@evident-ai/cli 3.1.1-dev.716aea1 → 3.1.1-dev.7972495

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 +417,23 @@ 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("Run `evident login` on a machine with a browser to get a token.");
421
+ console.log("Manage or revoke existing tokens under Settings \u2192 CLI tokens.");
377
422
  blank();
378
423
  process.stdout.write("Paste token: ");
379
- const token = await new Promise((resolve2) => {
424
+ const token = await new Promise((resolve3) => {
380
425
  let data = "";
381
426
  process.stdin.setEncoding("utf8");
382
427
  process.stdin.on("data", (chunk) => {
383
428
  data += chunk;
384
429
  });
385
430
  process.stdin.on("end", () => {
386
- resolve2(data.trim());
431
+ resolve3(data.trim());
387
432
  });
388
433
  if (process.stdin.isTTY) {
389
434
  process.stdin.once("data", (chunk) => {
390
435
  process.stdin.pause();
391
- resolve2(chunk.toString().trim());
436
+ resolve3(chunk.toString().trim());
392
437
  });
393
438
  process.stdin.resume();
394
439
  }
@@ -423,9 +468,22 @@ async function login(options) {
423
468
  }
424
469
 
425
470
  // src/commands/logout.ts
471
+ function describeFailure(failure) {
472
+ if (failure.type === "enumerate") {
473
+ return `could not list stored keychain entries (${failure.error.message})`;
474
+ }
475
+ return `${failure.account} (${failure.error.message})`;
476
+ }
426
477
  async function logout(options = {}) {
427
478
  if (options.all) {
428
- await deleteToken({ all: true });
479
+ const result = await deleteToken({ all: true });
480
+ if (result.failures.length > 0) {
481
+ printError(
482
+ `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.`
483
+ );
484
+ process.exitCode = 1;
485
+ return;
486
+ }
429
487
  printSuccess("Logged out of all endpoints.");
430
488
  return;
431
489
  }
@@ -467,9 +525,9 @@ async function whoami() {
467
525
  }
468
526
 
469
527
  // src/commands/run.ts
528
+ import { homedir as homedir2 } from "os";
529
+ import { isAbsolute as isAbsolute2, join as join2, parse, resolve as resolvePath } from "path";
470
530
  import chalk6 from "chalk";
471
- import ora3 from "ora";
472
- import { select as select3 } from "@inquirer/prompts";
473
531
 
474
532
  // ../../packages/types/src/telemetry/index.ts
475
533
  var TelemetryEventTypes = {
@@ -478,13 +536,20 @@ var TelemetryEventTypes = {
478
536
  AGENT_DISCONNECTED: "agent.disconnected",
479
537
  AGENT_MESSAGE_PROCESSING: "agent.message_processing",
480
538
  AGENT_MESSAGE_DONE: "agent.message_done",
481
- AGENT_MESSAGE_FAILED: "agent.message_failed"
539
+ AGENT_MESSAGE_FAILED: "agent.message_failed",
540
+ // A `warn`/`error` runner-side log line forwarded server-side for
541
+ // observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
542
+ RUNNER_ACTIVITY: "runner.activity"
482
543
  };
483
544
 
484
545
  // ../../packages/types/src/tunnel/index.ts
485
546
  var MAX_FRAME_BYTES = 256 * 1024;
486
547
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
487
548
 
549
+ // ../../packages/types/src/runner-files.ts
550
+ var MAX_FILE_PUSH_BYTES = 64 * 1024;
551
+ var MAX_FILE_SYNC_DIRECTORIES = 16;
552
+
488
553
  // ../../packages/types/src/logging/index.ts
489
554
  var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
555
  function log(level, event, fields) {
@@ -499,6 +564,12 @@ function log(level, event, fields) {
499
564
  );
500
565
  }
501
566
  }
567
+ function errorFields(err) {
568
+ if (err instanceof Error) {
569
+ return { error: err.message, error_name: err.name };
570
+ }
571
+ return { error: String(err) };
572
+ }
502
573
  function stripQuery(url) {
503
574
  try {
504
575
  return new URL(url).pathname;
@@ -508,6 +579,10 @@ function stripQuery(url) {
508
579
  }
509
580
  }
510
581
 
582
+ // src/commands/run.ts
583
+ import ora3 from "ora";
584
+ import { select as select3 } from "@inquirer/prompts";
585
+
511
586
  // src/lib/telemetry.ts
512
587
  var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
513
588
  function getCliVersion() {
@@ -519,6 +594,13 @@ var isShuttingDown = false;
519
594
  var FLUSH_INTERVAL_MS = 5e3;
520
595
  var MAX_BUFFER_SIZE = 50;
521
596
  var FLUSH_TIMEOUT_MS = 3e3;
597
+ var authProvider = null;
598
+ function setTelemetryAuthProvider(provider) {
599
+ authProvider = provider;
600
+ }
601
+ var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
602
+ var lastFlushFailureLoggedAt = 0;
603
+ var suppressedFlushFailureCount = 0;
522
604
  function logEvent(eventType, options = {}) {
523
605
  const event = {
524
606
  event_type: eventType,
@@ -553,9 +635,16 @@ async function flushEvents() {
553
635
  flushTimeout = null;
554
636
  }
555
637
  try {
556
- const credentials2 = await getToken();
557
- if (!credentials2) {
558
- return;
638
+ const providerContext = authProvider?.();
639
+ let authHeader;
640
+ if (providerContext?.authHeader) {
641
+ authHeader = providerContext.authHeader;
642
+ } else {
643
+ const credentials2 = await getToken();
644
+ if (!credentials2) {
645
+ return;
646
+ }
647
+ authHeader = `Bearer ${credentials2.token}`;
559
648
  }
560
649
  const apiUrl = getApiUrlConfig();
561
650
  const controller = new AbortController();
@@ -570,7 +659,7 @@ async function flushEvents() {
570
659
  method: "POST",
571
660
  headers: {
572
661
  "Content-Type": "application/json",
573
- Authorization: `Bearer ${credentials2.token}`
662
+ Authorization: authHeader
574
663
  },
575
664
  body: JSON.stringify(request),
576
665
  signal: controller.signal
@@ -582,8 +671,15 @@ async function flushEvents() {
582
671
  clearTimeout(timeout);
583
672
  }
584
673
  } catch (error2) {
585
- if (process.env.DEBUG) {
586
- console.error("Telemetry error:", error2);
674
+ const now = Date.now();
675
+ if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
676
+ const message = error2 instanceof Error ? error2.message : String(error2);
677
+ const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
678
+ console.error(`Telemetry flush error: ${message}${suffix}`);
679
+ lastFlushFailureLoggedAt = now;
680
+ suppressedFlushFailureCount = 0;
681
+ } else {
682
+ suppressedFlushFailureCount++;
587
683
  }
588
684
  }
589
685
  }
@@ -652,6 +748,69 @@ var EventTypes = {
652
748
  DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
653
749
  };
654
750
 
751
+ // src/lib/runner-activity-telemetry.ts
752
+ var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
753
+ var SEVERITY_BY_LEVEL = {
754
+ warn: "warning",
755
+ error: "error"
756
+ };
757
+ var MAX_MESSAGE_LENGTH = 500;
758
+ var TRUNCATION_MARKER = "\u2026";
759
+ function redact(message) {
760
+ return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
761
+ }
762
+ function truncate(message) {
763
+ if (message.length <= MAX_MESSAGE_LENGTH) return message;
764
+ return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
765
+ }
766
+ var RATE_LIMIT_WINDOW_MS = 6e4;
767
+ var RATE_LIMIT_MAX_EVENTS = 30;
768
+ var windowStartedAt = 0;
769
+ var windowCount = 0;
770
+ var windowDroppedCount = 0;
771
+ function admitUnderRateLimit(now) {
772
+ if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
773
+ if (windowDroppedCount > 0) {
774
+ console.error(
775
+ `[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
776
+ );
777
+ }
778
+ windowStartedAt = now;
779
+ windowCount = 0;
780
+ windowDroppedCount = 0;
781
+ }
782
+ if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
783
+ windowDroppedCount++;
784
+ if (windowDroppedCount === 1) {
785
+ console.error(
786
+ `[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
787
+ );
788
+ }
789
+ return false;
790
+ }
791
+ windowCount++;
792
+ return true;
793
+ }
794
+ function forwardRunnerActivity(entry, context) {
795
+ try {
796
+ if (!FORWARDED_LEVELS.has(entry.level)) return;
797
+ if (!context.agentId || !context.authHeader) return;
798
+ if (!admitUnderRateLimit(Date.now())) return;
799
+ const rawMessage = entry.error ?? entry.message ?? "";
800
+ const message = truncate(redact(rawMessage));
801
+ logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
802
+ severity: SEVERITY_BY_LEVEL[entry.level],
803
+ message,
804
+ metadata: { source: "cli.run" },
805
+ agentId: context.agentId
806
+ });
807
+ } catch (err) {
808
+ console.error(
809
+ `[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
810
+ );
811
+ }
812
+ }
813
+
655
814
  // src/lib/auth.ts
656
815
  async function getAuthCredentials() {
657
816
  const runnerKey = process.env.EVIDENT_RUNNER_KEY;
@@ -719,7 +878,7 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
719
878
  if (health.healthy) {
720
879
  return health;
721
880
  }
722
- await new Promise((resolve2) => setTimeout(resolve2, 1e3));
881
+ await new Promise((resolve3) => setTimeout(resolve3, 1e3));
723
882
  }
724
883
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
725
884
  }
@@ -1267,6 +1426,16 @@ async function buildFileParts(attachments, capable) {
1267
1426
  );
1268
1427
  dataUrl = null;
1269
1428
  }
1429
+ if (dataUrl !== null && typeof dataUrl === "object") {
1430
+ outcomes.push({
1431
+ index: a.index,
1432
+ mime: a.mime,
1433
+ filename: a.filename,
1434
+ status: "failed",
1435
+ reason: "needs_reauth"
1436
+ });
1437
+ continue;
1438
+ }
1270
1439
  if (dataUrl == null) {
1271
1440
  outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
1272
1441
  continue;
@@ -1348,7 +1517,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
1348
1517
  }
1349
1518
  }
1350
1519
  if (attempt < READ_BACK_ATTEMPTS - 1) {
1351
- await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1520
+ await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
1352
1521
  }
1353
1522
  }
1354
1523
  return null;
@@ -1476,6 +1645,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
1476
1645
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1477
1646
  return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1478
1647
  }
1648
+ function isB2AbandonmentConfirmed(params) {
1649
+ return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
1650
+ }
1479
1651
  function messageError(messages, userMessageId) {
1480
1652
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1481
1653
  const error2 = errorOf(reply);
@@ -1489,6 +1661,42 @@ function messageError(messages, userMessageId) {
1489
1661
  }
1490
1662
  return "The agent run failed.";
1491
1663
  }
1664
+ function messageFailure(messages, userMessageId) {
1665
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1666
+ const error2 = errorOf(reply);
1667
+ if (error2 == null || typeof error2 !== "object") return null;
1668
+ const e = error2;
1669
+ const replyProviderId = reply?.info?.providerID ?? null;
1670
+ const replyModelId = reply?.info?.modelID ?? null;
1671
+ if (e.name === "ProviderAuthError") {
1672
+ const data = e.data;
1673
+ const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
1674
+ return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
1675
+ }
1676
+ if (e.name === "APIError") {
1677
+ const data = e.data;
1678
+ const statusCode = data?.statusCode;
1679
+ if (statusCode === 401 || statusCode === 403) {
1680
+ return {
1681
+ kind: "model_auth",
1682
+ providerId: replyProviderId,
1683
+ modelId: replyModelId,
1684
+ reason: "rejected"
1685
+ };
1686
+ }
1687
+ }
1688
+ return null;
1689
+ }
1690
+ function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
1691
+ if (classified != null) return classified;
1692
+ if (hasConfiguredProvider !== false) return null;
1693
+ return {
1694
+ kind: "model_auth",
1695
+ providerId: replyProviderId,
1696
+ modelId: replyModelId,
1697
+ reason: "missing"
1698
+ };
1699
+ }
1492
1700
  function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1493
1701
  if (!messages || messages.length === 0) return false;
1494
1702
  return messages.some(
@@ -1684,10 +1892,11 @@ var StreamForwarder = class {
1684
1892
  * Abort every in-flight stream (e.g. on WebSocket close).
1685
1893
  */
1686
1894
  abortAll() {
1687
- for (const stream of this.inflight.values()) {
1895
+ for (const [sid, stream] of this.inflight.entries()) {
1688
1896
  try {
1689
1897
  stream.abort();
1690
- } catch {
1898
+ } catch (err) {
1899
+ log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
1691
1900
  }
1692
1901
  }
1693
1902
  this.inflight.clear();
@@ -1721,12 +1930,12 @@ var StreamForwarder = class {
1721
1930
  let endBody;
1722
1931
  if (has_body) {
1723
1932
  const chunks = [];
1724
- bodyPromise = new Promise((resolve2) => {
1933
+ bodyPromise = new Promise((resolve3) => {
1725
1934
  pushBody = (buf) => {
1726
1935
  chunks.push(buf);
1727
1936
  };
1728
1937
  endBody = () => {
1729
- resolve2(Buffer.concat(chunks));
1938
+ resolve3(Buffer.concat(chunks));
1730
1939
  };
1731
1940
  });
1732
1941
  }
@@ -1843,7 +2052,7 @@ function connectTunnel(options) {
1843
2052
  } = options;
1844
2053
  const tunnelUrl = getTunnelUrlConfig();
1845
2054
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1846
- return new Promise((resolve2, reject) => {
2055
+ return new Promise((resolve3, reject) => {
1847
2056
  const ws = new WebSocket2(url, {
1848
2057
  headers: {
1849
2058
  Authorization: authHeader
@@ -1898,7 +2107,7 @@ function connectTunnel(options) {
1898
2107
  clearTimeout(connectionTimeout);
1899
2108
  const connectedAgentId = message.agent_id ?? agentId;
1900
2109
  onConnected?.(connectedAgentId);
1901
- resolve2({
2110
+ resolve3({
1902
2111
  ws,
1903
2112
  close: () => ws.close(1e3, "CLI shutdown")
1904
2113
  });
@@ -1960,7 +2169,11 @@ var RunnerConnection = class {
1960
2169
  if (this.connection) {
1961
2170
  try {
1962
2171
  this.connection.close();
1963
- } catch {
2172
+ } catch (err) {
2173
+ log("error", "runner_connection_close_failed", {
2174
+ agent_id: this.resolvedAgentId,
2175
+ ...errorFields(err)
2176
+ });
1964
2177
  }
1965
2178
  this.connection = null;
1966
2179
  }
@@ -2012,6 +2225,416 @@ var RunnerConnection = class {
2012
2225
  }
2013
2226
  };
2014
2227
 
2228
+ // src/lib/tunnel/ready-marker.ts
2229
+ import { writeFileSync } from "fs";
2230
+ function writeTunnelReadyMarker(path, agentId) {
2231
+ try {
2232
+ writeFileSync(path, `${agentId}
2233
+ `);
2234
+ return { ok: true };
2235
+ } catch (error2) {
2236
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2237
+ }
2238
+ }
2239
+
2240
+ // src/lib/channels/driver.ts
2241
+ import { homedir } from "os";
2242
+
2243
+ // src/lib/file-push.ts
2244
+ import { randomUUID } from "crypto";
2245
+ import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
2246
+ import { basename, dirname as dirname2, isAbsolute, join, relative, resolve as resolve2, sep } from "path";
2247
+ var FILE_MODE = 384;
2248
+ var DIRECTORY_MODE = 448;
2249
+ async function writePushedFile(request) {
2250
+ const { requestedPath, content, allowedDirectories, homeDir } = request;
2251
+ const bytes = content.byteLength;
2252
+ if (allowedDirectories.length === 0) {
2253
+ return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
2254
+ path: requestedPath,
2255
+ bytes
2256
+ });
2257
+ }
2258
+ if (bytes > MAX_FILE_PUSH_BYTES) {
2259
+ return refuse(
2260
+ "file_too_large",
2261
+ `File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
2262
+ {
2263
+ path: requestedPath,
2264
+ bytes
2265
+ }
2266
+ );
2267
+ }
2268
+ const candidate = expandAndValidate(requestedPath, homeDir);
2269
+ if (candidate === null) {
2270
+ return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
2271
+ path: requestedPath,
2272
+ bytes
2273
+ });
2274
+ }
2275
+ try {
2276
+ const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
2277
+ dirname2(candidate)
2278
+ );
2279
+ const realTarget = join(existingAncestor, ...missingSegments, basename(candidate));
2280
+ const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
2281
+ if (allowedDirectory === null) {
2282
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2283
+ path: realTarget,
2284
+ bytes
2285
+ });
2286
+ }
2287
+ if (missingSegments.length > 0) {
2288
+ await createMissingDirectories(existingAncestor, missingSegments);
2289
+ const realParent = await realpath(dirname2(realTarget));
2290
+ if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
2291
+ return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
2292
+ path: realTarget,
2293
+ bytes,
2294
+ reason: "parent_changed_after_create"
2295
+ });
2296
+ }
2297
+ }
2298
+ await writeAtomically(realTarget, content);
2299
+ log("info", "file_push_written", { path: realTarget, bytes });
2300
+ return { ok: true, path: realTarget };
2301
+ } catch (err) {
2302
+ const errno = err.code ?? "UNKNOWN";
2303
+ return refuse("write_failed", `The runner could not write the file (${errno}).`, {
2304
+ path: candidate,
2305
+ bytes,
2306
+ errno,
2307
+ ...errorFields(err)
2308
+ });
2309
+ }
2310
+ }
2311
+ function expandAndValidate(requestedPath, homeDir) {
2312
+ if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
2313
+ return null;
2314
+ }
2315
+ const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join(homeDir, requestedPath.slice(2)) : requestedPath;
2316
+ if (expanded.split(/[/\\]/).includes("..")) {
2317
+ return null;
2318
+ }
2319
+ if (!isAbsolute(expanded)) {
2320
+ return null;
2321
+ }
2322
+ const candidate = resolve2(expanded);
2323
+ const name = basename(candidate);
2324
+ return name === "" || name === "." || name === ".." ? null : candidate;
2325
+ }
2326
+ async function resolveNearestExistingAncestor(directory) {
2327
+ const missingSegments = [];
2328
+ let current = directory;
2329
+ for (; ; ) {
2330
+ try {
2331
+ return { existingAncestor: await realpath(current), missingSegments };
2332
+ } catch (err) {
2333
+ const parent = dirname2(current);
2334
+ if (err.code !== "ENOENT" || parent === current) {
2335
+ throw err;
2336
+ }
2337
+ missingSegments.unshift(basename(current));
2338
+ current = parent;
2339
+ }
2340
+ }
2341
+ }
2342
+ async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
2343
+ for (const directory of allowedDirectories) {
2344
+ if (!isAbsolute(directory)) {
2345
+ log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
2346
+ continue;
2347
+ }
2348
+ const realDirectory = await realpathCreatingIfMissing(directory);
2349
+ if (realDirectory !== null && contains(realDirectory, realTarget)) {
2350
+ return realDirectory;
2351
+ }
2352
+ }
2353
+ return null;
2354
+ }
2355
+ async function realpathCreatingIfMissing(directory) {
2356
+ try {
2357
+ return await realpath(directory);
2358
+ } catch (err) {
2359
+ if (err.code !== "ENOENT") {
2360
+ log("warn", "file_push_allowed_directory_skipped", {
2361
+ directory,
2362
+ reason: "unresolvable",
2363
+ ...errorFields(err)
2364
+ });
2365
+ return null;
2366
+ }
2367
+ }
2368
+ try {
2369
+ await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
2370
+ await chmod(directory, DIRECTORY_MODE);
2371
+ return await realpath(directory);
2372
+ } catch (err) {
2373
+ log("warn", "file_push_allowed_directory_skipped", {
2374
+ directory,
2375
+ reason: "create_failed",
2376
+ ...errorFields(err)
2377
+ });
2378
+ return null;
2379
+ }
2380
+ }
2381
+ function contains(realDirectory, realTarget) {
2382
+ const rel = relative(realDirectory, realTarget);
2383
+ return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
2384
+ }
2385
+ async function createMissingDirectories(existingAncestor, missingSegments) {
2386
+ let current = existingAncestor;
2387
+ for (const segment of missingSegments) {
2388
+ current = join(current, segment);
2389
+ await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
2390
+ await chmod(current, DIRECTORY_MODE);
2391
+ }
2392
+ }
2393
+ async function writeAtomically(realTarget, content) {
2394
+ const temporaryPath = join(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
2395
+ let handle;
2396
+ try {
2397
+ handle = await open2(temporaryPath, "wx", FILE_MODE);
2398
+ await handle.writeFile(content);
2399
+ await handle.chmod(FILE_MODE);
2400
+ await handle.close();
2401
+ handle = void 0;
2402
+ await rename(temporaryPath, realTarget);
2403
+ } catch (err) {
2404
+ await discardTemporaryFile(temporaryPath, handle);
2405
+ throw err;
2406
+ }
2407
+ }
2408
+ async function discardTemporaryFile(temporaryPath, handle) {
2409
+ try {
2410
+ await handle?.close();
2411
+ } catch (err) {
2412
+ log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
2413
+ }
2414
+ try {
2415
+ await unlink(temporaryPath);
2416
+ } catch (err) {
2417
+ const errno = err.code;
2418
+ if (errno !== "ENOENT" && errno !== "ENOTDIR") {
2419
+ log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
2420
+ }
2421
+ }
2422
+ }
2423
+ function refuse(code, message, fields) {
2424
+ log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
2425
+ return { ok: false, code, message };
2426
+ }
2427
+
2428
+ // src/lib/runner-file-sync.ts
2429
+ var MAX_ACK_ATTEMPTS = 5;
2430
+ async function syncPendingRunnerFiles(options) {
2431
+ const pending = await listPendingFiles(options);
2432
+ const pendingIds = new Set(pending.map((file) => file.id));
2433
+ for (const id of options.ackFailures.keys()) {
2434
+ if (!pendingIds.has(id)) options.ackFailures.delete(id);
2435
+ }
2436
+ if (pending.length === 0) return 0;
2437
+ options.log({
2438
+ level: "info",
2439
+ message: `Runner file sync: ${pending.length} file(s) queued for this runner`
2440
+ });
2441
+ let applied = 0;
2442
+ for (const file of pending) {
2443
+ if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
2444
+ if (await applyOne(options, file)) applied += 1;
2445
+ }
2446
+ return applied;
2447
+ }
2448
+ async function listPendingFiles(options) {
2449
+ let res;
2450
+ try {
2451
+ res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
2452
+ headers: { Authorization: options.getAuthHeader() }
2453
+ });
2454
+ } catch (err) {
2455
+ options.log({
2456
+ level: "warn",
2457
+ message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
2458
+ });
2459
+ return [];
2460
+ }
2461
+ if (!res.ok) {
2462
+ options.log({
2463
+ level: res.status === 404 ? "debug" : "warn",
2464
+ message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
2465
+ });
2466
+ return [];
2467
+ }
2468
+ let body;
2469
+ try {
2470
+ body = await res.json();
2471
+ } catch (err) {
2472
+ options.log({
2473
+ level: "warn",
2474
+ message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
2475
+ });
2476
+ return [];
2477
+ }
2478
+ if (!Array.isArray(body)) {
2479
+ options.log({
2480
+ level: "warn",
2481
+ message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
2482
+ });
2483
+ return [];
2484
+ }
2485
+ const files = [];
2486
+ for (const entry of body) {
2487
+ const file = asPendingFile(entry);
2488
+ if (file === null) {
2489
+ options.log({
2490
+ level: "warn",
2491
+ message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
2492
+ });
2493
+ continue;
2494
+ }
2495
+ files.push(file);
2496
+ }
2497
+ return files;
2498
+ }
2499
+ function asPendingFile(entry) {
2500
+ if (entry === null || typeof entry !== "object") return null;
2501
+ const { id, path, size } = entry;
2502
+ if (typeof id !== "string" || id === "") return null;
2503
+ if (typeof path !== "string" || path === "") return null;
2504
+ if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
2505
+ return { id, path, size };
2506
+ }
2507
+ async function applyOne(options, file) {
2508
+ const label = `${file.id.slice(0, 8)} (${file.path})`;
2509
+ if (options.allowedDirectories.length === 0) {
2510
+ options.log({
2511
+ level: "warn",
2512
+ message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
2513
+ });
2514
+ await ack(options, file, "rejected", "file_sync_disabled");
2515
+ return false;
2516
+ }
2517
+ if (file.size > MAX_FILE_PUSH_BYTES) {
2518
+ options.log({
2519
+ level: "warn",
2520
+ message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
2521
+ });
2522
+ await ack(options, file, "rejected", "file_too_large");
2523
+ return false;
2524
+ }
2525
+ const download = await downloadContent(options, file, label);
2526
+ if (!download.ok) {
2527
+ if (download.terminal) await ack(options, file, "rejected", download.code);
2528
+ return false;
2529
+ }
2530
+ let outcome;
2531
+ try {
2532
+ outcome = await writePushedFile({
2533
+ requestedPath: file.path,
2534
+ content: download.content,
2535
+ allowedDirectories: options.allowedDirectories,
2536
+ homeDir: options.homeDir
2537
+ });
2538
+ } catch (err) {
2539
+ options.log({
2540
+ level: "error",
2541
+ message: `Runner file ${label} could not be written: ${describe(err)}`
2542
+ });
2543
+ await ack(options, file, "rejected", "write_failed");
2544
+ return false;
2545
+ }
2546
+ if (!outcome.ok) {
2547
+ options.log({
2548
+ level: "warn",
2549
+ message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
2550
+ });
2551
+ await ack(options, file, "rejected", outcome.code);
2552
+ return false;
2553
+ }
2554
+ options.log({
2555
+ level: "info",
2556
+ message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
2557
+ });
2558
+ await ack(options, file, "applied");
2559
+ return true;
2560
+ }
2561
+ function durableDownloadCode(status) {
2562
+ return status === 413 ? "file_too_large" : "write_failed";
2563
+ }
2564
+ async function downloadContent(options, file, label) {
2565
+ try {
2566
+ const res = await options.fetchImpl(
2567
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
2568
+ { headers: { Authorization: options.getAuthHeader() } }
2569
+ );
2570
+ if (!res.ok) {
2571
+ const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
2572
+ if (!terminal) {
2573
+ options.log({
2574
+ level: "warn",
2575
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
2576
+ });
2577
+ return { ok: false, terminal: false };
2578
+ }
2579
+ const code = durableDownloadCode(res.status);
2580
+ options.log({
2581
+ level: "error",
2582
+ message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
2583
+ });
2584
+ return { ok: false, terminal: true, code };
2585
+ }
2586
+ return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
2587
+ } catch (err) {
2588
+ options.log({
2589
+ level: "warn",
2590
+ message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
2591
+ });
2592
+ return { ok: false, terminal: false };
2593
+ }
2594
+ }
2595
+ async function ack(options, file, status, reason) {
2596
+ const outcome = `${status}${reason ? ` (${reason})` : ""}`;
2597
+ try {
2598
+ const res = await options.fetchImpl(
2599
+ `${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
2600
+ {
2601
+ method: "POST",
2602
+ headers: {
2603
+ Authorization: options.getAuthHeader(),
2604
+ "Content-Type": "application/json"
2605
+ },
2606
+ body: JSON.stringify(reason ? { status, reason } : { status })
2607
+ }
2608
+ );
2609
+ if (!res.ok) {
2610
+ recordAckFailure(
2611
+ options,
2612
+ file,
2613
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
2614
+ );
2615
+ return;
2616
+ }
2617
+ options.ackFailures.delete(file.id);
2618
+ } catch (err) {
2619
+ recordAckFailure(
2620
+ options,
2621
+ file,
2622
+ `Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
2623
+ );
2624
+ }
2625
+ }
2626
+ function recordAckFailure(options, file, what) {
2627
+ const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
2628
+ options.ackFailures.set(file.id, attempts);
2629
+ options.log({
2630
+ level: "error",
2631
+ message: attempts >= MAX_ACK_ATTEMPTS ? `${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})`
2632
+ });
2633
+ }
2634
+ function describe(err) {
2635
+ return err instanceof Error ? err.message : String(err);
2636
+ }
2637
+
2015
2638
  // src/lib/channels/driver.ts
2016
2639
  function messageIdOf(m) {
2017
2640
  if (!m || typeof m !== "object") return void 0;
@@ -2040,7 +2663,10 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
2040
2663
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
2041
2664
  var HEARTBEAT_MS = 6e4;
2042
2665
  var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
2666
+ var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
2667
+ var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
2043
2668
  var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
2669
+ var MAX_SUPERSEDED_CONVERSATIONS = 256;
2044
2670
  var ChannelAuthError = class extends Error {
2045
2671
  constructor(message) {
2046
2672
  super(message);
@@ -2077,8 +2703,38 @@ var ChannelDriver = class _ChannelDriver {
2077
2703
  pausedMaxWaitMs;
2078
2704
  stuckQueuedMs;
2079
2705
  now;
2706
+ fileSyncDirectories;
2707
+ homeDir;
2080
2708
  /** Cache of conversationId → opencode sessionId. */
2081
2709
  sessions = /* @__PURE__ */ new Map();
2710
+ /**
2711
+ * conversationId → the opencode session this runner has ABANDONED as that
2712
+ * conversation's binding (#553), after a genuine (`sessionExists === true`)
2713
+ * dispatch failure: the session still exists but is wedged, so #485's self-heal
2714
+ * must bind a fresh one.
2715
+ *
2716
+ * Dropping the local binding + clearing the server row is not enough on its own:
2717
+ * a SIBLING message dispatched earlier in the same drain is still in-flight under
2718
+ * the same session, and its watcher's routine status writes carry
2719
+ * `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
2720
+ * and `ensureSession`'s persisted-id fallback then reuses it, defeating the
2721
+ * self-heal. This map makes the runner authoritative instead of racing those
2722
+ * writes: *`ensureSession` never reuses an abandoned id for that conversation,
2723
+ * whatever the server row says* — which holds even when the resurrecting write
2724
+ * is one we deliberately keep (see `markDone`).
2725
+ *
2726
+ * Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
2727
+ * one conversation hold ONE entry (the newest abandonment replaces the older), and
2728
+ * hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
2729
+ * NEWEST abandoned id per conversation is guarded: after a second abandonment a
2730
+ * late sibling of the FIRST session can write that id back and `ensureSession`
2731
+ * will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
2732
+ * NOT dropped when the session's watcher tears down: `markDone` still writes the
2733
+ * abandoned id back (it must, or the reply is lost), so the guard has to outlive
2734
+ * the turn that resurrects it. In-memory only — a restart forgets it, at the same
2735
+ * bounded cost.
2736
+ */
2737
+ supersededSessions = /* @__PURE__ */ new Map();
2082
2738
  /**
2083
2739
  * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
2084
2740
  * longer idempotent (no caller-supplied `messageID`), and its read-back picks
@@ -2201,6 +2857,24 @@ var ChannelDriver = class _ChannelDriver {
2201
2857
  sessionTitles = /* @__PURE__ */ new Map();
2202
2858
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
2203
2859
  draining = false;
2860
+ /**
2861
+ * Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
2862
+ * drain ping don't download, write and ack the same file twice.
2863
+ */
2864
+ syncingFiles = false;
2865
+ /**
2866
+ * Consecutive failed acks per pending file (#559). Lives on the driver so it
2867
+ * survives across drains — without it, a file whose ack keeps failing is
2868
+ * re-downloaded and re-written every ~2s until the server expires it.
2869
+ */
2870
+ fileAckFailures = /* @__PURE__ */ new Map();
2871
+ /**
2872
+ * Monotonic count of files this runner has pulled and written (#559). Only
2873
+ * ever increases, so `run.ts` detects work by comparing it against the value
2874
+ * it saw on the previous cycle — including work that landed mid-sleep, the
2875
+ * same trick `lastProxiedActivityAt` uses.
2876
+ */
2877
+ appliedFileCount = 0;
2204
2878
  /**
2205
2879
  * The currently-executing `drainPending()` promise, or null when idle. Lets a
2206
2880
  * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
@@ -2231,6 +2905,8 @@ var ChannelDriver = class _ChannelDriver {
2231
2905
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2232
2906
  this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2233
2907
  this.now = config2.now ?? (() => Date.now());
2908
+ this.fileSyncDirectories = config2.fileSyncDirectories ?? [];
2909
+ this.homeDir = config2.homeDir ?? homedir();
2234
2910
  }
2235
2911
  /** The IPv4-loopback base URL for the local `opencode serve`. */
2236
2912
  get opencodeBase() {
@@ -2258,6 +2934,47 @@ var ChannelDriver = class _ChannelDriver {
2258
2934
  );
2259
2935
  return run2;
2260
2936
  }
2937
+ /**
2938
+ * Pull-and-apply any files Evident has queued for this runner (#559), riding
2939
+ * the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
2940
+ * and drain ping that call `drainPending()`. There is deliberately no channel,
2941
+ * control frame or poll loop of its own: worst-case latency is one poll tick.
2942
+ *
2943
+ * NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
2944
+ * cost a conversation turn. Failures are logged and either acked as a terminal
2945
+ * outcome or left pending for the next drain (see `runner-file-sync.ts`).
2946
+ *
2947
+ * Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
2948
+ *
2949
+ * @returns the number of files written to disk.
2950
+ */
2951
+ async syncPendingFiles() {
2952
+ if (this.stopped) return 0;
2953
+ if (this.syncingFiles) return 0;
2954
+ this.syncingFiles = true;
2955
+ try {
2956
+ const applied = await syncPendingRunnerFiles({
2957
+ agentId: this.agentId,
2958
+ apiUrl: this.apiUrl,
2959
+ getAuthHeader: this.getAuthHeader,
2960
+ fetchImpl: this.fetchImpl,
2961
+ allowedDirectories: this.fileSyncDirectories,
2962
+ homeDir: this.homeDir,
2963
+ ackFailures: this.fileAckFailures,
2964
+ log: this.log
2965
+ });
2966
+ this.appliedFileCount += applied;
2967
+ return applied;
2968
+ } catch (err) {
2969
+ this.log({
2970
+ level: "error",
2971
+ message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
2972
+ });
2973
+ return 0;
2974
+ } finally {
2975
+ this.syncingFiles = false;
2976
+ }
2977
+ }
2261
2978
  async runDrain() {
2262
2979
  let dispatched = 0;
2263
2980
  try {
@@ -2291,6 +3008,28 @@ var ChannelDriver = class _ChannelDriver {
2291
3008
  }
2292
3009
  return false;
2293
3010
  }
3011
+ /**
3012
+ * File-pull work, for `run.ts`'s idle accounting (#559).
3013
+ *
3014
+ * Pulling a file is real work that `drainPending()` knows nothing about, so
3015
+ * without this a near-idle runner counts a credential pull as an empty tick
3016
+ * and `--idle-timeout` can `process.exit` mid-pull — leaving a
3017
+ * `.evident-push-*.tmp` behind — or immediately after the write, before the
3018
+ * browser has run the authorize/callback that activates it (the user then sees
3019
+ * `saved_not_activated` for a runner that was fine).
3020
+ *
3021
+ * Two signals because one cannot cover both cases: `inFlight` is the pull
3022
+ * happening RIGHT NOW (it may outlive the tick that started it), and
3023
+ * `appliedFiles` is monotonic so a pull that started AND finished between two
3024
+ * idle checks still shows up as an advance.
3025
+ *
3026
+ * CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
3027
+ * the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
3028
+ * samples afterwards reads `true` every single cycle and can never idle out.
3029
+ */
3030
+ fileSyncActivity() {
3031
+ return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
3032
+ }
2294
3033
  /**
2295
3034
  * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2296
3035
  * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
@@ -2352,7 +3091,7 @@ var ChannelDriver = class _ChannelDriver {
2352
3091
  await this.sleep(step);
2353
3092
  }
2354
3093
  }
2355
- while (this.hasInFlightWatchers()) {
3094
+ while (this.hasInFlightWatchers() || this.syncingFiles) {
2356
3095
  if (this.now() >= deadline) return false;
2357
3096
  await this.sleep(step);
2358
3097
  }
@@ -2387,10 +3126,15 @@ var ChannelDriver = class _ChannelDriver {
2387
3126
  * @returns the count of messages NEWLY dispatched (not already in-flight).
2388
3127
  */
2389
3128
  async processConversation(conv) {
2390
- const sessionId = await this.ensureSession(conv);
3129
+ const { sessionId, refusedSessionId } = await this.ensureSession(conv);
2391
3130
  const messages = await this.getPendingMessages(conv.id);
2392
3131
  let dispatched = 0;
2393
3132
  let skippedAlreadyDispatched = 0;
3133
+ if (refusedSessionId && messages.length > 0) {
3134
+ void this.postSignal(conv.id, messages[0].id, "session_superseded", {
3135
+ superseded_session_id: refusedSessionId
3136
+ });
3137
+ }
2394
3138
  for (const message of messages) {
2395
3139
  if (this.stopped) break;
2396
3140
  if (this.dispatched.has(message.id)) {
@@ -2417,7 +3161,8 @@ var ChannelDriver = class _ChannelDriver {
2417
3161
  } catch (err) {
2418
3162
  if (err instanceof ChannelAuthError) throw err;
2419
3163
  this.dispatched.delete(message.id);
2420
- if (await sessionExists(this.port, sessionId) === false) {
3164
+ const exists = await sessionExists(this.port, sessionId);
3165
+ if (exists === false) {
2421
3166
  this.sessions.delete(conv.id);
2422
3167
  this.log({
2423
3168
  level: "warn",
@@ -2427,15 +3172,39 @@ var ChannelDriver = class _ChannelDriver {
2427
3172
  });
2428
3173
  break;
2429
3174
  }
2430
- await this.markFailed(conv.id, message.id).catch(() => {
3175
+ if (exists === null) {
3176
+ this.log({
3177
+ level: "warn",
3178
+ message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
3179
+ conversation_id: conv.id,
3180
+ message_id: message.id
3181
+ });
3182
+ break;
3183
+ }
3184
+ const errorMessage = err instanceof Error ? err.message : String(err);
3185
+ this.sessions.delete(conv.id);
3186
+ this.supersede(conv.id, sessionId);
3187
+ this.log({
3188
+ level: "warn",
3189
+ 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.`,
3190
+ conversation_id: conv.id,
3191
+ message_id: message.id
3192
+ });
3193
+ await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
3194
+ this.log({
3195
+ level: "warn",
3196
+ message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
3197
+ conversation_id: conv.id,
3198
+ message_id: message.id
3199
+ });
2431
3200
  });
2432
3201
  this.log({
2433
3202
  level: "error",
2434
- message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
3203
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
2435
3204
  conversation_id: conv.id,
2436
3205
  message_id: message.id
2437
3206
  });
2438
- continue;
3207
+ break;
2439
3208
  }
2440
3209
  if (opencodeMessageId === null) {
2441
3210
  this.log({
@@ -2461,8 +3230,42 @@ var ChannelDriver = class _ChannelDriver {
2461
3230
  this.ensureWatcherRunning(sessionId);
2462
3231
  return dispatched;
2463
3232
  }
3233
+ /**
3234
+ * Record that `sessionId` is no longer a valid binding for `conversationId`
3235
+ * (#553). Keyed by conversation and hard-capped, so it cannot grow with the
3236
+ * number of failures — see the `supersededSessions` field doc.
3237
+ */
3238
+ supersede(conversationId, sessionId) {
3239
+ this.supersededSessions.delete(conversationId);
3240
+ this.supersededSessions.set(conversationId, sessionId);
3241
+ while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
3242
+ const oldest = this.supersededSessions.keys().next().value;
3243
+ if (oldest === void 0) return;
3244
+ this.supersededSessions.delete(oldest);
3245
+ }
3246
+ }
3247
+ /** Whether `sessionId` is the session this conversation has abandoned (#553). */
3248
+ isSuperseded(conversationId, sessionId) {
3249
+ return this.supersededSessions.get(conversationId) === sessionId;
3250
+ }
3251
+ /**
3252
+ * Resolve the opencode session to run this conversation's turns in.
3253
+ *
3254
+ * `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
3255
+ * binding was an id this runner had abandoned, so a resurrection genuinely
3256
+ * happened and a fresh session was bound instead. The caller reports it.
3257
+ */
2464
3258
  async ensureSession(conv) {
2465
3259
  const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
3260
+ if (bound && this.isSuperseded(conv.id, bound)) {
3261
+ this.log({
3262
+ level: "warn",
3263
+ 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.`,
3264
+ conversation_id: conv.id
3265
+ });
3266
+ this.sessions.delete(conv.id);
3267
+ return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
3268
+ }
2466
3269
  if (bound) {
2467
3270
  const exists = await sessionExists(this.port, bound);
2468
3271
  if (exists === false) {
@@ -2472,12 +3275,12 @@ var ChannelDriver = class _ChannelDriver {
2472
3275
  conversation_id: conv.id
2473
3276
  });
2474
3277
  this.sessions.delete(conv.id);
2475
- return this.createAndBindSession(conv.id);
3278
+ return { sessionId: await this.createAndBindSession(conv.id) };
2476
3279
  }
2477
3280
  this.sessions.set(conv.id, bound);
2478
- return bound;
3281
+ return { sessionId: bound };
2479
3282
  }
2480
- return this.createAndBindSession(conv.id);
3283
+ return { sessionId: await this.createAndBindSession(conv.id) };
2481
3284
  }
2482
3285
  /**
2483
3286
  * Create a fresh OpenCode session for a conversation, cache the binding, and
@@ -2565,7 +3368,11 @@ var ChannelDriver = class _ChannelDriver {
2565
3368
  * (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
2566
3369
  * (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
2567
3370
  * OMITS that one image and the text turn still sends — NEVER throws the turn.
2568
- * Failures are logged with context (no silent swallow).
3371
+ * A 404 body carrying `{ reason: 'needs_reauth' }` (#547 the server CONFIRMED
3372
+ * a Slack `files:read` scope problem via `files.info`) instead resolves the
3373
+ * `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
3374
+ * user to reconnect Slack instead of a generic "unavailable". Failures are
3375
+ * logged with context (no silent swallow).
2569
3376
  */
2570
3377
  async fetchAttachmentDataUrl(messageId, index, mime) {
2571
3378
  try {
@@ -2574,6 +3381,25 @@ var ChannelDriver = class _ChannelDriver {
2574
3381
  { headers: { Authorization: this.getAuthHeader() } }
2575
3382
  );
2576
3383
  if (!res.ok) {
3384
+ let reason;
3385
+ try {
3386
+ const body = await res.json();
3387
+ if (body && typeof body.reason === "string") reason = body.reason;
3388
+ } catch (parseErr) {
3389
+ this.log({
3390
+ level: "debug",
3391
+ message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index}: error body was not JSON (${parseErr instanceof Error ? parseErr.message : String(parseErr)}) \u2014 treating as a plain failure`,
3392
+ message_id: messageId
3393
+ });
3394
+ }
3395
+ if (reason === "needs_reauth") {
3396
+ this.log({
3397
+ level: "error",
3398
+ message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 server confirmed a Slack reauth/scope problem \u2014 omitting this image (text turn proceeds)`,
3399
+ message_id: messageId
3400
+ });
3401
+ return { needsReauth: true };
3402
+ }
2577
3403
  this.log({
2578
3404
  level: "error",
2579
3405
  message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
@@ -2613,6 +3439,9 @@ var ChannelDriver = class _ChannelDriver {
2613
3439
  if (this.attachmentsSkippedSignalled.has(messageId)) return;
2614
3440
  this.attachmentsSkippedSignalled.add(messageId);
2615
3441
  const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
3442
+ const failedReason = outcomes.some(
3443
+ (o) => o.status === "failed" && o.reason === "needs_reauth"
3444
+ ) ? "needs_reauth" : void 0;
2616
3445
  this.log({
2617
3446
  level: "info",
2618
3447
  message: `Message ${messageId.slice(0, 8)}: ${skipped} image(s) skipped (${capabilityUnknown ? "capability was unreadable \u2014 failed open to text-only" : "model not attachment-capable"}), ${failed} image(s) unavailable (deleted-at-source or fetch failure) \u2014 noting to Evident`,
@@ -2622,7 +3451,8 @@ var ChannelDriver = class _ChannelDriver {
2622
3451
  void this.postSignal(conversationId, messageId, "attachments_skipped", {
2623
3452
  skipped,
2624
3453
  failed,
2625
- ...skipped > 0 ? { skipped_reason: skippedReason } : {}
3454
+ ...skipped > 0 ? { skipped_reason: skippedReason } : {},
3455
+ ...failedReason ? { failed_reason: failedReason } : {}
2626
3456
  });
2627
3457
  }
2628
3458
  /** Register a freshly-dispatched message with its session's watcher state. */
@@ -2653,12 +3483,17 @@ var ChannelDriver = class _ChannelDriver {
2653
3483
  stuckReported: false,
2654
3484
  lastAliveAt: 0,
2655
3485
  aliveInFlight: false,
3486
+ titleSynced: false,
3487
+ titleSyncInFlight: false,
2656
3488
  awaitingHumanLatched: false,
2657
3489
  pausedOnQuestion: false,
2658
3490
  pausedOnPermission: false,
2659
3491
  pausedClearConfirmed: false,
2660
3492
  pausedInFlight: false,
2661
- deliveryDeadlineAnchored: false
3493
+ deliveryDeadlineAnchored: false,
3494
+ b2PinnedSinceMs: 0,
3495
+ b2LastDescendantCheckMs: 0,
3496
+ b2AbandonedSignalled: false
2662
3497
  });
2663
3498
  }
2664
3499
  /**
@@ -2726,12 +3561,17 @@ var ChannelDriver = class _ChannelDriver {
2726
3561
  // with no extra `re_adopted` signal needed (folds old WI-6).
2727
3562
  lastAliveAt: 0,
2728
3563
  aliveInFlight: false,
3564
+ titleSynced: false,
3565
+ titleSyncInFlight: false,
2729
3566
  awaitingHumanLatched: false,
2730
3567
  pausedOnQuestion: false,
2731
3568
  pausedOnPermission: false,
2732
3569
  pausedClearConfirmed: false,
2733
3570
  pausedInFlight: false,
2734
- deliveryDeadlineAnchored: false
3571
+ deliveryDeadlineAnchored: false,
3572
+ b2PinnedSinceMs: 0,
3573
+ b2LastDescendantCheckMs: 0,
3574
+ b2AbandonedSignalled: false
2735
3575
  });
2736
3576
  }
2737
3577
  /**
@@ -2893,58 +3733,7 @@ var ChannelDriver = class _ChannelDriver {
2893
3733
  }
2894
3734
  }
2895
3735
  if (state === "done") {
2896
- this.anchorDeliveryDeadline(inFlight);
2897
- if (!inFlight.done) {
2898
- this.log({
2899
- level: "info",
2900
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2901
- conversation_id: conv.id,
2902
- message_id: inFlight.evidentMessageId
2903
- });
2904
- const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2905
- const usage = messageUsage(messages, inFlight.opencodeMessageId);
2906
- try {
2907
- await this.markDone(
2908
- conv.id,
2909
- inFlight.evidentMessageId,
2910
- sessionId,
2911
- inFlight.opencodeMessageId,
2912
- title,
2913
- usage
2914
- );
2915
- } catch (err) {
2916
- if (err instanceof ChannelAuthError) throw err;
2917
- if (err instanceof ChannelTerminalError) {
2918
- this.log({
2919
- level: "warn",
2920
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2921
- conversation_id: conv.id,
2922
- message_id: inFlight.evidentMessageId
2923
- });
2924
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2925
- return;
2926
- }
2927
- if (this.now() >= inFlight.deadline) {
2928
- this.log({
2929
- level: "warn",
2930
- 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)}`,
2931
- conversation_id: conv.id,
2932
- message_id: inFlight.evidentMessageId
2933
- });
2934
- this.removeInFlight(watcher, inFlight.evidentMessageId);
2935
- return;
2936
- }
2937
- this.log({
2938
- level: "warn",
2939
- message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2940
- conversation_id: conv.id,
2941
- message_id: inFlight.evidentMessageId
2942
- });
2943
- return;
2944
- }
2945
- inFlight.done = true;
2946
- }
2947
- this.removeInFlight(watcher, inFlight.evidentMessageId);
3736
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
2948
3737
  return;
2949
3738
  }
2950
3739
  if (state === "failed") {
@@ -2958,8 +3747,16 @@ var ChannelDriver = class _ChannelDriver {
2958
3747
  message_id: inFlight.evidentMessageId
2959
3748
  });
2960
3749
  const usage = messageUsage(messages, inFlight.opencodeMessageId);
3750
+ const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
2961
3751
  try {
2962
- await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2, usage);
3752
+ await this.markFailed(
3753
+ conv.id,
3754
+ inFlight.evidentMessageId,
3755
+ sessionId,
3756
+ error2,
3757
+ usage,
3758
+ failure
3759
+ );
2963
3760
  } catch (err) {
2964
3761
  if (err instanceof ChannelAuthError) throw err;
2965
3762
  if (err instanceof ChannelTerminalError) {
@@ -3004,6 +3801,44 @@ var ChannelDriver = class _ChannelDriver {
3004
3801
  });
3005
3802
  }
3006
3803
  const activelyRunning = state === "running" && !awaitingHuman;
3804
+ const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
3805
+ const snapshotReadable = messages != null && messages.length > 0;
3806
+ if (!pinnedNow) {
3807
+ if (snapshotReadable) {
3808
+ inFlight.b2PinnedSinceMs = 0;
3809
+ inFlight.b2LastDescendantCheckMs = 0;
3810
+ inFlight.b2AbandonedSignalled = false;
3811
+ }
3812
+ } else {
3813
+ if (inFlight.b2AbandonedSignalled) {
3814
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3815
+ return;
3816
+ }
3817
+ if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
3818
+ const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
3819
+ if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
3820
+ inFlight.b2LastDescendantCheckMs = this.now();
3821
+ const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
3822
+ if (isB2AbandonmentConfirmed({
3823
+ pinnedForMs,
3824
+ minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
3825
+ descendantOngoing
3826
+ })) {
3827
+ inFlight.b2AbandonedSignalled = true;
3828
+ this.log({
3829
+ level: "warn",
3830
+ 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`,
3831
+ conversation_id: conv.id,
3832
+ message_id: id
3833
+ });
3834
+ void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
3835
+ watched_for_ms: pinnedForMs
3836
+ });
3837
+ await this.settleMessageDone(sessionId, watcher, inFlight, messages);
3838
+ return;
3839
+ }
3840
+ }
3841
+ }
3007
3842
  if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
3008
3843
  this.log({
3009
3844
  level: "warn",
@@ -3023,6 +3858,18 @@ var ChannelDriver = class _ChannelDriver {
3023
3858
  inFlight.aliveInFlight = false;
3024
3859
  if (ok) inFlight.lastAliveAt = this.now();
3025
3860
  });
3861
+ if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
3862
+ inFlight.titleSyncInFlight = true;
3863
+ void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
3864
+ if (!title) {
3865
+ inFlight.titleSyncInFlight = false;
3866
+ return;
3867
+ }
3868
+ const ok = await this.patchConversationTitle(conv.id, title);
3869
+ inFlight.titleSyncInFlight = false;
3870
+ if (ok) inFlight.titleSynced = true;
3871
+ });
3872
+ }
3026
3873
  }
3027
3874
  if (awaitingHuman) {
3028
3875
  if (!inFlight.awaitingHumanLatched) {
@@ -3060,6 +3907,70 @@ var ChannelDriver = class _ChannelDriver {
3060
3907
  this.removeInFlight(watcher, inFlight.evidentMessageId);
3061
3908
  }
3062
3909
  }
3910
+ /**
3911
+ * Settle a message whose run-state has resolved `'done'` — extracted verbatim
3912
+ * (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
3913
+ * inline `state === 'done'` branch body, so a SECOND caller (the #721
3914
+ * b2-abandonment resolution) can reach the exact same completion behavior
3915
+ * (delivery-deadline anchoring, title resolution, usage extraction, and
3916
+ * `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
3917
+ * and risking the two copies silently drifting apart.
3918
+ */
3919
+ async settleMessageDone(sessionId, watcher, inFlight, messages) {
3920
+ const conv = watcher.conv;
3921
+ this.anchorDeliveryDeadline(inFlight);
3922
+ if (!inFlight.done) {
3923
+ this.log({
3924
+ level: "info",
3925
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
3926
+ conversation_id: conv.id,
3927
+ message_id: inFlight.evidentMessageId
3928
+ });
3929
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
3930
+ const usage = messageUsage(messages, inFlight.opencodeMessageId);
3931
+ try {
3932
+ await this.markDone(
3933
+ conv.id,
3934
+ inFlight.evidentMessageId,
3935
+ sessionId,
3936
+ inFlight.opencodeMessageId,
3937
+ title,
3938
+ usage
3939
+ );
3940
+ } catch (err) {
3941
+ if (err instanceof ChannelAuthError) throw err;
3942
+ if (err instanceof ChannelTerminalError) {
3943
+ this.log({
3944
+ level: "warn",
3945
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
3946
+ conversation_id: conv.id,
3947
+ message_id: inFlight.evidentMessageId
3948
+ });
3949
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3950
+ return;
3951
+ }
3952
+ if (this.now() >= inFlight.deadline) {
3953
+ this.log({
3954
+ level: "warn",
3955
+ 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)}`,
3956
+ conversation_id: conv.id,
3957
+ message_id: inFlight.evidentMessageId
3958
+ });
3959
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3960
+ return;
3961
+ }
3962
+ this.log({
3963
+ level: "warn",
3964
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
3965
+ conversation_id: conv.id,
3966
+ message_id: inFlight.evidentMessageId
3967
+ });
3968
+ return;
3969
+ }
3970
+ inFlight.done = true;
3971
+ }
3972
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
3973
+ }
3063
3974
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
3064
3975
  /**
3065
3976
  * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
@@ -3226,6 +4137,7 @@ var ChannelDriver = class _ChannelDriver {
3226
4137
  if (state === "failed") {
3227
4138
  const error2 = messageError(messages, ocId ?? "") ?? void 0;
3228
4139
  const usage = messageUsage(messages, ocId ?? "");
4140
+ const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
3229
4141
  this.log({
3230
4142
  level: "error",
3231
4143
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -3233,7 +4145,7 @@ var ChannelDriver = class _ChannelDriver {
3233
4145
  message_id: row.id
3234
4146
  });
3235
4147
  try {
3236
- await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
4148
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
3237
4149
  } catch (err) {
3238
4150
  if (err instanceof ChannelAuthError) throw err;
3239
4151
  if (err instanceof ChannelTerminalError) {
@@ -3639,6 +4551,47 @@ var ChannelDriver = class _ChannelDriver {
3639
4551
  }
3640
4552
  return false;
3641
4553
  }
4554
+ /**
4555
+ * Tri-state variant of the upward parentID membership walk (#721), used ONLY
4556
+ * by `isAnyDescendantSessionOngoing`. Walks the SAME cached
4557
+ * `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
4558
+ * `sessionBelongsTo`, which deliberately collapses "confirmed not a
4559
+ * descendant" and "the walk's fetch failed" into the same `false` (safe for
4560
+ * its OTHER callers: interaction attribution and the recovery-path
4561
+ * `isAnyDescendantSessionAlive`, both of which just retry next tick with no
4562
+ * safety consequence either way) — this variant keeps those two outcomes
4563
+ * SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
4564
+ * (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
4565
+ * not ongoing".
4566
+ *
4567
+ * Return contract:
4568
+ * - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
4569
+ * - `false` → the walk reached a definitive, parent-less root session
4570
+ * WITHOUT ever matching `rootSessionId` — `sessionId` is
4571
+ * CONFIRMED NOT a descendant of it.
4572
+ * - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
4573
+ * through the walk (`resolveSessionParent` returned `undefined`),
4574
+ * or the depth cap (32) was hit without a definitive answer (a
4575
+ * pathological/cyclic chain proves nothing either way). NEVER
4576
+ * treat this the same as `false` — see `sessionBelongsTo`'s own
4577
+ * doc comment above for why that collapse is safe THERE but not
4578
+ * here.
4579
+ *
4580
+ * `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
4581
+ * to the live-path descendant check, not a modification of shared code used
4582
+ * by interaction attribution or the recovery path.
4583
+ */
4584
+ async resolveSessionMembership(sessionId, rootSessionId) {
4585
+ let current = sessionId;
4586
+ for (let depth = 0; current && depth < 32; depth++) {
4587
+ if (current === rootSessionId) return true;
4588
+ const parent = await this.resolveSessionParent(current);
4589
+ if (parent === void 0) return null;
4590
+ if (parent === null) return false;
4591
+ current = parent;
4592
+ }
4593
+ return null;
4594
+ }
3642
4595
  /**
3643
4596
  * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3644
4597
  * `null` for a root session (no parent) and `undefined` when opencode is
@@ -3723,6 +4676,54 @@ var ChannelDriver = class _ChannelDriver {
3723
4676
  }
3724
4677
  return null;
3725
4678
  }
4679
+ /**
4680
+ * Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
4681
+ * session title onto the conversation via the PLAIN conversation-update
4682
+ * endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
4683
+ * message-status endpoint `markProcessing`/`markDone` use. Deliberately a
4684
+ * separate, lighter call: it carries no `status`, so it cannot re-trigger the
4685
+ * `processing`/`done` transition side effects (Slack notices, activity-log
4686
+ * rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
4687
+ * ever touches `conversations.title`. That route (`routes/conversations.ts`)
4688
+ * skips a title write matching the stored value, so a redundant call with the
4689
+ * same title is a real no-op — it does not bump `updated_at`, which the
4690
+ * conversation list sorts and paginates on. (Note this is a DIFFERENT guard
4691
+ * from `threads.ts`'s "non-empty AND changed" one, which only covers the
4692
+ * message-status PATCH; the non-empty half is enforced here instead, by
4693
+ * `resolveSessionTitle` never returning an empty/placeholder title.)
4694
+ *
4695
+ * Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
4696
+ * is logged and the title is simply retried on the next heartbeat tick (the
4697
+ * caller only latches `titleSynced` on `true`).
4698
+ */
4699
+ async patchConversationTitle(conversationId, title) {
4700
+ try {
4701
+ const res = await this.fetchImpl(
4702
+ `${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
4703
+ {
4704
+ method: "PATCH",
4705
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
4706
+ body: JSON.stringify({ title })
4707
+ }
4708
+ );
4709
+ if (!res.ok) {
4710
+ this.log({
4711
+ level: "debug",
4712
+ message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
4713
+ conversation_id: conversationId
4714
+ });
4715
+ return false;
4716
+ }
4717
+ return true;
4718
+ } catch (err) {
4719
+ this.log({
4720
+ level: "debug",
4721
+ 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)}`,
4722
+ conversation_id: conversationId
4723
+ });
4724
+ return false;
4725
+ }
4726
+ }
3726
4727
  /**
3727
4728
  * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3728
4729
  * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
@@ -3781,6 +4782,84 @@ var ChannelDriver = class _ChannelDriver {
3781
4782
  }
3782
4783
  return false;
3783
4784
  }
4785
+ /**
4786
+ * LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
4787
+ * sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
4788
+ * in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
4789
+ *
4790
+ * Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
4791
+ * cross-check above): that method judges liveness from the child's OWN
4792
+ * TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
4793
+ * on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
4794
+ * path the local opencode server IS running, so its in-memory status map is
4795
+ * live and authoritative — and per ADR-0047 §4a ("the child has its own entry
4796
+ * [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
4797
+ * ENTIRE turn (including any tool call it is itself executing), not a
4798
+ * per-message transcript snapshot. This sidesteps the "child's own tool is
4799
+ * executing, between its step's completion and the next generation step"
4800
+ * transcript gap that a transcript-based check would need a second,
4801
+ * sustained-window bound to guard against — it is simply not derived from
4802
+ * message timestamps at all.
4803
+ *
4804
+ * Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
4805
+ * status, as the recovery path does per §4a)? Because on the LIVE path the
4806
+ * root session can be shared: a SECOND, unrelated user message can land on the
4807
+ * SAME session (issue #721's own root cause) and keep the root `busy` for a
4808
+ * reason that has nothing to do with THIS message's delegation. A `task`
4809
+ * descendant session is spawned for exactly one delegated turn and never
4810
+ * reused, so its OWN status-map entry is unambiguous evidence about that one
4811
+ * delegation — which the root's status is not.
4812
+ *
4813
+ * Why membership is checked via `resolveSessionMembership`, NOT
4814
+ * `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
4815
+ * `GET /session/:id` fetch failure into "not a descendant", which would
4816
+ * silently drop a genuinely-live candidate from consideration on the one
4817
+ * unlucky tick its membership-walk fetch hiccups (#721).
4818
+ * `resolveSessionMembership` keeps that failure mode as a distinct `null`
4819
+ * (indeterminate) so it is folded into THIS method's own `indeterminate` flag
4820
+ * instead.
4821
+ *
4822
+ * Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
4823
+ * - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
4824
+ * - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
4825
+ * confirmed either way (`resolveSessionMembership` never
4826
+ * returned `null`), and every CONFIRMED descendant's status read
4827
+ * succeeded and is not ongoing (includes "no descendant session
4828
+ * exists at all" — e.g. a plain, non-`task` tool call).
4829
+ * - `null` → INDETERMINATE: `listSessions` failed, OR at least one
4830
+ * candidate's MEMBERSHIP could not be confirmed
4831
+ * (`resolveSessionMembership` returned `null` — a fetch failure
4832
+ * or pathological chain partway through the parent walk), OR at
4833
+ * least one CONFIRMED descendant's `isSessionOngoing` read
4834
+ * failed — and no OTHER candidate was already confirmed `true`.
4835
+ * The caller MUST NOT treat `null` the same as `false` here
4836
+ * (unlike the recovery cross-check's contract) — see
4837
+ * `isB2AbandonmentConfirmed`.
4838
+ */
4839
+ async isAnyDescendantSessionOngoing(rootSessionId) {
4840
+ const sessions = await listSessions(this.port);
4841
+ if (!sessions) {
4842
+ this.log({
4843
+ level: "warn",
4844
+ message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
4845
+ });
4846
+ return null;
4847
+ }
4848
+ let indeterminate = false;
4849
+ for (const candidate of sessions) {
4850
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
4851
+ const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
4852
+ if (membership === null) {
4853
+ indeterminate = true;
4854
+ continue;
4855
+ }
4856
+ if (membership === false) continue;
4857
+ const ongoing = await isSessionOngoing(this.port, candidate.id);
4858
+ if (ongoing === true) return true;
4859
+ if (ongoing === null) indeterminate = true;
4860
+ }
4861
+ return indeterminate ? null : false;
4862
+ }
3784
4863
  /**
3785
4864
  * Cheap decision-telemetry label for a running row's LAST correlated reply
3786
4865
  * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
@@ -3909,6 +4988,32 @@ var ChannelDriver = class _ChannelDriver {
3909
4988
  }
3910
4989
  return messages;
3911
4990
  }
4991
+ /**
4992
+ * The `opencode_session_id` fragment of a status PATCH body — `{}` when this
4993
+ * conversation has ABANDONED that session (#553). The field is optional
4994
+ * server-side and an absent one leaves the persisted binding untouched, so
4995
+ * omitting it is how a routine status write stops resurrecting it.
4996
+ *
4997
+ * ONLY for writes whose sole cost is a lost deep link. The `processing` notice
4998
+ * degrades to no "View in Evident" link (the reaction swap still fires) and the
4999
+ * turn-failure notice is built from the PATCH's own `error` text with a link off
5000
+ * the persisted row — neither loses content the user came for. `markDone`
5001
+ * deliberately does NOT use this helper: the server fetches the reply text
5002
+ * THROUGH the session id it is given, so suppressing there would replace the
5003
+ * agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
5004
+ * `ensureSession` guard, not this suppression, is what makes the self-heal
5005
+ * stick.
5006
+ */
5007
+ sessionIdBody(sessionId, conversationId, messageId, status) {
5008
+ if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
5009
+ this.log({
5010
+ level: "debug",
5011
+ 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)}`,
5012
+ conversation_id: conversationId,
5013
+ message_id: messageId
5014
+ });
5015
+ return {};
5016
+ }
3912
5017
  /**
3913
5018
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
3914
5019
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -3938,7 +5043,7 @@ var ChannelDriver = class _ChannelDriver {
3938
5043
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3939
5044
  body: JSON.stringify({
3940
5045
  status: "processing",
3941
- opencode_session_id: sessionId,
5046
+ ...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
3942
5047
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3943
5048
  ...title ? { title } : {}
3944
5049
  })
@@ -3987,6 +5092,11 @@ var ChannelDriver = class _ChannelDriver {
3987
5092
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3988
5093
  body: JSON.stringify({
3989
5094
  status: "done",
5095
+ // ALWAYS sent, even for a session this conversation has abandoned
5096
+ // (#553): the server reads the reply text back out of THIS session id
5097
+ // to deliver it. Omitting it would leave the user with "✅ Done!"
5098
+ // instead of the answer — a worse regression than the resurrection it
5099
+ // would prevent, which `ensureSession`'s guard handles anyway.
3990
5100
  opencode_session_id: sessionId,
3991
5101
  ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3992
5102
  ...title ? { title } : {},
@@ -4003,16 +5113,31 @@ var ChannelDriver = class _ChannelDriver {
4003
5113
  }
4004
5114
  /**
4005
5115
  * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
4006
- * when provided (issue #182): a bare `markFailed(conv, msg)` sends
4007
- * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
4008
- * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
4009
- * failure reason reaches the channel.
5116
+ * when provided (issue #182). Three states for `sessionId`:
5117
+ * - omitted (`undefined`) → don't send the field, leave the persisted
5118
+ * session untouched (unused today; kept for API symmetry).
5119
+ * - a real id (`string`) → send it, update the persisted session (the
5120
+ * turn-failure call sites: an errored OpenCode turn).
5121
+ * - explicit `null` → send it, CLEAR the persisted session (issue
5122
+ * #485's dispatch-handoff-failure call site: the session id still
5123
+ * exists but is wedged, so the next attempt must get a fresh one
5124
+ * instead of reusing it — see WI-1's server-side null-clearing PATCH).
4010
5125
  */
4011
- async markFailed(conversationId, messageId, sessionId, error2, usage) {
5126
+ async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
4012
5127
  const body = { status: "failed" };
4013
- if (sessionId !== void 0) body.opencode_session_id = sessionId;
5128
+ if (sessionId === null) {
5129
+ body.opencode_session_id = null;
5130
+ } else if (sessionId !== void 0) {
5131
+ Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
5132
+ }
4014
5133
  if (error2 !== void 0) body.error = error2;
4015
5134
  if (usage) Object.assign(body, usage);
5135
+ if (failure) {
5136
+ body.failure_kind = failure.kind;
5137
+ body.failure_provider_id = failure.providerId;
5138
+ body.failure_model_id = failure.modelId;
5139
+ body.failure_reason = failure.reason;
5140
+ }
4016
5141
  await this.callWithRetry(
4017
5142
  "marking message as failed",
4018
5143
  () => this.fetchImpl(
@@ -4025,6 +5150,29 @@ var ChannelDriver = class _ChannelDriver {
4025
5150
  )
4026
5151
  );
4027
5152
  }
5153
+ /**
5154
+ * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
5155
+ *
5156
+ * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
5157
+ * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
5158
+ * to the P1-2b zero-provider check — one extra loopback call to
5159
+ * `hasAnyConfiguredProvider`, only reached when the structured classifier
5160
+ * couldn't place it. Fails open (never throws): a fallback probe failure
5161
+ * (`null`/indeterminate) leaves the classification `null`, which produces
5162
+ * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
5163
+ */
5164
+ async classifyModelAuthFailure(messages, userMessageId) {
5165
+ const classified = messageFailure(messages, userMessageId);
5166
+ if (classified != null) return classified;
5167
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
5168
+ const hasProvider = await hasAnyConfiguredProvider(this.port);
5169
+ return applyZeroProviderFallback(
5170
+ classified,
5171
+ hasProvider,
5172
+ reply?.info?.providerID ?? null,
5173
+ reply?.info?.modelID ?? null
5174
+ );
5175
+ }
4028
5176
  /**
4029
5177
  * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
4030
5178
  * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
@@ -4201,7 +5349,7 @@ async function ensureOpenCodeRunning(ctx) {
4201
5349
  console.log(chalk5.yellow("Tip: Run with the correct port:"));
4202
5350
  console.log(
4203
5351
  chalk5.dim(
4204
- ` ${getCliName()} run --agent ${ctx.agentId} --port ${runningInstances[0].port}`
5352
+ ` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
4205
5353
  )
4206
5354
  );
4207
5355
  }
@@ -4331,19 +5479,21 @@ async function resolveAgentIdFromKey(authHeader) {
4331
5479
  return { agent_id: data.agent_id };
4332
5480
  }
4333
5481
  return {
4334
- error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
5482
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
4335
5483
  };
4336
5484
  } catch (error2) {
4337
5485
  const message = error2 instanceof Error ? error2.message : "Unknown error";
4338
5486
  return { error: `Failed to resolve runner from key: ${message}` };
4339
5487
  }
4340
5488
  }
5489
+ var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
4341
5490
  async function notifyAgentDisconnected(agentId, authHeader) {
4342
5491
  const apiUrl = getApiUrlConfig();
4343
5492
  try {
4344
5493
  const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
4345
5494
  method: "POST",
4346
- headers: { Authorization: authHeader }
5495
+ headers: { Authorization: authHeader },
5496
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
4347
5497
  });
4348
5498
  if (!response.ok) {
4349
5499
  const serverMessage = await readErrorMessage(response);
@@ -4354,7 +5504,35 @@ async function notifyAgentDisconnected(agentId, authHeader) {
4354
5504
  }
4355
5505
  return { ok: true };
4356
5506
  } catch (error2) {
4357
- return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
5507
+ return { ok: false, error: describeBestEffortError(error2) };
5508
+ }
5509
+ }
5510
+ function describeBestEffortError(error2) {
5511
+ const name = error2?.name;
5512
+ if (name === "TimeoutError" || name === "AbortError") {
5513
+ return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
5514
+ }
5515
+ return error2 instanceof Error ? error2.message : String(error2);
5516
+ }
5517
+ async function reportMicrovmId(agentId, authHeader, microvmId) {
5518
+ try {
5519
+ const apiUrl = getApiUrlConfig();
5520
+ const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
5521
+ method: "POST",
5522
+ headers: { Authorization: authHeader, "Content-Type": "application/json" },
5523
+ body: JSON.stringify({ microvm_id: microvmId }),
5524
+ signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
5525
+ });
5526
+ if (!response.ok) {
5527
+ const serverMessage = await readErrorMessage(response);
5528
+ return {
5529
+ ok: false,
5530
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
5531
+ };
5532
+ }
5533
+ return { ok: true };
5534
+ } catch (error2) {
5535
+ return { ok: false, error: describeBestEffortError(error2) };
4358
5536
  }
4359
5537
  }
4360
5538
  async function getAgentInfo(agentId, authHeader) {
@@ -4404,6 +5582,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
4404
5582
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
4405
5583
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4406
5584
  var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
5585
+ var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
4407
5586
  function resolveLogLevel(options) {
4408
5587
  const accepted = Object.keys(LOG_LEVELS);
4409
5588
  const validate = (value, source) => {
@@ -4427,6 +5606,34 @@ function resolveLogLevel(options) {
4427
5606
  }
4428
5607
  return "info";
4429
5608
  }
5609
+ function resolveFileSyncDirectories(raw, homeDir) {
5610
+ const directories = [];
5611
+ for (const entry of raw ?? []) {
5612
+ const trimmed = entry.trim();
5613
+ if (trimmed === "") {
5614
+ throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
5615
+ }
5616
+ const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join2(homeDir, trimmed.slice(2)) : trimmed;
5617
+ if (!isAbsolute2(expanded)) {
5618
+ throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
5619
+ }
5620
+ const normalized = resolvePath(expanded);
5621
+ if (parse(normalized).root === normalized) {
5622
+ throw new Error(
5623
+ `--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
5624
+ );
5625
+ }
5626
+ if (!directories.includes(normalized)) {
5627
+ directories.push(normalized);
5628
+ }
5629
+ }
5630
+ if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
5631
+ throw new Error(
5632
+ `--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
5633
+ );
5634
+ }
5635
+ return directories;
5636
+ }
4430
5637
  function meetsThreshold(state, level) {
4431
5638
  return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
4432
5639
  }
@@ -4448,6 +5655,10 @@ function log2(state, message, level = "info") {
4448
5655
  function logActivity(state, entry) {
4449
5656
  const level = entry.level ?? (entry.type === "error" ? "error" : "info");
4450
5657
  if (!meetsThreshold(state, level)) return;
5658
+ forwardRunnerActivity(
5659
+ { level, message: entry.message, error: entry.error },
5660
+ { agentId: state.agentId, authHeader: state.authHeader }
5661
+ );
4451
5662
  const fullEntry = {
4452
5663
  ...entry,
4453
5664
  level,
@@ -4548,18 +5759,29 @@ async function handleAuthError(state, error2) {
4548
5759
  async function driveChannels(state, driver) {
4549
5760
  let idlePolls = 0;
4550
5761
  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
5762
+ let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
4551
5763
  while (state.running) {
4552
5764
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
4553
5765
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
4554
5766
  if (state.interactive) displayStatus(state);
4555
5767
  await state.connection.reconnectPromise;
4556
5768
  }
5769
+ const carriedOverFileSync = driver.fileSyncActivity().inFlight;
5770
+ void driver.syncPendingFiles().catch(
5771
+ (error2) => logActivity(state, {
5772
+ type: "error",
5773
+ error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
5774
+ })
5775
+ );
4557
5776
  try {
4558
5777
  const processed = await driver.drainPending();
4559
5778
  state.messageCount += processed;
4560
5779
  const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4561
5780
  lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4562
- if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
5781
+ const appliedFiles = driver.fileSyncActivity().appliedFiles;
5782
+ const fileActivity = carriedOverFileSync || appliedFiles !== lastSeenAppliedFiles;
5783
+ lastSeenAppliedFiles = appliedFiles;
5784
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
4563
5785
  idlePolls = 0;
4564
5786
  if (processed > 0 && state.interactive) displayStatus(state);
4565
5787
  } else if (state.idleTimeout !== null) {
@@ -4588,7 +5810,7 @@ async function driveChannels(state, driver) {
4588
5810
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
4589
5811
  if (state.interactive) displayStatus(state);
4590
5812
  }
4591
- await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
5813
+ await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
4592
5814
  if (state.idleTimeout !== null && idlePolls >= 2) {
4593
5815
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
4594
5816
  if (idleMs > state.idleTimeout * 1e3) {
@@ -4691,7 +5913,18 @@ async function notifyOffline(state) {
4691
5913
  if (state.interactive) displayStatus(state);
4692
5914
  }
4693
5915
  }
5916
+ async function timeShutdownPhase(state, durations, name, run2) {
5917
+ const startedAt = Date.now();
5918
+ try {
5919
+ return await run2();
5920
+ } finally {
5921
+ const elapsedMs = Date.now() - startedAt;
5922
+ durations[name] = elapsedMs;
5923
+ log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
5924
+ }
5925
+ }
4694
5926
  async function cleanup(state, opts = {}) {
5927
+ const durations = {};
4695
5928
  state.running = false;
4696
5929
  for (const timer of state.sessionCleanupTimers) {
4697
5930
  clearInterval(timer);
@@ -4705,7 +5938,13 @@ async function cleanup(state, opts = {}) {
4705
5938
  logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4706
5939
  displayStatus(state);
4707
5940
  }
4708
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
5941
+ const driver = state.channelDriver;
5942
+ const settled = await timeShutdownPhase(
5943
+ state,
5944
+ durations,
5945
+ "drain",
5946
+ () => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
5947
+ );
4709
5948
  if (!settled) {
4710
5949
  logActivity(state, {
4711
5950
  type: "info",
@@ -4714,13 +5953,15 @@ async function cleanup(state, opts = {}) {
4714
5953
  if (state.interactive) displayStatus(state);
4715
5954
  }
4716
5955
  }
4717
- await notifyOffline(state);
5956
+ await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
4718
5957
  if (state.connection) {
4719
- state.connection.close();
5958
+ const connection = state.connection;
5959
+ await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
4720
5960
  state.connection = null;
4721
5961
  }
4722
5962
  if (state.opencodeProcess) {
4723
- stopOpenCode(state.opencodeProcess);
5963
+ const opencodeProcess = state.opencodeProcess;
5964
+ await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
4724
5965
  if (state.interactive) {
4725
5966
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
4726
5967
  displayStatus(state);
@@ -4729,12 +5970,15 @@ async function cleanup(state, opts = {}) {
4729
5970
  }
4730
5971
  state.opencodeProcess = null;
4731
5972
  }
5973
+ return durations;
4732
5974
  }
4733
5975
  async function run(options) {
4734
5976
  const interactive = isInteractive(options.json);
4735
5977
  let logLevel;
5978
+ let fileSyncDirectories;
4736
5979
  try {
4737
5980
  logLevel = resolveLogLevel(options);
5981
+ fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir2());
4738
5982
  } catch (error2) {
4739
5983
  const message = error2 instanceof Error ? error2.message : String(error2);
4740
5984
  if (options.json) {
@@ -4769,6 +6013,12 @@ async function run(options) {
4769
6013
  sessionCleanupTimers: [],
4770
6014
  authHeader: ""
4771
6015
  };
6016
+ setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
6017
+ if (fileSyncDirectories.length > 0) {
6018
+ log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
6019
+ } else {
6020
+ log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
6021
+ }
4772
6022
  if (!options.runner && options.agent) {
4773
6023
  telemetry.info(
4774
6024
  EventTypes.DEPRECATED_AGENT_FLAG_USED,
@@ -4792,14 +6042,38 @@ async function run(options) {
4792
6042
  const handleSignal = async () => {
4793
6043
  if (state.shuttingDown) return;
4794
6044
  state.shuttingDown = true;
6045
+ const shutdownStartedAt = Date.now();
4795
6046
  if (state.interactive) {
4796
6047
  logActivity(state, { type: "info", message: "Shutting down..." });
4797
6048
  displayStatus(state);
4798
6049
  } else {
4799
6050
  log2(state, "Shutting down...");
4800
6051
  }
4801
- await cleanup(state, { graceful: true });
4802
- await shutdownTelemetry();
6052
+ const durations = await cleanup(state, { graceful: true });
6053
+ const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
6054
+ await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
6055
+ let timer;
6056
+ const flushed = shutdownTelemetry().then(
6057
+ () => true,
6058
+ (error2) => {
6059
+ log2(
6060
+ state,
6061
+ `Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
6062
+ "warn"
6063
+ );
6064
+ return true;
6065
+ }
6066
+ );
6067
+ const timedOut = new Promise((resolve3) => {
6068
+ timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
6069
+ });
6070
+ if (!await Promise.race([flushed, timedOut])) {
6071
+ log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
6072
+ }
6073
+ clearTimeout(timer);
6074
+ });
6075
+ const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
6076
+ log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
4803
6077
  process.exit(0);
4804
6078
  };
4805
6079
  process.on("SIGINT", handleSignal);
@@ -4913,6 +6187,21 @@ async function run(options) {
4913
6187
  }
4914
6188
  spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
4915
6189
  state.agentName = validation.agent.name;
6190
+ const microvmId = process.env.MICROVM_ID?.trim();
6191
+ if (microvmId) {
6192
+ const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
6193
+ if (reported.ok) {
6194
+ log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
6195
+ } else {
6196
+ const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
6197
+ log2(state, message, "warn");
6198
+ if (state.interactive && !state.json) {
6199
+ logActivity(state, { type: "info", level: "warn", message });
6200
+ }
6201
+ }
6202
+ } else {
6203
+ log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
6204
+ }
4916
6205
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
4917
6206
  try {
4918
6207
  const oc = await ensureOpenCodeRunning({
@@ -4961,6 +6250,10 @@ async function run(options) {
4961
6250
  getAuthHeader: () => state.authHeader,
4962
6251
  conversationFilter: state.conversationFilter,
4963
6252
  stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
6253
+ // #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
6254
+ // REJECTED with `file_sync_disabled` on the ack, not silently ignored.
6255
+ fileSyncDirectories,
6256
+ homeDir: homedir2(),
4964
6257
  log: (entry) => (
4965
6258
  // Thread the driver's real level straight through so `debug`/`warn`
4966
6259
  // survive the sink filter (they no longer collapse to info). `type`
@@ -4987,6 +6280,18 @@ async function run(options) {
4987
6280
  type: "info",
4988
6281
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
4989
6282
  });
6283
+ if (options.tunnelReadyFile) {
6284
+ const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
6285
+ if (marker.ok) {
6286
+ log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
6287
+ } else {
6288
+ log2(
6289
+ state,
6290
+ `Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
6291
+ "error"
6292
+ );
6293
+ }
6294
+ }
4990
6295
  emitAgentConnected(state.agentId, {
4991
6296
  port: state.port,
4992
6297
  cli_version: getCliVersion(),
@@ -5042,6 +6347,12 @@ async function run(options) {
5042
6347
  onDrainPing: () => {
5043
6348
  if (!state.running) return;
5044
6349
  logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
6350
+ void channelDriver.syncPendingFiles().catch(
6351
+ (error2) => logActivity(state, {
6352
+ type: "error",
6353
+ error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
6354
+ })
6355
+ );
5045
6356
  channelDriver.drainPending().then((processed) => {
5046
6357
  if (processed > 0) {
5047
6358
  state.messageCount += processed;
@@ -5125,7 +6436,10 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
5125
6436
  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);
5126
6437
  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 }));
5127
6438
  program.command("whoami").description("Show the currently logged in user").action(whoami);
5128
- 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(
6439
+ 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(
6440
+ "-a, --agent [id]",
6441
+ "Deprecated alias for --runner (still supported; --runner wins if both are given)"
6442
+ ).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
5129
6443
  "--log-level <level>",
5130
6444
  "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
5131
6445
  ).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(
@@ -5137,6 +6451,14 @@ program.command("run").description("Connect to Evident and process messages").op
5137
6451
  ).option(
5138
6452
  "--session-cleanup-interval <duration>",
5139
6453
  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
6454
+ ).option(
6455
+ "--enable-file-sync-to <dir>",
6456
+ "Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
6457
+ (value, previous) => previous.concat([value]),
6458
+ []
6459
+ ).option(
6460
+ "--tunnel-ready-file <path>",
6461
+ "Path to write once the tunnel is connected (set by the MicroVM hooks; unused on a developer machine)"
5140
6462
  ).action(
5141
6463
  (options) => {
5142
6464
  run({
@@ -5153,7 +6475,11 @@ program.command("run").description("Connect to Evident and process messages").op
5153
6475
  // Raw strings — the resolver in run.ts single-sources parsing (M1).
5154
6476
  sessionCleanupMaxAge: options.sessionCleanupMaxAge,
5155
6477
  sessionCleanupMaxCount: options.sessionCleanupMaxCount,
5156
- sessionCleanupInterval: options.sessionCleanupInterval
6478
+ sessionCleanupInterval: options.sessionCleanupInterval,
6479
+ // Raw values — expansion/validation is single-sourced in run.ts's
6480
+ // resolveFileSyncDirectories.
6481
+ enableFileSyncTo: options.enableFileSyncTo,
6482
+ tunnelReadyFile: options.tunnelReadyFile
5157
6483
  });
5158
6484
  }
5159
6485
  );