@evident-ai/cli 3.1.1-dev.2f78b33 → 3.1.1-dev.2f79773

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 = {
@@ -47,8 +47,35 @@ var credentials = new Conf({
47
47
  projectName: "evident",
48
48
  projectSuffix: "",
49
49
  configName: "credentials",
50
- defaults: {}
50
+ defaults: {},
51
+ configFileMode: 384
51
52
  });
53
+ var CREDENTIALS_FILE_MODE = 384;
54
+ var CREDENTIALS_DIR_MODE = 448;
55
+ var permissionWarningEmitted = false;
56
+ function hardenCredentialsPermissions() {
57
+ if (process.platform === "win32") {
58
+ return;
59
+ }
60
+ const file = credentials.path;
61
+ for (const [path, mode] of [
62
+ [file, CREDENTIALS_FILE_MODE],
63
+ [dirname(file), CREDENTIALS_DIR_MODE]
64
+ ]) {
65
+ try {
66
+ if (existsSync(path) && (statSync(path).mode & 511) !== mode) {
67
+ chmodSync(path, mode);
68
+ }
69
+ } catch (err) {
70
+ if (!permissionWarningEmitted) {
71
+ permissionWarningEmitted = true;
72
+ console.error(
73
+ `[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)}`
74
+ );
75
+ }
76
+ }
77
+ }
78
+ }
52
79
  function getApiUrlConfig() {
53
80
  return getApiUrl();
54
81
  }
@@ -59,6 +86,7 @@ function credentialsKey() {
59
86
  return getApiUrl();
60
87
  }
61
88
  function getCredentials() {
89
+ hardenCredentialsPermissions();
62
90
  const byEndpoint = credentials.get("byEndpoint") ?? {};
63
91
  return byEndpoint[credentialsKey()] ?? {};
64
92
  }
@@ -70,14 +98,17 @@ function setCredentials(creds) {
70
98
  expiresAt: creds.expiresAt
71
99
  };
72
100
  credentials.set("byEndpoint", byEndpoint);
101
+ hardenCredentialsPermissions();
73
102
  }
74
103
  function clearCredentials() {
75
104
  const byEndpoint = credentials.get("byEndpoint") ?? {};
76
105
  delete byEndpoint[credentialsKey()];
77
106
  credentials.set("byEndpoint", byEndpoint);
107
+ hardenCredentialsPermissions();
78
108
  }
79
109
  function clearAllCredentials() {
80
110
  credentials.clear();
111
+ hardenCredentialsPermissions();
81
112
  }
82
113
  function getCliName() {
83
114
  const argv1 = process.argv[1] || "";
@@ -236,16 +267,28 @@ async function getToken() {
236
267
  }
237
268
  return null;
238
269
  }
270
+ function toError(err) {
271
+ return err instanceof Error ? err : new Error(String(err));
272
+ }
239
273
  async function deleteToken(options = {}) {
240
274
  const keytar = await getKeytar();
275
+ const failures = [];
241
276
  if (keytar) {
242
277
  if (options.all) {
243
- const all = await keytar.findCredentials(SERVICE_NAME).catch(() => []);
278
+ let accounts = [];
279
+ try {
280
+ accounts = await keytar.findCredentials(SERVICE_NAME);
281
+ } catch (err) {
282
+ failures.push({ type: "enumerate", error: toError(err) });
283
+ }
244
284
  await Promise.all(
245
- all.map(
246
- (entry) => keytar.deletePassword(SERVICE_NAME, entry.account).catch(() => {
247
- })
248
- )
285
+ accounts.map(async (entry) => {
286
+ try {
287
+ await keytar.deletePassword(SERVICE_NAME, entry.account);
288
+ } catch (err) {
289
+ failures.push({ type: "delete", account: entry.account, error: toError(err) });
290
+ }
291
+ })
249
292
  );
250
293
  } else {
251
294
  await keytar.deletePassword(SERVICE_NAME, keychainAccount());
@@ -256,6 +299,7 @@ async function deleteToken(options = {}) {
256
299
  } else {
257
300
  clearCredentials();
258
301
  }
302
+ return { failures };
259
303
  }
260
304
 
261
305
  // src/utils/ui.ts
@@ -285,14 +329,14 @@ function blank() {
285
329
  console.log();
286
330
  }
287
331
  function waitForEnter(prompt = "Press Enter to continue...") {
288
- return new Promise((resolve2) => {
332
+ return new Promise((resolve3) => {
289
333
  process.stdout.write(chalk.dim(prompt));
290
334
  const handler = () => {
291
335
  process.stdin.removeListener("data", handler);
292
336
  process.stdin.setRawMode?.(false);
293
337
  process.stdin.pause();
294
338
  console.log();
295
- resolve2();
339
+ resolve3();
296
340
  };
297
341
  if (process.stdin.isTTY) {
298
342
  process.stdin.setRawMode?.(true);
@@ -302,7 +346,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
302
346
  });
303
347
  }
304
348
  function sleep(ms) {
305
- return new Promise((resolve2) => setTimeout(resolve2, ms));
349
+ return new Promise((resolve3) => setTimeout(resolve3, ms));
306
350
  }
307
351
 
308
352
  // src/commands/login.ts
@@ -373,22 +417,25 @@ async function deviceFlowLogin(options) {
373
417
  }
374
418
  async function tokenLogin() {
375
419
  console.log("Token login mode.");
376
- console.log("Visit your Evident dashboard to generate a CLI token.");
420
+ console.log("Create a token under Settings \u2192 CLI tokens in the dashboard, then paste it below.");
421
+ console.log(
422
+ "(Alternatively, run `evident login` on a machine with a browser, or set EVIDENT_TOKEN for CI.)"
423
+ );
377
424
  blank();
378
425
  process.stdout.write("Paste token: ");
379
- const token = await new Promise((resolve2) => {
426
+ const token = await new Promise((resolve3) => {
380
427
  let data = "";
381
428
  process.stdin.setEncoding("utf8");
382
429
  process.stdin.on("data", (chunk) => {
383
430
  data += chunk;
384
431
  });
385
432
  process.stdin.on("end", () => {
386
- resolve2(data.trim());
433
+ resolve3(data.trim());
387
434
  });
388
435
  if (process.stdin.isTTY) {
389
436
  process.stdin.once("data", (chunk) => {
390
437
  process.stdin.pause();
391
- resolve2(chunk.toString().trim());
438
+ resolve3(chunk.toString().trim());
392
439
  });
393
440
  process.stdin.resume();
394
441
  }
@@ -397,13 +444,22 @@ async function tokenLogin() {
397
444
  printError("No token provided.");
398
445
  process.exit(1);
399
446
  }
447
+ await validateAndStoreToken(token);
448
+ }
449
+ async function validateAndStoreToken(token) {
400
450
  const spinner = ora("Validating token...").start();
401
451
  try {
402
- const result = await api.post("/auth/token/validate", { token });
452
+ const result = await api.get("/me", {
453
+ headers: { Authorization: `Bearer ${token}` }
454
+ });
455
+ if (!result.user) {
456
+ throw new Error(
457
+ "This token is not a user login (e.g. a runner key). Paste a CLI token instead."
458
+ );
459
+ }
403
460
  await storeToken({
404
461
  token,
405
- user: result.user,
406
- expiresAt: result.expires_at
462
+ user: { email: result.user.email }
407
463
  });
408
464
  spinner.stop();
409
465
  printSuccess(`Logged in as ${chalk2.bold(result.user.email)}`);
@@ -423,9 +479,22 @@ async function login(options) {
423
479
  }
424
480
 
425
481
  // src/commands/logout.ts
482
+ function describeFailure(failure) {
483
+ if (failure.type === "enumerate") {
484
+ return `could not list stored keychain entries (${failure.error.message})`;
485
+ }
486
+ return `${failure.account} (${failure.error.message})`;
487
+ }
426
488
  async function logout(options = {}) {
427
489
  if (options.all) {
428
- await deleteToken({ all: true });
490
+ const result = await deleteToken({ all: true });
491
+ if (result.failures.length > 0) {
492
+ printError(
493
+ `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.`
494
+ );
495
+ process.exitCode = 1;
496
+ return;
497
+ }
429
498
  printSuccess("Logged out of all endpoints.");
430
499
  return;
431
500
  }
@@ -450,7 +519,9 @@ async function whoami() {
450
519
  blank();
451
520
  console.log(keyValue("Endpoint", apiUrl));
452
521
  console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
453
- console.log(keyValue("User ID", credentials2.user.id));
522
+ if (credentials2.user.id) {
523
+ console.log(keyValue("User ID", credentials2.user.id));
524
+ }
454
525
  if (credentials2.expiresAt) {
455
526
  const expiresAt = new Date(credentials2.expiresAt);
456
527
  const now = /* @__PURE__ */ new Date();
@@ -467,9 +538,9 @@ async function whoami() {
467
538
  }
468
539
 
469
540
  // src/commands/run.ts
541
+ import { homedir as homedir2 } from "os";
542
+ import { isAbsolute as isAbsolute2, join as join2, parse, resolve as resolvePath } from "path";
470
543
  import chalk6 from "chalk";
471
- import ora3 from "ora";
472
- import { select as select3 } from "@inquirer/prompts";
473
544
 
474
545
  // ../../packages/types/src/telemetry/index.ts
475
546
  var TelemetryEventTypes = {
@@ -478,13 +549,20 @@ var TelemetryEventTypes = {
478
549
  AGENT_DISCONNECTED: "agent.disconnected",
479
550
  AGENT_MESSAGE_PROCESSING: "agent.message_processing",
480
551
  AGENT_MESSAGE_DONE: "agent.message_done",
481
- AGENT_MESSAGE_FAILED: "agent.message_failed"
552
+ AGENT_MESSAGE_FAILED: "agent.message_failed",
553
+ // A `warn`/`error` runner-side log line forwarded server-side for
554
+ // observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
555
+ RUNNER_ACTIVITY: "runner.activity"
482
556
  };
483
557
 
484
558
  // ../../packages/types/src/tunnel/index.ts
485
559
  var MAX_FRAME_BYTES = 256 * 1024;
486
560
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
487
561
 
562
+ // ../../packages/types/src/runner-files.ts
563
+ var MAX_FILE_PUSH_BYTES = 64 * 1024;
564
+ var MAX_FILE_SYNC_DIRECTORIES = 16;
565
+
488
566
  // ../../packages/types/src/logging/index.ts
489
567
  var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
568
  function log(level, event, fields) {
@@ -514,6 +592,10 @@ function stripQuery(url) {
514
592
  }
515
593
  }
516
594
 
595
+ // src/commands/run.ts
596
+ import ora3 from "ora";
597
+ import { select as select3 } from "@inquirer/prompts";
598
+
517
599
  // src/lib/telemetry.ts
518
600
  var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
519
601
  function getCliVersion() {
@@ -525,6 +607,13 @@ var isShuttingDown = false;
525
607
  var FLUSH_INTERVAL_MS = 5e3;
526
608
  var MAX_BUFFER_SIZE = 50;
527
609
  var FLUSH_TIMEOUT_MS = 3e3;
610
+ var authProvider = null;
611
+ function setTelemetryAuthProvider(provider) {
612
+ authProvider = provider;
613
+ }
614
+ var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
615
+ var lastFlushFailureLoggedAt = 0;
616
+ var suppressedFlushFailureCount = 0;
528
617
  function logEvent(eventType, options = {}) {
529
618
  const event = {
530
619
  event_type: eventType,
@@ -559,9 +648,16 @@ async function flushEvents() {
559
648
  flushTimeout = null;
560
649
  }
561
650
  try {
562
- const credentials2 = await getToken();
563
- if (!credentials2) {
564
- return;
651
+ const providerContext = authProvider?.();
652
+ let authHeader;
653
+ if (providerContext?.authHeader) {
654
+ authHeader = providerContext.authHeader;
655
+ } else {
656
+ const credentials2 = await getToken();
657
+ if (!credentials2) {
658
+ return;
659
+ }
660
+ authHeader = `Bearer ${credentials2.token}`;
565
661
  }
566
662
  const apiUrl = getApiUrlConfig();
567
663
  const controller = new AbortController();
@@ -576,7 +672,7 @@ async function flushEvents() {
576
672
  method: "POST",
577
673
  headers: {
578
674
  "Content-Type": "application/json",
579
- Authorization: `Bearer ${credentials2.token}`
675
+ Authorization: authHeader
580
676
  },
581
677
  body: JSON.stringify(request),
582
678
  signal: controller.signal
@@ -588,8 +684,15 @@ async function flushEvents() {
588
684
  clearTimeout(timeout);
589
685
  }
590
686
  } catch (error2) {
591
- if (process.env.DEBUG) {
592
- console.error("Telemetry error:", error2);
687
+ const now = Date.now();
688
+ if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
689
+ const message = error2 instanceof Error ? error2.message : String(error2);
690
+ const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
691
+ console.error(`Telemetry flush error: ${message}${suffix}`);
692
+ lastFlushFailureLoggedAt = now;
693
+ suppressedFlushFailureCount = 0;
694
+ } else {
695
+ suppressedFlushFailureCount++;
593
696
  }
594
697
  }
595
698
  }
@@ -658,6 +761,69 @@ var EventTypes = {
658
761
  DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
659
762
  };
660
763
 
764
+ // src/lib/runner-activity-telemetry.ts
765
+ var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
766
+ var SEVERITY_BY_LEVEL = {
767
+ warn: "warning",
768
+ error: "error"
769
+ };
770
+ var MAX_MESSAGE_LENGTH = 500;
771
+ var TRUNCATION_MARKER = "\u2026";
772
+ function redact(message) {
773
+ return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
774
+ }
775
+ function truncate(message) {
776
+ if (message.length <= MAX_MESSAGE_LENGTH) return message;
777
+ return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
778
+ }
779
+ var RATE_LIMIT_WINDOW_MS = 6e4;
780
+ var RATE_LIMIT_MAX_EVENTS = 30;
781
+ var windowStartedAt = 0;
782
+ var windowCount = 0;
783
+ var windowDroppedCount = 0;
784
+ function admitUnderRateLimit(now) {
785
+ if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
786
+ if (windowDroppedCount > 0) {
787
+ console.error(
788
+ `[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
789
+ );
790
+ }
791
+ windowStartedAt = now;
792
+ windowCount = 0;
793
+ windowDroppedCount = 0;
794
+ }
795
+ if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
796
+ windowDroppedCount++;
797
+ if (windowDroppedCount === 1) {
798
+ console.error(
799
+ `[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
800
+ );
801
+ }
802
+ return false;
803
+ }
804
+ windowCount++;
805
+ return true;
806
+ }
807
+ function forwardRunnerActivity(entry, context) {
808
+ try {
809
+ if (!FORWARDED_LEVELS.has(entry.level)) return;
810
+ if (!context.agentId || !context.authHeader) return;
811
+ if (!admitUnderRateLimit(Date.now())) return;
812
+ const rawMessage = entry.error ?? entry.message ?? "";
813
+ const message = truncate(redact(rawMessage));
814
+ logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
815
+ severity: SEVERITY_BY_LEVEL[entry.level],
816
+ message,
817
+ metadata: { source: "cli.run" },
818
+ agentId: context.agentId
819
+ });
820
+ } catch (err) {
821
+ console.error(
822
+ `[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
823
+ );
824
+ }
825
+ }
826
+
661
827
  // src/lib/auth.ts
662
828
  async function getAuthCredentials() {
663
829
  const runnerKey = process.env.EVIDENT_RUNNER_KEY;
@@ -725,7 +891,7 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
725
891
  if (health.healthy) {
726
892
  return health;
727
893
  }
728
- await new Promise((resolve2) => setTimeout(resolve2, 1e3));
894
+ await new Promise((resolve3) => setTimeout(resolve3, 1e3));
729
895
  }
730
896
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
731
897
  }
@@ -1364,7 +1530,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
1364
1530
  }
1365
1531
  }
1366
1532
  if (attempt < READ_BACK_ATTEMPTS - 1) {
1367
- await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1533
+ await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
1368
1534
  }
1369
1535
  }
1370
1536
  return null;
@@ -1492,6 +1658,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
1492
1658
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1493
1659
  return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1494
1660
  }
1661
+ function isB2AbandonmentConfirmed(params) {
1662
+ return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
1663
+ }
1495
1664
  function messageError(messages, userMessageId) {
1496
1665
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1497
1666
  const error2 = errorOf(reply);
@@ -1505,6 +1674,42 @@ function messageError(messages, userMessageId) {
1505
1674
  }
1506
1675
  return "The agent run failed.";
1507
1676
  }
1677
+ function messageFailure(messages, userMessageId) {
1678
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1679
+ const error2 = errorOf(reply);
1680
+ if (error2 == null || typeof error2 !== "object") return null;
1681
+ const e = error2;
1682
+ const replyProviderId = reply?.info?.providerID ?? null;
1683
+ const replyModelId = reply?.info?.modelID ?? null;
1684
+ if (e.name === "ProviderAuthError") {
1685
+ const data = e.data;
1686
+ const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
1687
+ return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
1688
+ }
1689
+ if (e.name === "APIError") {
1690
+ const data = e.data;
1691
+ const statusCode = data?.statusCode;
1692
+ if (statusCode === 401 || statusCode === 403) {
1693
+ return {
1694
+ kind: "model_auth",
1695
+ providerId: replyProviderId,
1696
+ modelId: replyModelId,
1697
+ reason: "rejected"
1698
+ };
1699
+ }
1700
+ }
1701
+ return null;
1702
+ }
1703
+ function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
1704
+ if (classified != null) return classified;
1705
+ if (hasConfiguredProvider !== false) return null;
1706
+ return {
1707
+ kind: "model_auth",
1708
+ providerId: replyProviderId,
1709
+ modelId: replyModelId,
1710
+ reason: "missing"
1711
+ };
1712
+ }
1508
1713
  function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1509
1714
  if (!messages || messages.length === 0) return false;
1510
1715
  return messages.some(
@@ -1738,12 +1943,12 @@ var StreamForwarder = class {
1738
1943
  let endBody;
1739
1944
  if (has_body) {
1740
1945
  const chunks = [];
1741
- bodyPromise = new Promise((resolve2) => {
1946
+ bodyPromise = new Promise((resolve3) => {
1742
1947
  pushBody = (buf) => {
1743
1948
  chunks.push(buf);
1744
1949
  };
1745
1950
  endBody = () => {
1746
- resolve2(Buffer.concat(chunks));
1951
+ resolve3(Buffer.concat(chunks));
1747
1952
  };
1748
1953
  });
1749
1954
  }
@@ -1860,7 +2065,7 @@ function connectTunnel(options) {
1860
2065
  } = options;
1861
2066
  const tunnelUrl = getTunnelUrlConfig();
1862
2067
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1863
- return new Promise((resolve2, reject) => {
2068
+ return new Promise((resolve3, reject) => {
1864
2069
  const ws = new WebSocket2(url, {
1865
2070
  headers: {
1866
2071
  Authorization: authHeader
@@ -1915,7 +2120,7 @@ function connectTunnel(options) {
1915
2120
  clearTimeout(connectionTimeout);
1916
2121
  const connectedAgentId = message.agent_id ?? agentId;
1917
2122
  onConnected?.(connectedAgentId);
1918
- resolve2({
2123
+ resolve3({
1919
2124
  ws,
1920
2125
  close: () => ws.close(1e3, "CLI shutdown")
1921
2126
  });
@@ -2033,6 +2238,416 @@ var RunnerConnection = class {
2033
2238
  }
2034
2239
  };
2035
2240
 
2241
+ // src/lib/tunnel/ready-marker.ts
2242
+ import { writeFileSync } from "fs";
2243
+ function writeTunnelReadyMarker(path, agentId) {
2244
+ try {
2245
+ writeFileSync(path, `${agentId}
2246
+ `);
2247
+ return { ok: true };
2248
+ } catch (error2) {
2249
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2250
+ }
2251
+ }
2252
+
2253
+ // src/lib/channels/driver.ts
2254
+ import { homedir } from "os";
2255
+
2256
+ // src/lib/file-push.ts
2257
+ import { randomUUID } from "crypto";
2258
+ import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
2259
+ import { basename, dirname as dirname2, isAbsolute, join, relative, resolve as resolve2, sep } from "path";
2260
+ var FILE_MODE = 384;
2261
+ var DIRECTORY_MODE = 448;
2262
+ async function writePushedFile(request) {
2263
+ const { requestedPath, content, allowedDirectories, homeDir } = request;
2264
+ const bytes = content.byteLength;
2265
+ if (allowedDirectories.length === 0) {
2266
+ return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
2267
+ path: requestedPath,
2268
+ bytes
2269
+ });
2270
+ }
2271
+ if (bytes > MAX_FILE_PUSH_BYTES) {
2272
+ return refuse(
2273
+ "file_too_large",
2274
+ `File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
2275
+ {
2276
+ path: requestedPath,
2277
+ bytes
2278
+ }
2279
+ );
2280
+ }
2281
+ const candidate = expandAndValidate(requestedPath, homeDir);
2282
+ if (candidate === null) {
2283
+ return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
2284
+ path: requestedPath,
2285
+ bytes
2286
+ });
2287
+ }
2288
+ try {
2289
+ const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
2290
+ dirname2(candidate)
2291
+ );
2292
+ const realTarget = join(existingAncestor, ...missingSegments, basename(candidate));
2293
+ const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
2294
+ if (allowedDirectory === null) {
2295
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2296
+ path: realTarget,
2297
+ bytes
2298
+ });
2299
+ }
2300
+ if (missingSegments.length > 0) {
2301
+ await createMissingDirectories(existingAncestor, missingSegments);
2302
+ const realParent = await realpath(dirname2(realTarget));
2303
+ if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
2304
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2305
+ path: realTarget,
2306
+ bytes,
2307
+ reason: "parent_changed_after_create"
2308
+ });
2309
+ }
2310
+ }
2311
+ await writeAtomically(realTarget, content);
2312
+ log("info", "file_push_written", { path: realTarget, bytes });
2313
+ return { ok: true, path: realTarget };
2314
+ } catch (err) {
2315
+ const errno = err.code ?? "UNKNOWN";
2316
+ return refuse("write_failed", `The runner could not write the file (${errno}).`, {
2317
+ path: candidate,
2318
+ bytes,
2319
+ errno,
2320
+ ...errorFields(err)
2321
+ });
2322
+ }
2323
+ }
2324
+ function expandAndValidate(requestedPath, homeDir) {
2325
+ if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
2326
+ return null;
2327
+ }
2328
+ const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join(homeDir, requestedPath.slice(2)) : requestedPath;
2329
+ if (expanded.split(/[/\\]/).includes("..")) {
2330
+ return null;
2331
+ }
2332
+ if (!isAbsolute(expanded)) {
2333
+ return null;
2334
+ }
2335
+ const candidate = resolve2(expanded);
2336
+ const name = basename(candidate);
2337
+ return name === "" || name === "." || name === ".." ? null : candidate;
2338
+ }
2339
+ async function resolveNearestExistingAncestor(directory) {
2340
+ const missingSegments = [];
2341
+ let current = directory;
2342
+ for (; ; ) {
2343
+ try {
2344
+ return { existingAncestor: await realpath(current), missingSegments };
2345
+ } catch (err) {
2346
+ const parent = dirname2(current);
2347
+ if (err.code !== "ENOENT" || parent === current) {
2348
+ throw err;
2349
+ }
2350
+ missingSegments.unshift(basename(current));
2351
+ current = parent;
2352
+ }
2353
+ }
2354
+ }
2355
+ async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
2356
+ for (const directory of allowedDirectories) {
2357
+ if (!isAbsolute(directory)) {
2358
+ log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
2359
+ continue;
2360
+ }
2361
+ const realDirectory = await realpathCreatingIfMissing(directory);
2362
+ if (realDirectory !== null && contains(realDirectory, realTarget)) {
2363
+ return realDirectory;
2364
+ }
2365
+ }
2366
+ return null;
2367
+ }
2368
+ async function realpathCreatingIfMissing(directory) {
2369
+ try {
2370
+ return await realpath(directory);
2371
+ } catch (err) {
2372
+ if (err.code !== "ENOENT") {
2373
+ log("warn", "file_push_allowed_directory_skipped", {
2374
+ directory,
2375
+ reason: "unresolvable",
2376
+ ...errorFields(err)
2377
+ });
2378
+ return null;
2379
+ }
2380
+ }
2381
+ try {
2382
+ await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
2383
+ await chmod(directory, DIRECTORY_MODE);
2384
+ return await realpath(directory);
2385
+ } catch (err) {
2386
+ log("warn", "file_push_allowed_directory_skipped", {
2387
+ directory,
2388
+ reason: "create_failed",
2389
+ ...errorFields(err)
2390
+ });
2391
+ return null;
2392
+ }
2393
+ }
2394
+ function contains(realDirectory, realTarget) {
2395
+ const rel = relative(realDirectory, realTarget);
2396
+ return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
2397
+ }
2398
+ async function createMissingDirectories(existingAncestor, missingSegments) {
2399
+ let current = existingAncestor;
2400
+ for (const segment of missingSegments) {
2401
+ current = join(current, segment);
2402
+ await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
2403
+ await chmod(current, DIRECTORY_MODE);
2404
+ }
2405
+ }
2406
+ async function writeAtomically(realTarget, content) {
2407
+ const temporaryPath = join(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2408
+ let handle;
2409
+ try {
2410
+ handle = await open2(temporaryPath, "wx", FILE_MODE);
2411
+ await handle.writeFile(content);
2412
+ await handle.chmod(FILE_MODE);
2413
+ await handle.close();
2414
+ handle = void 0;
2415
+ await rename(temporaryPath, realTarget);
2416
+ } catch (err) {
2417
+ await discardTemporaryFile(temporaryPath, handle);
2418
+ throw err;
2419
+ }
2420
+ }
2421
+ async function discardTemporaryFile(temporaryPath, handle) {
2422
+ try {
2423
+ await handle?.close();
2424
+ } catch (err) {
2425
+ log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
2426
+ }
2427
+ try {
2428
+ await unlink(temporaryPath);
2429
+ } catch (err) {
2430
+ const errno = err.code;
2431
+ if (errno !== "ENOENT" && errno !== "ENOTDIR") {
2432
+ log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
2433
+ }
2434
+ }
2435
+ }
2436
+ function refuse(code, message, fields) {
2437
+ log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
2438
+ return { ok: false, code, message };
2439
+ }
2440
+
2441
+ // src/lib/runner-file-sync.ts
2442
+ var MAX_ACK_ATTEMPTS = 5;
2443
+ async function syncPendingRunnerFiles(options) {
2444
+ const pending = await listPendingFiles(options);
2445
+ const pendingIds = new Set(pending.map((file) => file.id));
2446
+ for (const id of options.ackFailures.keys()) {
2447
+ if (!pendingIds.has(id)) options.ackFailures.delete(id);
2448
+ }
2449
+ if (pending.length === 0) return 0;
2450
+ options.log({
2451
+ level: "info",
2452
+ message: `Runner file sync: ${pending.length} file(s) queued for this runner`
2453
+ });
2454
+ let applied = 0;
2455
+ for (const file of pending) {
2456
+ if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
2457
+ if (await applyOne(options, file)) applied += 1;
2458
+ }
2459
+ return applied;
2460
+ }
2461
+ async function listPendingFiles(options) {
2462
+ let res;
2463
+ try {
2464
+ res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
2465
+ headers: { Authorization: options.getAuthHeader() }
2466
+ });
2467
+ } catch (err) {
2468
+ options.log({
2469
+ level: "warn",
2470
+ message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
2471
+ });
2472
+ return [];
2473
+ }
2474
+ if (!res.ok) {
2475
+ options.log({
2476
+ level: res.status === 404 ? "debug" : "warn",
2477
+ message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
2478
+ });
2479
+ return [];
2480
+ }
2481
+ let body;
2482
+ try {
2483
+ body = await res.json();
2484
+ } catch (err) {
2485
+ options.log({
2486
+ level: "warn",
2487
+ message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
2488
+ });
2489
+ return [];
2490
+ }
2491
+ if (!Array.isArray(body)) {
2492
+ options.log({
2493
+ level: "warn",
2494
+ message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
2495
+ });
2496
+ return [];
2497
+ }
2498
+ const files = [];
2499
+ for (const entry of body) {
2500
+ const file = asPendingFile(entry);
2501
+ if (file === null) {
2502
+ options.log({
2503
+ level: "warn",
2504
+ message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
2505
+ });
2506
+ continue;
2507
+ }
2508
+ files.push(file);
2509
+ }
2510
+ return files;
2511
+ }
2512
+ function asPendingFile(entry) {
2513
+ if (entry === null || typeof entry !== "object") return null;
2514
+ const { id, path, size } = entry;
2515
+ if (typeof id !== "string" || id === "") return null;
2516
+ if (typeof path !== "string" || path === "") return null;
2517
+ if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
2518
+ return { id, path, size };
2519
+ }
2520
+ async function applyOne(options, file) {
2521
+ const label = `${file.id.slice(0, 8)} (${file.path})`;
2522
+ if (options.allowedDirectories.length === 0) {
2523
+ options.log({
2524
+ level: "warn",
2525
+ message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
2526
+ });
2527
+ await ack(options, file, "rejected", "file_sync_disabled");
2528
+ return false;
2529
+ }
2530
+ if (file.size > MAX_FILE_PUSH_BYTES) {
2531
+ options.log({
2532
+ level: "warn",
2533
+ message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
2534
+ });
2535
+ await ack(options, file, "rejected", "file_too_large");
2536
+ return false;
2537
+ }
2538
+ const download = await downloadContent(options, file, label);
2539
+ if (!download.ok) {
2540
+ if (download.terminal) await ack(options, file, "rejected", download.code);
2541
+ return false;
2542
+ }
2543
+ let outcome;
2544
+ try {
2545
+ outcome = await writePushedFile({
2546
+ requestedPath: file.path,
2547
+ content: download.content,
2548
+ allowedDirectories: options.allowedDirectories,
2549
+ homeDir: options.homeDir
2550
+ });
2551
+ } catch (err) {
2552
+ options.log({
2553
+ level: "error",
2554
+ message: `Runner file ${label} could not be written: ${describe(err)}`
2555
+ });
2556
+ await ack(options, file, "rejected", "write_failed");
2557
+ return false;
2558
+ }
2559
+ if (!outcome.ok) {
2560
+ options.log({
2561
+ level: "warn",
2562
+ message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
2563
+ });
2564
+ await ack(options, file, "rejected", outcome.code);
2565
+ return false;
2566
+ }
2567
+ options.log({
2568
+ level: "info",
2569
+ message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
2570
+ });
2571
+ await ack(options, file, "applied");
2572
+ return true;
2573
+ }
2574
+ function durableDownloadCode(status) {
2575
+ return status === 413 ? "file_too_large" : "write_failed";
2576
+ }
2577
+ async function downloadContent(options, file, label) {
2578
+ try {
2579
+ const res = await options.fetchImpl(
2580
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
2581
+ { headers: { Authorization: options.getAuthHeader() } }
2582
+ );
2583
+ if (!res.ok) {
2584
+ const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
2585
+ if (!terminal) {
2586
+ options.log({
2587
+ level: "warn",
2588
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
2589
+ });
2590
+ return { ok: false, terminal: false };
2591
+ }
2592
+ const code = durableDownloadCode(res.status);
2593
+ options.log({
2594
+ level: "error",
2595
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
2596
+ });
2597
+ return { ok: false, terminal: true, code };
2598
+ }
2599
+ return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
2600
+ } catch (err) {
2601
+ options.log({
2602
+ level: "warn",
2603
+ message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
2604
+ });
2605
+ return { ok: false, terminal: false };
2606
+ }
2607
+ }
2608
+ async function ack(options, file, status, reason) {
2609
+ const outcome = `${status}${reason ? ` (${reason})` : ""}`;
2610
+ try {
2611
+ const res = await options.fetchImpl(
2612
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
2613
+ {
2614
+ method: "POST",
2615
+ headers: {
2616
+ Authorization: options.getAuthHeader(),
2617
+ "Content-Type": "application/json"
2618
+ },
2619
+ body: JSON.stringify(reason ? { status, reason } : { status })
2620
+ }
2621
+ );
2622
+ if (!res.ok) {
2623
+ recordAckFailure(
2624
+ options,
2625
+ file,
2626
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
2627
+ );
2628
+ return;
2629
+ }
2630
+ options.ackFailures.delete(file.id);
2631
+ } catch (err) {
2632
+ recordAckFailure(
2633
+ options,
2634
+ file,
2635
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
2636
+ );
2637
+ }
2638
+ }
2639
+ function recordAckFailure(options, file, what) {
2640
+ const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
2641
+ options.ackFailures.set(file.id, attempts);
2642
+ options.log({
2643
+ level: "error",
2644
+ message: attempts >= MAX_ACK_ATTEMPTS ? `${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})`
2645
+ });
2646
+ }
2647
+ function describe(err) {
2648
+ return err instanceof Error ? err.message : String(err);
2649
+ }
2650
+
2036
2651
  // src/lib/channels/driver.ts
2037
2652
  function messageIdOf(m) {
2038
2653
  if (!m || typeof m !== "object") return void 0;
@@ -2061,7 +2676,10 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
2061
2676
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
2062
2677
  var HEARTBEAT_MS = 6e4;
2063
2678
  var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
2679
+ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
2680
+ var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2064
2681
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
2682
+ var MAX_SUPERSEDED_CONVERSATIONS = 256;
2065
2683
  var ChannelAuthError = class extends Error {
2066
2684
  constructor(message) {
2067
2685
  super(message);
@@ -2098,8 +2716,38 @@ var ChannelDriver = class _ChannelDriver {
2098
2716
  pausedMaxWaitMs;
2099
2717
  stuckQueuedMs;
2100
2718
  now;
2719
+ fileSyncDirectories;
2720
+ homeDir;
2101
2721
  /** Cache of conversationId → opencode sessionId. */
2102
2722
  sessions = /* @__PURE__ */ new Map();
2723
+ /**
2724
+ * conversationId → the opencode session this runner has ABANDONED as that
2725
+ * conversation's binding (#553), after a genuine (`sessionExists === true`)
2726
+ * dispatch failure: the session still exists but is wedged, so #485's self-heal
2727
+ * must bind a fresh one.
2728
+ *
2729
+ * Dropping the local binding + clearing the server row is not enough on its own:
2730
+ * a SIBLING message dispatched earlier in the same drain is still in-flight under
2731
+ * the same session, and its watcher's routine status writes carry
2732
+ * `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
2733
+ * and `ensureSession`'s persisted-id fallback then reuses it, defeating the
2734
+ * self-heal. This map makes the runner authoritative instead of racing those
2735
+ * writes: *`ensureSession` never reuses an abandoned id for that conversation,
2736
+ * whatever the server row says* — which holds even when the resurrecting write
2737
+ * is one we deliberately keep (see `markDone`).
2738
+ *
2739
+ * Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
2740
+ * one conversation hold ONE entry (the newest abandonment replaces the older), and
2741
+ * hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
2742
+ * NEWEST abandoned id per conversation is guarded: after a second abandonment a
2743
+ * late sibling of the FIRST session can write that id back and `ensureSession`
2744
+ * will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
2745
+ * NOT dropped when the session's watcher tears down: `markDone` still writes the
2746
+ * abandoned id back (it must, or the reply is lost), so the guard has to outlive
2747
+ * the turn that resurrects it. In-memory only — a restart forgets it, at the same
2748
+ * bounded cost.
2749
+ */
2750
+ supersededSessions = /* @__PURE__ */ new Map();
2103
2751
  /**
2104
2752
  * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
2105
2753
  * longer idempotent (no caller-supplied `messageID`), and its read-back picks
@@ -2222,6 +2870,24 @@ var ChannelDriver = class _ChannelDriver {
2222
2870
  sessionTitles = /* @__PURE__ */ new Map();
2223
2871
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
2224
2872
  draining = false;
2873
+ /**
2874
+ * Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
2875
+ * drain ping don't download, write and ack the same file twice.
2876
+ */
2877
+ syncingFiles = false;
2878
+ /**
2879
+ * Consecutive failed acks per pending file (#559). Lives on the driver so it
2880
+ * survives across drains — without it, a file whose ack keeps failing is
2881
+ * re-downloaded and re-written every ~2s until the server expires it.
2882
+ */
2883
+ fileAckFailures = /* @__PURE__ */ new Map();
2884
+ /**
2885
+ * Monotonic count of files this runner has pulled and written (#559). Only
2886
+ * ever increases, so `run.ts` detects work by comparing it against the value
2887
+ * it saw on the previous cycle — including work that landed mid-sleep, the
2888
+ * same trick `lastProxiedActivityAt` uses.
2889
+ */
2890
+ appliedFileCount = 0;
2225
2891
  /**
2226
2892
  * The currently-executing `drainPending()` promise, or null when idle. Lets a
2227
2893
  * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
@@ -2252,6 +2918,8 @@ var ChannelDriver = class _ChannelDriver {
2252
2918
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2253
2919
  this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2254
2920
  this.now = config2.now ?? (() => Date.now());
2921
+ this.fileSyncDirectories = config2.fileSyncDirectories ?? [];
2922
+ this.homeDir = config2.homeDir ?? homedir();
2255
2923
  }
2256
2924
  /** The IPv4-loopback base URL for the local `opencode serve`. */
2257
2925
  get opencodeBase() {
@@ -2279,10 +2947,51 @@ var ChannelDriver = class _ChannelDriver {
2279
2947
  );
2280
2948
  return run2;
2281
2949
  }
2282
- async runDrain() {
2283
- let dispatched = 0;
2950
+ /**
2951
+ * Pull-and-apply any files Evident has queued for this runner (#559), riding
2952
+ * the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
2953
+ * and drain ping that call `drainPending()`. There is deliberately no channel,
2954
+ * control frame or poll loop of its own: worst-case latency is one poll tick.
2955
+ *
2956
+ * NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
2957
+ * cost a conversation turn. Failures are logged and either acked as a terminal
2958
+ * outcome or left pending for the next drain (see `runner-file-sync.ts`).
2959
+ *
2960
+ * Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
2961
+ *
2962
+ * @returns the number of files written to disk.
2963
+ */
2964
+ async syncPendingFiles() {
2965
+ if (this.stopped) return 0;
2966
+ if (this.syncingFiles) return 0;
2967
+ this.syncingFiles = true;
2284
2968
  try {
2285
- const conversations = await this.getPendingConversations();
2969
+ const applied = await syncPendingRunnerFiles({
2970
+ agentId: this.agentId,
2971
+ apiUrl: this.apiUrl,
2972
+ getAuthHeader: this.getAuthHeader,
2973
+ fetchImpl: this.fetchImpl,
2974
+ allowedDirectories: this.fileSyncDirectories,
2975
+ homeDir: this.homeDir,
2976
+ ackFailures: this.fileAckFailures,
2977
+ log: this.log
2978
+ });
2979
+ this.appliedFileCount += applied;
2980
+ return applied;
2981
+ } catch (err) {
2982
+ this.log({
2983
+ level: "error",
2984
+ message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
2985
+ });
2986
+ return 0;
2987
+ } finally {
2988
+ this.syncingFiles = false;
2989
+ }
2990
+ }
2991
+ async runDrain() {
2992
+ let dispatched = 0;
2993
+ try {
2994
+ const conversations = await this.getPendingConversations();
2286
2995
  if (conversations.length > 0) {
2287
2996
  const total = conversations.reduce((sum, c) => sum + (c.pending_message_count ?? 0), 0);
2288
2997
  this.log({
@@ -2312,6 +3021,28 @@ var ChannelDriver = class _ChannelDriver {
2312
3021
  }
2313
3022
  return false;
2314
3023
  }
3024
+ /**
3025
+ * File-pull work, for `run.ts`'s idle accounting (#559).
3026
+ *
3027
+ * Pulling a file is real work that `drainPending()` knows nothing about, so
3028
+ * without this a near-idle runner counts a credential pull as an empty tick
3029
+ * and `--idle-timeout` can `process.exit` mid-pull — leaving a
3030
+ * `.evident-push-*.tmp` behind — or immediately after the write, before the
3031
+ * browser has run the authorize/callback that activates it (the user then sees
3032
+ * `saved_not_activated` for a runner that was fine).
3033
+ *
3034
+ * Two signals because one cannot cover both cases: `inFlight` is the pull
3035
+ * happening RIGHT NOW (it may outlive the tick that started it), and
3036
+ * `appliedFiles` is monotonic so a pull that started AND finished between two
3037
+ * idle checks still shows up as an advance.
3038
+ *
3039
+ * CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
3040
+ * the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
3041
+ * samples afterwards reads `true` every single cycle and can never idle out.
3042
+ */
3043
+ fileSyncActivity() {
3044
+ return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
3045
+ }
2315
3046
  /**
2316
3047
  * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2317
3048
  * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
@@ -2373,7 +3104,7 @@ var ChannelDriver = class _ChannelDriver {
2373
3104
  await this.sleep(step);
2374
3105
  }
2375
3106
  }
2376
- while (this.hasInFlightWatchers()) {
3107
+ while (this.hasInFlightWatchers() || this.syncingFiles) {
2377
3108
  if (this.now() >= deadline) return false;
2378
3109
  await this.sleep(step);
2379
3110
  }
@@ -2408,10 +3139,15 @@ var ChannelDriver = class _ChannelDriver {
2408
3139
  * @returns the count of messages NEWLY dispatched (not already in-flight).
2409
3140
  */
2410
3141
  async processConversation(conv) {
2411
- const sessionId = await this.ensureSession(conv);
3142
+ const { sessionId, refusedSessionId } = await this.ensureSession(conv);
2412
3143
  const messages = await this.getPendingMessages(conv.id);
2413
3144
  let dispatched = 0;
2414
3145
  let skippedAlreadyDispatched = 0;
3146
+ if (refusedSessionId && messages.length > 0) {
3147
+ void this.postSignal(conv.id, messages[0].id, "session_superseded", {
3148
+ superseded_session_id: refusedSessionId
3149
+ });
3150
+ }
2415
3151
  for (const message of messages) {
2416
3152
  if (this.stopped) break;
2417
3153
  if (this.dispatched.has(message.id)) {
@@ -2460,6 +3196,13 @@ var ChannelDriver = class _ChannelDriver {
2460
3196
  }
2461
3197
  const errorMessage = err instanceof Error ? err.message : String(err);
2462
3198
  this.sessions.delete(conv.id);
3199
+ this.supersede(conv.id, sessionId);
3200
+ this.log({
3201
+ level: "warn",
3202
+ 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.`,
3203
+ conversation_id: conv.id,
3204
+ message_id: message.id
3205
+ });
2463
3206
  await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
2464
3207
  this.log({
2465
3208
  level: "warn",
@@ -2500,8 +3243,42 @@ var ChannelDriver = class _ChannelDriver {
2500
3243
  this.ensureWatcherRunning(sessionId);
2501
3244
  return dispatched;
2502
3245
  }
3246
+ /**
3247
+ * Record that `sessionId` is no longer a valid binding for `conversationId`
3248
+ * (#553). Keyed by conversation and hard-capped, so it cannot grow with the
3249
+ * number of failures — see the `supersededSessions` field doc.
3250
+ */
3251
+ supersede(conversationId, sessionId) {
3252
+ this.supersededSessions.delete(conversationId);
3253
+ this.supersededSessions.set(conversationId, sessionId);
3254
+ while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
3255
+ const oldest = this.supersededSessions.keys().next().value;
3256
+ if (oldest === void 0) return;
3257
+ this.supersededSessions.delete(oldest);
3258
+ }
3259
+ }
3260
+ /** Whether `sessionId` is the session this conversation has abandoned (#553). */
3261
+ isSuperseded(conversationId, sessionId) {
3262
+ return this.supersededSessions.get(conversationId) === sessionId;
3263
+ }
3264
+ /**
3265
+ * Resolve the opencode session to run this conversation's turns in.
3266
+ *
3267
+ * `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
3268
+ * binding was an id this runner had abandoned, so a resurrection genuinely
3269
+ * happened and a fresh session was bound instead. The caller reports it.
3270
+ */
2503
3271
  async ensureSession(conv) {
2504
3272
  const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
3273
+ if (bound && this.isSuperseded(conv.id, bound)) {
3274
+ this.log({
3275
+ level: "warn",
3276
+ 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.`,
3277
+ conversation_id: conv.id
3278
+ });
3279
+ this.sessions.delete(conv.id);
3280
+ return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
3281
+ }
2505
3282
  if (bound) {
2506
3283
  const exists = await sessionExists(this.port, bound);
2507
3284
  if (exists === false) {
@@ -2511,12 +3288,12 @@ var ChannelDriver = class _ChannelDriver {
2511
3288
  conversation_id: conv.id
2512
3289
  });
2513
3290
  this.sessions.delete(conv.id);
2514
- return this.createAndBindSession(conv.id);
3291
+ return { sessionId: await this.createAndBindSession(conv.id) };
2515
3292
  }
2516
3293
  this.sessions.set(conv.id, bound);
2517
- return bound;
3294
+ return { sessionId: bound };
2518
3295
  }
2519
- return this.createAndBindSession(conv.id);
3296
+ return { sessionId: await this.createAndBindSession(conv.id) };
2520
3297
  }
2521
3298
  /**
2522
3299
  * Create a fresh OpenCode session for a conversation, cache the binding, and
@@ -2527,7 +3304,12 @@ var ChannelDriver = class _ChannelDriver {
2527
3304
  const directory = await this.resolveOpenCodeDirectory();
2528
3305
  const sessionId = await createOpenCodeSession(this.port, directory);
2529
3306
  this.sessions.set(conversationId, sessionId);
2530
- await this.persistSession(conversationId, sessionId).catch(() => {
3307
+ await this.persistSession(conversationId, sessionId).catch((err) => {
3308
+ this.log({
3309
+ level: "warn",
3310
+ 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)}`,
3311
+ conversation_id: conversationId
3312
+ });
2531
3313
  });
2532
3314
  return sessionId;
2533
3315
  }
@@ -2719,12 +3501,17 @@ var ChannelDriver = class _ChannelDriver {
2719
3501
  stuckReported: false,
2720
3502
  lastAliveAt: 0,
2721
3503
  aliveInFlight: false,
3504
+ titleSynced: false,
3505
+ titleSyncInFlight: false,
2722
3506
  awaitingHumanLatched: false,
2723
3507
  pausedOnQuestion: false,
2724
3508
  pausedOnPermission: false,
2725
3509
  pausedClearConfirmed: false,
2726
3510
  pausedInFlight: false,
2727
- deliveryDeadlineAnchored: false
3511
+ deliveryDeadlineAnchored: false,
3512
+ b2PinnedSinceMs: 0,
3513
+ b2LastDescendantCheckMs: 0,
3514
+ b2AbandonedSignalled: false
2728
3515
  });
2729
3516
  }
2730
3517
  /**
@@ -2792,12 +3579,17 @@ var ChannelDriver = class _ChannelDriver {
2792
3579
  // with no extra `re_adopted` signal needed (folds old WI-6).
2793
3580
  lastAliveAt: 0,
2794
3581
  aliveInFlight: false,
3582
+ titleSynced: false,
3583
+ titleSyncInFlight: false,
2795
3584
  awaitingHumanLatched: false,
2796
3585
  pausedOnQuestion: false,
2797
3586
  pausedOnPermission: false,
2798
3587
  pausedClearConfirmed: false,
2799
3588
  pausedInFlight: false,
2800
- deliveryDeadlineAnchored: false
3589
+ deliveryDeadlineAnchored: false,
3590
+ b2PinnedSinceMs: 0,
3591
+ b2LastDescendantCheckMs: 0,
3592
+ b2AbandonedSignalled: false
2801
3593
  });
2802
3594
  }
2803
3595
  /**
@@ -2959,58 +3751,7 @@ var ChannelDriver = class _ChannelDriver {
2959
3751
  }
2960
3752
  }
2961
3753
  if (state === "done") {
2962
- this.anchorDeliveryDeadline(inFlight);
2963
- if (!inFlight.done) {
2964
- this.log({
2965
- level: "info",
2966
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2967
- conversation_id: conv.id,
2968
- message_id: inFlight.evidentMessageId
2969
- });
2970
- const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2971
- const usage = messageUsage(messages, inFlight.opencodeMessageId);
2972
- try {
2973
- await this.markDone(
2974
- conv.id,
2975
- inFlight.evidentMessageId,
2976
- sessionId,
2977
- inFlight.opencodeMessageId,
2978
- title,
2979
- usage
2980
- );
2981
- } catch (err) {
2982
- if (err instanceof ChannelAuthError) throw err;
2983
- if (err instanceof ChannelTerminalError) {
2984
- this.log({
2985
- level: "warn",
2986
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2987
- conversation_id: conv.id,
2988
- message_id: inFlight.evidentMessageId
2989
- });
2990
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2991
- return;
2992
- }
2993
- if (this.now() >= inFlight.deadline) {
2994
- this.log({
2995
- level: "warn",
2996
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
2997
- conversation_id: conv.id,
2998
- message_id: inFlight.evidentMessageId
2999
- });
3000
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3001
- return;
3002
- }
3003
- this.log({
3004
- level: "warn",
3005
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
3006
- conversation_id: conv.id,
3007
- message_id: inFlight.evidentMessageId
3008
- });
3009
- return;
3010
- }
3011
- inFlight.done = true;
3012
- }
3013
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3754
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3014
3755
  return;
3015
3756
  }
3016
3757
  if (state === "failed") {
@@ -3024,8 +3765,16 @@ var ChannelDriver = class _ChannelDriver {
3024
3765
  message_id: inFlight.evidentMessageId
3025
3766
  });
3026
3767
  const usage = messageUsage(messages, inFlight.opencodeMessageId);
3768
+ const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
3027
3769
  try {
3028
- await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2, usage);
3770
+ await this.markFailed(
3771
+ conv.id,
3772
+ inFlight.evidentMessageId,
3773
+ sessionId,
3774
+ error2,
3775
+ usage,
3776
+ failure
3777
+ );
3029
3778
  } catch (err) {
3030
3779
  if (err instanceof ChannelAuthError) throw err;
3031
3780
  if (err instanceof ChannelTerminalError) {
@@ -3070,6 +3819,44 @@ var ChannelDriver = class _ChannelDriver {
3070
3819
  });
3071
3820
  }
3072
3821
  const activelyRunning = state === "running" && !awaitingHuman;
3822
+ const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
3823
+ const snapshotReadable = messages != null && messages.length > 0;
3824
+ if (!pinnedNow) {
3825
+ if (snapshotReadable) {
3826
+ inFlight.b2PinnedSinceMs = 0;
3827
+ inFlight.b2LastDescendantCheckMs = 0;
3828
+ inFlight.b2AbandonedSignalled = false;
3829
+ }
3830
+ } else {
3831
+ if (inFlight.b2AbandonedSignalled) {
3832
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3833
+ return;
3834
+ }
3835
+ if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
3836
+ const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
3837
+ if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
3838
+ inFlight.b2LastDescendantCheckMs = this.now();
3839
+ const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
3840
+ if (isB2AbandonmentConfirmed({
3841
+ pinnedForMs,
3842
+ minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
3843
+ descendantOngoing
3844
+ })) {
3845
+ inFlight.b2AbandonedSignalled = true;
3846
+ this.log({
3847
+ level: "warn",
3848
+ 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`,
3849
+ conversation_id: conv.id,
3850
+ message_id: id
3851
+ });
3852
+ void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
3853
+ watched_for_ms: pinnedForMs
3854
+ });
3855
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3856
+ return;
3857
+ }
3858
+ }
3859
+ }
3073
3860
  if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
3074
3861
  this.log({
3075
3862
  level: "warn",
@@ -3089,6 +3876,18 @@ var ChannelDriver = class _ChannelDriver {
3089
3876
  inFlight.aliveInFlight = false;
3090
3877
  if (ok) inFlight.lastAliveAt = this.now();
3091
3878
  });
3879
+ if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
3880
+ inFlight.titleSyncInFlight = true;
3881
+ void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
3882
+ if (!title) {
3883
+ inFlight.titleSyncInFlight = false;
3884
+ return;
3885
+ }
3886
+ const ok = await this.patchConversationTitle(conv.id, title);
3887
+ inFlight.titleSyncInFlight = false;
3888
+ if (ok) inFlight.titleSynced = true;
3889
+ });
3890
+ }
3092
3891
  }
3093
3892
  if (awaitingHuman) {
3094
3893
  if (!inFlight.awaitingHumanLatched) {
@@ -3126,6 +3925,70 @@ var ChannelDriver = class _ChannelDriver {
3126
3925
  this.removeInFlight(watcher, inFlight.evidentMessageId);
3127
3926
  }
3128
3927
  }
3928
+ /**
3929
+ * Settle a message whose run-state has resolved `'done'` — extracted verbatim
3930
+ * (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
3931
+ * inline `state === 'done'` branch body, so a SECOND caller (the #721
3932
+ * b2-abandonment resolution) can reach the exact same completion behavior
3933
+ * (delivery-deadline anchoring, title resolution, usage extraction, and
3934
+ * `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
3935
+ * and risking the two copies silently drifting apart.
3936
+ */
3937
+ async settleMessageDone(sessionId, watcher, inFlight, messages) {
3938
+ const conv = watcher.conv;
3939
+ this.anchorDeliveryDeadline(inFlight);
3940
+ if (!inFlight.done) {
3941
+ this.log({
3942
+ level: "info",
3943
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
3944
+ conversation_id: conv.id,
3945
+ message_id: inFlight.evidentMessageId
3946
+ });
3947
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
3948
+ const usage = messageUsage(messages, inFlight.opencodeMessageId);
3949
+ try {
3950
+ await this.markDone(
3951
+ conv.id,
3952
+ inFlight.evidentMessageId,
3953
+ sessionId,
3954
+ inFlight.opencodeMessageId,
3955
+ title,
3956
+ usage
3957
+ );
3958
+ } catch (err) {
3959
+ if (err instanceof ChannelAuthError) throw err;
3960
+ if (err instanceof ChannelTerminalError) {
3961
+ this.log({
3962
+ level: "warn",
3963
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
3964
+ conversation_id: conv.id,
3965
+ message_id: inFlight.evidentMessageId
3966
+ });
3967
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3968
+ return;
3969
+ }
3970
+ if (this.now() >= inFlight.deadline) {
3971
+ this.log({
3972
+ level: "warn",
3973
+ 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)}`,
3974
+ conversation_id: conv.id,
3975
+ message_id: inFlight.evidentMessageId
3976
+ });
3977
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3978
+ return;
3979
+ }
3980
+ this.log({
3981
+ level: "warn",
3982
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
3983
+ conversation_id: conv.id,
3984
+ message_id: inFlight.evidentMessageId
3985
+ });
3986
+ return;
3987
+ }
3988
+ inFlight.done = true;
3989
+ }
3990
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3991
+ }
3129
3992
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
3130
3993
  /**
3131
3994
  * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
@@ -3292,6 +4155,7 @@ var ChannelDriver = class _ChannelDriver {
3292
4155
  if (state === "failed") {
3293
4156
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
3294
4157
  const usage = messageUsage(messages, ocId ?? "");
4158
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3295
4159
  this.log({
3296
4160
  level: "error",
3297
4161
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -3299,7 +4163,7 @@ var ChannelDriver = class _ChannelDriver {
3299
4163
  message_id: row.id
3300
4164
  });
3301
4165
  try {
3302
- await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
4166
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
3303
4167
  } catch (err) {
3304
4168
  if (err instanceof ChannelAuthError) throw err;
3305
4169
  if (err instanceof ChannelTerminalError) {
@@ -3705,6 +4569,47 @@ var ChannelDriver = class _ChannelDriver {
3705
4569
  }
3706
4570
  return false;
3707
4571
  }
4572
+ /**
4573
+ * Tri-state variant of the upward parentID membership walk (#721), used ONLY
4574
+ * by `isAnyDescendantSessionOngoing`. Walks the SAME cached
4575
+ * `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
4576
+ * `sessionBelongsTo`, which deliberately collapses "confirmed not a
4577
+ * descendant" and "the walk's fetch failed" into the same `false` (safe for
4578
+ * its OTHER callers: interaction attribution and the recovery-path
4579
+ * `isAnyDescendantSessionAlive`, both of which just retry next tick with no
4580
+ * safety consequence either way) — this variant keeps those two outcomes
4581
+ * SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
4582
+ * (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
4583
+ * not ongoing".
4584
+ *
4585
+ * Return contract:
4586
+ * - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
4587
+ * - `false` → the walk reached a definitive, parent-less root session
4588
+ * WITHOUT ever matching `rootSessionId` — `sessionId` is
4589
+ * CONFIRMED NOT a descendant of it.
4590
+ * - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
4591
+ * through the walk (`resolveSessionParent` returned `undefined`),
4592
+ * or the depth cap (32) was hit without a definitive answer (a
4593
+ * pathological/cyclic chain proves nothing either way). NEVER
4594
+ * treat this the same as `false` — see `sessionBelongsTo`'s own
4595
+ * doc comment above for why that collapse is safe THERE but not
4596
+ * here.
4597
+ *
4598
+ * `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
4599
+ * to the live-path descendant check, not a modification of shared code used
4600
+ * by interaction attribution or the recovery path.
4601
+ */
4602
+ async resolveSessionMembership(sessionId, rootSessionId) {
4603
+ let current = sessionId;
4604
+ for (let depth = 0; current && depth < 32; depth++) {
4605
+ if (current === rootSessionId) return true;
4606
+ const parent = await this.resolveSessionParent(current);
4607
+ if (parent === void 0) return null;
4608
+ if (parent === null) return false;
4609
+ current = parent;
4610
+ }
4611
+ return null;
4612
+ }
3708
4613
  /**
3709
4614
  * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3710
4615
  * `null` for a root session (no parent) and `undefined` when opencode is
@@ -3789,6 +4694,54 @@ var ChannelDriver = class _ChannelDriver {
3789
4694
  }
3790
4695
  return null;
3791
4696
  }
4697
+ /**
4698
+ * Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
4699
+ * session title onto the conversation via the PLAIN conversation-update
4700
+ * endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
4701
+ * message-status endpoint `markProcessing`/`markDone` use. Deliberately a
4702
+ * separate, lighter call: it carries no `status`, so it cannot re-trigger the
4703
+ * `processing`/`done` transition side effects (Slack notices, activity-log
4704
+ * rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
4705
+ * ever touches `conversations.title`. That route (`routes/conversations.ts`)
4706
+ * skips a title write matching the stored value, so a redundant call with the
4707
+ * same title is a real no-op — it does not bump `updated_at`, which the
4708
+ * conversation list sorts and paginates on. (Note this is a DIFFERENT guard
4709
+ * from `threads.ts`'s "non-empty AND changed" one, which only covers the
4710
+ * message-status PATCH; the non-empty half is enforced here instead, by
4711
+ * `resolveSessionTitle` never returning an empty/placeholder title.)
4712
+ *
4713
+ * Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
4714
+ * is logged and the title is simply retried on the next heartbeat tick (the
4715
+ * caller only latches `titleSynced` on `true`).
4716
+ */
4717
+ async patchConversationTitle(conversationId, title) {
4718
+ try {
4719
+ const res = await this.fetchImpl(
4720
+ `${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
4721
+ {
4722
+ method: "PATCH",
4723
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4724
+ body: JSON.stringify({ title })
4725
+ }
4726
+ );
4727
+ if (!res.ok) {
4728
+ this.log({
4729
+ level: "debug",
4730
+ message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
4731
+ conversation_id: conversationId
4732
+ });
4733
+ return false;
4734
+ }
4735
+ return true;
4736
+ } catch (err) {
4737
+ this.log({
4738
+ level: "debug",
4739
+ 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)}`,
4740
+ conversation_id: conversationId
4741
+ });
4742
+ return false;
4743
+ }
4744
+ }
3792
4745
  /**
3793
4746
  * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3794
4747
  * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
@@ -3847,6 +4800,84 @@ var ChannelDriver = class _ChannelDriver {
3847
4800
  }
3848
4801
  return false;
3849
4802
  }
4803
+ /**
4804
+ * LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
4805
+ * sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
4806
+ * in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
4807
+ *
4808
+ * Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
4809
+ * cross-check above): that method judges liveness from the child's OWN
4810
+ * TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
4811
+ * on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
4812
+ * path the local opencode server IS running, so its in-memory status map is
4813
+ * live and authoritative — and per ADR-0047 §4a ("the child has its own entry
4814
+ * [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
4815
+ * ENTIRE turn (including any tool call it is itself executing), not a
4816
+ * per-message transcript snapshot. This sidesteps the "child's own tool is
4817
+ * executing, between its step's completion and the next generation step"
4818
+ * transcript gap that a transcript-based check would need a second,
4819
+ * sustained-window bound to guard against — it is simply not derived from
4820
+ * message timestamps at all.
4821
+ *
4822
+ * Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
4823
+ * status, as the recovery path does per §4a)? Because on the LIVE path the
4824
+ * root session can be shared: a SECOND, unrelated user message can land on the
4825
+ * SAME session (issue #721's own root cause) and keep the root `busy` for a
4826
+ * reason that has nothing to do with THIS message's delegation. A `task`
4827
+ * descendant session is spawned for exactly one delegated turn and never
4828
+ * reused, so its OWN status-map entry is unambiguous evidence about that one
4829
+ * delegation — which the root's status is not.
4830
+ *
4831
+ * Why membership is checked via `resolveSessionMembership`, NOT
4832
+ * `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
4833
+ * `GET /session/:id` fetch failure into "not a descendant", which would
4834
+ * silently drop a genuinely-live candidate from consideration on the one
4835
+ * unlucky tick its membership-walk fetch hiccups (#721).
4836
+ * `resolveSessionMembership` keeps that failure mode as a distinct `null`
4837
+ * (indeterminate) so it is folded into THIS method's own `indeterminate` flag
4838
+ * instead.
4839
+ *
4840
+ * Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
4841
+ * - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
4842
+ * - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
4843
+ * confirmed either way (`resolveSessionMembership` never
4844
+ * returned `null`), and every CONFIRMED descendant's status read
4845
+ * succeeded and is not ongoing (includes "no descendant session
4846
+ * exists at all" — e.g. a plain, non-`task` tool call).
4847
+ * - `null` → INDETERMINATE: `listSessions` failed, OR at least one
4848
+ * candidate's MEMBERSHIP could not be confirmed
4849
+ * (`resolveSessionMembership` returned `null` — a fetch failure
4850
+ * or pathological chain partway through the parent walk), OR at
4851
+ * least one CONFIRMED descendant's `isSessionOngoing` read
4852
+ * failed — and no OTHER candidate was already confirmed `true`.
4853
+ * The caller MUST NOT treat `null` the same as `false` here
4854
+ * (unlike the recovery cross-check's contract) — see
4855
+ * `isB2AbandonmentConfirmed`.
4856
+ */
4857
+ async isAnyDescendantSessionOngoing(rootSessionId) {
4858
+ const sessions = await listSessions(this.port);
4859
+ if (!sessions) {
4860
+ this.log({
4861
+ level: "warn",
4862
+ message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
4863
+ });
4864
+ return null;
4865
+ }
4866
+ let indeterminate = false;
4867
+ for (const candidate of sessions) {
4868
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
4869
+ const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
4870
+ if (membership === null) {
4871
+ indeterminate = true;
4872
+ continue;
4873
+ }
4874
+ if (membership === false) continue;
4875
+ const ongoing = await isSessionOngoing(this.port, candidate.id);
4876
+ if (ongoing === true) return true;
4877
+ if (ongoing === null) indeterminate = true;
4878
+ }
4879
+ return indeterminate ? null : false;
4880
+ }
3850
4881
  /**
3851
4882
  * Cheap decision-telemetry label for a running row's LAST correlated reply
3852
4883
  * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
@@ -3975,6 +5006,32 @@ var ChannelDriver = class _ChannelDriver {
3975
5006
  }
3976
5007
  return messages;
3977
5008
  }
5009
+ /**
5010
+ * The `opencode_session_id` fragment of a status PATCH body — `{}` when this
5011
+ * conversation has ABANDONED that session (#553). The field is optional
5012
+ * server-side and an absent one leaves the persisted binding untouched, so
5013
+ * omitting it is how a routine status write stops resurrecting it.
5014
+ *
5015
+ * ONLY for writes whose sole cost is a lost deep link. The `processing` notice
5016
+ * degrades to no "View in Evident" link (the reaction swap still fires) and the
5017
+ * turn-failure notice is built from the PATCH's own `error` text with a link off
5018
+ * the persisted row — neither loses content the user came for. `markDone`
5019
+ * deliberately does NOT use this helper: the server fetches the reply text
5020
+ * THROUGH the session id it is given, so suppressing there would replace the
5021
+ * agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
5022
+ * `ensureSession` guard, not this suppression, is what makes the self-heal
5023
+ * stick.
5024
+ */
5025
+ sessionIdBody(sessionId, conversationId, messageId, status) {
5026
+ if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
5027
+ this.log({
5028
+ level: "debug",
5029
+ message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
5030
+ conversation_id: conversationId,
5031
+ message_id: messageId
5032
+ });
5033
+ return {};
5034
+ }
3978
5035
  /**
3979
5036
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
3980
5037
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -4004,7 +5061,7 @@ var ChannelDriver = class _ChannelDriver {
4004
5061
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4005
5062
  body: JSON.stringify({
4006
5063
  status: "processing",
4007
- opencode_session_id: sessionId,
5064
+ ...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
4008
5065
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
4009
5066
  ...title ? { title } : {}
4010
5067
  })
@@ -4053,6 +5110,11 @@ var ChannelDriver = class _ChannelDriver {
4053
5110
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4054
5111
  body: JSON.stringify({
4055
5112
  status: "done",
5113
+ // ALWAYS sent, even for a session this conversation has abandoned
5114
+ // (#553): the server reads the reply text back out of THIS session id
5115
+ // to deliver it. Omitting it would leave the user with "✅ Done!"
5116
+ // instead of the answer — a worse regression than the resurrection it
5117
+ // would prevent, which `ensureSession`'s guard handles anyway.
4056
5118
  opencode_session_id: sessionId,
4057
5119
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
4058
5120
  ...title ? { title } : {},
@@ -4079,11 +5141,21 @@ var ChannelDriver = class _ChannelDriver {
4079
5141
  * exists but is wedged, so the next attempt must get a fresh one
4080
5142
  * instead of reusing it — see WI-1's server-side null-clearing PATCH).
4081
5143
  */
4082
- async markFailed(conversationId, messageId, sessionId, error2, usage) {
5144
+ async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
4083
5145
  const body = { status: "failed" };
4084
- if (sessionId !== void 0) body.opencode_session_id = sessionId;
5146
+ if (sessionId === null) {
5147
+ body.opencode_session_id = null;
5148
+ } else if (sessionId !== void 0) {
5149
+ Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
5150
+ }
4085
5151
  if (error2 !== void 0) body.error = error2;
4086
5152
  if (usage) Object.assign(body, usage);
5153
+ if (failure) {
5154
+ body.failure_kind = failure.kind;
5155
+ body.failure_provider_id = failure.providerId;
5156
+ body.failure_model_id = failure.modelId;
5157
+ body.failure_reason = failure.reason;
5158
+ }
4087
5159
  await this.callWithRetry(
4088
5160
  "marking message as failed",
4089
5161
  () => this.fetchImpl(
@@ -4096,6 +5168,29 @@ var ChannelDriver = class _ChannelDriver {
4096
5168
  )
4097
5169
  );
4098
5170
  }
5171
+ /**
5172
+ * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
5173
+ *
5174
+ * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
5175
+ * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
5176
+ * to the P1-2b zero-provider check — one extra loopback call to
5177
+ * `hasAnyConfiguredProvider`, only reached when the structured classifier
5178
+ * couldn't place it. Fails open (never throws): a fallback probe failure
5179
+ * (`null`/indeterminate) leaves the classification `null`, which produces
5180
+ * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
5181
+ */
5182
+ async classifyModelAuthFailure(messages, userMessageId) {
5183
+ const classified = messageFailure(messages, userMessageId);
5184
+ if (classified != null) return classified;
5185
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
5186
+ const hasProvider = await hasAnyConfiguredProvider(this.port);
5187
+ return applyZeroProviderFallback(
5188
+ classified,
5189
+ hasProvider,
5190
+ reply?.info?.providerID ?? null,
5191
+ reply?.info?.modelID ?? null
5192
+ );
5193
+ }
4099
5194
  /**
4100
5195
  * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
4101
5196
  * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
@@ -4248,10 +5343,16 @@ var ChannelDriver = class _ChannelDriver {
4248
5343
  import chalk5 from "chalk";
4249
5344
  import ora2 from "ora";
4250
5345
  import { select as select2 } from "@inquirer/prompts";
5346
+ var INTERACTIVE_START_TIMEOUT_MS = 3e4;
4251
5347
  async function ensureOpenCodeRunning(ctx) {
4252
5348
  const healthCheck = await checkOpenCodeHealth(ctx.port);
4253
5349
  if (healthCheck.healthy) {
4254
- return { port: ctx.port, process: null, version: healthCheck.version ?? null };
5350
+ return {
5351
+ port: ctx.port,
5352
+ process: null,
5353
+ version: healthCheck.version ?? null,
5354
+ notReadyReason: null
5355
+ };
4255
5356
  }
4256
5357
  const runningInstances = await findHealthyOpenCodeInstances();
4257
5358
  if (runningInstances.length > 0) {
@@ -4272,7 +5373,7 @@ async function ensureOpenCodeRunning(ctx) {
4272
5373
  console.log(chalk5.yellow("Tip: Run with the correct port:"));
4273
5374
  console.log(
4274
5375
  chalk5.dim(
4275
- ` ${getCliName()} run --agent ${ctx.agentId} --port ${runningInstances[0].port}`
5376
+ ` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
4276
5377
  )
4277
5378
  );
4278
5379
  }
@@ -4292,14 +5393,22 @@ async function ensureOpenCodeRunning(ctx) {
4292
5393
  if (!ctx.interactive) {
4293
5394
  ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
4294
5395
  const proc = await startOpenCode(ctx.port);
4295
- const health = await waitForOpenCodeHealth(ctx.port, 3e4);
5396
+ const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
4296
5397
  if (!health.healthy) {
4297
- throw new Error(
4298
- `OpenCode failed to start on port ${ctx.port}. Install with: npm install -g opencode-ai`
4299
- );
5398
+ return {
5399
+ port: ctx.port,
5400
+ process: proc,
5401
+ version: null,
5402
+ notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
5403
+ };
4300
5404
  }
4301
5405
  ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
4302
- return { port: ctx.port, process: proc, version: health.version ?? null };
5406
+ return {
5407
+ port: ctx.port,
5408
+ process: proc,
5409
+ version: health.version ?? null,
5410
+ notReadyReason: null
5411
+ };
4303
5412
  }
4304
5413
  let port = ctx.port;
4305
5414
  if (isPortInUse(port)) {
@@ -4352,15 +5461,15 @@ Port ${port} is already in use.`));
4352
5461
  if (action === "start") {
4353
5462
  const spinner = ora2("Starting OpenCode...").start();
4354
5463
  const proc = await startOpenCode(port);
4355
- const health = await waitForOpenCodeHealth(port, 3e4);
5464
+ const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
4356
5465
  if (!health.healthy) {
4357
5466
  spinner.fail("Failed to start OpenCode");
4358
5467
  throw new Error("OpenCode failed to start");
4359
5468
  }
4360
5469
  spinner.stop();
4361
- return { port, process: proc, version: health.version ?? null };
5470
+ return { port, process: proc, version: health.version ?? null, notReadyReason: null };
4362
5471
  }
4363
- return { port, process: null, version: null };
5472
+ return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
4364
5473
  }
4365
5474
 
4366
5475
  // src/commands/agent-lookup.ts
@@ -4402,19 +5511,21 @@ async function resolveAgentIdFromKey(authHeader) {
4402
5511
  return { agent_id: data.agent_id };
4403
5512
  }
4404
5513
  return {
4405
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
5514
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
4406
5515
  };
4407
5516
  } catch (error2) {
4408
5517
  const message = error2 instanceof Error ? error2.message : "Unknown error";
4409
5518
  return { error: `Failed to resolve runner from key: ${message}` };
4410
5519
  }
4411
5520
  }
5521
+ var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
4412
5522
  async function notifyAgentDisconnected(agentId, authHeader) {
4413
5523
  const apiUrl = getApiUrlConfig();
4414
5524
  try {
4415
5525
  const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
4416
5526
  method: "POST",
4417
- headers: { Authorization: authHeader }
5527
+ headers: { Authorization: authHeader },
5528
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
4418
5529
  });
4419
5530
  if (!response.ok) {
4420
5531
  const serverMessage = await readErrorMessage(response);
@@ -4425,7 +5536,35 @@ async function notifyAgentDisconnected(agentId, authHeader) {
4425
5536
  }
4426
5537
  return { ok: true };
4427
5538
  } catch (error2) {
4428
- return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
5539
+ return { ok: false, error: describeBestEffortError(error2) };
5540
+ }
5541
+ }
5542
+ function describeBestEffortError(error2) {
5543
+ const name = error2?.name;
5544
+ if (name === "TimeoutError" || name === "AbortError") {
5545
+ return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
5546
+ }
5547
+ return error2 instanceof Error ? error2.message : String(error2);
5548
+ }
5549
+ async function reportMicrovmId(agentId, authHeader, microvmId) {
5550
+ try {
5551
+ const apiUrl = getApiUrlConfig();
5552
+ const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
5553
+ method: "POST",
5554
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
5555
+ body: JSON.stringify({ microvm_id: microvmId }),
5556
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
5557
+ });
5558
+ if (!response.ok) {
5559
+ const serverMessage = await readErrorMessage(response);
5560
+ return {
5561
+ ok: false,
5562
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
5563
+ };
5564
+ }
5565
+ return { ok: true };
5566
+ } catch (error2) {
5567
+ return { ok: false, error: describeBestEffortError(error2) };
4429
5568
  }
4430
5569
  }
4431
5570
  async function getAgentInfo(agentId, authHeader) {
@@ -4475,6 +5614,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
4475
5614
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
4476
5615
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4477
5616
  var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
5617
+ var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
4478
5618
  function resolveLogLevel(options) {
4479
5619
  const accepted = Object.keys(LOG_LEVELS);
4480
5620
  const validate = (value, source) => {
@@ -4498,6 +5638,63 @@ function resolveLogLevel(options) {
4498
5638
  }
4499
5639
  return "info";
4500
5640
  }
5641
+ function resolveFileSyncDirectories(raw, homeDir) {
5642
+ const directories = [];
5643
+ for (const entry of raw ?? []) {
5644
+ const trimmed = entry.trim();
5645
+ if (trimmed === "") {
5646
+ throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
5647
+ }
5648
+ const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join2(homeDir, trimmed.slice(2)) : trimmed;
5649
+ if (!isAbsolute2(expanded)) {
5650
+ throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
5651
+ }
5652
+ const normalized = resolvePath(expanded);
5653
+ if (parse(normalized).root === normalized) {
5654
+ throw new Error(
5655
+ `--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
5656
+ );
5657
+ }
5658
+ if (!directories.includes(normalized)) {
5659
+ directories.push(normalized);
5660
+ }
5661
+ }
5662
+ if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
5663
+ throw new Error(
5664
+ `--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
5665
+ );
5666
+ }
5667
+ return directories;
5668
+ }
5669
+ var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
5670
+ var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
5671
+ var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
5672
+ function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
5673
+ const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
5674
+ let raw;
5675
+ let source;
5676
+ if (options.opencodeStartTimeout !== void 0) {
5677
+ raw = options.opencodeStartTimeout;
5678
+ source = "--opencode-start-timeout";
5679
+ } else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
5680
+ raw = env[OPENCODE_START_TIMEOUT_ENV];
5681
+ source = OPENCODE_START_TIMEOUT_ENV;
5682
+ } else {
5683
+ return { timeoutMs: defaultMs, warnings: [] };
5684
+ }
5685
+ const trimmed = raw.trim();
5686
+ const seconds = Number(trimmed);
5687
+ const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
5688
+ if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
5689
+ return {
5690
+ timeoutMs: defaultMs,
5691
+ warnings: [
5692
+ `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`
5693
+ ]
5694
+ };
5695
+ }
5696
+ return { timeoutMs: seconds * 1e3, warnings: [] };
5697
+ }
4501
5698
  function meetsThreshold(state, level) {
4502
5699
  return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
4503
5700
  }
@@ -4519,6 +5716,10 @@ function log2(state, message, level = "info") {
4519
5716
  function logActivity(state, entry) {
4520
5717
  const level = entry.level ?? (entry.type === "error" ? "error" : "info");
4521
5718
  if (!meetsThreshold(state, level)) return;
5719
+ forwardRunnerActivity(
5720
+ { level, message: entry.message, error: entry.error },
5721
+ { agentId: state.agentId, authHeader: state.authHeader }
5722
+ );
4522
5723
  const fullEntry = {
4523
5724
  ...entry,
4524
5725
  level,
@@ -4619,18 +5820,29 @@ async function handleAuthError(state, error2) {
4619
5820
  async function driveChannels(state, driver) {
4620
5821
  let idlePolls = 0;
4621
5822
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
5823
+ let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
4622
5824
  while (state.running) {
4623
5825
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
4624
5826
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
4625
5827
  if (state.interactive) displayStatus(state);
4626
5828
  await state.connection.reconnectPromise;
4627
5829
  }
5830
+ const carriedOverFileSync = driver.fileSyncActivity().inFlight;
5831
+ void driver.syncPendingFiles().catch(
5832
+ (error2) => logActivity(state, {
5833
+ type: "error",
5834
+ error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
5835
+ })
5836
+ );
4628
5837
  try {
4629
5838
  const processed = await driver.drainPending();
4630
5839
  state.messageCount += processed;
4631
5840
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4632
5841
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4633
- if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
5842
+ const appliedFiles = driver.fileSyncActivity().appliedFiles;
5843
+ const fileActivity = carriedOverFileSync || appliedFiles !== lastSeenAppliedFiles;
5844
+ lastSeenAppliedFiles = appliedFiles;
5845
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
4634
5846
  idlePolls = 0;
4635
5847
  if (processed > 0 && state.interactive) displayStatus(state);
4636
5848
  } else if (state.idleTimeout !== null) {
@@ -4659,7 +5871,7 @@ async function driveChannels(state, driver) {
4659
5871
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
4660
5872
  if (state.interactive) displayStatus(state);
4661
5873
  }
4662
- await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
5874
+ await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
4663
5875
  if (state.idleTimeout !== null && idlePolls >= 2) {
4664
5876
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
4665
5877
  if (idleMs > state.idleTimeout * 1e3) {
@@ -4762,7 +5974,18 @@ async function notifyOffline(state) {
4762
5974
  if (state.interactive) displayStatus(state);
4763
5975
  }
4764
5976
  }
5977
+ async function timeShutdownPhase(state, durations, name, run2) {
5978
+ const startedAt = Date.now();
5979
+ try {
5980
+ return await run2();
5981
+ } finally {
5982
+ const elapsedMs = Date.now() - startedAt;
5983
+ durations[name] = elapsedMs;
5984
+ log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
5985
+ }
5986
+ }
4765
5987
  async function cleanup(state, opts = {}) {
5988
+ const durations = {};
4766
5989
  state.running = false;
4767
5990
  for (const timer of state.sessionCleanupTimers) {
4768
5991
  clearInterval(timer);
@@ -4776,7 +5999,13 @@ async function cleanup(state, opts = {}) {
4776
5999
  logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4777
6000
  displayStatus(state);
4778
6001
  }
4779
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
6002
+ const driver = state.channelDriver;
6003
+ const settled = await timeShutdownPhase(
6004
+ state,
6005
+ durations,
6006
+ "drain",
6007
+ () => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
6008
+ );
4780
6009
  if (!settled) {
4781
6010
  logActivity(state, {
4782
6011
  type: "info",
@@ -4785,13 +6014,15 @@ async function cleanup(state, opts = {}) {
4785
6014
  if (state.interactive) displayStatus(state);
4786
6015
  }
4787
6016
  }
4788
- await notifyOffline(state);
6017
+ await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
4789
6018
  if (state.connection) {
4790
- state.connection.close();
6019
+ const connection = state.connection;
6020
+ await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
4791
6021
  state.connection = null;
4792
6022
  }
4793
6023
  if (state.opencodeProcess) {
4794
- stopOpenCode(state.opencodeProcess);
6024
+ const opencodeProcess = state.opencodeProcess;
6025
+ await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
4795
6026
  if (state.interactive) {
4796
6027
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
4797
6028
  displayStatus(state);
@@ -4800,12 +6031,15 @@ async function cleanup(state, opts = {}) {
4800
6031
  }
4801
6032
  state.opencodeProcess = null;
4802
6033
  }
6034
+ return durations;
4803
6035
  }
4804
6036
  async function run(options) {
4805
6037
  const interactive = isInteractive(options.json);
4806
6038
  let logLevel;
6039
+ let fileSyncDirectories;
4807
6040
  try {
4808
6041
  logLevel = resolveLogLevel(options);
6042
+ fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir2());
4809
6043
  } catch (error2) {
4810
6044
  const message = error2 instanceof Error ? error2.message : String(error2);
4811
6045
  if (options.json) {
@@ -4840,6 +6074,12 @@ async function run(options) {
4840
6074
  sessionCleanupTimers: [],
4841
6075
  authHeader: ""
4842
6076
  };
6077
+ setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
6078
+ if (fileSyncDirectories.length > 0) {
6079
+ log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
6080
+ } else {
6081
+ log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
6082
+ }
4843
6083
  if (!options.runner && options.agent) {
4844
6084
  telemetry.info(
4845
6085
  EventTypes.DEPRECATED_AGENT_FLAG_USED,
@@ -4863,14 +6103,38 @@ async function run(options) {
4863
6103
  const handleSignal = async () => {
4864
6104
  if (state.shuttingDown) return;
4865
6105
  state.shuttingDown = true;
6106
+ const shutdownStartedAt = Date.now();
4866
6107
  if (state.interactive) {
4867
6108
  logActivity(state, { type: "info", message: "Shutting down..." });
4868
6109
  displayStatus(state);
4869
6110
  } else {
4870
6111
  log2(state, "Shutting down...");
4871
6112
  }
4872
- await cleanup(state, { graceful: true });
4873
- await shutdownTelemetry();
6113
+ const durations = await cleanup(state, { graceful: true });
6114
+ const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
6115
+ await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
6116
+ let timer;
6117
+ const flushed = shutdownTelemetry().then(
6118
+ () => true,
6119
+ (error2) => {
6120
+ log2(
6121
+ state,
6122
+ `Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
6123
+ "warn"
6124
+ );
6125
+ return true;
6126
+ }
6127
+ );
6128
+ const timedOut = new Promise((resolve3) => {
6129
+ timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
6130
+ });
6131
+ if (!await Promise.race([flushed, timedOut])) {
6132
+ log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
6133
+ }
6134
+ clearTimeout(timer);
6135
+ });
6136
+ const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
6137
+ log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
4874
6138
  process.exit(0);
4875
6139
  };
4876
6140
  process.on("SIGINT", handleSignal);
@@ -4984,40 +6248,67 @@ async function run(options) {
4984
6248
  }
4985
6249
  spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
4986
6250
  state.agentName = validation.agent.name;
6251
+ const microvmId = process.env.MICROVM_ID?.trim();
6252
+ if (microvmId) {
6253
+ const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
6254
+ if (reported.ok) {
6255
+ log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
6256
+ } else {
6257
+ const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
6258
+ log2(state, message, "warn");
6259
+ if (state.interactive && !state.json) {
6260
+ logActivity(state, { type: "info", level: "warn", message });
6261
+ }
6262
+ }
6263
+ } else {
6264
+ log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
6265
+ }
6266
+ const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
6267
+ for (const warning2 of opencodeStartTimeoutWarnings) {
6268
+ logActivity(state, { type: "info", level: "warn", message: warning2 });
6269
+ }
4987
6270
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
4988
6271
  try {
4989
6272
  const oc = await ensureOpenCodeRunning({
4990
6273
  port: state.port,
4991
6274
  interactive: state.interactive,
4992
6275
  agentId: state.agentId,
4993
- log: (message) => log2(state, message)
6276
+ log: (message) => log2(state, message),
6277
+ startTimeoutMs: opencodeStartTimeoutMs
4994
6278
  });
4995
6279
  state.port = oc.port;
4996
6280
  state.opencodeProcess = oc.process;
4997
6281
  state.opencodeVersion = oc.version;
4998
- state.opencodeConnected = oc.process !== null || oc.version !== null;
6282
+ state.opencodeConnected = oc.notReadyReason === null;
4999
6283
  const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
5000
6284
  ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
5001
- const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
5002
- if (versionWarning) {
5003
- log2(state, versionWarning, "warn");
5004
- if (state.interactive && !state.json) {
5005
- logActivity(state, { type: "info", level: "warn", message: versionWarning });
6285
+ if (!state.interactive && oc.notReadyReason !== null) {
6286
+ 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}).`;
6287
+ logActivity(state, { type: "info", level: "warn", message });
6288
+ } else {
6289
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
6290
+ if (versionWarning) {
6291
+ log2(state, versionWarning, "warn");
6292
+ if (state.interactive && !state.json) {
6293
+ logActivity(state, { type: "info", level: "warn", message: versionWarning });
6294
+ }
5006
6295
  }
5007
- }
5008
- const noProviderWarning = buildNoProviderWarning(await hasAnyConfiguredProvider(state.port));
5009
- if (noProviderWarning) {
5010
- log2(state, noProviderWarning, "warn");
5011
- if (state.interactive && !state.json) {
5012
- logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
5013
- blank();
5014
- console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
5015
- console.log(
5016
- chalk6.dim(
5017
- `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
5018
- )
5019
- );
5020
- blank();
6296
+ const noProviderWarning = buildNoProviderWarning(
6297
+ await hasAnyConfiguredProvider(state.port)
6298
+ );
6299
+ if (noProviderWarning) {
6300
+ log2(state, noProviderWarning, "warn");
6301
+ if (state.interactive && !state.json) {
6302
+ logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
6303
+ blank();
6304
+ console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
6305
+ console.log(
6306
+ chalk6.dim(
6307
+ `Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
6308
+ )
6309
+ );
6310
+ blank();
6311
+ }
5021
6312
  }
5022
6313
  }
5023
6314
  } catch (error2) {
@@ -5032,6 +6323,10 @@ async function run(options) {
5032
6323
  getAuthHeader: () => state.authHeader,
5033
6324
  conversationFilter: state.conversationFilter,
5034
6325
  stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
6326
+ // #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
6327
+ // REJECTED with `file_sync_disabled` on the ack, not silently ignored.
6328
+ fileSyncDirectories,
6329
+ homeDir: homedir2(),
5035
6330
  log: (entry) => (
5036
6331
  // Thread the driver's real level straight through so `debug`/`warn`
5037
6332
  // survive the sink filter (they no longer collapse to info). `type`
@@ -5058,6 +6353,18 @@ async function run(options) {
5058
6353
  type: "info",
5059
6354
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
5060
6355
  });
6356
+ if (options.tunnelReadyFile) {
6357
+ const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
6358
+ if (marker.ok) {
6359
+ log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
6360
+ } else {
6361
+ log2(
6362
+ state,
6363
+ `Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
6364
+ "error"
6365
+ );
6366
+ }
6367
+ }
5061
6368
  emitAgentConnected(state.agentId, {
5062
6369
  port: state.port,
5063
6370
  cli_version: getCliVersion(),
@@ -5113,6 +6420,12 @@ async function run(options) {
5113
6420
  onDrainPing: () => {
5114
6421
  if (!state.running) return;
5115
6422
  logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
6423
+ void channelDriver.syncPendingFiles().catch(
6424
+ (error2) => logActivity(state, {
6425
+ type: "error",
6426
+ error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
6427
+ })
6428
+ );
5116
6429
  channelDriver.drainPending().then((processed) => {
5117
6430
  if (processed > 0) {
5118
6431
  state.messageCount += processed;
@@ -5196,10 +6509,16 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
5196
6509
  program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
5197
6510
  program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
5198
6511
  program.command("whoami").description("Show the currently logged in user").action(whoami);
5199
- program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Runner ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("--runner [id]", "Alias for --agent (preferred name; wins if both are given)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
6512
+ 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(
6513
+ "-a, --agent [id]",
6514
+ "Deprecated alias for --runner (still supported; --runner wins if both are given)"
6515
+ ).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
5200
6516
  "--log-level <level>",
5201
6517
  "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
5202
- ).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").option(
6518
+ ).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(
6519
+ "--opencode-start-timeout <seconds>",
6520
+ "Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
6521
+ ).option("--json", "Output in JSON format").option(
5203
6522
  "--session-cleanup-max-age <duration>",
5204
6523
  "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
5205
6524
  ).option(
@@ -5208,6 +6527,14 @@ program.command("run").description("Connect to Evident and process messages").op
5208
6527
  ).option(
5209
6528
  "--session-cleanup-interval <duration>",
5210
6529
  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
6530
+ ).option(
6531
+ "--enable-file-sync-to <dir>",
6532
+ "Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
6533
+ (value, previous) => previous.concat([value]),
6534
+ []
6535
+ ).option(
6536
+ "--tunnel-ready-file <path>",
6537
+ "Path to write once the tunnel is connected (set by the MicroVM hooks; unused on a developer machine)"
5211
6538
  ).action(
5212
6539
  (options) => {
5213
6540
  run({
@@ -5220,11 +6547,18 @@ program.command("run").description("Connect to Evident and process messages").op
5220
6547
  verbose: options.verbose,
5221
6548
  conversation: options.conversation,
5222
6549
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
6550
+ // Raw string — validation/precedence is single-sourced in run.ts's
6551
+ // resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
6552
+ opencodeStartTimeout: options.opencodeStartTimeout,
5223
6553
  json: options.json,
5224
6554
  // Raw strings — the resolver in run.ts single-sources parsing (M1).
5225
6555
  sessionCleanupMaxAge: options.sessionCleanupMaxAge,
5226
6556
  sessionCleanupMaxCount: options.sessionCleanupMaxCount,
5227
- sessionCleanupInterval: options.sessionCleanupInterval
6557
+ sessionCleanupInterval: options.sessionCleanupInterval,
6558
+ // Raw values — expansion/validation is single-sourced in run.ts's
6559
+ // resolveFileSyncDirectories.
6560
+ enableFileSyncTo: options.enableFileSyncTo,
6561
+ tunnelReadyFile: options.tunnelReadyFile
5228
6562
  });
5229
6563
  }
5230
6564
  );