@evident-ai/cli 3.1.1-dev.3a785a9 → 3.1.1-dev.3b21f0c

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.js CHANGED
@@ -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();
@@ -466,192 +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
- // Deprecation telemetry (#412) — usage of the old `--agent`/`EVIDENT_AGENT_KEY`
650
- // names instead of the preferred `--runner`/`EVIDENT_RUNNER_KEY` (#409).
651
- DEPRECATED_AGENT_FLAG_USED: "cli.deprecated_agent_flag_used",
652
- DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
653
- };
654
-
655
535
  // src/lib/auth.ts
656
536
  async function getAuthCredentials() {
657
537
  const runnerKey = process.env.EVIDENT_RUNNER_KEY;
@@ -695,38 +575,745 @@ function isInteractive(jsonOutput) {
695
575
  return true;
696
576
  }
697
577
 
698
- // src/lib/opencode/health.ts
699
- 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;
700
582
  try {
701
- const response = await fetch(`http://127.0.0.1:${port}/global/health`, {
702
- signal: AbortSignal.timeout(2e3)
703
- // 2 second timeout
704
- });
705
- if (!response.ok) {
706
- return { healthy: false, error: `HTTP ${response.status}` };
583
+ const data = JSON.parse(text);
584
+ const message = data.message ?? data.error;
585
+ if (typeof message === "string" && message.trim()) {
586
+ return message;
707
587
  }
708
- const data = await response.json().catch(() => ({}));
709
- return { healthy: true, version: data.version };
710
- } catch (error2) {
711
- const message = error2 instanceof Error ? error2.message : "Unknown error";
712
- return { healthy: false, error: message };
588
+ } catch {
713
589
  }
590
+ return text.trim() || response.statusText || void 0;
714
591
  }
715
- async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
716
- const startTime = Date.now();
717
- while (Date.now() - startTime < timeoutMs) {
718
- const health = await checkOpenCodeHealth(port);
719
- if (health.healthy) {
720
- return health;
721
- }
722
- await new Promise((resolve2) => setTimeout(resolve2, 1e3));
723
- }
724
- return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
592
+ function authFailureHint(apiUrl, serverMessage) {
593
+ const reason = serverMessage ? `: ${serverMessage}` : "";
594
+ return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
725
595
  }
726
-
727
- // src/lib/opencode/opencode-version-gate.ts
728
- var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
729
- function isQueueValidatedVersion(version2) {
596
+ async function resolveAgentIdFromKey(authHeader) {
597
+ const apiUrl = getApiUrlConfig();
598
+ try {
599
+ const response = await fetch(`${apiUrl}/me`, {
600
+ headers: { Authorization: authHeader }
601
+ });
602
+ if (response.status === 401) {
603
+ const serverMessage = await readErrorMessage(response);
604
+ return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
605
+ }
606
+ if (!response.ok) {
607
+ const serverMessage = await readErrorMessage(response);
608
+ return {
609
+ error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
610
+ };
611
+ }
612
+ const data = await response.json();
613
+ if (data.auth_type === "agent_key" && data.agent_id) {
614
+ return { agent_id: data.agent_id };
615
+ }
616
+ return {
617
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
618
+ };
619
+ } catch (error2) {
620
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
621
+ return { error: `Failed to resolve runner from key: ${message}` };
622
+ }
623
+ }
624
+ var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
625
+ async function notifyAgentDisconnected(agentId, authHeader) {
626
+ const apiUrl = getApiUrlConfig();
627
+ try {
628
+ const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
629
+ method: "POST",
630
+ headers: { Authorization: authHeader },
631
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
632
+ });
633
+ if (!response.ok) {
634
+ const serverMessage = await readErrorMessage(response);
635
+ return {
636
+ ok: false,
637
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
638
+ };
639
+ }
640
+ return { ok: true };
641
+ } catch (error2) {
642
+ return { ok: false, error: describeBestEffortError(error2) };
643
+ }
644
+ }
645
+ function describeBestEffortError(error2) {
646
+ const name = error2?.name;
647
+ if (name === "TimeoutError" || name === "AbortError") {
648
+ return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
649
+ }
650
+ return error2 instanceof Error ? error2.message : String(error2);
651
+ }
652
+ async function reportMicrovmId(agentId, authHeader, microvmId) {
653
+ try {
654
+ const apiUrl = getApiUrlConfig();
655
+ const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
656
+ method: "POST",
657
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
658
+ body: JSON.stringify({ microvm_id: microvmId }),
659
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
660
+ });
661
+ if (!response.ok) {
662
+ const serverMessage = await readErrorMessage(response);
663
+ return {
664
+ ok: false,
665
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
666
+ };
667
+ }
668
+ return { ok: true };
669
+ } catch (error2) {
670
+ return { ok: false, error: describeBestEffortError(error2) };
671
+ }
672
+ }
673
+ function toReportedWindow(window) {
674
+ if (!window) return null;
675
+ return { utilization: window.utilization, resets_at: window.resetsAt };
676
+ }
677
+ async function reportClaudeUsage(agentId, authHeader, snapshot) {
678
+ try {
679
+ const apiUrl = getApiUrlConfig();
680
+ const response = await fetch(`${apiUrl}/runners/${agentId}/claude-usage`, {
681
+ method: "POST",
682
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
683
+ body: JSON.stringify({
684
+ five_hour: toReportedWindow(snapshot.fiveHour),
685
+ seven_day: toReportedWindow(snapshot.sevenDay)
686
+ }),
687
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
688
+ });
689
+ if (!response.ok) {
690
+ const serverMessage = await readErrorMessage(response);
691
+ return {
692
+ ok: false,
693
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
694
+ };
695
+ }
696
+ return { ok: true };
697
+ } catch (error2) {
698
+ return { ok: false, error: describeBestEffortError(error2) };
699
+ }
700
+ }
701
+ async function getAgentInfo(agentId, authHeader) {
702
+ const apiUrl = getApiUrlConfig();
703
+ try {
704
+ const response = await fetch(`${apiUrl}/runners/${agentId}`, {
705
+ headers: { Authorization: authHeader }
706
+ });
707
+ if (response.status === 401) {
708
+ const serverMessage = await readErrorMessage(response);
709
+ return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
710
+ }
711
+ if (response.status === 403) {
712
+ const serverMessage = await readErrorMessage(response);
713
+ return {
714
+ valid: false,
715
+ error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
716
+ };
717
+ }
718
+ if (response.status === 404) {
719
+ const serverMessage = await readErrorMessage(response);
720
+ return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
721
+ }
722
+ if (!response.ok) {
723
+ const serverMessage = await readErrorMessage(response);
724
+ return {
725
+ valid: false,
726
+ error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
727
+ };
728
+ }
729
+ const agent = await response.json();
730
+ if (agent.agent_type !== "local") {
731
+ return {
732
+ valid: false,
733
+ error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
734
+ };
735
+ }
736
+ return { valid: true, agent };
737
+ } catch (error2) {
738
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
739
+ return { valid: false, error: `Failed to validate runner: ${message}` };
740
+ }
741
+ }
742
+
743
+ // src/commands/status.ts
744
+ var STATUS_TIMEOUT_MS = 1e4;
745
+ function authLabelFor(credentials2) {
746
+ if (credentials2.authType === "agent_key") {
747
+ return credentials2.keySource === "agent_key" ? "runner key (EVIDENT_AGENT_KEY)" : "runner key (EVIDENT_RUNNER_KEY)";
748
+ }
749
+ return "user token";
750
+ }
751
+ function describeFetchError(error2) {
752
+ const name = error2?.name;
753
+ if (name === "TimeoutError" || name === "AbortError") {
754
+ return `timed out after ${STATUS_TIMEOUT_MS}ms waiting for a response`;
755
+ }
756
+ return error2 instanceof Error ? error2.message : String(error2);
757
+ }
758
+ async function checkStatus(jsonMode) {
759
+ const apiUrl = getApiUrlConfig();
760
+ const credentials2 = await getAuthCredentials();
761
+ if (!credentials2) {
762
+ return {
763
+ ok: false,
764
+ endpoint: apiUrl,
765
+ reason: "no_credentials",
766
+ error: "No credentials configured. Set EVIDENT_RUNNER_KEY (or EVIDENT_AGENT_KEY), or EVIDENT_TOKEN, or run `evident login`.",
767
+ exitCode: 1
768
+ };
769
+ }
770
+ if (credentials2.notice && !jsonMode) {
771
+ printWarning(credentials2.notice);
772
+ }
773
+ let response;
774
+ try {
775
+ response = await fetch(`${apiUrl}/me`, {
776
+ headers: { Authorization: getAuthHeader(credentials2) },
777
+ signal: AbortSignal.timeout(STATUS_TIMEOUT_MS)
778
+ });
779
+ } catch (error2) {
780
+ return {
781
+ ok: false,
782
+ endpoint: apiUrl,
783
+ authLabel: authLabelFor(credentials2),
784
+ reason: "unreachable",
785
+ error: `Could not reach ${apiUrl}: ${describeFetchError(error2)}. The credentials were NOT validated.`,
786
+ exitCode: 75
787
+ };
788
+ }
789
+ if (response.status === 401) {
790
+ const serverMessage = await readErrorMessage(response);
791
+ return {
792
+ ok: false,
793
+ endpoint: apiUrl,
794
+ authLabel: authLabelFor(credentials2),
795
+ reason: "unauthorized",
796
+ error: authFailureHint(apiUrl, serverMessage),
797
+ exitCode: 1
798
+ };
799
+ }
800
+ if (response.status >= 500) {
801
+ const serverMessage = await readErrorMessage(response);
802
+ return {
803
+ ok: false,
804
+ endpoint: apiUrl,
805
+ authLabel: authLabelFor(credentials2),
806
+ reason: "unreachable",
807
+ error: `${apiUrl} returned HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}. The credentials were NOT validated.`,
808
+ exitCode: 75
809
+ };
810
+ }
811
+ if (!response.ok) {
812
+ const serverMessage = await readErrorMessage(response);
813
+ return {
814
+ ok: false,
815
+ endpoint: apiUrl,
816
+ authLabel: authLabelFor(credentials2),
817
+ reason: "http_error",
818
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`,
819
+ exitCode: 1
820
+ };
821
+ }
822
+ const data = await response.json();
823
+ return {
824
+ ok: true,
825
+ endpoint: apiUrl,
826
+ authType: data.auth_type,
827
+ authLabel: authLabelFor(credentials2),
828
+ runnerId: data.auth_type === "agent_key" ? data.agent_id : void 0,
829
+ reason: "ok",
830
+ exitCode: 0
831
+ };
832
+ }
833
+ function printJson(result) {
834
+ const payload = {
835
+ ok: result.ok,
836
+ endpoint: result.endpoint
837
+ };
838
+ if (result.authType) payload.auth_type = result.authType;
839
+ if (result.runnerId) payload.runner_id = result.runnerId;
840
+ if (result.reason) payload.reason = result.reason;
841
+ if (result.error) payload.error = result.error;
842
+ console.log(JSON.stringify(payload));
843
+ }
844
+ function printHuman(result) {
845
+ blank();
846
+ console.log(keyValue("Endpoint", result.endpoint));
847
+ if (result.ok) {
848
+ console.log(keyValue("Auth", result.authLabel ?? "\u2014"));
849
+ if (result.runnerId) {
850
+ console.log(keyValue("Runner", result.runnerId));
851
+ }
852
+ console.log(keyValue("Status", "OK \u2014 credentials accepted"));
853
+ blank();
854
+ return;
855
+ }
856
+ if (result.authLabel) {
857
+ console.log(keyValue("Auth", result.authLabel));
858
+ }
859
+ blank();
860
+ printError(result.error ?? "Unknown error");
861
+ }
862
+ async function status(options = {}) {
863
+ const result = await checkStatus(Boolean(options.json));
864
+ if (options.json) {
865
+ printJson(result);
866
+ } else {
867
+ printHuman(result);
868
+ }
869
+ process.exit(result.exitCode);
870
+ }
871
+
872
+ // src/lib/claude-usage.ts
873
+ import { execFileSync } from "child_process";
874
+ import { readFileSync } from "fs";
875
+ import { homedir } from "os";
876
+ import { join } from "path";
877
+ var CLAUDE_USAGE_URL = "https://api.anthropic.com/api/oauth/usage";
878
+ var KEYCHAIN_SERVICE = "Claude Code-credentials";
879
+ function parseClaudeCliCredentials(raw) {
880
+ let parsed;
881
+ try {
882
+ parsed = JSON.parse(raw);
883
+ } catch {
884
+ return null;
885
+ }
886
+ const data = parsed.claudeAiOauth ?? parsed;
887
+ const creds = data;
888
+ if (typeof creds.accessToken !== "string" || typeof creds.expiresAt !== "number") {
889
+ return null;
890
+ }
891
+ return { accessToken: creds.accessToken, expiresAt: creds.expiresAt };
892
+ }
893
+ function readClaudeCliCredentials() {
894
+ if (process.platform === "darwin") {
895
+ try {
896
+ const raw = execFileSync(
897
+ "/usr/bin/security",
898
+ ["find-generic-password", "-s", KEYCHAIN_SERVICE, "-w"],
899
+ { encoding: "utf-8", timeout: 2e3, stdio: ["pipe", "pipe", "ignore"] }
900
+ );
901
+ return parseClaudeCliCredentials(raw);
902
+ } catch {
903
+ return null;
904
+ }
905
+ }
906
+ try {
907
+ const raw = readFileSync(join(homedir(), ".claude", ".credentials.json"), "utf-8");
908
+ return parseClaudeCliCredentials(raw);
909
+ } catch {
910
+ return null;
911
+ }
912
+ }
913
+ var ClaudeUsageError = class extends Error {
914
+ constructor(message, reason) {
915
+ super(message);
916
+ this.reason = reason;
917
+ }
918
+ };
919
+ function isLocalCredentialProblem(err) {
920
+ return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
921
+ }
922
+ function normalizeResetsAt(value) {
923
+ const ms = Date.parse(value);
924
+ return Number.isNaN(ms) ? null : new Date(ms).toISOString();
925
+ }
926
+ function toWindow(value) {
927
+ if (!value || typeof value !== "object") {
928
+ return null;
929
+ }
930
+ const window = value;
931
+ if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
932
+ return null;
933
+ }
934
+ const resetsAt = normalizeResetsAt(window.resets_at);
935
+ if (resetsAt === null) {
936
+ return null;
937
+ }
938
+ return { utilization: window.utilization, resetsAt };
939
+ }
940
+ async function getClaudeUsage() {
941
+ const credentials2 = readClaudeCliCredentials();
942
+ if (!credentials2) {
943
+ throw new ClaudeUsageError(
944
+ "No local Claude Code login found. Run `claude` once to sign in with your Claude subscription.",
945
+ "no_credentials"
946
+ );
947
+ }
948
+ if (credentials2.expiresAt < Date.now()) {
949
+ throw new ClaudeUsageError(
950
+ "Claude Code credentials have expired. Run `claude` to refresh them.",
951
+ "credentials_expired"
952
+ );
953
+ }
954
+ const res = await fetch(CLAUDE_USAGE_URL, {
955
+ headers: {
956
+ Authorization: `Bearer ${credentials2.accessToken}`,
957
+ "Content-Type": "application/json",
958
+ "anthropic-version": "2023-06-01"
959
+ }
960
+ });
961
+ if (!res.ok) {
962
+ throw new ClaudeUsageError(`Claude usage request failed: HTTP ${res.status}`, "request_failed");
963
+ }
964
+ const body = await res.json();
965
+ return {
966
+ fiveHour: toWindow(body.five_hour),
967
+ sevenDay: toWindow(body.seven_day)
968
+ };
969
+ }
970
+
971
+ // src/commands/claude-usage.ts
972
+ function formatWindow(label, window) {
973
+ if (!window) {
974
+ return keyValue(label, "not available for this plan");
975
+ }
976
+ const resetsAt = new Date(window.resetsAt);
977
+ return keyValue(label, `${window.utilization}% used, resets ${resetsAt.toLocaleString()}`);
978
+ }
979
+ async function claudeUsage() {
980
+ try {
981
+ const usage = await getClaudeUsage();
982
+ blank();
983
+ console.log(formatWindow("5-hour session", usage.fiveHour));
984
+ console.log(formatWindow("7-day", usage.sevenDay));
985
+ blank();
986
+ } catch (err) {
987
+ if (err instanceof ClaudeUsageError) {
988
+ printError(err.message);
989
+ process.exit(1);
990
+ }
991
+ throw err;
992
+ }
993
+ }
994
+
995
+ // src/commands/run.ts
996
+ import { homedir as homedir3 } from "os";
997
+ import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
998
+ import chalk6 from "chalk";
999
+
1000
+ // ../../packages/types/src/agents/index.ts
1001
+ var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
1002
+
1003
+ // ../../packages/types/src/telemetry/index.ts
1004
+ var TelemetryEventTypes = {
1005
+ // Agent activity events (shown in web UI activity log)
1006
+ AGENT_CONNECTED: "agent.connected",
1007
+ AGENT_DISCONNECTED: "agent.disconnected",
1008
+ AGENT_MESSAGE_PROCESSING: "agent.message_processing",
1009
+ AGENT_MESSAGE_DONE: "agent.message_done",
1010
+ AGENT_MESSAGE_FAILED: "agent.message_failed",
1011
+ // A `warn`/`error` runner-side log line forwarded server-side for
1012
+ // observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
1013
+ RUNNER_ACTIVITY: "runner.activity"
1014
+ };
1015
+
1016
+ // ../../packages/types/src/tunnel/index.ts
1017
+ var MAX_FRAME_BYTES = 256 * 1024;
1018
+ var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
1019
+
1020
+ // ../../packages/types/src/runner-files.ts
1021
+ var MAX_FILE_PUSH_BYTES = 64 * 1024;
1022
+ var MAX_FILE_SYNC_DIRECTORIES = 16;
1023
+
1024
+ // ../../packages/types/src/logging/index.ts
1025
+ var CORRELATION_ID_HEADER = "x-evident-correlation-id";
1026
+ function log(level, event, fields) {
1027
+ const method = level === "debug" ? "log" : level;
1028
+ try {
1029
+ console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
1030
+ } catch (err) {
1031
+ console.error(
1032
+ "[evident] log_serialize_failed",
1033
+ event,
1034
+ err instanceof Error ? err.message : String(err)
1035
+ );
1036
+ }
1037
+ }
1038
+ function errorFields(err) {
1039
+ if (err instanceof Error) {
1040
+ return { error: err.message, error_name: err.name };
1041
+ }
1042
+ return { error: String(err) };
1043
+ }
1044
+ function stripQuery(url) {
1045
+ try {
1046
+ return new URL(url).pathname;
1047
+ } catch {
1048
+ const q = url.indexOf("?");
1049
+ return q === -1 ? url : url.slice(0, q);
1050
+ }
1051
+ }
1052
+
1053
+ // src/commands/run.ts
1054
+ import ora3 from "ora";
1055
+ import { select as select3 } from "@inquirer/prompts";
1056
+
1057
+ // src/lib/telemetry.ts
1058
+ var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
1059
+ function getCliVersion() {
1060
+ return CLI_VERSION;
1061
+ }
1062
+ var eventBuffer = [];
1063
+ var flushTimeout = null;
1064
+ var isShuttingDown = false;
1065
+ var FLUSH_INTERVAL_MS = 5e3;
1066
+ var MAX_BUFFER_SIZE = 50;
1067
+ var FLUSH_TIMEOUT_MS = 3e3;
1068
+ var authProvider = null;
1069
+ function setTelemetryAuthProvider(provider) {
1070
+ authProvider = provider;
1071
+ }
1072
+ var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
1073
+ var lastFlushFailureLoggedAt = 0;
1074
+ var suppressedFlushFailureCount = 0;
1075
+ function logEvent(eventType, options = {}) {
1076
+ const event = {
1077
+ event_type: eventType,
1078
+ severity: options.severity || "info",
1079
+ message: options.message,
1080
+ metadata: options.metadata,
1081
+ agent_id: options.agentId,
1082
+ timestamp: (/* @__PURE__ */ new Date()).toISOString()
1083
+ };
1084
+ eventBuffer.push(event);
1085
+ if (options.severity === "error" || eventBuffer.length >= MAX_BUFFER_SIZE) {
1086
+ void flushEvents();
1087
+ } else if (!flushTimeout && !isShuttingDown) {
1088
+ flushTimeout = setTimeout(() => {
1089
+ flushTimeout = null;
1090
+ void flushEvents();
1091
+ }, FLUSH_INTERVAL_MS);
1092
+ }
1093
+ }
1094
+ var telemetry = {
1095
+ debug: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "debug", message, metadata, agentId }),
1096
+ info: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "info", message, metadata, agentId }),
1097
+ warn: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "warning", message, metadata, agentId }),
1098
+ error: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "error", message, metadata, agentId })
1099
+ };
1100
+ async function flushEvents() {
1101
+ if (eventBuffer.length === 0) return;
1102
+ const events = eventBuffer;
1103
+ eventBuffer = [];
1104
+ if (flushTimeout) {
1105
+ clearTimeout(flushTimeout);
1106
+ flushTimeout = null;
1107
+ }
1108
+ try {
1109
+ const providerContext = authProvider?.();
1110
+ let authHeader;
1111
+ if (providerContext?.authHeader) {
1112
+ authHeader = providerContext.authHeader;
1113
+ } else {
1114
+ const credentials2 = await getToken();
1115
+ if (!credentials2) {
1116
+ return;
1117
+ }
1118
+ authHeader = `Bearer ${credentials2.token}`;
1119
+ }
1120
+ const apiUrl = getApiUrlConfig();
1121
+ const controller = new AbortController();
1122
+ const timeout = setTimeout(() => controller.abort(), FLUSH_TIMEOUT_MS);
1123
+ try {
1124
+ const request = {
1125
+ events,
1126
+ client_type: "cli",
1127
+ client_version: CLI_VERSION
1128
+ };
1129
+ const response = await fetch(`${apiUrl}/telemetry/events`, {
1130
+ method: "POST",
1131
+ headers: {
1132
+ "Content-Type": "application/json",
1133
+ Authorization: authHeader
1134
+ },
1135
+ body: JSON.stringify(request),
1136
+ signal: controller.signal
1137
+ });
1138
+ if (!response.ok) {
1139
+ console.error(`Telemetry flush failed: ${response.status}`);
1140
+ }
1141
+ } finally {
1142
+ clearTimeout(timeout);
1143
+ }
1144
+ } catch (error2) {
1145
+ const now = Date.now();
1146
+ if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
1147
+ const message = error2 instanceof Error ? error2.message : String(error2);
1148
+ const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
1149
+ console.error(`Telemetry flush error: ${message}${suffix}`);
1150
+ lastFlushFailureLoggedAt = now;
1151
+ suppressedFlushFailureCount = 0;
1152
+ } else {
1153
+ suppressedFlushFailureCount++;
1154
+ }
1155
+ }
1156
+ }
1157
+ async function shutdownTelemetry() {
1158
+ isShuttingDown = true;
1159
+ if (flushTimeout) {
1160
+ clearTimeout(flushTimeout);
1161
+ flushTimeout = null;
1162
+ }
1163
+ await flushEvents();
1164
+ }
1165
+ function emitEvent(event) {
1166
+ logEvent(event.event_type, {
1167
+ severity: event.severity,
1168
+ message: event.message,
1169
+ metadata: event.metadata,
1170
+ agentId: event.agent_id
1171
+ });
1172
+ }
1173
+ function emitAgentConnected(agentId, metadata) {
1174
+ emitEvent({
1175
+ event_type: TelemetryEventTypes.AGENT_CONNECTED,
1176
+ severity: "info",
1177
+ message: "Agent CLI connected",
1178
+ metadata,
1179
+ agent_id: agentId
1180
+ });
1181
+ }
1182
+ function emitAgentDisconnected(agentId, metadata) {
1183
+ emitEvent({
1184
+ event_type: TelemetryEventTypes.AGENT_DISCONNECTED,
1185
+ severity: "info",
1186
+ message: `Agent CLI disconnected (code: ${metadata.code})`,
1187
+ metadata,
1188
+ agent_id: agentId
1189
+ });
1190
+ }
1191
+ var EventTypes = {
1192
+ // Tunnel lifecycle
1193
+ TUNNEL_STARTING: "tunnel.starting",
1194
+ TUNNEL_CONNECTED: "tunnel.connected",
1195
+ TUNNEL_DISCONNECTED: "tunnel.disconnected",
1196
+ TUNNEL_RECONNECTING: "tunnel.reconnecting",
1197
+ TUNNEL_ERROR: "tunnel.error",
1198
+ // OpenCode communication
1199
+ OPENCODE_HEALTH_CHECK: "opencode.health_check",
1200
+ OPENCODE_HEALTH_OK: "opencode.health_ok",
1201
+ OPENCODE_HEALTH_FAILED: "opencode.health_failed",
1202
+ OPENCODE_REQUEST_RECEIVED: "opencode.request_received",
1203
+ OPENCODE_REQUEST_FORWARDED: "opencode.request_forwarded",
1204
+ OPENCODE_RESPONSE_SENT: "opencode.response_sent",
1205
+ OPENCODE_UNREACHABLE: "opencode.unreachable",
1206
+ OPENCODE_ERROR: "opencode.error",
1207
+ // Authentication
1208
+ AUTH_LOGIN_STARTED: "auth.login_started",
1209
+ AUTH_LOGIN_SUCCESS: "auth.login_success",
1210
+ AUTH_LOGIN_FAILED: "auth.login_failed",
1211
+ AUTH_LOGOUT: "auth.logout",
1212
+ // CLI lifecycle
1213
+ CLI_STARTED: "cli.started",
1214
+ CLI_COMMAND: "cli.command",
1215
+ CLI_ERROR: "cli.error",
1216
+ // Deprecation telemetry (#412) — usage of the old `--agent`/`EVIDENT_AGENT_KEY`
1217
+ // names instead of the preferred `--runner`/`EVIDENT_RUNNER_KEY` (#409).
1218
+ DEPRECATED_AGENT_FLAG_USED: "cli.deprecated_agent_flag_used",
1219
+ DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
1220
+ };
1221
+
1222
+ // src/lib/runner-activity-telemetry.ts
1223
+ var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
1224
+ var SEVERITY_BY_LEVEL = {
1225
+ warn: "warning",
1226
+ error: "error"
1227
+ };
1228
+ var MAX_MESSAGE_LENGTH = 500;
1229
+ var TRUNCATION_MARKER = "\u2026";
1230
+ function redact(message) {
1231
+ return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
1232
+ }
1233
+ function truncate(message) {
1234
+ if (message.length <= MAX_MESSAGE_LENGTH) return message;
1235
+ return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
1236
+ }
1237
+ var RATE_LIMIT_WINDOW_MS = 6e4;
1238
+ var RATE_LIMIT_MAX_EVENTS = 30;
1239
+ var windowStartedAt = 0;
1240
+ var windowCount = 0;
1241
+ var windowDroppedCount = 0;
1242
+ function admitUnderRateLimit(now) {
1243
+ if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
1244
+ if (windowDroppedCount > 0) {
1245
+ console.error(
1246
+ `[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
1247
+ );
1248
+ }
1249
+ windowStartedAt = now;
1250
+ windowCount = 0;
1251
+ windowDroppedCount = 0;
1252
+ }
1253
+ if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
1254
+ windowDroppedCount++;
1255
+ if (windowDroppedCount === 1) {
1256
+ console.error(
1257
+ `[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
1258
+ );
1259
+ }
1260
+ return false;
1261
+ }
1262
+ windowCount++;
1263
+ return true;
1264
+ }
1265
+ function forwardRunnerActivity(entry, context) {
1266
+ try {
1267
+ if (!FORWARDED_LEVELS.has(entry.level)) return;
1268
+ if (!context.agentId || !context.authHeader) return;
1269
+ if (!admitUnderRateLimit(Date.now())) return;
1270
+ const rawMessage = entry.error ?? entry.message ?? "";
1271
+ const message = truncate(redact(rawMessage));
1272
+ logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
1273
+ severity: SEVERITY_BY_LEVEL[entry.level],
1274
+ message,
1275
+ metadata: { source: "cli.run" },
1276
+ agentId: context.agentId
1277
+ });
1278
+ } catch (err) {
1279
+ console.error(
1280
+ `[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
1281
+ );
1282
+ }
1283
+ }
1284
+
1285
+ // src/lib/opencode/health.ts
1286
+ async function checkOpenCodeHealth(port) {
1287
+ try {
1288
+ const response = await fetch(`http://127.0.0.1:${port}/global/health`, {
1289
+ signal: AbortSignal.timeout(2e3)
1290
+ // 2 second timeout
1291
+ });
1292
+ if (!response.ok) {
1293
+ return { healthy: false, error: `HTTP ${response.status}` };
1294
+ }
1295
+ const data = await response.json().catch(() => ({}));
1296
+ return { healthy: true, version: data.version };
1297
+ } catch (error2) {
1298
+ const message = error2 instanceof Error ? error2.message : "Unknown error";
1299
+ return { healthy: false, error: message };
1300
+ }
1301
+ }
1302
+ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
1303
+ const startTime = Date.now();
1304
+ while (Date.now() - startTime < timeoutMs) {
1305
+ const health = await checkOpenCodeHealth(port);
1306
+ if (health.healthy) {
1307
+ return health;
1308
+ }
1309
+ await new Promise((resolve3) => setTimeout(resolve3, 1e3));
1310
+ }
1311
+ return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
1312
+ }
1313
+
1314
+ // src/lib/opencode/opencode-version-gate.ts
1315
+ var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
1316
+ function isQueueValidatedVersion(version2) {
730
1317
  if (!version2) return false;
731
1318
  return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
732
1319
  }
@@ -1026,6 +1613,12 @@ async function promptOpenCodeInstall(interactive) {
1026
1613
  return action;
1027
1614
  }
1028
1615
 
1616
+ // src/lib/opencode/provider-check.ts
1617
+ function buildNoProviderWarning(hasProvider) {
1618
+ if (hasProvider !== false) return null;
1619
+ 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).";
1620
+ }
1621
+
1029
1622
  // src/lib/opencode/session.ts
1030
1623
  function opencodeBase(port) {
1031
1624
  return `http://127.0.0.1:${port}`;
@@ -1261,6 +1854,16 @@ async function buildFileParts(attachments, capable) {
1261
1854
  );
1262
1855
  dataUrl = null;
1263
1856
  }
1857
+ if (dataUrl !== null && typeof dataUrl === "object") {
1858
+ outcomes.push({
1859
+ index: a.index,
1860
+ mime: a.mime,
1861
+ filename: a.filename,
1862
+ status: "failed",
1863
+ reason: "needs_reauth"
1864
+ });
1865
+ continue;
1866
+ }
1264
1867
  if (dataUrl == null) {
1265
1868
  outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
1266
1869
  continue;
@@ -1342,7 +1945,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
1342
1945
  }
1343
1946
  }
1344
1947
  if (attempt < READ_BACK_ATTEMPTS - 1) {
1345
- await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1948
+ await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
1346
1949
  }
1347
1950
  }
1348
1951
  return null;
@@ -1470,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
1470
2073
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1471
2074
  return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1472
2075
  }
2076
+ function isB2AbandonmentConfirmed(params) {
2077
+ return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
2078
+ }
1473
2079
  function messageError(messages, userMessageId) {
1474
2080
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1475
2081
  const error2 = errorOf(reply);
@@ -1483,12 +2089,94 @@ function messageError(messages, userMessageId) {
1483
2089
  }
1484
2090
  return "The agent run failed.";
1485
2091
  }
2092
+ function isAbortedTerminalReply(messages, userMessageId) {
2093
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
2094
+ const error2 = errorOf(reply);
2095
+ if (error2 == null) return false;
2096
+ if (typeof error2 === "string") return error2.trim() === "Aborted";
2097
+ if (typeof error2 === "object") {
2098
+ const e = error2;
2099
+ if (e.name === "MessageAbortedError") return true;
2100
+ if (e.name === "AbortError") return true;
2101
+ const dataMessage = e.data?.message;
2102
+ const rendered = typeof dataMessage === "string" ? dataMessage : typeof e.message === "string" ? e.message : null;
2103
+ return rendered != null && rendered.trim() === "Aborted";
2104
+ }
2105
+ return false;
2106
+ }
2107
+ function messageFailure(messages, userMessageId) {
2108
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
2109
+ const error2 = errorOf(reply);
2110
+ if (error2 == null || typeof error2 !== "object") return null;
2111
+ const e = error2;
2112
+ const replyProviderId = reply?.info?.providerID ?? null;
2113
+ const replyModelId = reply?.info?.modelID ?? null;
2114
+ if (e.name === "ProviderAuthError") {
2115
+ const data = e.data;
2116
+ const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
2117
+ return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
2118
+ }
2119
+ if (e.name === "APIError") {
2120
+ const data = e.data;
2121
+ const statusCode = data?.statusCode;
2122
+ if (statusCode === 401 || statusCode === 403) {
2123
+ return {
2124
+ kind: "model_auth",
2125
+ providerId: replyProviderId,
2126
+ modelId: replyModelId,
2127
+ reason: "rejected"
2128
+ };
2129
+ }
2130
+ }
2131
+ return null;
2132
+ }
2133
+ function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
2134
+ if (classified != null) return classified;
2135
+ if (hasConfiguredProvider !== false) return null;
2136
+ return {
2137
+ kind: "model_auth",
2138
+ providerId: replyProviderId,
2139
+ modelId: replyModelId,
2140
+ reason: "missing"
2141
+ };
2142
+ }
1486
2143
  function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1487
2144
  if (!messages || messages.length === 0) return false;
1488
2145
  return messages.some(
1489
2146
  (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1490
2147
  );
1491
2148
  }
2149
+ async function hasAnyConfiguredProvider(port) {
2150
+ try {
2151
+ const res = await fetch(`${opencodeBase(port)}/config/providers`);
2152
+ if (!res.ok) {
2153
+ console.error(
2154
+ `[hasAnyConfiguredProvider] GET /config/providers returned HTTP ${res.status} (port ${port})`
2155
+ );
2156
+ return null;
2157
+ }
2158
+ const body = await res.json();
2159
+ if (!body || typeof body !== "object" || Array.isArray(body)) {
2160
+ console.error(
2161
+ `[hasAnyConfiguredProvider] GET /config/providers body was not a plain object (port ${port})`
2162
+ );
2163
+ return null;
2164
+ }
2165
+ const defaults2 = body.default;
2166
+ if (!defaults2 || typeof defaults2 !== "object" || Array.isArray(defaults2)) {
2167
+ console.error(
2168
+ `[hasAnyConfiguredProvider] GET /config/providers body had no \`default\` object (port ${port})`
2169
+ );
2170
+ return null;
2171
+ }
2172
+ return Object.keys(defaults2).length > 0;
2173
+ } catch (err) {
2174
+ console.error(
2175
+ `[hasAnyConfiguredProvider] GET /config/providers failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
2176
+ );
2177
+ return null;
2178
+ }
2179
+ }
1492
2180
 
1493
2181
  // src/lib/opencode/session-cleanup.ts
1494
2182
  var DURATION_UNIT_MS = {
@@ -1647,10 +2335,11 @@ var StreamForwarder = class {
1647
2335
  * Abort every in-flight stream (e.g. on WebSocket close).
1648
2336
  */
1649
2337
  abortAll() {
1650
- for (const stream of this.inflight.values()) {
2338
+ for (const [sid, stream] of this.inflight.entries()) {
1651
2339
  try {
1652
2340
  stream.abort();
1653
- } catch {
2341
+ } catch (err) {
2342
+ log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
1654
2343
  }
1655
2344
  }
1656
2345
  this.inflight.clear();
@@ -1684,12 +2373,12 @@ var StreamForwarder = class {
1684
2373
  let endBody;
1685
2374
  if (has_body) {
1686
2375
  const chunks = [];
1687
- bodyPromise = new Promise((resolve2) => {
2376
+ bodyPromise = new Promise((resolve3) => {
1688
2377
  pushBody = (buf) => {
1689
2378
  chunks.push(buf);
1690
2379
  };
1691
2380
  endBody = () => {
1692
- resolve2(Buffer.concat(chunks));
2381
+ resolve3(Buffer.concat(chunks));
1693
2382
  };
1694
2383
  });
1695
2384
  }
@@ -1806,7 +2495,7 @@ function connectTunnel(options) {
1806
2495
  } = options;
1807
2496
  const tunnelUrl = getTunnelUrlConfig();
1808
2497
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1809
- return new Promise((resolve2, reject) => {
2498
+ return new Promise((resolve3, reject) => {
1810
2499
  const ws = new WebSocket2(url, {
1811
2500
  headers: {
1812
2501
  Authorization: authHeader
@@ -1861,7 +2550,7 @@ function connectTunnel(options) {
1861
2550
  clearTimeout(connectionTimeout);
1862
2551
  const connectedAgentId = message.agent_id ?? agentId;
1863
2552
  onConnected?.(connectedAgentId);
1864
- resolve2({
2553
+ resolve3({
1865
2554
  ws,
1866
2555
  close: () => ws.close(1e3, "CLI shutdown")
1867
2556
  });
@@ -1923,57 +2612,502 @@ var RunnerConnection = class {
1923
2612
  if (this.connection) {
1924
2613
  try {
1925
2614
  this.connection.close();
1926
- } catch {
2615
+ } catch (err) {
2616
+ log("error", "runner_connection_close_failed", {
2617
+ agent_id: this.resolvedAgentId,
2618
+ ...errorFields(err)
2619
+ });
2620
+ }
2621
+ this.connection = null;
2622
+ }
2623
+ }
2624
+ async connectWithRetry(isReconnect) {
2625
+ if (isReconnect && this.reconnecting) return;
2626
+ this.reconnecting = true;
2627
+ this.close();
2628
+ const { events } = this.opts;
2629
+ while (this.opts.isRunning()) {
2630
+ try {
2631
+ this.connection = await connectTunnel({
2632
+ agentId: this.resolvedAgentId,
2633
+ authHeader: this.opts.getAuthHeader(),
2634
+ port: this.opts.port,
2635
+ onConnected: (agentId) => {
2636
+ this.reconnectAttempt = 0;
2637
+ this.reconnecting = false;
2638
+ this.resolvedAgentId = agentId;
2639
+ events.onConnected(agentId, isReconnect);
2640
+ },
2641
+ onDisconnected: (code, reason) => {
2642
+ events.onDisconnected(code, reason);
2643
+ if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
2644
+ this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
2645
+ events.onError?.(`Reconnection failed: ${err.message}`);
2646
+ });
2647
+ }
2648
+ },
2649
+ onError: (error2) => events.onError?.(error2),
2650
+ onResponse: () => events.onResponse?.(),
2651
+ onDrainPing: () => events.onDrainPing?.(),
2652
+ onInfo: (message) => events.onInfo?.(message)
2653
+ });
2654
+ return;
2655
+ } catch (error2) {
2656
+ this.reconnectAttempt++;
2657
+ if (error2.message === "Unauthorized") {
2658
+ this.reconnecting = false;
2659
+ throw error2;
2660
+ }
2661
+ const delay = getReconnectDelay(this.reconnectAttempt);
2662
+ events.onReconnecting?.(this.reconnectAttempt);
2663
+ events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
2664
+ await this.sleep(delay);
2665
+ }
2666
+ }
2667
+ this.reconnecting = false;
2668
+ }
2669
+ };
2670
+
2671
+ // src/lib/tunnel/ready-marker.ts
2672
+ import { writeFileSync } from "fs";
2673
+ function writeTunnelReadyMarker(path, agentId) {
2674
+ try {
2675
+ writeFileSync(path, `${agentId}
2676
+ `);
2677
+ return { ok: true };
2678
+ } catch (error2) {
2679
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2680
+ }
2681
+ }
2682
+
2683
+ // src/lib/claude-usage-reporting.ts
2684
+ var VALID_MODES = ["auto", "on", "off"];
2685
+ function resolveClaudeUsageReportingMode(flagValue, env) {
2686
+ const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
2687
+ if (raw === void 0 || raw === "") {
2688
+ return { mode: "auto", warnings: [] };
2689
+ }
2690
+ const normalized = raw.trim().toLowerCase();
2691
+ if (VALID_MODES.includes(normalized)) {
2692
+ return { mode: normalized, warnings: [] };
2693
+ }
2694
+ const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
2695
+ return {
2696
+ mode: "auto",
2697
+ warnings: [
2698
+ `Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
2699
+ ]
2700
+ };
2701
+ }
2702
+ var BASE_REPORT_DELAY_MS = 10 * 6e4;
2703
+ var REPORT_DELAY_JITTER_FRACTION = 0.2;
2704
+ function nextReportDelayMs(random = Math.random) {
2705
+ const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
2706
+ return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
2707
+ }
2708
+ var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
2709
+ var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
2710
+ function claudeUsageFailureLogLevel(consecutiveFailures) {
2711
+ return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
2712
+ }
2713
+
2714
+ // src/lib/channels/driver.ts
2715
+ import { homedir as homedir2 } from "os";
2716
+
2717
+ // src/lib/file-push.ts
2718
+ import { randomUUID } from "crypto";
2719
+ import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
2720
+ import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
2721
+ var FILE_MODE = 384;
2722
+ var DIRECTORY_MODE = 448;
2723
+ async function writePushedFile(request) {
2724
+ const { requestedPath, content, allowedDirectories, homeDir } = request;
2725
+ const bytes = content.byteLength;
2726
+ if (allowedDirectories.length === 0) {
2727
+ return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
2728
+ path: requestedPath,
2729
+ bytes
2730
+ });
2731
+ }
2732
+ if (bytes > MAX_FILE_PUSH_BYTES) {
2733
+ return refuse(
2734
+ "file_too_large",
2735
+ `File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
2736
+ {
2737
+ path: requestedPath,
2738
+ bytes
2739
+ }
2740
+ );
2741
+ }
2742
+ const candidate = expandAndValidate(requestedPath, homeDir);
2743
+ if (candidate === null) {
2744
+ return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
2745
+ path: requestedPath,
2746
+ bytes
2747
+ });
2748
+ }
2749
+ try {
2750
+ const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
2751
+ dirname2(candidate)
2752
+ );
2753
+ const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
2754
+ const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
2755
+ if (allowedDirectory === null) {
2756
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2757
+ path: realTarget,
2758
+ bytes
2759
+ });
2760
+ }
2761
+ if (missingSegments.length > 0) {
2762
+ await createMissingDirectories(existingAncestor, missingSegments);
2763
+ const realParent = await realpath(dirname2(realTarget));
2764
+ if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
2765
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2766
+ path: realTarget,
2767
+ bytes,
2768
+ reason: "parent_changed_after_create"
2769
+ });
2770
+ }
2771
+ }
2772
+ await writeAtomically(realTarget, content);
2773
+ log("info", "file_push_written", { path: realTarget, bytes });
2774
+ return { ok: true, path: realTarget };
2775
+ } catch (err) {
2776
+ const errno = err.code ?? "UNKNOWN";
2777
+ return refuse("write_failed", `The runner could not write the file (${errno}).`, {
2778
+ path: candidate,
2779
+ bytes,
2780
+ errno,
2781
+ ...errorFields(err)
2782
+ });
2783
+ }
2784
+ }
2785
+ function expandAndValidate(requestedPath, homeDir) {
2786
+ if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
2787
+ return null;
2788
+ }
2789
+ const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
2790
+ if (expanded.split(/[/\\]/).includes("..")) {
2791
+ return null;
2792
+ }
2793
+ if (!isAbsolute(expanded)) {
2794
+ return null;
2795
+ }
2796
+ const candidate = resolve2(expanded);
2797
+ const name = basename(candidate);
2798
+ return name === "" || name === "." || name === ".." ? null : candidate;
2799
+ }
2800
+ async function resolveNearestExistingAncestor(directory) {
2801
+ const missingSegments = [];
2802
+ let current = directory;
2803
+ for (; ; ) {
2804
+ try {
2805
+ return { existingAncestor: await realpath(current), missingSegments };
2806
+ } catch (err) {
2807
+ const parent = dirname2(current);
2808
+ if (err.code !== "ENOENT" || parent === current) {
2809
+ throw err;
2810
+ }
2811
+ missingSegments.unshift(basename(current));
2812
+ current = parent;
2813
+ }
2814
+ }
2815
+ }
2816
+ async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
2817
+ for (const directory of allowedDirectories) {
2818
+ if (!isAbsolute(directory)) {
2819
+ log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
2820
+ continue;
2821
+ }
2822
+ const realDirectory = await realpathCreatingIfMissing(directory);
2823
+ if (realDirectory !== null && contains(realDirectory, realTarget)) {
2824
+ return realDirectory;
2825
+ }
2826
+ }
2827
+ return null;
2828
+ }
2829
+ async function realpathCreatingIfMissing(directory) {
2830
+ try {
2831
+ return await realpath(directory);
2832
+ } catch (err) {
2833
+ if (err.code !== "ENOENT") {
2834
+ log("warn", "file_push_allowed_directory_skipped", {
2835
+ directory,
2836
+ reason: "unresolvable",
2837
+ ...errorFields(err)
2838
+ });
2839
+ return null;
2840
+ }
2841
+ }
2842
+ try {
2843
+ await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
2844
+ await chmod(directory, DIRECTORY_MODE);
2845
+ return await realpath(directory);
2846
+ } catch (err) {
2847
+ log("warn", "file_push_allowed_directory_skipped", {
2848
+ directory,
2849
+ reason: "create_failed",
2850
+ ...errorFields(err)
2851
+ });
2852
+ return null;
2853
+ }
2854
+ }
2855
+ function contains(realDirectory, realTarget) {
2856
+ const rel = relative(realDirectory, realTarget);
2857
+ return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
2858
+ }
2859
+ async function createMissingDirectories(existingAncestor, missingSegments) {
2860
+ let current = existingAncestor;
2861
+ for (const segment of missingSegments) {
2862
+ current = join2(current, segment);
2863
+ await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
2864
+ await chmod(current, DIRECTORY_MODE);
2865
+ }
2866
+ }
2867
+ async function writeAtomically(realTarget, content) {
2868
+ const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2869
+ let handle;
2870
+ try {
2871
+ handle = await open2(temporaryPath, "wx", FILE_MODE);
2872
+ await handle.writeFile(content);
2873
+ await handle.chmod(FILE_MODE);
2874
+ await handle.close();
2875
+ handle = void 0;
2876
+ await rename(temporaryPath, realTarget);
2877
+ } catch (err) {
2878
+ await discardTemporaryFile(temporaryPath, handle);
2879
+ throw err;
2880
+ }
2881
+ }
2882
+ async function discardTemporaryFile(temporaryPath, handle) {
2883
+ try {
2884
+ await handle?.close();
2885
+ } catch (err) {
2886
+ log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
2887
+ }
2888
+ try {
2889
+ await unlink(temporaryPath);
2890
+ } catch (err) {
2891
+ const errno = err.code;
2892
+ if (errno !== "ENOENT" && errno !== "ENOTDIR") {
2893
+ log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
2894
+ }
2895
+ }
2896
+ }
2897
+ function refuse(code, message, fields) {
2898
+ log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
2899
+ return { ok: false, code, message };
2900
+ }
2901
+
2902
+ // src/lib/runner-file-sync.ts
2903
+ var MAX_ACK_ATTEMPTS = 5;
2904
+ async function syncPendingRunnerFiles(options) {
2905
+ const pending = await listPendingFiles(options);
2906
+ const pendingIds = new Set(pending.map((file) => file.id));
2907
+ for (const id of options.ackFailures.keys()) {
2908
+ if (!pendingIds.has(id)) options.ackFailures.delete(id);
2909
+ }
2910
+ if (pending.length === 0) return 0;
2911
+ options.log({
2912
+ level: "info",
2913
+ message: `Runner file sync: ${pending.length} file(s) queued for this runner`
2914
+ });
2915
+ let applied = 0;
2916
+ for (const file of pending) {
2917
+ if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
2918
+ if (await applyOne(options, file)) applied += 1;
2919
+ }
2920
+ return applied;
2921
+ }
2922
+ async function listPendingFiles(options) {
2923
+ let res;
2924
+ try {
2925
+ res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
2926
+ headers: { Authorization: options.getAuthHeader() }
2927
+ });
2928
+ } catch (err) {
2929
+ options.log({
2930
+ level: "warn",
2931
+ message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
2932
+ });
2933
+ return [];
2934
+ }
2935
+ if (!res.ok) {
2936
+ options.log({
2937
+ level: res.status === 404 ? "debug" : "warn",
2938
+ message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
2939
+ });
2940
+ return [];
2941
+ }
2942
+ let body;
2943
+ try {
2944
+ body = await res.json();
2945
+ } catch (err) {
2946
+ options.log({
2947
+ level: "warn",
2948
+ message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
2949
+ });
2950
+ return [];
2951
+ }
2952
+ if (!Array.isArray(body)) {
2953
+ options.log({
2954
+ level: "warn",
2955
+ message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
2956
+ });
2957
+ return [];
2958
+ }
2959
+ const files = [];
2960
+ for (const entry of body) {
2961
+ const file = asPendingFile(entry);
2962
+ if (file === null) {
2963
+ options.log({
2964
+ level: "warn",
2965
+ message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
2966
+ });
2967
+ continue;
2968
+ }
2969
+ files.push(file);
2970
+ }
2971
+ return files;
2972
+ }
2973
+ function asPendingFile(entry) {
2974
+ if (entry === null || typeof entry !== "object") return null;
2975
+ const { id, path, size } = entry;
2976
+ if (typeof id !== "string" || id === "") return null;
2977
+ if (typeof path !== "string" || path === "") return null;
2978
+ if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
2979
+ return { id, path, size };
2980
+ }
2981
+ async function applyOne(options, file) {
2982
+ const label = `${file.id.slice(0, 8)} (${file.path})`;
2983
+ if (options.allowedDirectories.length === 0) {
2984
+ options.log({
2985
+ level: "warn",
2986
+ message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
2987
+ });
2988
+ await ack(options, file, "rejected", "file_sync_disabled");
2989
+ return false;
2990
+ }
2991
+ if (file.size > MAX_FILE_PUSH_BYTES) {
2992
+ options.log({
2993
+ level: "warn",
2994
+ message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
2995
+ });
2996
+ await ack(options, file, "rejected", "file_too_large");
2997
+ return false;
2998
+ }
2999
+ const download = await downloadContent(options, file, label);
3000
+ if (!download.ok) {
3001
+ if (download.terminal) await ack(options, file, "rejected", download.code);
3002
+ return false;
3003
+ }
3004
+ let outcome;
3005
+ try {
3006
+ outcome = await writePushedFile({
3007
+ requestedPath: file.path,
3008
+ content: download.content,
3009
+ allowedDirectories: options.allowedDirectories,
3010
+ homeDir: options.homeDir
3011
+ });
3012
+ } catch (err) {
3013
+ options.log({
3014
+ level: "error",
3015
+ message: `Runner file ${label} could not be written: ${describe(err)}`
3016
+ });
3017
+ await ack(options, file, "rejected", "write_failed");
3018
+ return false;
3019
+ }
3020
+ if (!outcome.ok) {
3021
+ options.log({
3022
+ level: "warn",
3023
+ message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
3024
+ });
3025
+ await ack(options, file, "rejected", outcome.code);
3026
+ return false;
3027
+ }
3028
+ options.log({
3029
+ level: "info",
3030
+ message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
3031
+ });
3032
+ await ack(options, file, "applied");
3033
+ return true;
3034
+ }
3035
+ function durableDownloadCode(status2) {
3036
+ return status2 === 413 ? "file_too_large" : "write_failed";
3037
+ }
3038
+ async function downloadContent(options, file, label) {
3039
+ try {
3040
+ const res = await options.fetchImpl(
3041
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
3042
+ { headers: { Authorization: options.getAuthHeader() } }
3043
+ );
3044
+ if (!res.ok) {
3045
+ const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
3046
+ if (!terminal) {
3047
+ options.log({
3048
+ level: "warn",
3049
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
3050
+ });
3051
+ return { ok: false, terminal: false };
1927
3052
  }
1928
- this.connection = null;
3053
+ const code = durableDownloadCode(res.status);
3054
+ options.log({
3055
+ level: "error",
3056
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
3057
+ });
3058
+ return { ok: false, terminal: true, code };
1929
3059
  }
3060
+ return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
3061
+ } catch (err) {
3062
+ options.log({
3063
+ level: "warn",
3064
+ message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
3065
+ });
3066
+ return { ok: false, terminal: false };
1930
3067
  }
1931
- async connectWithRetry(isReconnect) {
1932
- if (isReconnect && this.reconnecting) return;
1933
- this.reconnecting = true;
1934
- this.close();
1935
- const { events } = this.opts;
1936
- while (this.opts.isRunning()) {
1937
- try {
1938
- this.connection = await connectTunnel({
1939
- agentId: this.resolvedAgentId,
1940
- authHeader: this.opts.getAuthHeader(),
1941
- port: this.opts.port,
1942
- onConnected: (agentId) => {
1943
- this.reconnectAttempt = 0;
1944
- this.reconnecting = false;
1945
- this.resolvedAgentId = agentId;
1946
- events.onConnected(agentId, isReconnect);
1947
- },
1948
- onDisconnected: (code, reason) => {
1949
- events.onDisconnected(code, reason);
1950
- if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
1951
- this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
1952
- events.onError?.(`Reconnection failed: ${err.message}`);
1953
- });
1954
- }
1955
- },
1956
- onError: (error2) => events.onError?.(error2),
1957
- onResponse: () => events.onResponse?.(),
1958
- onDrainPing: () => events.onDrainPing?.(),
1959
- onInfo: (message) => events.onInfo?.(message)
1960
- });
1961
- return;
1962
- } catch (error2) {
1963
- this.reconnectAttempt++;
1964
- if (error2.message === "Unauthorized") {
1965
- this.reconnecting = false;
1966
- throw error2;
1967
- }
1968
- const delay = getReconnectDelay(this.reconnectAttempt);
1969
- events.onReconnecting?.(this.reconnectAttempt);
1970
- events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
1971
- await this.sleep(delay);
3068
+ }
3069
+ async function ack(options, file, status2, reason) {
3070
+ const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
3071
+ try {
3072
+ const res = await options.fetchImpl(
3073
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
3074
+ {
3075
+ method: "POST",
3076
+ headers: {
3077
+ Authorization: options.getAuthHeader(),
3078
+ "Content-Type": "application/json"
3079
+ },
3080
+ body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
1972
3081
  }
3082
+ );
3083
+ if (!res.ok) {
3084
+ recordAckFailure(
3085
+ options,
3086
+ file,
3087
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
3088
+ );
3089
+ return;
1973
3090
  }
1974
- this.reconnecting = false;
3091
+ options.ackFailures.delete(file.id);
3092
+ } catch (err) {
3093
+ recordAckFailure(
3094
+ options,
3095
+ file,
3096
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
3097
+ );
1975
3098
  }
1976
- };
3099
+ }
3100
+ function recordAckFailure(options, file, what) {
3101
+ const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
3102
+ options.ackFailures.set(file.id, attempts);
3103
+ options.log({
3104
+ level: "error",
3105
+ message: attempts >= MAX_ACK_ATTEMPTS ? `${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})`
3106
+ });
3107
+ }
3108
+ function describe(err) {
3109
+ return err instanceof Error ? err.message : String(err);
3110
+ }
1977
3111
 
1978
3112
  // src/lib/channels/driver.ts
1979
3113
  function messageIdOf(m) {
@@ -2003,7 +3137,10 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
2003
3137
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
2004
3138
  var HEARTBEAT_MS = 6e4;
2005
3139
  var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
3140
+ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
3141
+ var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2006
3142
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
3143
+ var MAX_SUPERSEDED_CONVERSATIONS = 256;
2007
3144
  var ChannelAuthError = class extends Error {
2008
3145
  constructor(message) {
2009
3146
  super(message);
@@ -2012,10 +3149,10 @@ var ChannelAuthError = class extends Error {
2012
3149
  };
2013
3150
  var ChannelTerminalError = class extends Error {
2014
3151
  status;
2015
- constructor(message, status) {
3152
+ constructor(message, status2) {
2016
3153
  super(message);
2017
3154
  this.name = "ChannelTerminalError";
2018
- this.status = status;
3155
+ this.status = status2;
2019
3156
  }
2020
3157
  };
2021
3158
  function backoffDelay(attempt, policy) {
@@ -2023,10 +3160,10 @@ function backoffDelay(attempt, policy) {
2023
3160
  const capped = Math.min(policy.maxDelayMs, exp);
2024
3161
  return Math.floor(Math.random() * capped);
2025
3162
  }
2026
- function isRetryableStatus(status) {
2027
- return status === 429 || status >= 500 && status <= 599;
3163
+ function isRetryableStatus(status2) {
3164
+ return status2 === 429 || status2 >= 500 && status2 <= 599;
2028
3165
  }
2029
- var ChannelDriver = class {
3166
+ var ChannelDriver = class _ChannelDriver {
2030
3167
  agentId;
2031
3168
  port;
2032
3169
  apiUrl;
@@ -2040,8 +3177,38 @@ var ChannelDriver = class {
2040
3177
  pausedMaxWaitMs;
2041
3178
  stuckQueuedMs;
2042
3179
  now;
3180
+ fileSyncDirectories;
3181
+ homeDir;
2043
3182
  /** Cache of conversationId → opencode sessionId. */
2044
3183
  sessions = /* @__PURE__ */ new Map();
3184
+ /**
3185
+ * conversationId → the opencode session this runner has ABANDONED as that
3186
+ * conversation's binding (#553), after a genuine (`sessionExists === true`)
3187
+ * dispatch failure: the session still exists but is wedged, so #485's self-heal
3188
+ * must bind a fresh one.
3189
+ *
3190
+ * Dropping the local binding + clearing the server row is not enough on its own:
3191
+ * a SIBLING message dispatched earlier in the same drain is still in-flight under
3192
+ * the same session, and its watcher's routine status writes carry
3193
+ * `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
3194
+ * and `ensureSession`'s persisted-id fallback then reuses it, defeating the
3195
+ * self-heal. This map makes the runner authoritative instead of racing those
3196
+ * writes: *`ensureSession` never reuses an abandoned id for that conversation,
3197
+ * whatever the server row says* — which holds even when the resurrecting write
3198
+ * is one we deliberately keep (see `markDone`).
3199
+ *
3200
+ * Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
3201
+ * one conversation hold ONE entry (the newest abandonment replaces the older), and
3202
+ * hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
3203
+ * NEWEST abandoned id per conversation is guarded: after a second abandonment a
3204
+ * late sibling of the FIRST session can write that id back and `ensureSession`
3205
+ * will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
3206
+ * NOT dropped when the session's watcher tears down: `markDone` still writes the
3207
+ * abandoned id back (it must, or the reply is lost), so the guard has to outlive
3208
+ * the turn that resurrects it. In-memory only — a restart forgets it, at the same
3209
+ * bounded cost.
3210
+ */
3211
+ supersededSessions = /* @__PURE__ */ new Map();
2045
3212
  /**
2046
3213
  * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
2047
3214
  * longer idempotent (no caller-supplied `messageID`), and its read-back picks
@@ -2109,6 +3276,24 @@ var ChannelDriver = class {
2109
3276
  * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
2110
3277
  */
2111
3278
  readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
3279
+ /**
3280
+ * "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
3281
+ * `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
3282
+ * every ~2s drain until opencode's status becomes readable, but the
3283
+ * server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
3284
+ * on any non-`unresolved` outcome so the set cannot grow beyond the currently
3285
+ * unresolvable rows.
3286
+ */
3287
+ redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
3288
+ /**
3289
+ * First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
3290
+ * `pending` row is invisible to every cron arm (all require `status =
3291
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
3292
+ * nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
3293
+ * takes `dispatch` instead of `unresolved` (reusing the existing knob — see
3294
+ * ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
3295
+ */
3296
+ redriveUnresolvedSince = /* @__PURE__ */ new Map();
2112
3297
  /**
2113
3298
  * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
2114
3299
  * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
@@ -2151,9 +3336,12 @@ var ChannelDriver = class {
2151
3336
  sessionParents = /* @__PURE__ */ new Map();
2152
3337
  /**
2153
3338
  * Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
2154
- * NON-EMPTY name is stored (terminal — a real session name won't later un-name),
2155
- * so we do NOT re-GET `/session/:id` every tick. A missing entry = not yet
2156
- * resolved OR resolved-but-still-empty re-fetch on next need, since OpenCode
3339
+ * NON-EMPTY, non-placeholder name is stored (terminal — a real session name
3340
+ * won't later un-name), so we do NOT re-GET `/session/:id` every tick. "Non-empty"
3341
+ * excludes OpenCode's synchronous default title (see
3342
+ * `OPENCODE_DEFAULT_TITLE_PREFIX`, #549) — that placeholder is treated the same
3343
+ * as an empty title so it never latches. A missing entry = not yet resolved OR
3344
+ * resolved-but-still-empty/placeholder → re-fetch on next need, since OpenCode
2157
3345
  * names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
2158
3346
  * the watcher completion path AND the restart-recovery re-adopt path (which has
2159
3347
  * no watcher) can resolve the title.
@@ -2161,6 +3349,24 @@ var ChannelDriver = class {
2161
3349
  sessionTitles = /* @__PURE__ */ new Map();
2162
3350
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
2163
3351
  draining = false;
3352
+ /**
3353
+ * Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
3354
+ * drain ping don't download, write and ack the same file twice.
3355
+ */
3356
+ syncingFiles = false;
3357
+ /**
3358
+ * Consecutive failed acks per pending file (#559). Lives on the driver so it
3359
+ * survives across drains — without it, a file whose ack keeps failing is
3360
+ * re-downloaded and re-written every ~2s until the server expires it.
3361
+ */
3362
+ fileAckFailures = /* @__PURE__ */ new Map();
3363
+ /**
3364
+ * Monotonic count of files this runner has pulled and written (#559). Only
3365
+ * ever increases, so `run.ts` detects work by comparing it against the value
3366
+ * it saw on the previous cycle — including work that landed mid-sleep, the
3367
+ * same trick `lastProxiedActivityAt` uses.
3368
+ */
3369
+ appliedFileCount = 0;
2164
3370
  /**
2165
3371
  * The currently-executing `drainPending()` promise, or null when idle. Lets a
2166
3372
  * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
@@ -2176,21 +3382,23 @@ var ChannelDriver = class {
2176
3382
  * and stops opencode.
2177
3383
  */
2178
3384
  stopped = false;
2179
- constructor(config2) {
2180
- this.agentId = config2.agentId;
2181
- this.port = config2.port;
2182
- this.apiUrl = config2.apiUrl.replace(/\/$/, "");
2183
- this.getAuthHeader = config2.getAuthHeader;
2184
- this.conversationFilter = config2.conversationFilter ?? null;
2185
- this.retry = { ...DEFAULT_RETRY_POLICY, ...config2.retry };
2186
- this.log = config2.log ?? (() => {
3385
+ constructor(config) {
3386
+ this.agentId = config.agentId;
3387
+ this.port = config.port;
3388
+ this.apiUrl = config.apiUrl.replace(/\/$/, "");
3389
+ this.getAuthHeader = config.getAuthHeader;
3390
+ this.conversationFilter = config.conversationFilter ?? null;
3391
+ this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
3392
+ this.log = config.log ?? (() => {
2187
3393
  });
2188
- this.fetchImpl = config2.fetchImpl ?? fetch;
2189
- this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
2190
- this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
2191
- this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2192
- this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2193
- this.now = config2.now ?? (() => Date.now());
3394
+ this.fetchImpl = config.fetchImpl ?? fetch;
3395
+ this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
3396
+ this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
3397
+ this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
3398
+ this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
3399
+ this.now = config.now ?? (() => Date.now());
3400
+ this.fileSyncDirectories = config.fileSyncDirectories ?? [];
3401
+ this.homeDir = config.homeDir ?? homedir2();
2194
3402
  }
2195
3403
  /** The IPv4-loopback base URL for the local `opencode serve`. */
2196
3404
  get opencodeBase() {
@@ -2218,6 +3426,47 @@ var ChannelDriver = class {
2218
3426
  );
2219
3427
  return run2;
2220
3428
  }
3429
+ /**
3430
+ * Pull-and-apply any files Evident has queued for this runner (#559), riding
3431
+ * the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
3432
+ * and drain ping that call `drainPending()`. There is deliberately no channel,
3433
+ * control frame or poll loop of its own: worst-case latency is one poll tick.
3434
+ *
3435
+ * NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
3436
+ * cost a conversation turn. Failures are logged and either acked as a terminal
3437
+ * outcome or left pending for the next drain (see `runner-file-sync.ts`).
3438
+ *
3439
+ * Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
3440
+ *
3441
+ * @returns the number of files written to disk.
3442
+ */
3443
+ async syncPendingFiles() {
3444
+ if (this.stopped) return 0;
3445
+ if (this.syncingFiles) return 0;
3446
+ this.syncingFiles = true;
3447
+ try {
3448
+ const applied = await syncPendingRunnerFiles({
3449
+ agentId: this.agentId,
3450
+ apiUrl: this.apiUrl,
3451
+ getAuthHeader: this.getAuthHeader,
3452
+ fetchImpl: this.fetchImpl,
3453
+ allowedDirectories: this.fileSyncDirectories,
3454
+ homeDir: this.homeDir,
3455
+ ackFailures: this.fileAckFailures,
3456
+ log: this.log
3457
+ });
3458
+ this.appliedFileCount += applied;
3459
+ return applied;
3460
+ } catch (err) {
3461
+ this.log({
3462
+ level: "error",
3463
+ message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
3464
+ });
3465
+ return 0;
3466
+ } finally {
3467
+ this.syncingFiles = false;
3468
+ }
3469
+ }
2221
3470
  async runDrain() {
2222
3471
  let dispatched = 0;
2223
3472
  try {
@@ -2251,6 +3500,28 @@ var ChannelDriver = class {
2251
3500
  }
2252
3501
  return false;
2253
3502
  }
3503
+ /**
3504
+ * File-pull work, for `run.ts`'s idle accounting (#559).
3505
+ *
3506
+ * Pulling a file is real work that `drainPending()` knows nothing about, so
3507
+ * without this a near-idle runner counts a credential pull as an empty tick
3508
+ * and `--idle-timeout` can `process.exit` mid-pull — leaving a
3509
+ * `.evident-push-*.tmp` behind — or immediately after the write, before the
3510
+ * browser has run the authorize/callback that activates it (the user then sees
3511
+ * `saved_not_activated` for a runner that was fine).
3512
+ *
3513
+ * Two signals because one cannot cover both cases: `inFlight` is the pull
3514
+ * happening RIGHT NOW (it may outlive the tick that started it), and
3515
+ * `appliedFiles` is monotonic so a pull that started AND finished between two
3516
+ * idle checks still shows up as an advance.
3517
+ *
3518
+ * CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
3519
+ * the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
3520
+ * samples afterwards reads `true` every single cycle and can never idle out.
3521
+ */
3522
+ fileSyncActivity() {
3523
+ return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
3524
+ }
2254
3525
  /**
2255
3526
  * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2256
3527
  * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
@@ -2312,7 +3583,7 @@ var ChannelDriver = class {
2312
3583
  await this.sleep(step);
2313
3584
  }
2314
3585
  }
2315
- while (this.hasInFlightWatchers()) {
3586
+ while (this.hasInFlightWatchers() || this.syncingFiles) {
2316
3587
  if (this.now() >= deadline) return false;
2317
3588
  await this.sleep(step);
2318
3589
  }
@@ -2347,16 +3618,27 @@ var ChannelDriver = class {
2347
3618
  * @returns the count of messages NEWLY dispatched (not already in-flight).
2348
3619
  */
2349
3620
  async processConversation(conv) {
2350
- const sessionId = await this.ensureSession(conv);
3621
+ const { sessionId, refusedSessionId } = await this.ensureSession(conv);
2351
3622
  const messages = await this.getPendingMessages(conv.id);
2352
3623
  let dispatched = 0;
2353
3624
  let skippedAlreadyDispatched = 0;
3625
+ if (refusedSessionId && messages.length > 0) {
3626
+ void this.postSignal(conv.id, messages[0].id, "session_superseded", {
3627
+ superseded_session_id: refusedSessionId
3628
+ });
3629
+ }
2354
3630
  for (const message of messages) {
2355
3631
  if (this.stopped) break;
2356
3632
  if (this.dispatched.has(message.id)) {
2357
3633
  skippedAlreadyDispatched += 1;
2358
3634
  continue;
2359
3635
  }
3636
+ if (message.opencode_message_id) {
3637
+ const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
3638
+ if (outcome !== "dispatch") {
3639
+ break;
3640
+ }
3641
+ }
2360
3642
  const options = {
2361
3643
  agent: message.opencode_agent ?? void 0,
2362
3644
  model: message.opencode_model ?? void 0
@@ -2365,64 +3647,351 @@ var ChannelDriver = class {
2365
3647
  try {
2366
3648
  this.log({
2367
3649
  level: "info",
2368
- message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
3650
+ message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
3651
+ conversation_id: conv.id,
3652
+ message_id: message.id
3653
+ });
3654
+ const sendAttachments = this.buildSendAttachments(conv, message);
3655
+ opencodeMessageId = await this.dispatchLocked(
3656
+ sessionId,
3657
+ () => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
3658
+ );
3659
+ } catch (err) {
3660
+ if (err instanceof ChannelAuthError) throw err;
3661
+ this.dispatched.delete(message.id);
3662
+ const exists = await sessionExists(this.port, sessionId);
3663
+ if (exists === false) {
3664
+ this.sessions.delete(conv.id);
3665
+ this.log({
3666
+ level: "warn",
3667
+ message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
3668
+ conversation_id: conv.id,
3669
+ message_id: message.id
3670
+ });
3671
+ break;
3672
+ }
3673
+ if (exists === null) {
3674
+ this.log({
3675
+ level: "warn",
3676
+ 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.`,
3677
+ conversation_id: conv.id,
3678
+ message_id: message.id
3679
+ });
3680
+ break;
3681
+ }
3682
+ const errorMessage = err instanceof Error ? err.message : String(err);
3683
+ this.sessions.delete(conv.id);
3684
+ this.supersede(conv.id, sessionId);
3685
+ this.log({
3686
+ level: "warn",
3687
+ 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.`,
3688
+ conversation_id: conv.id,
3689
+ message_id: message.id
3690
+ });
3691
+ await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
3692
+ this.log({
3693
+ level: "warn",
3694
+ 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)}`,
3695
+ conversation_id: conv.id,
3696
+ message_id: message.id
3697
+ });
3698
+ });
3699
+ this.log({
3700
+ level: "error",
3701
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
3702
+ conversation_id: conv.id,
3703
+ message_id: message.id
3704
+ });
3705
+ break;
3706
+ }
3707
+ if (opencodeMessageId === null) {
3708
+ this.log({
3709
+ level: "warn",
3710
+ message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
3711
+ conversation_id: conv.id,
3712
+ message_id: message.id
3713
+ });
3714
+ continue;
3715
+ }
3716
+ this.dispatched.add(message.id);
3717
+ this.registerInFlight(conv, sessionId, message, opencodeMessageId);
3718
+ dispatched += 1;
3719
+ void this.postSignal(conv.id, message.id, "dispatched");
3720
+ }
3721
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3722
+ this.log({
3723
+ level: "warn",
3724
+ message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
3725
+ conversation_id: conv.id
3726
+ });
3727
+ }
3728
+ this.ensureWatcherRunning(sessionId);
3729
+ return dispatched;
3730
+ }
3731
+ /**
3732
+ * Poll a session's message list for the re-drive fence (#965), via the
3733
+ * INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
3734
+ * hits the global `fetch` and would bypass the same override every other
3735
+ * opencode poll in this file respects. Mirrors `readoptProcessing`'s own
3736
+ * snapshot fetch (`:3081-3111`). `null` = unreadable (non-OK response,
3737
+ * non-array body, or a network exception) — treated as "can't observe",
3738
+ * never as "confirmed gone".
3739
+ */
3740
+ async pollSessionMessagesForRedrive(conv, message, sessionId) {
3741
+ try {
3742
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
3743
+ if (!res.ok) {
3744
+ this.log({
3745
+ level: "warn",
3746
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status} \u2014 treating as unreadable this tick`,
3747
+ conversation_id: conv.id,
3748
+ message_id: message.id
3749
+ });
3750
+ return null;
3751
+ }
3752
+ const body = await res.json();
3753
+ if (!Array.isArray(body)) {
3754
+ this.log({
3755
+ level: "warn",
3756
+ message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
3757
+ conversation_id: conv.id,
3758
+ message_id: message.id
3759
+ });
3760
+ return null;
3761
+ }
3762
+ return body;
3763
+ } catch (err) {
3764
+ this.log({
3765
+ level: "warn",
3766
+ message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3767
+ conversation_id: conv.id,
3768
+ message_id: message.id
3769
+ });
3770
+ return null;
3771
+ }
3772
+ }
3773
+ /**
3774
+ * The re-drive fence for a `pending` row that already carries a stored
3775
+ * `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
3776
+ * least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
3777
+ * The lifecycle cron can falsely reclaim a `processing` row back to `pending`
3778
+ * mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
3779
+ * without this fence the drain loop would re-`prompt_async` the SAME turn a
3780
+ * second time against live GitHub state. Mirrors `readoptOne`'s job for the
3781
+ * `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
3782
+ * needed here because `refusedSessionId` already handles the one case
3783
+ * (#553 abandoned session) that path exists for.
3784
+ *
3785
+ * Only `ChannelAuthError` propagates; every other failure resolves to
3786
+ * `unresolved` and is retried whole on the next ~2s drain tick.
3787
+ */
3788
+ async resolveRedrive(conv, sessionId, message, refusedSessionId) {
3789
+ const ocId = message.opencode_message_id ?? null;
3790
+ if (refusedSessionId) {
3791
+ this.clearRedriveUnresolved(message.id);
3792
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3793
+ return "dispatch";
3794
+ }
3795
+ const messages = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
3796
+ if (messages == null || messages.length === 0) {
3797
+ return this.resolveRedriveUnresolved(conv, message);
3798
+ }
3799
+ const state = messageRunState(messages, ocId ?? "");
3800
+ if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
3801
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3802
+ if (ongoing === false) {
3803
+ this.log({
3804
+ level: "info",
3805
+ message: `Re-drive: message ${message.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
3806
+ conversation_id: conv.id,
3807
+ message_id: message.id
3808
+ });
3809
+ this.clearRedriveUnresolved(message.id);
3810
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3811
+ return "dispatch";
3812
+ }
3813
+ }
3814
+ if (state === "done" || state === "failed") {
3815
+ return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
3816
+ }
3817
+ if (state === "running" || state === "queued") {
3818
+ const ongoing = await isSessionOngoing(this.port, sessionId);
3819
+ if (ongoing === true) {
3820
+ return this.reattachRedrive(conv, sessionId, message, ocId);
3821
+ }
3822
+ if (ongoing === false) {
3823
+ this.clearRedriveUnresolved(message.id);
3824
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3825
+ return "dispatch";
3826
+ }
3827
+ return this.resolveRedriveUnresolved(conv, message);
3828
+ }
3829
+ this.clearRedriveUnresolved(message.id);
3830
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3831
+ return "dispatch";
3832
+ }
3833
+ /**
3834
+ * The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
3835
+ * opencode's own status map — undo the false reclaim instead of starting a
3836
+ * second turn.
3837
+ */
3838
+ async reattachRedrive(conv, sessionId, message, ocId) {
3839
+ let anchorMs;
3840
+ const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
3841
+ if (!Number.isNaN(parsed)) {
3842
+ anchorMs = parsed;
3843
+ } else {
3844
+ anchorMs = this.now();
3845
+ this.log({
3846
+ level: "error",
3847
+ message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
3848
+ conversation_id: conv.id,
3849
+ message_id: message.id
3850
+ });
3851
+ }
3852
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3853
+ try {
3854
+ await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
3855
+ } catch (err) {
3856
+ if (err instanceof ChannelAuthError) throw err;
3857
+ this.log({
3858
+ level: "warn",
3859
+ message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3860
+ conversation_id: conv.id,
3861
+ message_id: message.id
3862
+ });
3863
+ return "unresolved";
3864
+ }
3865
+ this.clearRedriveUnresolved(message.id);
3866
+ this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
3867
+ this.dispatched.add(message.id);
3868
+ this.readopted.add(message.id);
3869
+ this.ensureWatcherRunning(sessionId);
3870
+ const watchedForMs = this.now() - anchorMs;
3871
+ void this.postSignal(conv.id, message.id, "redrive_reattached", {
3872
+ watched_for_ms: watchedForMs
3873
+ });
3874
+ this.log({
3875
+ level: "warn",
3876
+ message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
3877
+ conversation_id: conv.id,
3878
+ message_id: message.id
3879
+ });
3880
+ return "reattached";
3881
+ }
3882
+ /**
3883
+ * The `settled` outcome (Task 3.2): the prior turn already finished (or
3884
+ * errored) while nobody was watching — deliver/report it instead of re-running.
3885
+ * Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
3886
+ * (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
3887
+ * drain, same as any other non-auth failure). The restart-abort carve-out that
3888
+ * keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
3889
+ * so a row reaching this `failed` branch is a GENUINE failure.
3890
+ */
3891
+ async settleRedrive(conv, sessionId, message, ocId, messages, state) {
3892
+ try {
3893
+ if (state === "done") {
3894
+ const title = await this.resolveSessionTitle(sessionId, conv.id);
3895
+ const usage = messageUsage(messages, ocId ?? "");
3896
+ this.log({
3897
+ level: "info",
3898
+ message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
2369
3899
  conversation_id: conv.id,
2370
3900
  message_id: message.id
2371
3901
  });
2372
- const sendAttachments = this.buildSendAttachments(conv, message);
2373
- opencodeMessageId = await this.dispatchLocked(
2374
- sessionId,
2375
- () => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
2376
- );
2377
- } catch (err) {
2378
- if (err instanceof ChannelAuthError) throw err;
2379
- this.dispatched.delete(message.id);
2380
- if (await sessionExists(this.port, sessionId) === false) {
2381
- this.sessions.delete(conv.id);
2382
- this.log({
2383
- level: "warn",
2384
- message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
2385
- conversation_id: conv.id,
2386
- message_id: message.id
2387
- });
2388
- break;
2389
- }
2390
- await this.markFailed(conv.id, message.id).catch(() => {
2391
- });
3902
+ await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
3903
+ } else {
3904
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
3905
+ const usage = messageUsage(messages, ocId ?? "");
3906
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
2392
3907
  this.log({
2393
3908
  level: "error",
2394
- message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
2395
- conversation_id: conv.id,
2396
- message_id: message.id
2397
- });
2398
- continue;
2399
- }
2400
- if (opencodeMessageId === null) {
2401
- this.log({
2402
- level: "warn",
2403
- 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`,
3909
+ message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
2404
3910
  conversation_id: conv.id,
2405
3911
  message_id: message.id
2406
3912
  });
2407
- continue;
3913
+ await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
2408
3914
  }
2409
- this.dispatched.add(message.id);
2410
- this.registerInFlight(conv, sessionId, message, opencodeMessageId);
2411
- dispatched += 1;
2412
- void this.postSignal(conv.id, message.id, "dispatched");
2413
- }
2414
- if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
3915
+ } catch (err) {
3916
+ if (err instanceof ChannelAuthError) throw err;
2415
3917
  this.log({
2416
3918
  level: "warn",
2417
- 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).`,
2418
- conversation_id: conv.id
3919
+ message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3920
+ conversation_id: conv.id,
3921
+ message_id: message.id
2419
3922
  });
3923
+ return "unresolved";
2420
3924
  }
2421
- this.ensureWatcherRunning(sessionId);
2422
- return dispatched;
3925
+ this.clearRedriveUnresolved(message.id);
3926
+ void this.postSignal(conv.id, message.id, "redrive_settled");
3927
+ return "settled";
2423
3928
  }
3929
+ /**
3930
+ * The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
3931
+ * observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
3932
+ * A `pending` row is invisible to every cron arm (all require `status =
3933
+ * 'processing'`), so an indefinitely-`unresolved` row would be stranded with
3934
+ * nothing driving it — bound it to the existing `pausedMaxWaitMs` window
3935
+ * (reusing the knob, not a new constant) and take `dispatch` once elapsed.
3936
+ */
3937
+ resolveRedriveUnresolved(conv, message) {
3938
+ const now = this.now();
3939
+ const since = this.redriveUnresolvedSince.get(message.id);
3940
+ if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
3941
+ this.clearRedriveUnresolved(message.id);
3942
+ void this.postSignal(conv.id, message.id, "redrive_redispatched");
3943
+ return "dispatch";
3944
+ }
3945
+ if (since === void 0) {
3946
+ this.redriveUnresolvedSince.set(message.id, now);
3947
+ }
3948
+ if (!this.redriveUnresolvedSignalled.has(message.id)) {
3949
+ this.redriveUnresolvedSignalled.add(message.id);
3950
+ void this.postSignal(conv.id, message.id, "redrive_unresolved");
3951
+ }
3952
+ return "unresolved";
3953
+ }
3954
+ /** Clear both `unresolved`-bound trackers for a row (any non-`unresolved` outcome). */
3955
+ clearRedriveUnresolved(messageId) {
3956
+ this.redriveUnresolvedSince.delete(messageId);
3957
+ this.redriveUnresolvedSignalled.delete(messageId);
3958
+ }
3959
+ /**
3960
+ * Record that `sessionId` is no longer a valid binding for `conversationId`
3961
+ * (#553). Keyed by conversation and hard-capped, so it cannot grow with the
3962
+ * number of failures — see the `supersededSessions` field doc.
3963
+ */
3964
+ supersede(conversationId, sessionId) {
3965
+ this.supersededSessions.delete(conversationId);
3966
+ this.supersededSessions.set(conversationId, sessionId);
3967
+ while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
3968
+ const oldest = this.supersededSessions.keys().next().value;
3969
+ if (oldest === void 0) return;
3970
+ this.supersededSessions.delete(oldest);
3971
+ }
3972
+ }
3973
+ /** Whether `sessionId` is the session this conversation has abandoned (#553). */
3974
+ isSuperseded(conversationId, sessionId) {
3975
+ return this.supersededSessions.get(conversationId) === sessionId;
3976
+ }
3977
+ /**
3978
+ * Resolve the opencode session to run this conversation's turns in.
3979
+ *
3980
+ * `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
3981
+ * binding was an id this runner had abandoned, so a resurrection genuinely
3982
+ * happened and a fresh session was bound instead. The caller reports it.
3983
+ */
2424
3984
  async ensureSession(conv) {
2425
3985
  const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
3986
+ if (bound && this.isSuperseded(conv.id, bound)) {
3987
+ this.log({
3988
+ level: "warn",
3989
+ 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.`,
3990
+ conversation_id: conv.id
3991
+ });
3992
+ this.sessions.delete(conv.id);
3993
+ return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
3994
+ }
2426
3995
  if (bound) {
2427
3996
  const exists = await sessionExists(this.port, bound);
2428
3997
  if (exists === false) {
@@ -2432,12 +4001,12 @@ var ChannelDriver = class {
2432
4001
  conversation_id: conv.id
2433
4002
  });
2434
4003
  this.sessions.delete(conv.id);
2435
- return this.createAndBindSession(conv.id);
4004
+ return { sessionId: await this.createAndBindSession(conv.id) };
2436
4005
  }
2437
4006
  this.sessions.set(conv.id, bound);
2438
- return bound;
4007
+ return { sessionId: bound };
2439
4008
  }
2440
- return this.createAndBindSession(conv.id);
4009
+ return { sessionId: await this.createAndBindSession(conv.id) };
2441
4010
  }
2442
4011
  /**
2443
4012
  * Create a fresh OpenCode session for a conversation, cache the binding, and
@@ -2448,7 +4017,12 @@ var ChannelDriver = class {
2448
4017
  const directory = await this.resolveOpenCodeDirectory();
2449
4018
  const sessionId = await createOpenCodeSession(this.port, directory);
2450
4019
  this.sessions.set(conversationId, sessionId);
2451
- await this.persistSession(conversationId, sessionId).catch(() => {
4020
+ await this.persistSession(conversationId, sessionId).catch((err) => {
4021
+ this.log({
4022
+ level: "warn",
4023
+ message: `Persisting the OpenCode session binding ${sessionId.slice(0, 8)} for conversation ${conversationId.slice(0, 8)} failed (best-effort, not retried) \u2014 the completion PATCH also carries opencode_session_id, so the binding is repaired when the turn finishes: ${err instanceof Error ? err.message : String(err)}`,
4024
+ conversation_id: conversationId
4025
+ });
2452
4026
  });
2453
4027
  return sessionId;
2454
4028
  }
@@ -2525,7 +4099,11 @@ var ChannelDriver = class {
2525
4099
  * (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
2526
4100
  * (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
2527
4101
  * OMITS that one image and the text turn still sends — NEVER throws the turn.
2528
- * Failures are logged with context (no silent swallow).
4102
+ * A 404 body carrying `{ reason: 'needs_reauth' }` (#547 the server CONFIRMED
4103
+ * a Slack `files:read` scope problem via `files.info`) instead resolves the
4104
+ * `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
4105
+ * user to reconnect Slack instead of a generic "unavailable". Failures are
4106
+ * logged with context (no silent swallow).
2529
4107
  */
2530
4108
  async fetchAttachmentDataUrl(messageId, index, mime) {
2531
4109
  try {
@@ -2534,6 +4112,25 @@ var ChannelDriver = class {
2534
4112
  { headers: { Authorization: this.getAuthHeader() } }
2535
4113
  );
2536
4114
  if (!res.ok) {
4115
+ let reason;
4116
+ try {
4117
+ const body = await res.json();
4118
+ if (body && typeof body.reason === "string") reason = body.reason;
4119
+ } catch (parseErr) {
4120
+ this.log({
4121
+ level: "debug",
4122
+ 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`,
4123
+ message_id: messageId
4124
+ });
4125
+ }
4126
+ if (reason === "needs_reauth") {
4127
+ this.log({
4128
+ level: "error",
4129
+ 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)`,
4130
+ message_id: messageId
4131
+ });
4132
+ return { needsReauth: true };
4133
+ }
2537
4134
  this.log({
2538
4135
  level: "error",
2539
4136
  message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
@@ -2573,6 +4170,9 @@ var ChannelDriver = class {
2573
4170
  if (this.attachmentsSkippedSignalled.has(messageId)) return;
2574
4171
  this.attachmentsSkippedSignalled.add(messageId);
2575
4172
  const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
4173
+ const failedReason = outcomes.some(
4174
+ (o) => o.status === "failed" && o.reason === "needs_reauth"
4175
+ ) ? "needs_reauth" : void 0;
2576
4176
  this.log({
2577
4177
  level: "info",
2578
4178
  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`,
@@ -2582,7 +4182,8 @@ var ChannelDriver = class {
2582
4182
  void this.postSignal(conversationId, messageId, "attachments_skipped", {
2583
4183
  skipped,
2584
4184
  failed,
2585
- ...skipped > 0 ? { skipped_reason: skippedReason } : {}
4185
+ ...skipped > 0 ? { skipped_reason: skippedReason } : {},
4186
+ ...failedReason ? { failed_reason: failedReason } : {}
2586
4187
  });
2587
4188
  }
2588
4189
  /** Register a freshly-dispatched message with its session's watcher state. */
@@ -2613,12 +4214,17 @@ var ChannelDriver = class {
2613
4214
  stuckReported: false,
2614
4215
  lastAliveAt: 0,
2615
4216
  aliveInFlight: false,
4217
+ titleSynced: false,
4218
+ titleSyncInFlight: false,
2616
4219
  awaitingHumanLatched: false,
2617
4220
  pausedOnQuestion: false,
2618
4221
  pausedOnPermission: false,
2619
4222
  pausedClearConfirmed: false,
2620
4223
  pausedInFlight: false,
2621
- deliveryDeadlineAnchored: false
4224
+ deliveryDeadlineAnchored: false,
4225
+ b2PinnedSinceMs: 0,
4226
+ b2LastDescendantCheckMs: 0,
4227
+ b2AbandonedSignalled: false
2622
4228
  });
2623
4229
  }
2624
4230
  /**
@@ -2633,9 +4239,10 @@ var ChannelDriver = class {
2633
4239
  * opencode reports ACTIVELY `running` is watched to completion (its liveness
2634
4240
  * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
2635
4241
  * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
2636
- * handed to the cron. The old "the `deadline` must settle before the ~15-min
2637
- * cron or they double-drive" reasoning is superseded: liveness now settles the
2638
- * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
4242
+ * handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
4243
+ * runner still holds; a reclaimed row that already ran is never re-dispatched
4244
+ * while opencode reports its turn ongoing (readopt's own gate here, and the
4245
+ * `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
2639
4246
  * (only the appear-guard uses it).
2640
4247
  *
2641
4248
  * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
@@ -2686,12 +4293,17 @@ var ChannelDriver = class {
2686
4293
  // with no extra `re_adopted` signal needed (folds old WI-6).
2687
4294
  lastAliveAt: 0,
2688
4295
  aliveInFlight: false,
4296
+ titleSynced: false,
4297
+ titleSyncInFlight: false,
2689
4298
  awaitingHumanLatched: false,
2690
4299
  pausedOnQuestion: false,
2691
4300
  pausedOnPermission: false,
2692
4301
  pausedClearConfirmed: false,
2693
4302
  pausedInFlight: false,
2694
- deliveryDeadlineAnchored: false
4303
+ deliveryDeadlineAnchored: false,
4304
+ b2PinnedSinceMs: 0,
4305
+ b2LastDescendantCheckMs: 0,
4306
+ b2AbandonedSignalled: false
2695
4307
  });
2696
4308
  }
2697
4309
  /**
@@ -2853,58 +4465,7 @@ var ChannelDriver = class {
2853
4465
  }
2854
4466
  }
2855
4467
  if (state === "done") {
2856
- this.anchorDeliveryDeadline(inFlight);
2857
- if (!inFlight.done) {
2858
- this.log({
2859
- level: "info",
2860
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2861
- conversation_id: conv.id,
2862
- message_id: inFlight.evidentMessageId
2863
- });
2864
- const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2865
- const usage = messageUsage(messages, inFlight.opencodeMessageId);
2866
- try {
2867
- await this.markDone(
2868
- conv.id,
2869
- inFlight.evidentMessageId,
2870
- sessionId,
2871
- inFlight.opencodeMessageId,
2872
- title,
2873
- usage
2874
- );
2875
- } catch (err) {
2876
- if (err instanceof ChannelAuthError) throw err;
2877
- if (err instanceof ChannelTerminalError) {
2878
- this.log({
2879
- level: "warn",
2880
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2881
- conversation_id: conv.id,
2882
- message_id: inFlight.evidentMessageId
2883
- });
2884
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2885
- return;
2886
- }
2887
- if (this.now() >= inFlight.deadline) {
2888
- this.log({
2889
- level: "warn",
2890
- 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)}`,
2891
- conversation_id: conv.id,
2892
- message_id: inFlight.evidentMessageId
2893
- });
2894
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2895
- return;
2896
- }
2897
- this.log({
2898
- level: "warn",
2899
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2900
- conversation_id: conv.id,
2901
- message_id: inFlight.evidentMessageId
2902
- });
2903
- return;
2904
- }
2905
- inFlight.done = true;
2906
- }
2907
- this.removeInFlight(watcher, inFlight.evidentMessageId);
4468
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
2908
4469
  return;
2909
4470
  }
2910
4471
  if (state === "failed") {
@@ -2918,8 +4479,16 @@ var ChannelDriver = class {
2918
4479
  message_id: inFlight.evidentMessageId
2919
4480
  });
2920
4481
  const usage = messageUsage(messages, inFlight.opencodeMessageId);
4482
+ const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
2921
4483
  try {
2922
- await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2, usage);
4484
+ await this.markFailed(
4485
+ conv.id,
4486
+ inFlight.evidentMessageId,
4487
+ sessionId,
4488
+ error2,
4489
+ usage,
4490
+ failure
4491
+ );
2923
4492
  } catch (err) {
2924
4493
  if (err instanceof ChannelAuthError) throw err;
2925
4494
  if (err instanceof ChannelTerminalError) {
@@ -2964,6 +4533,44 @@ var ChannelDriver = class {
2964
4533
  });
2965
4534
  }
2966
4535
  const activelyRunning = state === "running" && !awaitingHuman;
4536
+ const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
4537
+ const snapshotReadable = messages != null && messages.length > 0;
4538
+ if (!pinnedNow) {
4539
+ if (snapshotReadable) {
4540
+ inFlight.b2PinnedSinceMs = 0;
4541
+ inFlight.b2LastDescendantCheckMs = 0;
4542
+ inFlight.b2AbandonedSignalled = false;
4543
+ }
4544
+ } else {
4545
+ if (inFlight.b2AbandonedSignalled) {
4546
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
4547
+ return;
4548
+ }
4549
+ if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
4550
+ const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
4551
+ if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
4552
+ inFlight.b2LastDescendantCheckMs = this.now();
4553
+ const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
4554
+ if (isB2AbandonmentConfirmed({
4555
+ pinnedForMs,
4556
+ minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
4557
+ descendantOngoing
4558
+ })) {
4559
+ inFlight.b2AbandonedSignalled = true;
4560
+ this.log({
4561
+ level: "warn",
4562
+ message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
4563
+ conversation_id: conv.id,
4564
+ message_id: id
4565
+ });
4566
+ void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
4567
+ watched_for_ms: pinnedForMs
4568
+ });
4569
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
4570
+ return;
4571
+ }
4572
+ }
4573
+ }
2967
4574
  if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
2968
4575
  this.log({
2969
4576
  level: "warn",
@@ -2983,6 +4590,18 @@ var ChannelDriver = class {
2983
4590
  inFlight.aliveInFlight = false;
2984
4591
  if (ok) inFlight.lastAliveAt = this.now();
2985
4592
  });
4593
+ if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
4594
+ inFlight.titleSyncInFlight = true;
4595
+ void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
4596
+ if (!title) {
4597
+ inFlight.titleSyncInFlight = false;
4598
+ return;
4599
+ }
4600
+ const ok = await this.patchConversationTitle(conv.id, title);
4601
+ inFlight.titleSyncInFlight = false;
4602
+ if (ok) inFlight.titleSynced = true;
4603
+ });
4604
+ }
2986
4605
  }
2987
4606
  if (awaitingHuman) {
2988
4607
  if (!inFlight.awaitingHumanLatched) {
@@ -3020,6 +4639,70 @@ var ChannelDriver = class {
3020
4639
  this.removeInFlight(watcher, inFlight.evidentMessageId);
3021
4640
  }
3022
4641
  }
4642
+ /**
4643
+ * Settle a message whose run-state has resolved `'done'` — extracted verbatim
4644
+ * (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
4645
+ * inline `state === 'done'` branch body, so a SECOND caller (the #721
4646
+ * b2-abandonment resolution) can reach the exact same completion behavior
4647
+ * (delivery-deadline anchoring, title resolution, usage extraction, and
4648
+ * `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
4649
+ * and risking the two copies silently drifting apart.
4650
+ */
4651
+ async settleMessageDone(sessionId, watcher, inFlight, messages) {
4652
+ const conv = watcher.conv;
4653
+ this.anchorDeliveryDeadline(inFlight);
4654
+ if (!inFlight.done) {
4655
+ this.log({
4656
+ level: "info",
4657
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
4658
+ conversation_id: conv.id,
4659
+ message_id: inFlight.evidentMessageId
4660
+ });
4661
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
4662
+ const usage = messageUsage(messages, inFlight.opencodeMessageId);
4663
+ try {
4664
+ await this.markDone(
4665
+ conv.id,
4666
+ inFlight.evidentMessageId,
4667
+ sessionId,
4668
+ inFlight.opencodeMessageId,
4669
+ title,
4670
+ usage
4671
+ );
4672
+ } catch (err) {
4673
+ if (err instanceof ChannelAuthError) throw err;
4674
+ if (err instanceof ChannelTerminalError) {
4675
+ this.log({
4676
+ level: "warn",
4677
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
4678
+ conversation_id: conv.id,
4679
+ message_id: inFlight.evidentMessageId
4680
+ });
4681
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4682
+ return;
4683
+ }
4684
+ if (this.now() >= inFlight.deadline) {
4685
+ this.log({
4686
+ level: "warn",
4687
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
4688
+ conversation_id: conv.id,
4689
+ message_id: inFlight.evidentMessageId
4690
+ });
4691
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4692
+ return;
4693
+ }
4694
+ this.log({
4695
+ level: "warn",
4696
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
4697
+ conversation_id: conv.id,
4698
+ message_id: inFlight.evidentMessageId
4699
+ });
4700
+ return;
4701
+ }
4702
+ inFlight.done = true;
4703
+ }
4704
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
4705
+ }
3023
4706
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
3024
4707
  /**
3025
4708
  * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
@@ -3119,7 +4802,10 @@ var ChannelDriver = class {
3119
4802
  * re-dispatched (at most once, see `forceReadoptRun`):
3120
4803
  * - `done` → `markDone` now (guarded like the watcher's done branch);
3121
4804
  * - `failed` → `markFailed` with the surfaced error (issue #182), so an
3122
- * errored turn is reported failed on restart, NOT re-dispatched;
4805
+ * errored turn is reported failed on restart, NOT re-dispatched
4806
+ * EXCEPT a restart-ABORTED turn under a not-ongoing session,
4807
+ * which is a restart orphan wearing a terminal error and is
4808
+ * re-dispatched instead (issue #1310, see the branch below);
3123
4809
  * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
3124
4810
  * tracking the stored id so the reply correlates by it;
3125
4811
  * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
@@ -3183,9 +4869,19 @@ var ChannelDriver = class {
3183
4869
  void this.postSignal(row.conversation_id, row.id, "readopt_done");
3184
4870
  return;
3185
4871
  }
3186
- if (state === "failed") {
4872
+ const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
4873
+ if (restartAborted) {
4874
+ this.log({
4875
+ level: "info",
4876
+ message: `Re-adopt: message ${row.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
4877
+ conversation_id: row.conversation_id,
4878
+ message_id: row.id
4879
+ });
4880
+ }
4881
+ if (state === "failed" && !restartAborted) {
3187
4882
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
3188
4883
  const usage = messageUsage(messages, ocId ?? "");
4884
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3189
4885
  this.log({
3190
4886
  level: "error",
3191
4887
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -3193,7 +4889,7 @@ var ChannelDriver = class {
3193
4889
  message_id: row.id
3194
4890
  });
3195
4891
  try {
3196
- await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
4892
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
3197
4893
  } catch (err) {
3198
4894
  if (err instanceof ChannelAuthError) throw err;
3199
4895
  if (err instanceof ChannelTerminalError) {
@@ -3460,7 +5156,8 @@ var ChannelDriver = class {
3460
5156
  opencode_model: row.opencode_model,
3461
5157
  source_message_id: row.source_message_id,
3462
5158
  slack_user_id: row.slack_user_id,
3463
- attachments: row.attachments ?? null
5159
+ attachments: row.attachments ?? null,
5160
+ opencode_message_id: row.opencode_message_id
3464
5161
  };
3465
5162
  }
3466
5163
  /**
@@ -3599,6 +5296,47 @@ var ChannelDriver = class {
3599
5296
  }
3600
5297
  return false;
3601
5298
  }
5299
+ /**
5300
+ * Tri-state variant of the upward parentID membership walk (#721), used ONLY
5301
+ * by `isAnyDescendantSessionOngoing`. Walks the SAME cached
5302
+ * `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
5303
+ * `sessionBelongsTo`, which deliberately collapses "confirmed not a
5304
+ * descendant" and "the walk's fetch failed" into the same `false` (safe for
5305
+ * its OTHER callers: interaction attribution and the recovery-path
5306
+ * `isAnyDescendantSessionAlive`, both of which just retry next tick with no
5307
+ * safety consequence either way) — this variant keeps those two outcomes
5308
+ * SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
5309
+ * (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
5310
+ * not ongoing".
5311
+ *
5312
+ * Return contract:
5313
+ * - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
5314
+ * - `false` → the walk reached a definitive, parent-less root session
5315
+ * WITHOUT ever matching `rootSessionId` — `sessionId` is
5316
+ * CONFIRMED NOT a descendant of it.
5317
+ * - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
5318
+ * through the walk (`resolveSessionParent` returned `undefined`),
5319
+ * or the depth cap (32) was hit without a definitive answer (a
5320
+ * pathological/cyclic chain proves nothing either way). NEVER
5321
+ * treat this the same as `false` — see `sessionBelongsTo`'s own
5322
+ * doc comment above for why that collapse is safe THERE but not
5323
+ * here.
5324
+ *
5325
+ * `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
5326
+ * to the live-path descendant check, not a modification of shared code used
5327
+ * by interaction attribution or the recovery path.
5328
+ */
5329
+ async resolveSessionMembership(sessionId, rootSessionId) {
5330
+ let current = sessionId;
5331
+ for (let depth = 0; current && depth < 32; depth++) {
5332
+ if (current === rootSessionId) return true;
5333
+ const parent = await this.resolveSessionParent(current);
5334
+ if (parent === void 0) return null;
5335
+ if (parent === null) return false;
5336
+ current = parent;
5337
+ }
5338
+ return null;
5339
+ }
3602
5340
  /**
3603
5341
  * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3604
5342
  * `null` for a root session (no parent) and `undefined` when opencode is
@@ -3621,19 +5359,36 @@ var ChannelDriver = class {
3621
5359
  if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3622
5360
  return parent;
3623
5361
  }
5362
+ /**
5363
+ * OpenCode's synchronous default session title (e.g.
5364
+ * `"New session - 1737800000000"`), assigned immediately when a session is
5365
+ * created — before OpenCode's async LLM-based auto-titling later renames it
5366
+ * mid-turn (#549). Matched by this literal, case-sensitive prefix only; the
5367
+ * timestamp suffix's exact format is deliberately NOT matched, since the prefix
5368
+ * alone is the stable, cheap signal and over-anchoring on the timestamp
5369
+ * representation risks silently breaking if OpenCode ever changes it. Accepted
5370
+ * trade-off: a genuine LLM-assigned title that happens to literally start with
5371
+ * this prefix would also fail to latch (see `resolveSessionTitle`) —
5372
+ * vanishingly unlikely in practice, and deliberately not engineered around.
5373
+ */
5374
+ static OPENCODE_DEFAULT_TITLE_PREFIX = /^New session - /;
3624
5375
  /**
3625
5376
  * Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
3626
5377
  * status PATCH can carry it into the "Live sessions" list. Driver-level cache so
3627
5378
  * BOTH the watcher completion path and the restart-recovery re-adopt path (which
3628
5379
  * has no watcher) can use it. `conversationId` is passed only for log context.
3629
5380
  * Best-effort:
3630
- * - a resolved NON-EMPTY title is cached and terminal (a real session name
5381
+ * - a resolved NON-EMPTY title that does NOT match
5382
+ * `OPENCODE_DEFAULT_TITLE_PREFIX` is cached and terminal (a real session name
3631
5383
  * won't later un-name), so we do NOT re-GET `/session/:id` every tick;
3632
- * - while the title is still absent/empty we do NOT latch it — OpenCode names
3633
- * sessions asynchronously mid-turn, so an early call (e.g. at `processing`)
3634
- * must leave the cache unresolved and re-fetch on the next need so a later
3635
- * call (e.g. at `done`) picks up the name assigned in the meantime. Such a
3636
- * call returns `null` (omit the title on THIS PATCH) without caching;
5384
+ * - while the title is still absent, empty, or matches the OpenCode
5385
+ * placeholder prefix (#549) we do NOT latch it OpenCode names sessions
5386
+ * asynchronously mid-turn, so an early call (e.g. at `processing`) must leave
5387
+ * the cache unresolved and re-fetch on the next need so a later call (e.g. at
5388
+ * `done`) picks up the name assigned in the meantime. Such a call returns
5389
+ * `null` (omit the title on THIS PATCH) without caching. If a session is
5390
+ * never renamed, the title is omitted forever rather than ever persisting
5391
+ * the placeholder as a last resort;
3637
5392
  * - a failed request likewise leaves the cache unresolved (retry next need)
3638
5393
  * and returns `null` — it must NEVER throw or block completion.
3639
5394
  * A failure is logged with agent/session context (no silent catch).
@@ -3646,7 +5401,7 @@ var ChannelDriver = class {
3646
5401
  if (res.ok) {
3647
5402
  const body = await res.json();
3648
5403
  const title = body && typeof body.title === "string" ? body.title.trim() : "";
3649
- if (title.length > 0) {
5404
+ if (title.length > 0 && !_ChannelDriver.OPENCODE_DEFAULT_TITLE_PREFIX.test(title)) {
3650
5405
  this.sessionTitles.set(sessionId, title);
3651
5406
  return title;
3652
5407
  }
@@ -3666,6 +5421,54 @@ var ChannelDriver = class {
3666
5421
  }
3667
5422
  return null;
3668
5423
  }
5424
+ /**
5425
+ * Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
5426
+ * session title onto the conversation via the PLAIN conversation-update
5427
+ * endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
5428
+ * message-status endpoint `markProcessing`/`markDone` use. Deliberately a
5429
+ * separate, lighter call: it carries no `status`, so it cannot re-trigger the
5430
+ * `processing`/`done` transition side effects (Slack notices, activity-log
5431
+ * rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
5432
+ * ever touches `conversations.title`. That route (`routes/conversations.ts`)
5433
+ * skips a title write matching the stored value, so a redundant call with the
5434
+ * same title is a real no-op — it does not bump `updated_at`, which the
5435
+ * conversation list sorts and paginates on. (Note this is a DIFFERENT guard
5436
+ * from `threads.ts`'s "non-empty AND changed" one, which only covers the
5437
+ * message-status PATCH; the non-empty half is enforced here instead, by
5438
+ * `resolveSessionTitle` never returning an empty/placeholder title.)
5439
+ *
5440
+ * Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
5441
+ * is logged and the title is simply retried on the next heartbeat tick (the
5442
+ * caller only latches `titleSynced` on `true`).
5443
+ */
5444
+ async patchConversationTitle(conversationId, title) {
5445
+ try {
5446
+ const res = await this.fetchImpl(
5447
+ `${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
5448
+ {
5449
+ method: "PATCH",
5450
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
5451
+ body: JSON.stringify({ title })
5452
+ }
5453
+ );
5454
+ if (!res.ok) {
5455
+ this.log({
5456
+ level: "debug",
5457
+ message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
5458
+ conversation_id: conversationId
5459
+ });
5460
+ return false;
5461
+ }
5462
+ return true;
5463
+ } catch (err) {
5464
+ this.log({
5465
+ level: "debug",
5466
+ message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
5467
+ conversation_id: conversationId
5468
+ });
5469
+ return false;
5470
+ }
5471
+ }
3669
5472
  /**
3670
5473
  * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3671
5474
  * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
@@ -3724,6 +5527,84 @@ var ChannelDriver = class {
3724
5527
  }
3725
5528
  return false;
3726
5529
  }
5530
+ /**
5531
+ * LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
5532
+ * sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
5533
+ * in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
5534
+ *
5535
+ * Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
5536
+ * cross-check above): that method judges liveness from the child's OWN
5537
+ * TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
5538
+ * on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
5539
+ * path the local opencode server IS running, so its in-memory status map is
5540
+ * live and authoritative — and per ADR-0047 §4a ("the child has its own entry
5541
+ * [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
5542
+ * ENTIRE turn (including any tool call it is itself executing), not a
5543
+ * per-message transcript snapshot. This sidesteps the "child's own tool is
5544
+ * executing, between its step's completion and the next generation step"
5545
+ * transcript gap that a transcript-based check would need a second,
5546
+ * sustained-window bound to guard against — it is simply not derived from
5547
+ * message timestamps at all.
5548
+ *
5549
+ * Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
5550
+ * status, as the recovery path does per §4a)? Because on the LIVE path the
5551
+ * root session can be shared: a SECOND, unrelated user message can land on the
5552
+ * SAME session (issue #721's own root cause) and keep the root `busy` for a
5553
+ * reason that has nothing to do with THIS message's delegation. A `task`
5554
+ * descendant session is spawned for exactly one delegated turn and never
5555
+ * reused, so its OWN status-map entry is unambiguous evidence about that one
5556
+ * delegation — which the root's status is not.
5557
+ *
5558
+ * Why membership is checked via `resolveSessionMembership`, NOT
5559
+ * `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
5560
+ * `GET /session/:id` fetch failure into "not a descendant", which would
5561
+ * silently drop a genuinely-live candidate from consideration on the one
5562
+ * unlucky tick its membership-walk fetch hiccups (#721).
5563
+ * `resolveSessionMembership` keeps that failure mode as a distinct `null`
5564
+ * (indeterminate) so it is folded into THIS method's own `indeterminate` flag
5565
+ * instead.
5566
+ *
5567
+ * Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
5568
+ * - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
5569
+ * - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
5570
+ * confirmed either way (`resolveSessionMembership` never
5571
+ * returned `null`), and every CONFIRMED descendant's status read
5572
+ * succeeded and is not ongoing (includes "no descendant session
5573
+ * exists at all" — e.g. a plain, non-`task` tool call).
5574
+ * - `null` → INDETERMINATE: `listSessions` failed, OR at least one
5575
+ * candidate's MEMBERSHIP could not be confirmed
5576
+ * (`resolveSessionMembership` returned `null` — a fetch failure
5577
+ * or pathological chain partway through the parent walk), OR at
5578
+ * least one CONFIRMED descendant's `isSessionOngoing` read
5579
+ * failed — and no OTHER candidate was already confirmed `true`.
5580
+ * The caller MUST NOT treat `null` the same as `false` here
5581
+ * (unlike the recovery cross-check's contract) — see
5582
+ * `isB2AbandonmentConfirmed`.
5583
+ */
5584
+ async isAnyDescendantSessionOngoing(rootSessionId) {
5585
+ const sessions = await listSessions(this.port);
5586
+ if (!sessions) {
5587
+ this.log({
5588
+ level: "warn",
5589
+ message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
5590
+ });
5591
+ return null;
5592
+ }
5593
+ let indeterminate = false;
5594
+ for (const candidate of sessions) {
5595
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
5596
+ const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
5597
+ if (membership === null) {
5598
+ indeterminate = true;
5599
+ continue;
5600
+ }
5601
+ if (membership === false) continue;
5602
+ const ongoing = await isSessionOngoing(this.port, candidate.id);
5603
+ if (ongoing === true) return true;
5604
+ if (ongoing === null) indeterminate = true;
5605
+ }
5606
+ return indeterminate ? null : false;
5607
+ }
3727
5608
  /**
3728
5609
  * Cheap decision-telemetry label for a running row's LAST correlated reply
3729
5610
  * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
@@ -3852,6 +5733,32 @@ var ChannelDriver = class {
3852
5733
  }
3853
5734
  return messages;
3854
5735
  }
5736
+ /**
5737
+ * The `opencode_session_id` fragment of a status PATCH body — `{}` when this
5738
+ * conversation has ABANDONED that session (#553). The field is optional
5739
+ * server-side and an absent one leaves the persisted binding untouched, so
5740
+ * omitting it is how a routine status write stops resurrecting it.
5741
+ *
5742
+ * ONLY for writes whose sole cost is a lost deep link. The `processing` notice
5743
+ * degrades to no "View in Evident" link (the reaction swap still fires) and the
5744
+ * turn-failure notice is built from the PATCH's own `error` text with a link off
5745
+ * the persisted row — neither loses content the user came for. `markDone`
5746
+ * deliberately does NOT use this helper: the server fetches the reply text
5747
+ * THROUGH the session id it is given, so suppressing there would replace the
5748
+ * agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
5749
+ * `ensureSession` guard, not this suppression, is what makes the self-heal
5750
+ * stick.
5751
+ */
5752
+ sessionIdBody(sessionId, conversationId, messageId, status2) {
5753
+ if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
5754
+ this.log({
5755
+ level: "debug",
5756
+ message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
5757
+ conversation_id: conversationId,
5758
+ message_id: messageId
5759
+ });
5760
+ return {};
5761
+ }
3855
5762
  /**
3856
5763
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
3857
5764
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -3881,7 +5788,7 @@ var ChannelDriver = class {
3881
5788
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3882
5789
  body: JSON.stringify({
3883
5790
  status: "processing",
3884
- opencode_session_id: sessionId,
5791
+ ...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
3885
5792
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3886
5793
  ...title ? { title } : {}
3887
5794
  })
@@ -3930,6 +5837,11 @@ var ChannelDriver = class {
3930
5837
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3931
5838
  body: JSON.stringify({
3932
5839
  status: "done",
5840
+ // ALWAYS sent, even for a session this conversation has abandoned
5841
+ // (#553): the server reads the reply text back out of THIS session id
5842
+ // to deliver it. Omitting it would leave the user with "✅ Done!"
5843
+ // instead of the answer — a worse regression than the resurrection it
5844
+ // would prevent, which `ensureSession`'s guard handles anyway.
3933
5845
  opencode_session_id: sessionId,
3934
5846
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3935
5847
  ...title ? { title } : {},
@@ -3946,16 +5858,31 @@ var ChannelDriver = class {
3946
5858
  }
3947
5859
  /**
3948
5860
  * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3949
- * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3950
- * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3951
- * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3952
- * failure reason reaches the channel.
5861
+ * when provided (issue #182). Three states for `sessionId`:
5862
+ * - omitted (`undefined`) → don't send the field, leave the persisted
5863
+ * session untouched (unused today; kept for API symmetry).
5864
+ * - a real id (`string`) → send it, update the persisted session (the
5865
+ * turn-failure call sites: an errored OpenCode turn).
5866
+ * - explicit `null` → send it, CLEAR the persisted session (issue
5867
+ * #485's dispatch-handoff-failure call site: the session id still
5868
+ * exists but is wedged, so the next attempt must get a fresh one
5869
+ * instead of reusing it — see WI-1's server-side null-clearing PATCH).
3953
5870
  */
3954
- async markFailed(conversationId, messageId, sessionId, error2, usage) {
5871
+ async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
3955
5872
  const body = { status: "failed" };
3956
- if (sessionId !== void 0) body.opencode_session_id = sessionId;
5873
+ if (sessionId === null) {
5874
+ body.opencode_session_id = null;
5875
+ } else if (sessionId !== void 0) {
5876
+ Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
5877
+ }
3957
5878
  if (error2 !== void 0) body.error = error2;
3958
5879
  if (usage) Object.assign(body, usage);
5880
+ if (failure) {
5881
+ body.failure_kind = failure.kind;
5882
+ body.failure_provider_id = failure.providerId;
5883
+ body.failure_model_id = failure.modelId;
5884
+ body.failure_reason = failure.reason;
5885
+ }
3959
5886
  await this.callWithRetry(
3960
5887
  "marking message as failed",
3961
5888
  () => this.fetchImpl(
@@ -3968,6 +5895,29 @@ var ChannelDriver = class {
3968
5895
  )
3969
5896
  );
3970
5897
  }
5898
+ /**
5899
+ * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
5900
+ *
5901
+ * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
5902
+ * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
5903
+ * to the P1-2b zero-provider check — one extra loopback call to
5904
+ * `hasAnyConfiguredProvider`, only reached when the structured classifier
5905
+ * couldn't place it. Fails open (never throws): a fallback probe failure
5906
+ * (`null`/indeterminate) leaves the classification `null`, which produces
5907
+ * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
5908
+ */
5909
+ async classifyModelAuthFailure(messages, userMessageId) {
5910
+ const classified = messageFailure(messages, userMessageId);
5911
+ if (classified != null) return classified;
5912
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
5913
+ const hasProvider = await hasAnyConfiguredProvider(this.port);
5914
+ return applyZeroProviderFallback(
5915
+ classified,
5916
+ hasProvider,
5917
+ reply?.info?.providerID ?? null,
5918
+ reply?.info?.modelID ?? null
5919
+ );
5920
+ }
3971
5921
  /**
3972
5922
  * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3973
5923
  * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
@@ -4120,10 +6070,16 @@ var ChannelDriver = class {
4120
6070
  import chalk5 from "chalk";
4121
6071
  import ora2 from "ora";
4122
6072
  import { select as select2 } from "@inquirer/prompts";
6073
+ var INTERACTIVE_START_TIMEOUT_MS = 3e4;
4123
6074
  async function ensureOpenCodeRunning(ctx) {
4124
6075
  const healthCheck = await checkOpenCodeHealth(ctx.port);
4125
6076
  if (healthCheck.healthy) {
4126
- return { port: ctx.port, process: null, version: healthCheck.version ?? null };
6077
+ return {
6078
+ port: ctx.port,
6079
+ process: null,
6080
+ version: healthCheck.version ?? null,
6081
+ notReadyReason: null
6082
+ };
4127
6083
  }
4128
6084
  const runningInstances = await findHealthyOpenCodeInstances();
4129
6085
  if (runningInstances.length > 0) {
@@ -4144,7 +6100,7 @@ async function ensureOpenCodeRunning(ctx) {
4144
6100
  console.log(chalk5.yellow("Tip: Run with the correct port:"));
4145
6101
  console.log(
4146
6102
  chalk5.dim(
4147
- ` ${getCliName()} run --agent ${ctx.agentId} --port ${runningInstances[0].port}`
6103
+ ` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
4148
6104
  )
4149
6105
  );
4150
6106
  }
@@ -4164,14 +6120,22 @@ async function ensureOpenCodeRunning(ctx) {
4164
6120
  if (!ctx.interactive) {
4165
6121
  ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
4166
6122
  const proc = await startOpenCode(ctx.port);
4167
- const health = await waitForOpenCodeHealth(ctx.port, 3e4);
6123
+ const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
4168
6124
  if (!health.healthy) {
4169
- throw new Error(
4170
- `OpenCode failed to start on port ${ctx.port}. Install with: npm install -g opencode-ai`
4171
- );
6125
+ return {
6126
+ port: ctx.port,
6127
+ process: proc,
6128
+ version: null,
6129
+ notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
6130
+ };
4172
6131
  }
4173
6132
  ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
4174
- return { port: ctx.port, process: proc, version: health.version ?? null };
6133
+ return {
6134
+ port: ctx.port,
6135
+ process: proc,
6136
+ version: health.version ?? null,
6137
+ notReadyReason: null
6138
+ };
4175
6139
  }
4176
6140
  let port = ctx.port;
4177
6141
  if (isPortInUse(port)) {
@@ -4224,122 +6188,15 @@ Port ${port} is already in use.`));
4224
6188
  if (action === "start") {
4225
6189
  const spinner = ora2("Starting OpenCode...").start();
4226
6190
  const proc = await startOpenCode(port);
4227
- const health = await waitForOpenCodeHealth(port, 3e4);
6191
+ const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
4228
6192
  if (!health.healthy) {
4229
6193
  spinner.fail("Failed to start OpenCode");
4230
6194
  throw new Error("OpenCode failed to start");
4231
6195
  }
4232
6196
  spinner.stop();
4233
- return { port, process: proc, version: health.version ?? null };
4234
- }
4235
- return { port, process: null, version: null };
4236
- }
4237
-
4238
- // src/commands/agent-lookup.ts
4239
- async function readErrorMessage(response) {
4240
- const text = await response.text().catch(() => "");
4241
- if (!text) return response.statusText || void 0;
4242
- try {
4243
- const data = JSON.parse(text);
4244
- const message = data.message ?? data.error;
4245
- if (typeof message === "string" && message.trim()) {
4246
- return message;
4247
- }
4248
- } catch {
4249
- }
4250
- return text.trim() || response.statusText || void 0;
4251
- }
4252
- function authFailureHint(apiUrl, serverMessage) {
4253
- const reason = serverMessage ? `: ${serverMessage}` : "";
4254
- return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
4255
- }
4256
- async function resolveAgentIdFromKey(authHeader) {
4257
- const apiUrl = getApiUrlConfig();
4258
- try {
4259
- const response = await fetch(`${apiUrl}/me`, {
4260
- headers: { Authorization: authHeader }
4261
- });
4262
- if (response.status === 401) {
4263
- const serverMessage = await readErrorMessage(response);
4264
- return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
4265
- }
4266
- if (!response.ok) {
4267
- const serverMessage = await readErrorMessage(response);
4268
- return {
4269
- error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
4270
- };
4271
- }
4272
- const data = await response.json();
4273
- if (data.auth_type === "agent_key" && data.agent_id) {
4274
- return { agent_id: data.agent_id };
4275
- }
4276
- return {
4277
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
4278
- };
4279
- } catch (error2) {
4280
- const message = error2 instanceof Error ? error2.message : "Unknown error";
4281
- return { error: `Failed to resolve runner from key: ${message}` };
4282
- }
4283
- }
4284
- async function notifyAgentDisconnected(agentId, authHeader) {
4285
- const apiUrl = getApiUrlConfig();
4286
- try {
4287
- const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
4288
- method: "POST",
4289
- headers: { Authorization: authHeader }
4290
- });
4291
- if (!response.ok) {
4292
- const serverMessage = await readErrorMessage(response);
4293
- return {
4294
- ok: false,
4295
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
4296
- };
4297
- }
4298
- return { ok: true };
4299
- } catch (error2) {
4300
- return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
4301
- }
4302
- }
4303
- async function getAgentInfo(agentId, authHeader) {
4304
- const apiUrl = getApiUrlConfig();
4305
- try {
4306
- const response = await fetch(`${apiUrl}/runners/${agentId}`, {
4307
- headers: { Authorization: authHeader }
4308
- });
4309
- if (response.status === 401) {
4310
- const serverMessage = await readErrorMessage(response);
4311
- return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
4312
- }
4313
- if (response.status === 403) {
4314
- const serverMessage = await readErrorMessage(response);
4315
- return {
4316
- valid: false,
4317
- error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
4318
- };
4319
- }
4320
- if (response.status === 404) {
4321
- const serverMessage = await readErrorMessage(response);
4322
- return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
4323
- }
4324
- if (!response.ok) {
4325
- const serverMessage = await readErrorMessage(response);
4326
- return {
4327
- valid: false,
4328
- error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
4329
- };
4330
- }
4331
- const agent = await response.json();
4332
- if (agent.agent_type !== "local") {
4333
- return {
4334
- valid: false,
4335
- error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
4336
- };
4337
- }
4338
- return { valid: true, agent };
4339
- } catch (error2) {
4340
- const message = error2 instanceof Error ? error2.message : "Unknown error";
4341
- return { valid: false, error: `Failed to validate runner: ${message}` };
6197
+ return { port, process: proc, version: health.version ?? null, notReadyReason: null };
4342
6198
  }
6199
+ return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
4343
6200
  }
4344
6201
 
4345
6202
  // src/commands/run.ts
@@ -4347,6 +6204,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
4347
6204
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
4348
6205
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4349
6206
  var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
6207
+ var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
4350
6208
  function resolveLogLevel(options) {
4351
6209
  const accepted = Object.keys(LOG_LEVELS);
4352
6210
  const validate = (value, source) => {
@@ -4370,6 +6228,63 @@ function resolveLogLevel(options) {
4370
6228
  }
4371
6229
  return "info";
4372
6230
  }
6231
+ function resolveFileSyncDirectories(raw, homeDir) {
6232
+ const directories = [];
6233
+ for (const entry of raw ?? []) {
6234
+ const trimmed = entry.trim();
6235
+ if (trimmed === "") {
6236
+ throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
6237
+ }
6238
+ const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
6239
+ if (!isAbsolute2(expanded)) {
6240
+ throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
6241
+ }
6242
+ const normalized = resolvePath(expanded);
6243
+ if (parse(normalized).root === normalized) {
6244
+ throw new Error(
6245
+ `--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
6246
+ );
6247
+ }
6248
+ if (!directories.includes(normalized)) {
6249
+ directories.push(normalized);
6250
+ }
6251
+ }
6252
+ if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
6253
+ throw new Error(
6254
+ `--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
6255
+ );
6256
+ }
6257
+ return directories;
6258
+ }
6259
+ var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
6260
+ var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
6261
+ var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
6262
+ function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
6263
+ const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
6264
+ let raw;
6265
+ let source;
6266
+ if (options.opencodeStartTimeout !== void 0) {
6267
+ raw = options.opencodeStartTimeout;
6268
+ source = "--opencode-start-timeout";
6269
+ } else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
6270
+ raw = env[OPENCODE_START_TIMEOUT_ENV];
6271
+ source = OPENCODE_START_TIMEOUT_ENV;
6272
+ } else {
6273
+ return { timeoutMs: defaultMs, warnings: [] };
6274
+ }
6275
+ const trimmed = raw.trim();
6276
+ const seconds = Number(trimmed);
6277
+ const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
6278
+ if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
6279
+ return {
6280
+ timeoutMs: defaultMs,
6281
+ warnings: [
6282
+ `Ignoring invalid ${source} "${raw}": expected a positive integer number of seconds (at most ${MAX_OPENCODE_START_TIMEOUT_SECONDS}); using the default ${DEFAULT_OPENCODE_START_TIMEOUT_SECONDS}s`
6283
+ ]
6284
+ };
6285
+ }
6286
+ return { timeoutMs: seconds * 1e3, warnings: [] };
6287
+ }
4373
6288
  function meetsThreshold(state, level) {
4374
6289
  return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
4375
6290
  }
@@ -4391,6 +6306,10 @@ function log2(state, message, level = "info") {
4391
6306
  function logActivity(state, entry) {
4392
6307
  const level = entry.level ?? (entry.type === "error" ? "error" : "info");
4393
6308
  if (!meetsThreshold(state, level)) return;
6309
+ forwardRunnerActivity(
6310
+ { level, message: entry.message, error: entry.error },
6311
+ { agentId: state.agentId, authHeader: state.authHeader }
6312
+ );
4394
6313
  const fullEntry = {
4395
6314
  ...entry,
4396
6315
  level,
@@ -4490,23 +6409,46 @@ async function handleAuthError(state, error2) {
4490
6409
  }
4491
6410
  async function driveChannels(state, driver) {
4492
6411
  let idlePolls = 0;
6412
+ let idleMs = 0;
6413
+ let consecutiveDrainFailures = 0;
6414
+ let unreachableMs = 0;
4493
6415
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
6416
+ let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
4494
6417
  while (state.running) {
6418
+ const cycleStartedAtMs = performance.now();
6419
+ let idleThisCycle = false;
6420
+ let unreachableThisCycle = false;
4495
6421
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
4496
6422
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
4497
6423
  if (state.interactive) displayStatus(state);
4498
6424
  await state.connection.reconnectPromise;
4499
6425
  }
6426
+ const carriedOverFileSync = driver.fileSyncActivity().inFlight;
6427
+ void driver.syncPendingFiles().catch(
6428
+ (error2) => logActivity(state, {
6429
+ type: "error",
6430
+ error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
6431
+ })
6432
+ );
4500
6433
  try {
4501
6434
  const processed = await driver.drainPending();
6435
+ consecutiveDrainFailures = 0;
6436
+ unreachableMs = 0;
4502
6437
  state.messageCount += processed;
4503
6438
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4504
6439
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4505
- if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
6440
+ const appliedFiles = driver.fileSyncActivity().appliedFiles;
6441
+ const filesApplied = appliedFiles !== lastSeenAppliedFiles;
6442
+ const fileActivity = carriedOverFileSync || filesApplied;
6443
+ lastSeenAppliedFiles = appliedFiles;
6444
+ if (filesApplied) state.claudeUsageRearm?.();
6445
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
4506
6446
  idlePolls = 0;
6447
+ idleMs = 0;
4507
6448
  if (processed > 0 && state.interactive) displayStatus(state);
4508
6449
  } else if (state.idleTimeout !== null) {
4509
6450
  idlePolls++;
6451
+ idleThisCycle = true;
4510
6452
  if (idlePolls === 1) {
4511
6453
  logActivity(state, {
4512
6454
  type: "info",
@@ -4530,21 +6472,44 @@ async function driveChannels(state, driver) {
4530
6472
  const errorMessage = error2 instanceof Error ? error2.message : String(error2);
4531
6473
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
4532
6474
  if (state.interactive) displayStatus(state);
4533
- }
4534
- await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
4535
- if (state.idleTimeout !== null && idlePolls >= 2) {
4536
- const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
4537
- if (idleMs > state.idleTimeout * 1e3) {
4538
- logActivity(state, { type: "info", message: "Idle timeout reached" });
4539
- if (state.interactive) displayStatus(state);
4540
- break;
6475
+ if (driver.hasInFlightWatchers()) {
6476
+ consecutiveDrainFailures = 0;
6477
+ unreachableMs = 0;
6478
+ } else if (state.idleTimeout !== null) {
6479
+ consecutiveDrainFailures++;
6480
+ unreachableThisCycle = true;
6481
+ if (consecutiveDrainFailures === 1) {
6482
+ logActivity(state, {
6483
+ type: "info",
6484
+ message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
6485
+ });
6486
+ if (state.interactive) displayStatus(state);
6487
+ }
4541
6488
  }
4542
6489
  }
6490
+ await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
6491
+ const cycleMs = performance.now() - cycleStartedAtMs;
6492
+ if (idleThisCycle) idleMs += cycleMs;
6493
+ if (unreachableThisCycle) unreachableMs += cycleMs;
6494
+ if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
6495
+ logActivity(state, {
6496
+ type: "info",
6497
+ level: "warn",
6498
+ message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
6499
+ });
6500
+ if (state.interactive) displayStatus(state);
6501
+ break;
6502
+ }
6503
+ if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
6504
+ logActivity(state, { type: "info", message: "Idle timeout reached" });
6505
+ if (state.interactive) displayStatus(state);
6506
+ break;
6507
+ }
4543
6508
  }
4544
6509
  }
4545
6510
  var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
4546
- async function runSweep(state, driver, config2) {
4547
- const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
6511
+ async function runSweep(state, driver, config) {
6512
+ const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
4548
6513
  try {
4549
6514
  const sessions = await listSessions(state.port);
4550
6515
  if (sessions === null) {
@@ -4557,8 +6522,8 @@ async function runSweep(state, driver, config2) {
4557
6522
  const toDelete = selectSessionsToDelete(
4558
6523
  sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
4559
6524
  {
4560
- maxAgeMs: config2.maxAgeMs,
4561
- maxCount: config2.maxCount,
6525
+ maxAgeMs: config.maxAgeMs,
6526
+ maxCount: config.maxCount,
4562
6527
  nowMs: Date.now(),
4563
6528
  protectedIds: driver.protectedSessionIds()
4564
6529
  }
@@ -4594,7 +6559,7 @@ async function runSweep(state, driver, config2) {
4594
6559
  }
4595
6560
  }
4596
6561
  function scheduleSessionCleanup(state, driver, options) {
4597
- const config2 = resolveSessionCleanupConfig(
6562
+ const config = resolveSessionCleanupConfig(
4598
6563
  {
4599
6564
  maxAge: options.sessionCleanupMaxAge,
4600
6565
  maxCount: options.sessionCleanupMaxCount,
@@ -4602,21 +6567,129 @@ function scheduleSessionCleanup(state, driver, options) {
4602
6567
  },
4603
6568
  process.env
4604
6569
  );
4605
- for (const warning2 of config2.warnings) {
6570
+ for (const warning2 of config.warnings) {
4606
6571
  logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
4607
6572
  }
4608
- if (!config2.enabled) return;
6573
+ if (!config.enabled) return;
4609
6574
  logActivity(state, {
4610
6575
  type: "info",
4611
- message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
6576
+ message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
4612
6577
  });
4613
- const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
6578
+ const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
4614
6579
  const firstSweep = setTimeout(
4615
- () => void runSweep(state, driver, config2),
6580
+ () => void runSweep(state, driver, config),
4616
6581
  SESSION_CLEANUP_FIRST_SWEEP_MS
4617
6582
  );
4618
6583
  state.sessionCleanupTimers.push(interval, firstSweep);
4619
6584
  }
6585
+ function claudeUsageFailureStreakSuffix(consecutiveFailures) {
6586
+ return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
6587
+ }
6588
+ function scheduleClaudeUsageReporting(state, options) {
6589
+ const { mode, warnings } = resolveClaudeUsageReportingMode(
6590
+ options.claudeUsageReporting,
6591
+ process.env
6592
+ );
6593
+ for (const warning2 of warnings) {
6594
+ logActivity(state, {
6595
+ type: "info",
6596
+ level: "warn",
6597
+ message: `Claude usage reporting: ${warning2}`
6598
+ });
6599
+ }
6600
+ if (mode === "off") {
6601
+ logActivity(state, {
6602
+ type: "info",
6603
+ level: "debug",
6604
+ message: "Claude usage reporting is off (--claude-usage-reporting off)"
6605
+ });
6606
+ return null;
6607
+ }
6608
+ let consecutiveFailures = 0;
6609
+ let armed = false;
6610
+ let rearmRequested = false;
6611
+ const scheduleNextTick = () => {
6612
+ armed = true;
6613
+ rearmRequested = false;
6614
+ state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
6615
+ };
6616
+ const rearm = () => {
6617
+ if (armed) {
6618
+ rearmRequested = true;
6619
+ return;
6620
+ }
6621
+ rearmRequested = false;
6622
+ armed = true;
6623
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6624
+ };
6625
+ const tick = async (isProbe) => {
6626
+ try {
6627
+ const usage = await getClaudeUsage();
6628
+ const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
6629
+ if (result.ok) {
6630
+ if (consecutiveFailures > 0) {
6631
+ logActivity(state, {
6632
+ type: "info",
6633
+ level: "info",
6634
+ message: "Claude usage reporting recovered"
6635
+ });
6636
+ }
6637
+ consecutiveFailures = 0;
6638
+ logActivity(state, {
6639
+ type: "info",
6640
+ level: "debug",
6641
+ message: "Reported Claude usage to Evident"
6642
+ });
6643
+ } else {
6644
+ consecutiveFailures++;
6645
+ logActivity(state, {
6646
+ type: "info",
6647
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
6648
+ message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
6649
+ });
6650
+ }
6651
+ scheduleNextTick();
6652
+ } catch (error2) {
6653
+ if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
6654
+ if (mode === "on") {
6655
+ logActivity(state, {
6656
+ type: "info",
6657
+ level: "warn",
6658
+ message: "Claude usage reporting is forced on but no usable Claude Code login was found \u2014 run `claude` to sign in; reporting will keep retrying"
6659
+ });
6660
+ scheduleNextTick();
6661
+ } else if (isProbe) {
6662
+ logActivity(state, {
6663
+ type: "info",
6664
+ level: "debug",
6665
+ message: `Claude usage reporting: ${error2.message}`
6666
+ });
6667
+ armed = false;
6668
+ if (rearmRequested) rearm();
6669
+ } else {
6670
+ logActivity(state, {
6671
+ type: "info",
6672
+ level: "debug",
6673
+ message: `Claude usage reporting: ${error2.message}`
6674
+ });
6675
+ scheduleNextTick();
6676
+ }
6677
+ } else {
6678
+ consecutiveFailures++;
6679
+ const message = error2 instanceof Error ? error2.message : String(error2);
6680
+ logActivity(state, {
6681
+ type: "info",
6682
+ level: claudeUsageFailureLogLevel(consecutiveFailures),
6683
+ message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
6684
+ });
6685
+ scheduleNextTick();
6686
+ }
6687
+ }
6688
+ };
6689
+ armed = true;
6690
+ state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
6691
+ return rearm;
6692
+ }
4620
6693
  async function notifyOffline(state) {
4621
6694
  if (!state.agentId || !state.authHeader) return;
4622
6695
  if (!state.connected) {
@@ -4634,13 +6707,29 @@ async function notifyOffline(state) {
4634
6707
  if (state.interactive) displayStatus(state);
4635
6708
  }
4636
6709
  }
6710
+ async function timeShutdownPhase(state, durations, name, run2) {
6711
+ const startedAt = Date.now();
6712
+ try {
6713
+ return await run2();
6714
+ } finally {
6715
+ const elapsedMs = Date.now() - startedAt;
6716
+ durations[name] = elapsedMs;
6717
+ log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
6718
+ }
6719
+ }
4637
6720
  async function cleanup(state, opts = {}) {
6721
+ const durations = {};
4638
6722
  state.running = false;
4639
6723
  for (const timer of state.sessionCleanupTimers) {
4640
6724
  clearInterval(timer);
4641
6725
  clearTimeout(timer);
4642
6726
  }
4643
6727
  state.sessionCleanupTimers = [];
6728
+ if (state.claudeUsageTimer) {
6729
+ clearTimeout(state.claudeUsageTimer);
6730
+ state.claudeUsageTimer = null;
6731
+ }
6732
+ state.claudeUsageRearm = null;
4644
6733
  if (opts.graceful && state.channelDriver) {
4645
6734
  state.channelDriver.stop();
4646
6735
  log2(state, "Draining in-flight channel work before shutdown...");
@@ -4648,7 +6737,13 @@ async function cleanup(state, opts = {}) {
4648
6737
  logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4649
6738
  displayStatus(state);
4650
6739
  }
4651
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
6740
+ const driver = state.channelDriver;
6741
+ const settled = await timeShutdownPhase(
6742
+ state,
6743
+ durations,
6744
+ "drain",
6745
+ () => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
6746
+ );
4652
6747
  if (!settled) {
4653
6748
  logActivity(state, {
4654
6749
  type: "info",
@@ -4657,13 +6752,15 @@ async function cleanup(state, opts = {}) {
4657
6752
  if (state.interactive) displayStatus(state);
4658
6753
  }
4659
6754
  }
4660
- await notifyOffline(state);
6755
+ await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
4661
6756
  if (state.connection) {
4662
- state.connection.close();
6757
+ const connection = state.connection;
6758
+ await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
4663
6759
  state.connection = null;
4664
6760
  }
4665
6761
  if (state.opencodeProcess) {
4666
- stopOpenCode(state.opencodeProcess);
6762
+ const opencodeProcess = state.opencodeProcess;
6763
+ await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
4667
6764
  if (state.interactive) {
4668
6765
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
4669
6766
  displayStatus(state);
@@ -4672,12 +6769,15 @@ async function cleanup(state, opts = {}) {
4672
6769
  }
4673
6770
  state.opencodeProcess = null;
4674
6771
  }
6772
+ return durations;
4675
6773
  }
4676
6774
  async function run(options) {
4677
6775
  const interactive = isInteractive(options.json);
4678
6776
  let logLevel;
6777
+ let fileSyncDirectories;
4679
6778
  try {
4680
6779
  logLevel = resolveLogLevel(options);
6780
+ fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
4681
6781
  } catch (error2) {
4682
6782
  const message = error2 instanceof Error ? error2.message : String(error2);
4683
6783
  if (options.json) {
@@ -4710,8 +6810,16 @@ async function run(options) {
4710
6810
  messageCount: 0,
4711
6811
  lastProxiedActivityAt: null,
4712
6812
  sessionCleanupTimers: [],
6813
+ claudeUsageTimer: null,
6814
+ claudeUsageRearm: null,
4713
6815
  authHeader: ""
4714
6816
  };
6817
+ setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
6818
+ if (fileSyncDirectories.length > 0) {
6819
+ log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
6820
+ } else {
6821
+ log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
6822
+ }
4715
6823
  if (!options.runner && options.agent) {
4716
6824
  telemetry.info(
4717
6825
  EventTypes.DEPRECATED_AGENT_FLAG_USED,
@@ -4735,14 +6843,38 @@ async function run(options) {
4735
6843
  const handleSignal = async () => {
4736
6844
  if (state.shuttingDown) return;
4737
6845
  state.shuttingDown = true;
6846
+ const shutdownStartedAt = Date.now();
4738
6847
  if (state.interactive) {
4739
6848
  logActivity(state, { type: "info", message: "Shutting down..." });
4740
6849
  displayStatus(state);
4741
6850
  } else {
4742
6851
  log2(state, "Shutting down...");
4743
6852
  }
4744
- await cleanup(state, { graceful: true });
4745
- await shutdownTelemetry();
6853
+ const durations = await cleanup(state, { graceful: true });
6854
+ const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
6855
+ await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
6856
+ let timer;
6857
+ const flushed = shutdownTelemetry().then(
6858
+ () => true,
6859
+ (error2) => {
6860
+ log2(
6861
+ state,
6862
+ `Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
6863
+ "warn"
6864
+ );
6865
+ return true;
6866
+ }
6867
+ );
6868
+ const timedOut = new Promise((resolve3) => {
6869
+ timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
6870
+ });
6871
+ if (!await Promise.race([flushed, timedOut])) {
6872
+ log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
6873
+ }
6874
+ clearTimeout(timer);
6875
+ });
6876
+ const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
6877
+ log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
4746
6878
  process.exit(0);
4747
6879
  };
4748
6880
  process.on("SIGINT", handleSignal);
@@ -4759,6 +6891,7 @@ async function run(options) {
4759
6891
  console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
4760
6892
  blank();
4761
6893
  process.exit(1);
6894
+ return;
4762
6895
  }
4763
6896
  blank();
4764
6897
  console.log(chalk6.yellow("You are not logged in to Evident."));
@@ -4803,6 +6936,7 @@ async function run(options) {
4803
6936
  } else {
4804
6937
  printError(resolved.error || "Failed to resolve runner ID from key");
4805
6938
  process.exit(1);
6939
+ return;
4806
6940
  }
4807
6941
  } else {
4808
6942
  printError(
@@ -4816,6 +6950,7 @@ async function run(options) {
4816
6950
  );
4817
6951
  blank();
4818
6952
  process.exit(1);
6953
+ return;
4819
6954
  }
4820
6955
  }
4821
6956
  telemetry.info(
@@ -4856,25 +6991,67 @@ async function run(options) {
4856
6991
  }
4857
6992
  spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
4858
6993
  state.agentName = validation.agent.name;
6994
+ const microvmId = process.env.MICROVM_ID?.trim();
6995
+ if (microvmId) {
6996
+ const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
6997
+ if (reported.ok) {
6998
+ log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
6999
+ } else {
7000
+ const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
7001
+ log2(state, message, "warn");
7002
+ if (state.interactive && !state.json) {
7003
+ logActivity(state, { type: "info", level: "warn", message });
7004
+ }
7005
+ }
7006
+ } else {
7007
+ log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
7008
+ }
7009
+ const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
7010
+ for (const warning2 of opencodeStartTimeoutWarnings) {
7011
+ logActivity(state, { type: "info", level: "warn", message: warning2 });
7012
+ }
4859
7013
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
4860
7014
  try {
4861
7015
  const oc = await ensureOpenCodeRunning({
4862
7016
  port: state.port,
4863
7017
  interactive: state.interactive,
4864
7018
  agentId: state.agentId,
4865
- log: (message) => log2(state, message)
7019
+ log: (message) => log2(state, message),
7020
+ startTimeoutMs: opencodeStartTimeoutMs
4866
7021
  });
4867
7022
  state.port = oc.port;
4868
7023
  state.opencodeProcess = oc.process;
4869
7024
  state.opencodeVersion = oc.version;
4870
- state.opencodeConnected = oc.process !== null || oc.version !== null;
7025
+ state.opencodeConnected = oc.notReadyReason === null;
4871
7026
  const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
4872
7027
  ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
4873
- const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
4874
- if (versionWarning) {
4875
- log2(state, versionWarning, "warn");
4876
- if (state.interactive && !state.json) {
4877
- logActivity(state, { type: "info", level: "warn", message: versionWarning });
7028
+ if (!state.interactive && oc.notReadyReason !== null) {
7029
+ const message = `OpenCode is not ready on port ${state.port}: ${oc.notReadyReason}. The runner will still come online, but messages will fail until opencode answers \u2014 raise the wait with --opencode-start-timeout <seconds> (env ${OPENCODE_START_TIMEOUT_ENV}).`;
7030
+ logActivity(state, { type: "info", level: "warn", message });
7031
+ } else {
7032
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
7033
+ if (versionWarning) {
7034
+ log2(state, versionWarning, "warn");
7035
+ if (state.interactive && !state.json) {
7036
+ logActivity(state, { type: "info", level: "warn", message: versionWarning });
7037
+ }
7038
+ }
7039
+ const noProviderWarning = buildNoProviderWarning(
7040
+ await hasAnyConfiguredProvider(state.port)
7041
+ );
7042
+ if (noProviderWarning) {
7043
+ log2(state, noProviderWarning, "warn");
7044
+ if (state.interactive && !state.json) {
7045
+ logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
7046
+ blank();
7047
+ console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
7048
+ console.log(
7049
+ chalk6.dim(
7050
+ `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
7051
+ )
7052
+ );
7053
+ blank();
7054
+ }
4878
7055
  }
4879
7056
  }
4880
7057
  } catch (error2) {
@@ -4889,6 +7066,10 @@ async function run(options) {
4889
7066
  getAuthHeader: () => state.authHeader,
4890
7067
  conversationFilter: state.conversationFilter,
4891
7068
  stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
7069
+ // #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
7070
+ // REJECTED with `file_sync_disabled` on the ack, not silently ignored.
7071
+ fileSyncDirectories,
7072
+ homeDir: homedir3(),
4892
7073
  log: (entry) => (
4893
7074
  // Thread the driver's real level straight through so `debug`/`warn`
4894
7075
  // survive the sink filter (they no longer collapse to info). `type`
@@ -4915,6 +7096,18 @@ async function run(options) {
4915
7096
  type: "info",
4916
7097
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
4917
7098
  });
7099
+ if (options.tunnelReadyFile) {
7100
+ const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
7101
+ if (marker.ok) {
7102
+ log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
7103
+ } else {
7104
+ log2(
7105
+ state,
7106
+ `Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
7107
+ "error"
7108
+ );
7109
+ }
7110
+ }
4918
7111
  emitAgentConnected(state.agentId, {
4919
7112
  port: state.port,
4920
7113
  cli_version: getCliVersion(),
@@ -4970,6 +7163,12 @@ async function run(options) {
4970
7163
  onDrainPing: () => {
4971
7164
  if (!state.running) return;
4972
7165
  logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
7166
+ void channelDriver.syncPendingFiles().catch(
7167
+ (error2) => logActivity(state, {
7168
+ type: "error",
7169
+ error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
7170
+ })
7171
+ );
4973
7172
  channelDriver.drainPending().then((processed) => {
4974
7173
  if (processed > 0) {
4975
7174
  state.messageCount += processed;
@@ -4999,6 +7198,7 @@ async function run(options) {
4999
7198
  throw error2;
5000
7199
  }
5001
7200
  scheduleSessionCleanup(state, channelDriver, options);
7201
+ state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
5002
7202
  if (!interactive || state.json) {
5003
7203
  log2(state, "Driving channel messages...");
5004
7204
  }
@@ -5053,10 +7253,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
5053
7253
  program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
5054
7254
  program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
5055
7255
  program.command("whoami").description("Show the currently logged in user").action(whoami);
5056
- 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(
7256
+ program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
7257
+ program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
7258
+ program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
7259
+ "-a, --agent [id]",
7260
+ "Deprecated alias for --runner (still supported; --runner wins if both are given)"
7261
+ ).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
5057
7262
  "--log-level <level>",
5058
7263
  "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
5059
- ).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").option(
7264
+ ).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
7265
+ "--opencode-start-timeout <seconds>",
7266
+ "Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
7267
+ ).option("--json", "Output in JSON format").option(
5060
7268
  "--session-cleanup-max-age <duration>",
5061
7269
  "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
5062
7270
  ).option(
@@ -5065,6 +7273,17 @@ program.command("run").description("Connect to Evident and process messages").op
5065
7273
  ).option(
5066
7274
  "--session-cleanup-interval <duration>",
5067
7275
  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
7276
+ ).option(
7277
+ "--claude-usage-reporting <mode>",
7278
+ "Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
7279
+ ).option(
7280
+ "--enable-file-sync-to <dir>",
7281
+ "Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
7282
+ (value, previous) => previous.concat([value]),
7283
+ []
7284
+ ).option(
7285
+ "--tunnel-ready-file <path>",
7286
+ "Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
5068
7287
  ).action(
5069
7288
  (options) => {
5070
7289
  run({
@@ -5077,11 +7296,21 @@ program.command("run").description("Connect to Evident and process messages").op
5077
7296
  verbose: options.verbose,
5078
7297
  conversation: options.conversation,
5079
7298
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
7299
+ // Raw string — validation/precedence is single-sourced in run.ts's
7300
+ // resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
7301
+ opencodeStartTimeout: options.opencodeStartTimeout,
5080
7302
  json: options.json,
5081
7303
  // Raw strings — the resolver in run.ts single-sources parsing (M1).
5082
7304
  sessionCleanupMaxAge: options.sessionCleanupMaxAge,
5083
7305
  sessionCleanupMaxCount: options.sessionCleanupMaxCount,
5084
- sessionCleanupInterval: options.sessionCleanupInterval
7306
+ sessionCleanupInterval: options.sessionCleanupInterval,
7307
+ // Raw string — the resolver in run.ts single-sources parsing
7308
+ // (resolveClaudeUsageReportingMode).
7309
+ claudeUsageReporting: options.claudeUsageReporting,
7310
+ // Raw values — expansion/validation is single-sourced in run.ts's
7311
+ // resolveFileSyncDirectories.
7312
+ enableFileSyncTo: options.enableFileSyncTo,
7313
+ tunnelReadyFile: options.tunnelReadyFile
5085
7314
  });
5086
7315
  }
5087
7316
  );