)
3636
- * must leave the cache unresolved and re-fetch on the next need so a later
3637
- * call (e.g. at `done`) picks up the name assigned in the meantime. Such a
3638
- * call returns `null` (omit the title on THIS PATCH) without caching;
5743
+ * - while the title is still absent, empty, or matches the OpenCode
5744
+ * placeholder prefix (#549) we do NOT latch it OpenCode names sessions
5745
+ * asynchronously mid-turn, so an early call (e.g. at `processing`) must leave
5746
+ * the cache unresolved and re-fetch on the next need so a later call (e.g. at
5747
+ * `done`) picks up the name assigned in the meantime. Such a call returns
5748
+ * `null` (omit the title on THIS PATCH) without caching. If a session is
5749
+ * never renamed, the title is omitted forever rather than ever persisting
5750
+ * the placeholder as a last resort;
3639
5751
  * - a failed request likewise leaves the cache unresolved (retry next need)
3640
5752
  * and returns `null` — it must NEVER throw or block completion.
3641
5753
  * A failure is logged with agent/session context (no silent catch).
@@ -3648,7 +5760,7 @@ var ChannelDriver = class {
3648
5760
  if (res.ok) {
3649
5761
  const body = await res.json();
3650
5762
  const title = body && typeof body.title === "string" ? body.title.trim() : "";
3651
- if (title.length > 0) {
5763
+ if (title.length > 0 && !_ChannelDriver.OPENCODE_DEFAULT_TITLE_PREFIX.test(title)) {
3652
5764
  this.sessionTitles.set(sessionId, title);
3653
5765
  return title;
3654
5766
  }
@@ -3668,6 +5780,54 @@ var ChannelDriver = class {
3668
5780
  }
3669
5781
  return null;
3670
5782
  }
5783
+ /**
5784
+ * Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
5785
+ * session title onto the conversation via the PLAIN conversation-update
5786
+ * endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
5787
+ * message-status endpoint `markProcessing`/`markDone` use. Deliberately a
5788
+ * separate, lighter call: it carries no `status`, so it cannot re-trigger the
5789
+ * `processing`/`done` transition side effects (Slack notices, activity-log
5790
+ * rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
5791
+ * ever touches `conversations.title`. That route (`routes/conversations.ts`)
5792
+ * skips a title write matching the stored value, so a redundant call with the
5793
+ * same title is a real no-op — it does not bump `updated_at`, which the
5794
+ * conversation list sorts and paginates on. (Note this is a DIFFERENT guard
5795
+ * from `threads.ts`'s "non-empty AND changed" one, which only covers the
5796
+ * message-status PATCH; the non-empty half is enforced here instead, by
5797
+ * `resolveSessionTitle` never returning an empty/placeholder title.)
5798
+ *
5799
+ * Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
5800
+ * is logged and the title is simply retried on the next heartbeat tick (the
5801
+ * caller only latches `titleSynced` on `true`).
5802
+ */
5803
+ async patchConversationTitle(conversationId, title) {
5804
+ try {
5805
+ const res = await this.fetchImpl(
5806
+ `${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
5807
+ {
5808
+ method: "PATCH",
5809
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
5810
+ body: JSON.stringify({ title })
5811
+ }
5812
+ );
5813
+ if (!res.ok) {
5814
+ this.log({
5815
+ level: "debug",
5816
+ message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
5817
+ conversation_id: conversationId
5818
+ });
5819
+ return false;
5820
+ }
5821
+ return true;
5822
+ } catch (err) {
5823
+ this.log({
5824
+ level: "debug",
5825
+ 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)}`,
5826
+ conversation_id: conversationId
5827
+ });
5828
+ return false;
5829
+ }
5830
+ }
3671
5831
  /**
3672
5832
  * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3673
5833
  * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
@@ -3726,6 +5886,84 @@ var ChannelDriver = class {
3726
5886
  }
3727
5887
  return false;
3728
5888
  }
5889
+ /**
5890
+ * LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
5891
+ * sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
5892
+ * in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
5893
+ *
5894
+ * Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
5895
+ * cross-check above): that method judges liveness from the child's OWN
5896
+ * TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
5897
+ * on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
5898
+ * path the local opencode server IS running, so its in-memory status map is
5899
+ * live and authoritative — and per ADR-0047 §4a ("the child has its own entry
5900
+ * [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
5901
+ * ENTIRE turn (including any tool call it is itself executing), not a
5902
+ * per-message transcript snapshot. This sidesteps the "child's own tool is
5903
+ * executing, between its step's completion and the next generation step"
5904
+ * transcript gap that a transcript-based check would need a second,
5905
+ * sustained-window bound to guard against — it is simply not derived from
5906
+ * message timestamps at all.
5907
+ *
5908
+ * Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
5909
+ * status, as the recovery path does per §4a)? Because on the LIVE path the
5910
+ * root session can be shared: a SECOND, unrelated user message can land on the
5911
+ * SAME session (issue #721's own root cause) and keep the root `busy` for a
5912
+ * reason that has nothing to do with THIS message's delegation. A `task`
5913
+ * descendant session is spawned for exactly one delegated turn and never
5914
+ * reused, so its OWN status-map entry is unambiguous evidence about that one
5915
+ * delegation — which the root's status is not.
5916
+ *
5917
+ * Why membership is checked via `resolveSessionMembership`, NOT
5918
+ * `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
5919
+ * `GET /session/:id` fetch failure into "not a descendant", which would
5920
+ * silently drop a genuinely-live candidate from consideration on the one
5921
+ * unlucky tick its membership-walk fetch hiccups (#721).
5922
+ * `resolveSessionMembership` keeps that failure mode as a distinct `null`
5923
+ * (indeterminate) so it is folded into THIS method's own `indeterminate` flag
5924
+ * instead.
5925
+ *
5926
+ * Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
5927
+ * - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
5928
+ * - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
5929
+ * confirmed either way (`resolveSessionMembership` never
5930
+ * returned `null`), and every CONFIRMED descendant's status read
5931
+ * succeeded and is not ongoing (includes "no descendant session
5932
+ * exists at all" — e.g. a plain, non-`task` tool call).
5933
+ * - `null` → INDETERMINATE: `listSessions` failed, OR at least one
5934
+ * candidate's MEMBERSHIP could not be confirmed
5935
+ * (`resolveSessionMembership` returned `null` — a fetch failure
5936
+ * or pathological chain partway through the parent walk), OR at
5937
+ * least one CONFIRMED descendant's `isSessionOngoing` read
5938
+ * failed — and no OTHER candidate was already confirmed `true`.
5939
+ * The caller MUST NOT treat `null` the same as `false` here
5940
+ * (unlike the recovery cross-check's contract) — see
5941
+ * `isB2AbandonmentConfirmed`.
5942
+ */
5943
+ async isAnyDescendantSessionOngoing(rootSessionId) {
5944
+ const sessions = await listSessions(this.port);
5945
+ if (!sessions) {
5946
+ this.log({
5947
+ level: "warn",
5948
+ message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
5949
+ });
5950
+ return null;
5951
+ }
5952
+ let indeterminate = false;
5953
+ for (const candidate of sessions) {
5954
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
5955
+ const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
5956
+ if (membership === null) {
5957
+ indeterminate = true;
5958
+ continue;
5959
+ }
5960
+ if (membership === false) continue;
5961
+ const ongoing = await isSessionOngoing(this.port, candidate.id);
5962
+ if (ongoing === true) return true;
5963
+ if (ongoing === null) indeterminate = true;
5964
+ }
5965
+ return indeterminate ? null : false;
5966
+ }
3729
5967
  /**
3730
5968
  * Cheap decision-telemetry label for a running row's LAST correlated reply
3731
5969
  * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
@@ -3854,24 +6092,54 @@ var ChannelDriver = class {
3854
6092
  }
3855
6093
  return messages;
3856
6094
  }
6095
+ /**
6096
+ * The `opencode_session_id` fragment of a status PATCH body — `{}` when this
6097
+ * conversation has ABANDONED that session (#553). The field is optional
6098
+ * server-side and an absent one leaves the persisted binding untouched, so
6099
+ * omitting it is how a routine status write stops resurrecting it.
6100
+ *
6101
+ * ONLY for writes whose sole cost is a lost deep link. The `processing` notice
6102
+ * degrades to no "View in Evident" link (the reaction swap still fires) and the
6103
+ * turn-failure notice is built from the PATCH's own `error` text with a link off
6104
+ * the persisted row — neither loses content the user came for. `markDone`
6105
+ * deliberately does NOT use this helper: the server fetches the reply text
6106
+ * THROUGH the session id it is given, so suppressing there would replace the
6107
+ * agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
6108
+ * `ensureSession` guard, not this suppression, is what makes the self-heal
6109
+ * stick.
6110
+ */
6111
+ sessionIdBody(sessionId, conversationId, messageId, status2) {
6112
+ if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
6113
+ this.log({
6114
+ level: "debug",
6115
+ message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
6116
+ conversation_id: conversationId,
6117
+ message_id: messageId
6118
+ });
6119
+ return {};
6120
+ }
3857
6121
  /**
3858
6122
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
3859
6123
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
3860
6124
  * opencode_session_id}` → `notifyMessageStarted` (hourglass→runner swap +
3861
6125
  * deep-linked "View in Evident" notice).
3862
6126
  *
3863
- * Return/throw contract (consumed by the watcher's swap-to-running guard):
3864
- * - returns `true` → the server transitioned the row to processing;
3865
- * - returns `false` → the server gave a DEFINITIVE "already-processing"
3866
- * answer (a non-retryable, non-auth status e.g. a
3867
- * conflict because a duplicate already transitioned it),
3868
- * so the caller treats it as already-started and does NOT
3869
- * retry;
3870
- * - throws `ChannelAuthError` on 401/403 (terminal auth failure);
3871
- * - throws on a TRANSIENT failure (retryable 5xx/429 status, or a
3872
- * network-level error from `fetch`) — i.e. NO definitive server response —
3873
- * so the caller leaves the message un-started and retries the swap on the
3874
- * next tick.
6127
+ * Outcome contract (consumed by the watcher's swap-to-running guard):
6128
+ * - resolves (`void`) → the server transitioned the row to
6129
+ * processing (or idempotently confirmed
6130
+ * already-processing that answer is
6131
+ * still a 200, never a refusal);
6132
+ * - throws `ChannelAuthError` → 401/403 (terminal auth failure);
6133
+ * - throws `ChannelTerminalError` → a definitive non-retryable, non-auth 4xx
6134
+ * (404 the row or its conversation is
6135
+ * gone, 400 the update was rejected).
6136
+ * Retrying cannot help;
6137
+ * - throws a plain `Error` → a TRANSIENT failure (retryable 5xx/429
6138
+ * status, or a network-level error from
6139
+ * `fetch`) — i.e. NO definitive server
6140
+ * response — so the caller leaves the
6141
+ * message un-started and retries the swap
6142
+ * on the next tick.
3875
6143
  * A single attempt (no internal retry): the watcher's per-tick loop is the
3876
6144
  * retry vehicle for the swap-to-running.
3877
6145
  */
@@ -3883,18 +6151,18 @@ var ChannelDriver = class {
3883
6151
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3884
6152
  body: JSON.stringify({
3885
6153
  status: "processing",
3886
- opencode_session_id: sessionId,
6154
+ ...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
3887
6155
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3888
6156
  ...title ? { title } : {}
3889
6157
  })
3890
6158
  }
3891
6159
  );
3892
6160
  this.assertAuth(res, "marking message as processing");
3893
- if (res.ok) return true;
6161
+ if (res.ok) return;
3894
6162
  if (isRetryableStatus(res.status)) {
3895
6163
  throw new Error(`marking message as processing: HTTP ${res.status}`);
3896
6164
  }
3897
- return false;
6165
+ throw new ChannelTerminalError(`marking message as processing: HTTP ${res.status}`, res.status);
3898
6166
  }
3899
6167
  /**
3900
6168
  * EXISTING combinedAuth completion route — idempotent (WI-CHAN-2). `PATCH
@@ -3932,6 +6200,11 @@ var ChannelDriver = class {
3932
6200
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3933
6201
  body: JSON.stringify({
3934
6202
  status: "done",
6203
+ // ALWAYS sent, even for a session this conversation has abandoned
6204
+ // (#553): the server reads the reply text back out of THIS session id
6205
+ // to deliver it. Omitting it would leave the user with "✅ Done!"
6206
+ // instead of the answer — a worse regression than the resurrection it
6207
+ // would prevent, which `ensureSession`'s guard handles anyway.
3935
6208
  opencode_session_id: sessionId,
3936
6209
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3937
6210
  ...title ? { title } : {},
@@ -3948,16 +6221,31 @@ var ChannelDriver = class {
3948
6221
  }
3949
6222
  /**
3950
6223
  * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3951
- * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3952
- * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3953
- * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3954
- * failure reason reaches the channel.
6224
+ * when provided (issue #182). Three states for `sessionId`:
6225
+ * - omitted (`undefined`) → don't send the field, leave the persisted
6226
+ * session untouched (unused today; kept for API symmetry).
6227
+ * - a real id (`string`) → send it, update the persisted session (the
6228
+ * turn-failure call sites: an errored OpenCode turn).
6229
+ * - explicit `null` → send it, CLEAR the persisted session (issue
6230
+ * #485's dispatch-handoff-failure call site: the session id still
6231
+ * exists but is wedged, so the next attempt must get a fresh one
6232
+ * instead of reusing it — see WI-1's server-side null-clearing PATCH).
3955
6233
  */
3956
- async markFailed(conversationId, messageId, sessionId, error2, usage) {
6234
+ async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
3957
6235
  const body = { status: "failed" };
3958
- if (sessionId !== void 0) body.opencode_session_id = sessionId;
6236
+ if (sessionId === null) {
6237
+ body.opencode_session_id = null;
6238
+ } else if (sessionId !== void 0) {
6239
+ Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
6240
+ }
3959
6241
  if (error2 !== void 0) body.error = error2;
3960
6242
  if (usage) Object.assign(body, usage);
6243
+ if (failure) {
6244
+ body.failure_kind = failure.kind;
6245
+ body.failure_provider_id = failure.providerId;
6246
+ body.failure_model_id = failure.modelId;
6247
+ body.failure_reason = failure.reason;
6248
+ }
3961
6249
  await this.callWithRetry(
3962
6250
  "marking message as failed",
3963
6251
  () => this.fetchImpl(
@@ -3970,6 +6258,29 @@ var ChannelDriver = class {
3970
6258
  )
3971
6259
  );
3972
6260
  }
6261
+ /**
6262
+ * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
6263
+ *
6264
+ * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
6265
+ * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
6266
+ * to the P1-2b zero-provider check — one extra loopback call to
6267
+ * `hasAnyConfiguredProvider`, only reached when the structured classifier
6268
+ * couldn't place it. Fails open (never throws): a fallback probe failure
6269
+ * (`null`/indeterminate) leaves the classification `null`, which produces
6270
+ * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
6271
+ */
6272
+ async classifyModelAuthFailure(messages, userMessageId) {
6273
+ const classified = messageFailure(messages, userMessageId);
6274
+ if (classified != null) return classified;
6275
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
6276
+ const hasProvider = await hasAnyConfiguredProvider(this.port);
6277
+ return applyZeroProviderFallback(
6278
+ classified,
6279
+ hasProvider,
6280
+ reply?.info?.providerID ?? null,
6281
+ reply?.info?.modelID ?? null
6282
+ );
6283
+ }
3973
6284
  /**
3974
6285
  * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3975
6286
  * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
@@ -4122,10 +6433,16 @@ var ChannelDriver = class {
4122
6433
  import chalk5 from "chalk";
4123
6434
  import ora2 from "ora";
4124
6435
  import { select as select2 } from "@inquirer/prompts";
6436
+ var INTERACTIVE_START_TIMEOUT_MS = 3e4;
4125
6437
  async function ensureOpenCodeRunning(ctx) {
4126
6438
  const healthCheck = await checkOpenCodeHealth(ctx.port);
4127
6439
  if (healthCheck.healthy) {
4128
- return { port: ctx.port, process: null, version: healthCheck.version ?? null };
6440
+ return {
6441
+ port: ctx.port,
6442
+ process: null,
6443
+ version: healthCheck.version ?? null,
6444
+ notReadyReason: null
6445
+ };
4129
6446
  }
4130
6447
  const runningInstances = await findHealthyOpenCodeInstances();
4131
6448
  if (runningInstances.length > 0) {
@@ -4146,7 +6463,7 @@ async function ensureOpenCodeRunning(ctx) {
4146
6463
  console.log(chalk5.yellow("Tip: Run with the correct port:"));
4147
6464
  console.log(
4148
6465
  chalk5.dim(
4149
- ` ${getCliName()} run --agent ${ctx.agentId} --port ${runningInstances[0].port}`
6466
+ ` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
4150
6467
  )
4151
6468
  );
4152
6469
  }
@@ -4166,14 +6483,22 @@ async function ensureOpenCodeRunning(ctx) {
4166
6483
  if (!ctx.interactive) {
4167
6484
  ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
4168
6485
  const proc = await startOpenCode(ctx.port);
4169
- const health = await waitForOpenCodeHealth(ctx.port, 3e4);
6486
+ const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
4170
6487
  if (!health.healthy) {
4171
- throw new Error(
4172
- `OpenCode failed to start on port ${ctx.port}. Install with: npm install -g opencode-ai`
4173
- );
6488
+ return {
6489
+ port: ctx.port,
6490
+ process: proc,
6491
+ version: null,
6492
+ notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
6493
+ };
4174
6494
  }
4175
6495
  ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
4176
- return { port: ctx.port, process: proc, version: health.version ?? null };
6496
+ return {
6497
+ port: ctx.port,
6498
+ process: proc,
6499
+ version: health.version ?? null,
6500
+ notReadyReason: null
6501
+ };
4177
6502
  }
4178
6503
  let port = ctx.port;
4179
6504
  if (isPortInUse(port)) {
@@ -4226,122 +6551,15 @@ Port ${port} is already in use.`));
4226
6551
  if (action === "start") {
4227
6552
  const spinner = ora2("Starting OpenCode...").start();
4228
6553
  const proc = await startOpenCode(port);
4229
- const health = await waitForOpenCodeHealth(port, 3e4);
6554
+ const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
4230
6555
  if (!health.healthy) {
4231
6556
  spinner.fail("Failed to start OpenCode");
4232
6557
  throw new Error("OpenCode failed to start");
4233
6558
  }
4234
6559
  spinner.stop();
4235
- return { port, process: proc, version: health.version ?? null };
4236
- }
4237
- return { port, process: null, version: null };
4238
- }
4239
-
4240
- // src/commands/agent-lookup.ts
4241
- async function readErrorMessage(response) {
4242
- const text = await response.text().catch(() => "");
4243
- if (!text) return response.statusText || void 0;
4244
- try {
4245
- const data = JSON.parse(text);
4246
- const message = data.message ?? data.error;
4247
- if (typeof message === "string" && message.trim()) {
4248
- return message;
4249
- }
4250
- } catch {
4251
- }
4252
- return text.trim() || response.statusText || void 0;
4253
- }
4254
- function authFailureHint(apiUrl, serverMessage) {
4255
- const reason = serverMessage ? `: ${serverMessage}` : "";
4256
- return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
4257
- }
4258
- async function resolveAgentIdFromKey(authHeader) {
4259
- const apiUrl = getApiUrlConfig();
4260
- try {
4261
- const response = await fetch(`${apiUrl}/me`, {
4262
- headers: { Authorization: authHeader }
4263
- });
4264
- if (response.status === 401) {
4265
- const serverMessage = await readErrorMessage(response);
4266
- return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
4267
- }
4268
- if (!response.ok) {
4269
- const serverMessage = await readErrorMessage(response);
4270
- return {
4271
- error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
4272
- };
4273
- }
4274
- const data = await response.json();
4275
- if (data.auth_type === "agent_key" && data.agent_id) {
4276
- return { agent_id: data.agent_id };
4277
- }
4278
- return {
4279
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
4280
- };
4281
- } catch (error2) {
4282
- const message = error2 instanceof Error ? error2.message : "Unknown error";
4283
- return { error: `Failed to resolve runner from key: ${message}` };
4284
- }
4285
- }
4286
- async function notifyAgentDisconnected(agentId, authHeader) {
4287
- const apiUrl = getApiUrlConfig();
4288
- try {
4289
- const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
4290
- method: "POST",
4291
- headers: { Authorization: authHeader }
4292
- });
4293
- if (!response.ok) {
4294
- const serverMessage = await readErrorMessage(response);
4295
- return {
4296
- ok: false,
4297
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
4298
- };
4299
- }
4300
- return { ok: true };
4301
- } catch (error2) {
4302
- return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
4303
- }
4304
- }
4305
- async function getAgentInfo(agentId, authHeader) {
4306
- const apiUrl = getApiUrlConfig();
4307
- try {
4308
- const response = await fetch(`${apiUrl}/runners/${agentId}`, {
4309
- headers: { Authorization: authHeader }
4310
- });
4311
- if (response.status === 401) {
4312
- const serverMessage = await readErrorMessage(response);
4313
- return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
4314
- }
4315
- if (response.status === 403) {
4316
- const serverMessage = await readErrorMessage(response);
4317
- return {
4318
- valid: false,
4319
- error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
4320
- };
4321
- }
4322
- if (response.status === 404) {
4323
- const serverMessage = await readErrorMessage(response);
4324
- return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
4325
- }
4326
- if (!response.ok) {
4327
- const serverMessage = await readErrorMessage(response);
4328
- return {
4329
- valid: false,
4330
- error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
4331
- };
4332
- }
4333
- const agent = await response.json();
4334
- if (agent.agent_type !== "local") {
4335
- return {
4336
- valid: false,
4337
- error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
4338
- };
4339
- }
4340
- return { valid: true, agent };
4341
- } catch (error2) {
4342
- const message = error2 instanceof Error ? error2.message : "Unknown error";
4343
- return { valid: false, error: `Failed to validate runner: ${message}` };
6560
+ return { port, process: proc, version: health.version ?? null, notReadyReason: null };
4344
6561
  }
6562
+ return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
4345
6563
  }
4346
6564
 
4347
6565
  // src/commands/run.ts
@@ -4349,6 +6567,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
4349
6567
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
4350
6568
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4351
6569
  var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
6570
+ var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
4352
6571
  function resolveLogLevel(options) {
4353
6572
  const accepted = Object.keys(LOG_LEVELS);
4354
6573
  const validate = (value, source) => {
@@ -4372,6 +6591,63 @@ function resolveLogLevel(options) {
4372
6591
  }
4373
6592
  return "info";
4374
6593
  }
6594
+ function resolveFileSyncDirectories(raw, homeDir) {
6595
+ const directories = [];
6596
+ for (const entry of raw ?? []) {
6597
+ const trimmed = entry.trim();
6598
+ if (trimmed === "") {
6599
+ throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
6600
+ }
6601
+ const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
6602
+ if (!isAbsolute2(expanded)) {
6603
+ throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
6604
+ }
6605
+ const normalized = resolvePath(expanded);
6606
+ if (parse(normalized).root === normalized) {
6607
+ throw new Error(
6608
+ `--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
6609
+ );
6610
+ }
6611
+ if (!directories.includes(normalized)) {
6612
+ directories.push(normalized);
6613
+ }
6614
+ }
6615
+ if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
6616
+ throw new Error(
6617
+ `--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
6618
+ );
6619
+ }
6620
+ return directories;
6621
+ }
6622
+ var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
6623
+ var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
6624
+ var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
6625
+ function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
6626
+ const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
6627
+ let raw;
6628
+ let source;
6629
+ if (options.opencodeStartTimeout !== void 0) {
6630
+ raw = options.opencodeStartTimeout;
6631
+ source = "--opencode-start-timeout";
6632
+ } else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
6633
+ raw = env[OPENCODE_START_TIMEOUT_ENV];
6634
+ source = OPENCODE_START_TIMEOUT_ENV;
6635
+ } else {
6636
+ return { timeoutMs: defaultMs, warnings: [] };
6637
+ }
6638
+ const trimmed = raw.trim();
6639
+ const seconds = Number(trimmed);
6640
+ const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
6641
+ if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
6642
+ return {
6643
+ timeoutMs: defaultMs,
6644
+ warnings: [
6645
+ `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`
6646
+ ]
6647
+ };
6648
+ }
6649
+ return { timeoutMs: seconds * 1e3, warnings: [] };
6650
+ }
4375
6651
  function meetsThreshold(state, level) {
4376
6652
  return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
4377
6653
  }
@@ -4393,6 +6669,10 @@ function log2(state, message, level = "info") {
4393
6669
  function logActivity(state, entry) {
4394
6670
  const level = entry.level ?? (entry.type === "error" ? "error" : "info");
4395
6671
  if (!meetsThreshold(state, level)) return;
6672
+ forwardRunnerActivity(
6673
+ { level, message: entry.message, error: entry.error },
6674
+ { agentId: state.agentId, authHeader: state.authHeader }
6675
+ );
4396
6676
  const fullEntry = {
4397
6677
  ...entry,
4398
6678
  level,
@@ -4492,23 +6772,46 @@ async function handleAuthError(state, error2) {
4492
6772
  }
4493
6773
  async function driveChannels(state, driver) {
4494
6774
  let idlePolls = 0;
6775
+ let idleMs = 0;
6776
+ let consecutiveDrainFailures = 0;
6777
+ let unreachableMs = 0;
4495
6778
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
6779
+ let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
4496
6780
  while (state.running) {
6781
+ const cycleStartedAtMs = performance.now();
6782
+ let idleThisCycle = false;
6783
+ let unreachableThisCycle = false;
4497
6784
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
4498
6785
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
4499
6786
  if (state.interactive) displayStatus(state);
4500
6787
  await state.connection.reconnectPromise;
4501
6788
  }
6789
+ const carriedOverFileSync = driver.fileSyncActivity().inFlight;
6790
+ void driver.syncPendingFiles().catch(
6791
+ (error2) => logActivity(state, {
6792
+ type: "error",
6793
+ error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
6794
+ })
6795
+ );
4502
6796
  try {
4503
6797
  const processed = await driver.drainPending();
6798
+ consecutiveDrainFailures = 0;
6799
+ unreachableMs = 0;
4504
6800
  state.messageCount += processed;
4505
6801
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4506
6802
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4507
- if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
6803
+ const appliedFiles = driver.fileSyncActivity().appliedFiles;
6804
+ const filesApplied = appliedFiles !== lastSeenAppliedFiles;
6805
+ const fileActivity = carriedOverFileSync || filesApplied;
6806
+ lastSeenAppliedFiles = appliedFiles;
6807
+ if (filesApplied) state.claudeUsageRearm?.();
6808
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
4508
6809
  idlePolls = 0;
6810
+ idleMs = 0;
4509
6811
  if (processed > 0 && state.interactive) displayStatus(state);
4510
6812
  } else if (state.idleTimeout !== null) {
4511
6813
  idlePolls++;
6814
+ idleThisCycle = true;
4512
6815
  if (idlePolls === 1) {
4513
6816
  logActivity(state, {
4514
6817
  type: "info",
@@ -4532,21 +6835,44 @@ async function driveChannels(state, driver) {
4532
6835
  const errorMessage = error2 instanceof Error ? error2.message : String(error2);
4533
6836
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
4534
6837
  if (state.interactive) displayStatus(state);
4535
- }
4536
- await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
4537
- if (state.idleTimeout !== null && idlePolls >= 2) {
4538
- const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
4539
- if (idleMs > state.idleTimeout * 1e3) {
4540
- logActivity(state, { type: "info", message: "Idle timeout reached" });
4541
- if (state.interactive) displayStatus(state);
4542
- break;
6838
+ if (driver.hasInFlightWatchers()) {
6839
+ consecutiveDrainFailures = 0;
6840
+ unreachableMs = 0;
6841
+ } else if (state.idleTimeout !== null) {
6842
+ consecutiveDrainFailures++;
6843
+ unreachableThisCycle = true;
6844
+ if (consecutiveDrainFailures === 1) {
6845
+ logActivity(state, {
6846
+ type: "info",
6847
+ message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
6848
+ });
6849
+ if (state.interactive) displayStatus(state);
6850
+ }
4543
6851
  }
4544
6852
  }
6853
+ await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
6854
+ const cycleMs = performance.now() - cycleStartedAtMs;
6855
+ if (idleThisCycle) idleMs += cycleMs;
6856
+ if (unreachableThisCycle) unreachableMs += cycleMs;
6857
+ if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
6858
+ logActivity(state, {
6859
+ type: "info",
6860
+ level: "warn",
6861
+ message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
6862
+ });
6863
+ if (state.interactive) displayStatus(state);
6864
+ break;
6865
+ }
6866
+ if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
6867
+ logActivity(state, { type: "info", message: "Idle timeout reached" });
6868
+ if (state.interactive) displayStatus(state);
6869
+ break;
6870
+ }
4545
6871
  }
4546
6872
  }
4547
6873
  var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
4548
- async function runSweep(state, driver, config2) {
4549
- const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
6874
+ async function runSweep(state, driver, config) {
6875
+ const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
4550
6876
  try {
4551
6877
  const sessions = await listSessions(state.port);
4552
6878
  if (sessions === null) {
@@ -4559,8 +6885,8 @@ async function runSweep(state, driver, config2) {
4559
6885
  const toDelete = selectSessionsToDelete(
4560
6886
  sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
4561
6887
  {
4562
- maxAgeMs: config2.maxAgeMs,
4563
- maxCount: config2.maxCount,
6888
+ maxAgeMs: config.maxAgeMs,
6889
+ maxCount: config.maxCount,
4564
6890
  nowMs: Date.now(),
4565
6891
  protectedIds: driver.protectedSessionIds()
4566
6892
  }
@@ -4596,7 +6922,7 @@ async function runSweep(state, driver, config2) {
4596
6922
  }
4597
6923
  }
4598
6924
  function scheduleSessionCleanup(state, driver, options) {
4599
- const config2 = resolveSessionCleanupConfig(
6925
+ const config = resolveSessionCleanupConfig(
4600
6926
  {
4601
6927
  maxAge: options.sessionCleanupMaxAge,
4602
6928
  maxCount: options.sessionCleanupMaxCount,
@@ -4604,21 +6930,129 @@ function scheduleSessionCleanup(state, driver, options) {
4604
6930
  },
4605
6931
  process.env
4606
6932
  );
4607
- for (const warning2 of config2.warnings) {
6933
+ for (const warning2 of config.warnings) {
4608
6934
  logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
4609
6935
  }
4610
- if (!config2.enabled) return;
6936
+ if (!config.enabled) return;
4611
6937
  logActivity(state, {
4612
6938
  type: "info",
4613
- message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
6939
+ message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
4614
6940
  });
4615
- const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
6941
+ const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
4616
6942
  const firstSweep = setTimeout(
4617
- () => void runSweep(state, driver, config2),
6943
+ () => void runSweep(state, driver, config),
4618
6944
  SESSION_CLEANUP_FIRST_SWEEP_MS
4619
6945
  );
4620
6946
  state.sessionCleanupTimers.push(interval, firstSweep);
4621
6947
  }
6948
+ function claudeUsageFailureStreakSuffix(consecutiveFailures) {
6949
+ return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
6950
+ }
6951
+ function scheduleClaudeUsageReporting(state, options) {
6952
+ const { mode, warnings } = resolveClaudeUsageReportingMode(
6953
+ options.claudeUsageReporting,
6954
+ process.env
6955
+ );
6956
+ for (const warning2 of warnings) {
6957
+ logActivity(state, {
6958
+ type: "info",
6959
+ level: "warn",
6960
+ message: `Claude usage reporting: ${warning2}`
6961
+ });
6962
+ }
6963
+ if (mode === "off") {
6964
+ logActivity(state, {
6965
+ type: "info",
6966
+ level: "debug",
6967
+ message: "Claude usage reporting is off (--claude-usage-reporting off)"
6968
+ });
6969
+ return null;
6970
+ }
6971
+ let consecutiveFailures = 0;
6972
+ let armed = false;
6973
+ let rearmRequested = false;
6974
+ const scheduleNextTick = () => {
6975
+ armed = true;
6976
+ rearmRequested = false;
6977
+ state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
6978
+ };
6979
+ const rearm = () => {
6980
+ if (armed) {
6981
+ rearmRequested = true;
6982
+ return;
6983
+ }
6984
+ rearmRequested = false;
6985
+ armed = true;
6986
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6987
+ };
6988
+ const tick = async (isProbe) => {
6989
+ try {
6990
+ const usage = await getClaudeUsage();
6991
+ const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
6992
+ if (result.ok) {
6993
+ if (consecutiveFailures > 0) {
6994
+ logActivity(state, {
6995
+ type: "info",
6996
+ level: "info",
6997
+ message: "Claude usage reporting recovered"
6998
+ });
6999
+ }
7000
+ consecutiveFailures = 0;
7001
+ logActivity(state, {
7002
+ type: "info",
7003
+ level: "debug",
7004
+ message: "Reported Claude usage to Evident"
7005
+ });
7006
+ } else {
7007
+ consecutiveFailures++;
7008
+ logActivity(state, {
7009
+ type: "info",
7010
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
7011
+ message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
7012
+ });
7013
+ }
7014
+ scheduleNextTick();
7015
+ } catch (error2) {
7016
+ if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
7017
+ if (mode === "on") {
7018
+ logActivity(state, {
7019
+ type: "info",
7020
+ level: "warn",
7021
+ 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"
7022
+ });
7023
+ scheduleNextTick();
7024
+ } else if (isProbe) {
7025
+ logActivity(state, {
7026
+ type: "info",
7027
+ level: "debug",
7028
+ message: `Claude usage reporting: ${error2.message}`
7029
+ });
7030
+ armed = false;
7031
+ if (rearmRequested) rearm();
7032
+ } else {
7033
+ logActivity(state, {
7034
+ type: "info",
7035
+ level: "debug",
7036
+ message: `Claude usage reporting: ${error2.message}`
7037
+ });
7038
+ scheduleNextTick();
7039
+ }
7040
+ } else {
7041
+ consecutiveFailures++;
7042
+ const message = error2 instanceof Error ? error2.message : String(error2);
7043
+ logActivity(state, {
7044
+ type: "info",
7045
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
7046
+ message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
7047
+ });
7048
+ scheduleNextTick();
7049
+ }
7050
+ }
7051
+ };
7052
+ armed = true;
7053
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
7054
+ return rearm;
7055
+ }
4622
7056
  async function notifyOffline(state) {
4623
7057
  if (!state.agentId || !state.authHeader) return;
4624
7058
  if (!state.connected) {
@@ -4636,13 +7070,29 @@ async function notifyOffline(state) {
4636
7070
  if (state.interactive) displayStatus(state);
4637
7071
  }
4638
7072
  }
7073
+ async function timeShutdownPhase(state, durations, name, run2) {
7074
+ const startedAt = Date.now();
7075
+ try {
7076
+ return await run2();
7077
+ } finally {
7078
+ const elapsedMs = Date.now() - startedAt;
7079
+ durations[name] = elapsedMs;
7080
+ log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
7081
+ }
7082
+ }
4639
7083
  async function cleanup(state, opts = {}) {
7084
+ const durations = {};
4640
7085
  state.running = false;
4641
7086
  for (const timer of state.sessionCleanupTimers) {
4642
7087
  clearInterval(timer);
4643
7088
  clearTimeout(timer);
4644
7089
  }
4645
7090
  state.sessionCleanupTimers = [];
7091
+ if (state.claudeUsageTimer) {
7092
+ clearTimeout(state.claudeUsageTimer);
7093
+ state.claudeUsageTimer = null;
7094
+ }
7095
+ state.claudeUsageRearm = null;
4646
7096
  if (opts.graceful && state.channelDriver) {
4647
7097
  state.channelDriver.stop();
4648
7098
  log2(state, "Draining in-flight channel work before shutdown...");
@@ -4650,7 +7100,13 @@ async function cleanup(state, opts = {}) {
4650
7100
  logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4651
7101
  displayStatus(state);
4652
7102
  }
4653
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
7103
+ const driver = state.channelDriver;
7104
+ const settled = await timeShutdownPhase(
7105
+ state,
7106
+ durations,
7107
+ "drain",
7108
+ () => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
7109
+ );
4654
7110
  if (!settled) {
4655
7111
  logActivity(state, {
4656
7112
  type: "info",
@@ -4659,13 +7115,15 @@ async function cleanup(state, opts = {}) {
4659
7115
  if (state.interactive) displayStatus(state);
4660
7116
  }
4661
7117
  }
4662
- await notifyOffline(state);
7118
+ await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
4663
7119
  if (state.connection) {
4664
- state.connection.close();
7120
+ const connection = state.connection;
7121
+ await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
4665
7122
  state.connection = null;
4666
7123
  }
4667
7124
  if (state.opencodeProcess) {
4668
- stopOpenCode(state.opencodeProcess);
7125
+ const opencodeProcess = state.opencodeProcess;
7126
+ await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
4669
7127
  if (state.interactive) {
4670
7128
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
4671
7129
  displayStatus(state);
@@ -4674,12 +7132,15 @@ async function cleanup(state, opts = {}) {
4674
7132
  }
4675
7133
  state.opencodeProcess = null;
4676
7134
  }
7135
+ return durations;
4677
7136
  }
4678
7137
  async function run(options) {
4679
7138
  const interactive = isInteractive(options.json);
4680
7139
  let logLevel;
7140
+ let fileSyncDirectories;
4681
7141
  try {
4682
7142
  logLevel = resolveLogLevel(options);
7143
+ fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
4683
7144
  } catch (error2) {
4684
7145
  const message = error2 instanceof Error ? error2.message : String(error2);
4685
7146
  if (options.json) {
@@ -4712,8 +7173,29 @@ async function run(options) {
4712
7173
  messageCount: 0,
4713
7174
  lastProxiedActivityAt: null,
4714
7175
  sessionCleanupTimers: [],
7176
+ claudeUsageTimer: null,
7177
+ claudeUsageRearm: null,
4715
7178
  authHeader: ""
4716
7179
  };
7180
+ setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
7181
+ if (fileSyncDirectories.length > 0) {
7182
+ log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
7183
+ } else {
7184
+ log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
7185
+ }
7186
+ if (!options.runner && options.agent) {
7187
+ telemetry.info(
7188
+ EventTypes.DEPRECATED_AGENT_FLAG_USED,
7189
+ "Deprecated --agent flag used instead of --runner",
7190
+ { command: "run" },
7191
+ state.agentId
7192
+ );
7193
+ const agentFlagNotice = "--agent is deprecated, use --runner instead; will be removed in a future release.";
7194
+ log2(state, agentFlagNotice, "warn");
7195
+ if (state.interactive && !state.json) {
7196
+ logActivity(state, { type: "info", level: "warn", message: agentFlagNotice });
7197
+ }
7198
+ }
4717
7199
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
4718
7200
  log2(
4719
7201
  state,
@@ -4724,14 +7206,38 @@ async function run(options) {
4724
7206
  const handleSignal = async () => {
4725
7207
  if (state.shuttingDown) return;
4726
7208
  state.shuttingDown = true;
7209
+ const shutdownStartedAt = Date.now();
4727
7210
  if (state.interactive) {
4728
7211
  logActivity(state, { type: "info", message: "Shutting down..." });
4729
7212
  displayStatus(state);
4730
7213
  } else {
4731
7214
  log2(state, "Shutting down...");
4732
7215
  }
4733
- await cleanup(state, { graceful: true });
4734
- await shutdownTelemetry();
7216
+ const durations = await cleanup(state, { graceful: true });
7217
+ const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
7218
+ await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
7219
+ let timer;
7220
+ const flushed = shutdownTelemetry().then(
7221
+ () => true,
7222
+ (error2) => {
7223
+ log2(
7224
+ state,
7225
+ `Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
7226
+ "warn"
7227
+ );
7228
+ return true;
7229
+ }
7230
+ );
7231
+ const timedOut = new Promise((resolve3) => {
7232
+ timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
7233
+ });
7234
+ if (!await Promise.race([flushed, timedOut])) {
7235
+ log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
7236
+ }
7237
+ clearTimeout(timer);
7238
+ });
7239
+ const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
7240
+ log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
4735
7241
  process.exit(0);
4736
7242
  };
4737
7243
  process.on("SIGINT", handleSignal);
@@ -4748,6 +7254,7 @@ async function run(options) {
4748
7254
  console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
4749
7255
  blank();
4750
7256
  process.exit(1);
7257
+ return;
4751
7258
  }
4752
7259
  blank();
4753
7260
  console.log(chalk6.yellow("You are not logged in to Evident."));
@@ -4764,6 +7271,19 @@ async function run(options) {
4764
7271
  logActivity(state, { type: "info", level: "warn", message: credentials2.notice });
4765
7272
  }
4766
7273
  }
7274
+ if (credentials2.keySource === "agent_key") {
7275
+ telemetry.info(
7276
+ EventTypes.DEPRECATED_AGENT_KEY_ENV_USED,
7277
+ "Deprecated EVIDENT_AGENT_KEY env var used instead of EVIDENT_RUNNER_KEY",
7278
+ { command: "run" },
7279
+ state.agentId
7280
+ );
7281
+ const agentKeyNotice = "EVIDENT_AGENT_KEY is deprecated, use EVIDENT_RUNNER_KEY instead; will be removed in a future release.";
7282
+ log2(state, agentKeyNotice, "warn");
7283
+ if (state.interactive && !state.json) {
7284
+ logActivity(state, { type: "info", level: "warn", message: agentKeyNotice });
7285
+ }
7286
+ }
4767
7287
  if (!state.agentId) {
4768
7288
  if (credentials2.authType === "agent_key") {
4769
7289
  const resolved = await resolveAgentIdFromKey(state.authHeader);
@@ -4779,6 +7299,7 @@ async function run(options) {
4779
7299
  } else {
4780
7300
  printError(resolved.error || "Failed to resolve runner ID from key");
4781
7301
  process.exit(1);
7302
+ return;
4782
7303
  }
4783
7304
  } else {
4784
7305
  printError(
@@ -4792,6 +7313,7 @@ async function run(options) {
4792
7313
  );
4793
7314
  blank();
4794
7315
  process.exit(1);
7316
+ return;
4795
7317
  }
4796
7318
  }
4797
7319
  telemetry.info(
@@ -4832,25 +7354,67 @@ async function run(options) {
4832
7354
  }
4833
7355
  spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
4834
7356
  state.agentName = validation.agent.name;
7357
+ const microvmId = process.env.MICROVM_ID?.trim();
7358
+ if (microvmId) {
7359
+ const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
7360
+ if (reported.ok) {
7361
+ log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
7362
+ } else {
7363
+ const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
7364
+ log2(state, message, "warn");
7365
+ if (state.interactive && !state.json) {
7366
+ logActivity(state, { type: "info", level: "warn", message });
7367
+ }
7368
+ }
7369
+ } else {
7370
+ log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
7371
+ }
7372
+ const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
7373
+ for (const warning2 of opencodeStartTimeoutWarnings) {
7374
+ logActivity(state, { type: "info", level: "warn", message: warning2 });
7375
+ }
4835
7376
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
4836
7377
  try {
4837
7378
  const oc = await ensureOpenCodeRunning({
4838
7379
  port: state.port,
4839
7380
  interactive: state.interactive,
4840
7381
  agentId: state.agentId,
4841
- log: (message) => log2(state, message)
7382
+ log: (message) => log2(state, message),
7383
+ startTimeoutMs: opencodeStartTimeoutMs
4842
7384
  });
4843
7385
  state.port = oc.port;
4844
7386
  state.opencodeProcess = oc.process;
4845
7387
  state.opencodeVersion = oc.version;
4846
- state.opencodeConnected = oc.process !== null || oc.version !== null;
7388
+ state.opencodeConnected = oc.notReadyReason === null;
4847
7389
  const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
4848
7390
  ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
4849
- const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
4850
- if (versionWarning) {
4851
- log2(state, versionWarning, "warn");
4852
- if (state.interactive && !state.json) {
4853
- logActivity(state, { type: "info", level: "warn", message: versionWarning });
7391
+ if (!state.interactive && oc.notReadyReason !== null) {
7392
+ 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}).`;
7393
+ logActivity(state, { type: "info", level: "warn", message });
7394
+ } else {
7395
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
7396
+ if (versionWarning) {
7397
+ log2(state, versionWarning, "warn");
7398
+ if (state.interactive && !state.json) {
7399
+ logActivity(state, { type: "info", level: "warn", message: versionWarning });
7400
+ }
7401
+ }
7402
+ const noProviderWarning = buildNoProviderWarning(
7403
+ await hasAnyConfiguredProvider(state.port)
7404
+ );
7405
+ if (noProviderWarning) {
7406
+ log2(state, noProviderWarning, "warn");
7407
+ if (state.interactive && !state.json) {
7408
+ logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
7409
+ blank();
7410
+ console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
7411
+ console.log(
7412
+ chalk6.dim(
7413
+ `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
7414
+ )
7415
+ );
7416
+ blank();
7417
+ }
4854
7418
  }
4855
7419
  }
4856
7420
  } catch (error2) {
@@ -4865,6 +7429,10 @@ async function run(options) {
4865
7429
  getAuthHeader: () => state.authHeader,
4866
7430
  conversationFilter: state.conversationFilter,
4867
7431
  stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
7432
+ // #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
7433
+ // REJECTED with `file_sync_disabled` on the ack, not silently ignored.
7434
+ fileSyncDirectories,
7435
+ homeDir: homedir3(),
4868
7436
  log: (entry) => (
4869
7437
  // Thread the driver's real level straight through so `debug`/`warn`
4870
7438
  // survive the sink filter (they no longer collapse to info). `type`
@@ -4891,6 +7459,18 @@ async function run(options) {
4891
7459
  type: "info",
4892
7460
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
4893
7461
  });
7462
+ if (options.tunnelReadyFile) {
7463
+ const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
7464
+ if (marker.ok) {
7465
+ log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
7466
+ } else {
7467
+ log2(
7468
+ state,
7469
+ `Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
7470
+ "error"
7471
+ );
7472
+ }
7473
+ }
4894
7474
  emitAgentConnected(state.agentId, {
4895
7475
  port: state.port,
4896
7476
  cli_version: getCliVersion(),
@@ -4946,6 +7526,12 @@ async function run(options) {
4946
7526
  onDrainPing: () => {
4947
7527
  if (!state.running) return;
4948
7528
  logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
7529
+ void channelDriver.syncPendingFiles().catch(
7530
+ (error2) => logActivity(state, {
7531
+ type: "error",
7532
+ error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
7533
+ })
7534
+ );
4949
7535
  channelDriver.drainPending().then((processed) => {
4950
7536
  if (processed > 0) {
4951
7537
  state.messageCount += processed;
@@ -4975,6 +7561,7 @@ async function run(options) {
4975
7561
  throw error2;
4976
7562
  }
4977
7563
  scheduleSessionCleanup(state, channelDriver, options);
7564
+ state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
4978
7565
  if (!interactive || state.json) {
4979
7566
  log2(state, "Driving channel messages...");
4980
7567
  }
@@ -5029,10 +7616,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
5029
7616
  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);
5030
7617
  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 }));
5031
7618
  program.command("whoami").description("Show the currently logged in user").action(whoami);
5032
- 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(
7619
+ program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
7620
+ program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
7621
+ 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(
7622
+ "-a, --agent [id]",
7623
+ "Deprecated alias for --runner (still supported; --runner wins if both are given)"
7624
+ ).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
5033
7625
  "--log-level <level>",
5034
7626
  "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
5035
- ).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(
7627
+ ).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(
7628
+ "--opencode-start-timeout <seconds>",
7629
+ "Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
7630
+ ).option("--json", "Output in JSON format").option(
5036
7631
  "--session-cleanup-max-age <duration>",
5037
7632
  "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
5038
7633
  ).option(
@@ -5041,6 +7636,17 @@ program.command("run").description("Connect to Evident and process messages").op
5041
7636
  ).option(
5042
7637
  "--session-cleanup-interval <duration>",
5043
7638
  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
7639
+ ).option(
7640
+ "--claude-usage-reporting <mode>",
7641
+ "Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
7642
+ ).option(
7643
+ "--enable-file-sync-to <dir>",
7644
+ "Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
7645
+ (value, previous) => previous.concat([value]),
7646
+ []
7647
+ ).option(
7648
+ "--tunnel-ready-file <path>",
7649
+ "Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
5044
7650
  ).action(
5045
7651
  (options) => {
5046
7652
  run({
@@ -5053,11 +7659,21 @@ program.command("run").description("Connect to Evident and process messages").op
5053
7659
  verbose: options.verbose,
5054
7660
  conversation: options.conversation,
5055
7661
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
7662
+ // Raw string — validation/precedence is single-sourced in run.ts's
7663
+ // resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
7664
+ opencodeStartTimeout: options.opencodeStartTimeout,
5056
7665
  json: options.json,
5057
7666
  // Raw strings — the resolver in run.ts single-sources parsing (M1).
5058
7667
  sessionCleanupMaxAge: options.sessionCleanupMaxAge,
5059
7668
  sessionCleanupMaxCount: options.sessionCleanupMaxCount,
5060
- sessionCleanupInterval: options.sessionCleanupInterval
7669
+ sessionCleanupInterval: options.sessionCleanupInterval,
7670
+ // Raw string — the resolver in run.ts single-sources parsing
7671
+ // (resolveClaudeUsageReportingMode).
7672
+ claudeUsageReporting: options.claudeUsageReporting,
7673
+ // Raw values — expansion/validation is single-sourced in run.ts's
7674
+ // resolveFileSyncDirectories.
7675
+ enableFileSyncTo: options.enableFileSyncTo,
7676
+ tunnelReadyFile: options.tunnelReadyFile
5061
7677
  });
5062
7678
  }
5063
7679
  );