@evident-ai/cli 3.0.1-dev.fb77f49 → 3.0.1-dev.fbf37f6

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
@@ -1,6 +1,7 @@
1
1
  #!/usr/bin/env node
2
2
 
3
3
  // src/index.ts
4
+ import { createRequire } from "module";
4
5
  import { Command } from "commander";
5
6
 
6
7
  // src/commands/login.ts
@@ -284,14 +285,14 @@ function blank() {
284
285
  console.log();
285
286
  }
286
287
  function waitForEnter(prompt = "Press Enter to continue...") {
287
- return new Promise((resolve) => {
288
+ return new Promise((resolve2) => {
288
289
  process.stdout.write(chalk.dim(prompt));
289
290
  const handler = () => {
290
291
  process.stdin.removeListener("data", handler);
291
292
  process.stdin.setRawMode?.(false);
292
293
  process.stdin.pause();
293
294
  console.log();
294
- resolve();
295
+ resolve2();
295
296
  };
296
297
  if (process.stdin.isTTY) {
297
298
  process.stdin.setRawMode?.(true);
@@ -301,7 +302,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
301
302
  });
302
303
  }
303
304
  function sleep(ms) {
304
- return new Promise((resolve) => setTimeout(resolve, ms));
305
+ return new Promise((resolve2) => setTimeout(resolve2, ms));
305
306
  }
306
307
 
307
308
  // src/commands/login.ts
@@ -375,19 +376,19 @@ async function tokenLogin() {
375
376
  console.log("Visit your Evident dashboard to generate a CLI token.");
376
377
  blank();
377
378
  process.stdout.write("Paste token: ");
378
- const token = await new Promise((resolve) => {
379
+ const token = await new Promise((resolve2) => {
379
380
  let data = "";
380
381
  process.stdin.setEncoding("utf8");
381
382
  process.stdin.on("data", (chunk) => {
382
383
  data += chunk;
383
384
  });
384
385
  process.stdin.on("end", () => {
385
- resolve(data.trim());
386
+ resolve2(data.trim());
386
387
  });
387
388
  if (process.stdin.isTTY) {
388
389
  process.stdin.once("data", (chunk) => {
389
390
  process.stdin.pause();
390
- resolve(chunk.toString().trim());
391
+ resolve2(chunk.toString().trim());
391
392
  });
392
393
  process.stdin.resume();
393
394
  }
@@ -484,8 +485,34 @@ var TelemetryEventTypes = {
484
485
  var MAX_FRAME_BYTES = 256 * 1024;
485
486
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
486
487
 
488
+ // ../../packages/types/src/logging/index.ts
489
+ var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
+ function log(level, event, fields) {
491
+ const method = level === "debug" ? "log" : level;
492
+ try {
493
+ console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
494
+ } catch (err) {
495
+ console.error(
496
+ "[evident] log_serialize_failed",
497
+ event,
498
+ err instanceof Error ? err.message : String(err)
499
+ );
500
+ }
501
+ }
502
+ function stripQuery(url) {
503
+ try {
504
+ return new URL(url).pathname;
505
+ } catch {
506
+ const q = url.indexOf("?");
507
+ return q === -1 ? url : url.slice(0, q);
508
+ }
509
+ }
510
+
487
511
  // src/lib/telemetry.ts
488
- var CLI_VERSION = process.env.npm_package_version || "unknown";
512
+ var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
513
+ function getCliVersion() {
514
+ return CLI_VERSION;
515
+ }
489
516
  var eventBuffer = [];
490
517
  var flushTimeout = null;
491
518
  var isShuttingDown = false;
@@ -679,20 +706,20 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
679
706
  if (health.healthy) {
680
707
  return health;
681
708
  }
682
- await new Promise((resolve) => setTimeout(resolve, 1e3));
709
+ await new Promise((resolve2) => setTimeout(resolve2, 1e3));
683
710
  }
684
711
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
685
712
  }
686
713
 
687
714
  // src/lib/opencode/opencode-version-gate.ts
688
- var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
689
- function isQueueValidatedVersion(version) {
690
- if (!version) return false;
691
- return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version);
692
- }
693
- function buildOpenCodeVersionWarning(version) {
694
- if (isQueueValidatedVersion(version)) return null;
695
- const detected = version ? `v${version}` : "unknown";
715
+ var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
716
+ function isQueueValidatedVersion(version2) {
717
+ if (!version2) return false;
718
+ return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
719
+ }
720
+ function buildOpenCodeVersionWarning(version2) {
721
+ if (isQueueValidatedVersion(version2)) return null;
722
+ const detected = version2 ? `v${version2}` : "unknown";
696
723
  const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
697
724
  return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
698
725
  }
@@ -1009,7 +1036,11 @@ function roleOf(m) {
1009
1036
  }
1010
1037
  function completedOf(m) {
1011
1038
  if (!m || typeof m !== "object") return void 0;
1012
- return m.info?.time?.completed;
1039
+ return m.info?.time?.completed ?? m.time?.completed;
1040
+ }
1041
+ function createdOf(m) {
1042
+ if (!m || typeof m !== "object") return void 0;
1043
+ return m.info?.time?.created ?? m.time?.created;
1013
1044
  }
1014
1045
  function idOf(m) {
1015
1046
  if (!m || typeof m !== "object") return void 0;
@@ -1029,6 +1060,102 @@ function finishOf(m) {
1029
1060
  const infoFinish = m.info?.finish;
1030
1061
  return typeof infoFinish === "string" ? infoFinish : void 0;
1031
1062
  }
1063
+ function errorOf(m) {
1064
+ if (!m || typeof m !== "object") return void 0;
1065
+ return m.info?.error ?? m.error;
1066
+ }
1067
+ function isAssistantInFlight(m) {
1068
+ if (completedOf(m) == null) return true;
1069
+ return finishOf(m) === "tool-calls";
1070
+ }
1071
+ async function getSessionMessages(port, sessionId) {
1072
+ try {
1073
+ const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
1074
+ if (!res.ok) return null;
1075
+ const body = await res.json();
1076
+ return Array.isArray(body) ? body : null;
1077
+ } catch {
1078
+ return null;
1079
+ }
1080
+ }
1081
+ function isSessionActivelyGenerating(messages) {
1082
+ if (!messages || messages.length === 0) return false;
1083
+ const last = messages[messages.length - 1];
1084
+ if (roleOf(last) !== "assistant") return false;
1085
+ return completedOf(last) == null;
1086
+ }
1087
+ function sessionLastActivityMs(session) {
1088
+ const candidates = [
1089
+ session.time?.updated,
1090
+ session.time?.created,
1091
+ session.time_updated,
1092
+ session.time_created,
1093
+ session.updated,
1094
+ session.created
1095
+ ];
1096
+ for (const c of candidates) {
1097
+ if (typeof c === "number" && Number.isFinite(c)) return c;
1098
+ }
1099
+ return null;
1100
+ }
1101
+ async function listSessions(port) {
1102
+ try {
1103
+ const res = await fetch(`${opencodeBase(port)}/session`);
1104
+ if (!res.ok) return null;
1105
+ const body = await res.json();
1106
+ return Array.isArray(body) ? body : null;
1107
+ } catch {
1108
+ return null;
1109
+ }
1110
+ }
1111
+ async function deleteSession(port, id) {
1112
+ try {
1113
+ const res = await fetch(`${opencodeBase(port)}/session/${id}`, { method: "DELETE" });
1114
+ return res.status >= 200 && res.status < 300;
1115
+ } catch {
1116
+ return false;
1117
+ }
1118
+ }
1119
+ async function sessionExists(port, id) {
1120
+ try {
1121
+ const res = await fetch(`${opencodeBase(port)}/session/${id}`);
1122
+ if (res.status >= 200 && res.status < 300) return true;
1123
+ if (res.status === 404) return false;
1124
+ return null;
1125
+ } catch {
1126
+ return null;
1127
+ }
1128
+ }
1129
+ async function getSessionStatuses(port) {
1130
+ try {
1131
+ const res = await fetch(`${opencodeBase(port)}/session/status`);
1132
+ if (!res.ok) {
1133
+ console.error(
1134
+ `[getSessionStatuses] GET /session/status returned HTTP ${res.status} (port ${port})`
1135
+ );
1136
+ return null;
1137
+ }
1138
+ const body = await res.json();
1139
+ if (body == null || typeof body !== "object" || Array.isArray(body)) {
1140
+ console.error(
1141
+ `[getSessionStatuses] GET /session/status body was not a plain object (port ${port})`
1142
+ );
1143
+ return null;
1144
+ }
1145
+ return body;
1146
+ } catch (err) {
1147
+ console.error(
1148
+ `[getSessionStatuses] GET /session/status failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
1149
+ );
1150
+ return null;
1151
+ }
1152
+ }
1153
+ async function isSessionOngoing(port, id) {
1154
+ const map = await getSessionStatuses(port);
1155
+ if (map == null) return null;
1156
+ const entry = map[id];
1157
+ return entry != null && entry.type !== "idle";
1158
+ }
1032
1159
  async function createOpenCodeSession(port, directory) {
1033
1160
  const url = new URL(`${opencodeBase(port)}/session`);
1034
1161
  if (directory && directory.trim()) {
@@ -1046,10 +1173,113 @@ async function createOpenCodeSession(port, directory) {
1046
1173
  const data = await response.json();
1047
1174
  return data.id;
1048
1175
  }
1049
- async function sendPromptAsync(port, sessionId, content, options, messageId) {
1176
+ async function getModelAttachmentCapability(port, model) {
1177
+ try {
1178
+ const res = await fetch(`${opencodeBase(port)}/config/providers`);
1179
+ if (!res.ok) {
1180
+ console.error(
1181
+ `[getModelAttachmentCapability] GET /config/providers returned HTTP ${res.status} (port ${port})`
1182
+ );
1183
+ return null;
1184
+ }
1185
+ const body = await res.json();
1186
+ const providers = Array.isArray(body?.providers) ? body.providers : null;
1187
+ if (!providers) {
1188
+ console.error(
1189
+ `[getModelAttachmentCapability] GET /config/providers body had no providers array (port ${port})`
1190
+ );
1191
+ return null;
1192
+ }
1193
+ const slash = model ? model.indexOf("/") : -1;
1194
+ const providerId = slash > 0 ? model.slice(0, slash) : void 0;
1195
+ let modelId = slash > 0 ? model.slice(slash + 1) : void 0;
1196
+ const defaults2 = body?.default && typeof body.default === "object" ? body.default : void 0;
1197
+ let provider = providerId ? providers.find((p) => p?.id === providerId) : void 0;
1198
+ if (!provider && !providerId) {
1199
+ const defaultProviderIds = defaults2 ? Object.keys(defaults2) : [];
1200
+ if (defaultProviderIds.length === 1) {
1201
+ provider = providers.find((p) => p?.id === defaultProviderIds[0]);
1202
+ }
1203
+ }
1204
+ if (!provider || !provider.models) return null;
1205
+ if (!modelId && defaults2 && typeof provider.id === "string") {
1206
+ const def = defaults2[provider.id];
1207
+ if (typeof def === "string") modelId = def;
1208
+ }
1209
+ if (!modelId) {
1210
+ if (providerId) {
1211
+ const keys = Object.keys(provider.models);
1212
+ if (keys.length === 1) modelId = keys[0];
1213
+ }
1214
+ if (!modelId) return null;
1215
+ }
1216
+ const entry = provider.models[modelId];
1217
+ if (!entry || typeof entry !== "object") return null;
1218
+ return typeof entry.attachment === "boolean" ? entry.attachment : null;
1219
+ } catch (err) {
1220
+ console.error(
1221
+ `[getModelAttachmentCapability] GET /config/providers failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
1222
+ );
1223
+ return null;
1224
+ }
1225
+ }
1226
+ async function buildFileParts(attachments, capable) {
1227
+ const outcomes = [];
1228
+ const parts = [];
1229
+ const capabilityUnknown = capable === null;
1230
+ if (capable !== true) {
1231
+ for (const a of attachments.inputs) {
1232
+ outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "skipped" });
1233
+ }
1234
+ return { parts, outcomes, capabilityUnknown };
1235
+ }
1236
+ for (const a of attachments.inputs) {
1237
+ let dataUrl = null;
1238
+ try {
1239
+ dataUrl = await attachments.fetchDataUrl(a.index);
1240
+ } catch (err) {
1241
+ console.error(
1242
+ `[buildFileParts] attachment ${a.index} (${a.mime}) fetch threw \u2014 omitting: ${err instanceof Error ? err.message : String(err)}`
1243
+ );
1244
+ dataUrl = null;
1245
+ }
1246
+ if (dataUrl == null) {
1247
+ outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
1248
+ continue;
1249
+ }
1250
+ parts.push({
1251
+ type: "file",
1252
+ mime: a.mime,
1253
+ url: dataUrl,
1254
+ ...a.filename ? { filename: a.filename } : {}
1255
+ });
1256
+ outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "sent" });
1257
+ }
1258
+ return { parts, outcomes, capabilityUnknown };
1259
+ }
1260
+ function messageText(m) {
1261
+ if (!m || !Array.isArray(m.parts)) return "";
1262
+ return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
1263
+ }
1264
+ async function sendPromptAsync(port, sessionId, content, options, attachments) {
1265
+ const before = await getSessionMessages(port, sessionId);
1266
+ const knownUserIds = new Set(
1267
+ (before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
1268
+ );
1269
+ const parts = [{ type: "text", text: content }];
1270
+ let pendingOutcomes = null;
1271
+ if (attachments && attachments.inputs.length > 0) {
1272
+ const capable = await getModelAttachmentCapability(port, options?.model);
1273
+ const {
1274
+ parts: fileParts,
1275
+ outcomes,
1276
+ capabilityUnknown
1277
+ } = await buildFileParts(attachments, capable);
1278
+ parts.push(...fileParts);
1279
+ if (attachments.onOutcomes) pendingOutcomes = { outcomes, capabilityUnknown };
1280
+ }
1050
1281
  const body = {
1051
- messageID: messageId,
1052
- parts: [{ type: "text", text: content }]
1282
+ parts
1053
1283
  };
1054
1284
  if (options?.agent) {
1055
1285
  body.agent = options.agent;
@@ -1072,6 +1302,32 @@ async function sendPromptAsync(port, sessionId, content, options, messageId) {
1072
1302
  const text = await res.text().catch(() => "");
1073
1303
  throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1074
1304
  }
1305
+ const READ_BACK_ATTEMPTS = 5;
1306
+ const READ_BACK_DELAY_MS = 150;
1307
+ for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
1308
+ const after = await getSessionMessages(port, sessionId);
1309
+ if (after) {
1310
+ let best = null;
1311
+ for (const m of after) {
1312
+ if (roleOf(m) !== "user") continue;
1313
+ const id = idOf(m);
1314
+ if (typeof id !== "string" || knownUserIds.has(id)) continue;
1315
+ if (messageText(m) !== content) continue;
1316
+ const created = createdOf(m) ?? 0;
1317
+ if (best === null || created > best.created) {
1318
+ best = { id, created };
1319
+ }
1320
+ }
1321
+ if (best) {
1322
+ if (pendingOutcomes && attachments?.onOutcomes) attachments.onOutcomes(pendingOutcomes);
1323
+ return best.id;
1324
+ }
1325
+ }
1326
+ if (attempt < READ_BACK_ATTEMPTS - 1) {
1327
+ await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1328
+ }
1329
+ }
1330
+ return null;
1075
1331
  }
1076
1332
  function findAssistantReplyAfter(messages, userMessageId) {
1077
1333
  if (!messages || messages.length === 0) return null;
@@ -1088,19 +1344,31 @@ function findAssistantReplyAfter(messages, userMessageId) {
1088
1344
  }
1089
1345
  function findLastAssistantReplyFor(messages, userMessageId) {
1090
1346
  if (!messages || messages.length === 0) return null;
1347
+ let lastCorrelated = null;
1348
+ let lastNonErrored = null;
1091
1349
  for (let i = messages.length - 1; i >= 0; i--) {
1092
1350
  const m = messages[i];
1093
- if (roleOf(m) === "assistant" && parentIdOf(m) === userMessageId) return m;
1351
+ if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
1352
+ if (lastCorrelated === null) lastCorrelated = m;
1353
+ if (errorOf(m) == null) {
1354
+ lastNonErrored = m;
1355
+ break;
1356
+ }
1094
1357
  }
1358
+ if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
1095
1359
  const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1096
1360
  if (userIndex === -1) return null;
1097
1361
  let last = null;
1362
+ let lastOk = null;
1098
1363
  for (let i = userIndex + 1; i < messages.length; i++) {
1099
1364
  const role = roleOf(messages[i]);
1100
1365
  if (role === "user") break;
1101
- if (role === "assistant") last = messages[i];
1366
+ if (role === "assistant") {
1367
+ last = messages[i];
1368
+ if (errorOf(messages[i]) == null) lastOk = messages[i];
1369
+ }
1102
1370
  }
1103
- return last;
1371
+ return lastOk ?? last;
1104
1372
  }
1105
1373
  function messageRunState(messages, userMessageId) {
1106
1374
  if (!messages || messages.length === 0) return "unknown";
@@ -1110,13 +1378,136 @@ function messageRunState(messages, userMessageId) {
1110
1378
  if (!reply) return "unknown";
1111
1379
  }
1112
1380
  if (!reply) return "queued";
1113
- if (completedOf(reply) == null) return "running";
1114
- if (finishOf(reply) === "tool-calls") return "running";
1115
- return "done";
1381
+ if (isAssistantInFlight(reply)) return "running";
1382
+ return errorOf(reply) != null ? "failed" : "done";
1383
+ }
1384
+ function isPreamblePinnedRunning(messages, userMessageId) {
1385
+ if (messageRunState(messages, userMessageId) !== "running") return false;
1386
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1387
+ return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1388
+ }
1389
+ function messageError(messages, userMessageId) {
1390
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1391
+ const error2 = errorOf(reply);
1392
+ if (error2 == null) return null;
1393
+ if (typeof error2 === "string") return error2;
1394
+ if (typeof error2 === "object") {
1395
+ const e = error2;
1396
+ const dataMessage = e.data?.message;
1397
+ if (typeof dataMessage === "string") return dataMessage;
1398
+ if (typeof e.message === "string") return e.message;
1399
+ }
1400
+ return "The agent run failed.";
1401
+ }
1402
+ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1403
+ if (!messages || messages.length === 0) return false;
1404
+ return messages.some(
1405
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1406
+ );
1407
+ }
1408
+
1409
+ // src/lib/opencode/session-cleanup.ts
1410
+ var DURATION_UNIT_MS = {
1411
+ s: 1e3,
1412
+ m: 60 * 1e3,
1413
+ h: 60 * 60 * 1e3,
1414
+ d: 24 * 60 * 60 * 1e3
1415
+ };
1416
+ function parseDurationMs(input) {
1417
+ const trimmed = input.trim();
1418
+ const match = /^(\d+)([smhd])$/.exec(trimmed);
1419
+ if (!match) {
1420
+ throw new Error(
1421
+ `Invalid duration "${input}": expected <number><unit> where unit is one of s, m, h, d (e.g. "7d", "24h", "30m", "90s").`
1422
+ );
1423
+ }
1424
+ const value = Number(match[1]);
1425
+ if (value <= 0) {
1426
+ throw new Error(`Invalid duration "${input}": must be a positive value.`);
1427
+ }
1428
+ return value * DURATION_UNIT_MS[match[2]];
1116
1429
  }
1117
- function opencodeMessageIdFor(queuedMessageId) {
1118
- const sanitized = queuedMessageId.replace(/[^a-zA-Z0-9]/g, "_");
1119
- return `msg_${sanitized}`;
1430
+ function selectSessionsToDelete(sessions, opts) {
1431
+ const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
1432
+ if (maxAgeMs === void 0 && maxCount === void 0) return [];
1433
+ const ageEligible = (s) => {
1434
+ if (maxAgeMs === void 0) return false;
1435
+ if (s.lastActivityMs === null) return true;
1436
+ return nowMs - s.lastActivityMs > maxAgeMs;
1437
+ };
1438
+ const countEligibleIds = /* @__PURE__ */ new Set();
1439
+ if (maxCount !== void 0) {
1440
+ const byActivityDesc = [...sessions].sort(
1441
+ (a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
1442
+ );
1443
+ for (const s of byActivityDesc.slice(maxCount)) {
1444
+ countEligibleIds.add(s.id);
1445
+ }
1446
+ }
1447
+ const toDelete = [];
1448
+ for (const s of sessions) {
1449
+ if (protectedIds.has(s.id)) continue;
1450
+ if (ageEligible(s) || countEligibleIds.has(s.id)) {
1451
+ toDelete.push(s.id);
1452
+ }
1453
+ }
1454
+ return toDelete;
1455
+ }
1456
+ var DEFAULT_INTERVAL = "1h";
1457
+ function resolve(flag, envValue, fallback) {
1458
+ return flag ?? envValue ?? fallback;
1459
+ }
1460
+ function parseMaxCount(input) {
1461
+ const trimmed = input.trim();
1462
+ if (!/^\d+$/.test(trimmed)) {
1463
+ throw new Error(`Invalid max-count "${input}": expected a positive integer.`);
1464
+ }
1465
+ const value = Number(trimmed);
1466
+ if (value <= 0) {
1467
+ throw new Error(`Invalid max-count "${input}": must be greater than 0.`);
1468
+ }
1469
+ return value;
1470
+ }
1471
+ function resolveSessionCleanupConfig(flags, env = process.env) {
1472
+ const warnings = [];
1473
+ const maxAgeRaw = resolve(flags.maxAge, env.EVIDENT_SESSION_CLEANUP_MAX_AGE);
1474
+ const maxCountRaw = resolve(flags.maxCount, env.EVIDENT_SESSION_CLEANUP_MAX_COUNT);
1475
+ const intervalRaw = resolve(
1476
+ flags.interval,
1477
+ env.EVIDENT_SESSION_CLEANUP_INTERVAL,
1478
+ DEFAULT_INTERVAL
1479
+ );
1480
+ let maxAgeMs;
1481
+ if (maxAgeRaw !== void 0) {
1482
+ try {
1483
+ maxAgeMs = parseDurationMs(maxAgeRaw);
1484
+ } catch (err) {
1485
+ warnings.push(
1486
+ `Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
1487
+ );
1488
+ }
1489
+ }
1490
+ let maxCount;
1491
+ if (maxCountRaw !== void 0) {
1492
+ try {
1493
+ maxCount = parseMaxCount(maxCountRaw);
1494
+ } catch (err) {
1495
+ warnings.push(
1496
+ `Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
1497
+ );
1498
+ }
1499
+ }
1500
+ let intervalMs;
1501
+ try {
1502
+ intervalMs = parseDurationMs(intervalRaw ?? DEFAULT_INTERVAL);
1503
+ } catch (err) {
1504
+ warnings.push(
1505
+ `Ignoring invalid --session-cleanup-interval, using default ${DEFAULT_INTERVAL}: ${err instanceof Error ? err.message : String(err)}`
1506
+ );
1507
+ intervalMs = parseDurationMs(DEFAULT_INTERVAL);
1508
+ }
1509
+ const enabled = maxAgeMs !== void 0 || maxCount !== void 0;
1510
+ return { enabled, maxAgeMs, maxCount, intervalMs, warnings };
1120
1511
  }
1121
1512
 
1122
1513
  // src/lib/tunnel/connection.ts
@@ -1187,24 +1578,34 @@ var StreamForwarder = class {
1187
1578
  }
1188
1579
  async handleOpen(frame) {
1189
1580
  const { sid, method, path, headers, has_body } = frame;
1581
+ const correlationId = headers?.[CORRELATION_ID_HEADER];
1582
+ const startedAt = Date.now();
1190
1583
  if (path === TUNNEL_DRAIN_PING_PATH) {
1191
1584
  this.callbacks.onDrainPing?.();
1192
1585
  this.send({ type: "head", sid, status: 204, headers: {} });
1193
1586
  this.send({ type: "res_end", sid });
1194
1587
  return;
1195
1588
  }
1589
+ if (process.env.DEBUG) {
1590
+ log("debug", "agent_request", {
1591
+ correlation_id: correlationId,
1592
+ sid,
1593
+ method,
1594
+ path: stripQuery(path)
1595
+ });
1596
+ }
1196
1597
  const ac = new AbortController();
1197
1598
  let bodyPromise;
1198
1599
  let pushBody;
1199
1600
  let endBody;
1200
1601
  if (has_body) {
1201
1602
  const chunks = [];
1202
- bodyPromise = new Promise((resolve) => {
1603
+ bodyPromise = new Promise((resolve2) => {
1203
1604
  pushBody = (buf) => {
1204
1605
  chunks.push(buf);
1205
1606
  };
1206
1607
  endBody = () => {
1207
- resolve(Buffer.concat(chunks));
1608
+ resolve2(Buffer.concat(chunks));
1208
1609
  };
1209
1610
  });
1210
1611
  }
@@ -1239,6 +1640,14 @@ var StreamForwarder = class {
1239
1640
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1240
1641
  });
1241
1642
  this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
1643
+ if (process.env.DEBUG) {
1644
+ log("debug", "agent_response", {
1645
+ correlation_id: correlationId,
1646
+ sid,
1647
+ status: upstream.status,
1648
+ duration_ms: Date.now() - startedAt
1649
+ });
1650
+ }
1242
1651
  this.callbacks.onHead?.(sid, upstream.status);
1243
1652
  try {
1244
1653
  if (upstream.body) {
@@ -1314,7 +1723,7 @@ function connectTunnel(options) {
1314
1723
  } = options;
1315
1724
  const tunnelUrl = getTunnelUrlConfig();
1316
1725
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1317
- return new Promise((resolve, reject) => {
1726
+ return new Promise((resolve2, reject) => {
1318
1727
  const ws = new WebSocket2(url, {
1319
1728
  headers: {
1320
1729
  Authorization: authHeader
@@ -1379,7 +1788,7 @@ function connectTunnel(options) {
1379
1788
  clearTimeout(connectionTimeout);
1380
1789
  const connectedAgentId = message.agent_id ?? agentId;
1381
1790
  onConnected?.(connectedAgentId);
1382
- resolve({
1791
+ resolve2({
1383
1792
  ws,
1384
1793
  close: () => ws.close(1e3, "CLI shutdown")
1385
1794
  });
@@ -1501,6 +1910,17 @@ function messageIdOf(m) {
1501
1910
  const infoId = m.info?.id;
1502
1911
  return typeof infoId === "string" ? infoId : void 0;
1503
1912
  }
1913
+ function cleanImageMime(contentType) {
1914
+ if (!contentType) return null;
1915
+ const media = contentType.split(";")[0].trim().toLowerCase();
1916
+ return /^image\/[a-z0-9.+-]+$/.test(media) ? media : null;
1917
+ }
1918
+ var LOG_LEVELS = {
1919
+ debug: 0,
1920
+ info: 1,
1921
+ warn: 2,
1922
+ error: 3
1923
+ };
1504
1924
  var DEFAULT_RETRY_POLICY = {
1505
1925
  maxAttempts: 6,
1506
1926
  baseDelayMs: 500,
@@ -1508,7 +1928,10 @@ var DEFAULT_RETRY_POLICY = {
1508
1928
  };
1509
1929
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1510
1930
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1511
- var DEFAULT_DISPATCH_CONFIRM_MS = 6e3;
1931
+ var DEFAULT_STUCK_QUEUED_MS = 6e4;
1932
+ var HEARTBEAT_MS = 6e4;
1933
+ var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
1934
+ var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
1512
1935
  var ChannelAuthError = class extends Error {
1513
1936
  constructor(message) {
1514
1937
  super(message);
@@ -1543,10 +1966,18 @@ var ChannelDriver = class {
1543
1966
  sleep;
1544
1967
  pausedPollIntervalMs;
1545
1968
  pausedMaxWaitMs;
1546
- dispatchConfirmMs;
1969
+ stuckQueuedMs;
1547
1970
  now;
1548
1971
  /** Cache of conversationId → opencode sessionId. */
1549
1972
  sessions = /* @__PURE__ */ new Map();
1973
+ /**
1974
+ * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
1975
+ * longer idempotent (no caller-supplied `messageID`), and its read-back picks
1976
+ * "the one new user row" — which is only unambiguous if no OTHER dispatch into
1977
+ * the SAME session interleaves its snapshot→POST→read-back. This map chains each
1978
+ * session's dispatches so they run serially; distinct sessions stay concurrent.
1979
+ */
1980
+ sessionDispatchLocks = /* @__PURE__ */ new Map();
1550
1981
  /**
1551
1982
  * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1552
1983
  * session: one polling loop services all of that session's in-flight messages.
@@ -1563,6 +1994,72 @@ var ChannelDriver = class {
1563
1994
  * a steady-state-poll re-dispatch will not double-run the message.
1564
1995
  */
1565
1996
  dispatched = /* @__PURE__ */ new Set();
1997
+ /**
1998
+ * Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
1999
+ * Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
2000
+ * so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
2001
+ * it is re-adopted and removed when its watcher settles or it is observed off
2002
+ * the processing list.
2003
+ */
2004
+ readopted = /* @__PURE__ */ new Set();
2005
+ /**
2006
+ * "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
2007
+ * re-adopted running/orphan row's watcher hit its `processed_at`-anchored
2008
+ * deadline (or an orphan whose window already elapsed): the still-`processing`
2009
+ * server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
2010
+ * drain until the 15-min cron resets it — spamming new turns.
2011
+ *
2012
+ * CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
2013
+ * does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
2014
+ * in opencode must still be delivered via `markDone` on the next drain — so
2015
+ * `readoptOne` computes `state` FIRST and this set is checked only on the
2016
+ * non-done path. It is cleared once the row leaves the processing list (cron
2017
+ * reset → it drains normally as `pending`), so it can never leak.
2018
+ */
2019
+ dontRedispatch = /* @__PURE__ */ new Set();
2020
+ /**
2021
+ * "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
2022
+ * re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
2023
+ * never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
2024
+ * that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
2025
+ * markDone failure must NOT land here (it must still retry next drain). Separate
2026
+ * from `dontRedispatch` because the two concerns are independent: a row can need
2027
+ * "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
2028
+ * the row leaves the processing list, exactly like `dontRedispatch`.
2029
+ */
2030
+ doneUndeliverable = /* @__PURE__ */ new Set();
2031
+ /**
2032
+ * "Already emitted `readopt_poll_unresolved` for this row" (#229). The b1 /
2033
+ * unreadable-status re-evaluate leaf leaves the row UN-tracked so it is re-read
2034
+ * every ~2s drain until the status map becomes readable — but the server-visible
2035
+ * signal is an OUTCOME, so it must fire at most ONCE per row, not once per drain
2036
+ * (Bugbot "Re-adopt signals flood every drain"). Cleared when the row leaves the
2037
+ * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
2038
+ */
2039
+ readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
2040
+ /**
2041
+ * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
2042
+ * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
2043
+ * is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
2044
+ * persist hasn't landed before tick N+1 re-reads the still-null
2045
+ * `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
2046
+ * A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
2047
+ * short-circuits while it is present, so a null-id row is re-dispatched AT MOST
2048
+ * ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
2049
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
2050
+ * re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
2051
+ * so the NEXT tick may retry exactly once more).
2052
+ */
2053
+ awaitingReadopt = /* @__PURE__ */ new Set();
2054
+ /**
2055
+ * "Already signalled `attachments_skipped` for this Evident message id" (#376).
2056
+ * The in-thread skip note is an OUTCOME, so it must fire AT MOST ONCE per message
2057
+ * — never re-post on a re-dispatch of the same row (`forceReadoptRun` or the
2058
+ * next-tick null-id retry both re-run `sendPromptAsync`, which re-fires
2059
+ * `onOutcomes`). Mirrors `readoptPollUnresolvedSignalled`: a local dedup on the
2060
+ * outcome, not the dispatch. Not cleared (a message is signalled once for life).
2061
+ */
2062
+ attachmentsSkippedSignalled = /* @__PURE__ */ new Set();
1566
2063
  /**
1567
2064
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1568
2065
  * first session creation so drain-created sessions are rooted at the project
@@ -1570,8 +2067,43 @@ var ChannelDriver = class {
1570
2067
  * not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
1571
2068
  */
1572
2069
  opencodeDirectory = void 0;
2070
+ /**
2071
+ * Cache of opencode `sessionId → parentID` (its parent session, or `null` when
2072
+ * the session is a root with no parent). Sub-agents spawned via the `task` tool
2073
+ * run in CHILD sessions whose `parentID` chains up to the Evident-created
2074
+ * (watched) session; we resolve this once per session so a child-session
2075
+ * question/permission can be attributed to the watched session's subtree
2076
+ * (`sessionBelongsTo`) instead of being dropped by an exact-id filter. A missing
2077
+ * entry = not yet resolved; `null` = resolved root (stop walking).
2078
+ */
2079
+ sessionParents = /* @__PURE__ */ new Map();
2080
+ /**
2081
+ * Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
2082
+ * NON-EMPTY name is stored (terminal — a real session name won't later un-name),
2083
+ * so we do NOT re-GET `/session/:id` every tick. A missing entry = not yet
2084
+ * resolved OR resolved-but-still-empty → re-fetch on next need, since OpenCode
2085
+ * names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
2086
+ * the watcher completion path AND the restart-recovery re-adopt path (which has
2087
+ * no watcher) can resolve the title.
2088
+ */
2089
+ sessionTitles = /* @__PURE__ */ new Map();
1573
2090
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
1574
2091
  draining = false;
2092
+ /**
2093
+ * The currently-executing `drainPending()` promise, or null when idle. Lets a
2094
+ * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
2095
+ * is about to dispatch is not missed by the `hasInFlightWatchers()` check (a
2096
+ * drain that entered before `stop()` still registers its watcher).
2097
+ */
2098
+ activeDrain = null;
2099
+ /**
2100
+ * Set by `stop()` on graceful shutdown. Once stopped, `drainPending` no longer
2101
+ * dispatches NEW work (it returns 0 immediately) — but the per-session watcher
2102
+ * loops already running keep going so in-flight turns can finish and deliver
2103
+ * their reply. `run.ts` awaits `waitForInFlight()` before it closes the tunnel
2104
+ * and stops opencode.
2105
+ */
2106
+ stopped = false;
1575
2107
  constructor(config2) {
1576
2108
  this.agentId = config2.agentId;
1577
2109
  this.port = config2.port;
@@ -1585,16 +2117,13 @@ var ChannelDriver = class {
1585
2117
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1586
2118
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1587
2119
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
1588
- this.dispatchConfirmMs = config2.dispatchConfirmMs ?? DEFAULT_DISPATCH_CONFIRM_MS;
2120
+ this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
1589
2121
  this.now = config2.now ?? (() => Date.now());
1590
2122
  }
1591
2123
  /** The IPv4-loopback base URL for the local `opencode serve`. */
1592
2124
  get opencodeBase() {
1593
2125
  return `http://127.0.0.1:${this.port}`;
1594
2126
  }
1595
- // -------------------------------------------------------------------------
1596
- // Public API
1597
- // -------------------------------------------------------------------------
1598
2127
  /**
1599
2128
  * Drain all pending channel conversations once: poll → dispatch → register.
1600
2129
  * Called on tunnel `connected` (WI-CHAN-4) and on each poll tick by `run.ts`.
@@ -1603,8 +2132,21 @@ var ChannelDriver = class {
1603
2132
  * @returns the number of messages NEWLY dispatched to opencode's native queue.
1604
2133
  */
1605
2134
  async drainPending() {
2135
+ if (this.stopped) return 0;
1606
2136
  if (this.draining) return 0;
1607
2137
  this.draining = true;
2138
+ const run2 = this.runDrain();
2139
+ this.activeDrain = run2.then(
2140
+ () => {
2141
+ this.activeDrain = null;
2142
+ },
2143
+ () => {
2144
+ this.activeDrain = null;
2145
+ }
2146
+ );
2147
+ return run2;
2148
+ }
2149
+ async runDrain() {
1608
2150
  let dispatched = 0;
1609
2151
  try {
1610
2152
  const conversations = await this.getPendingConversations();
@@ -1616,8 +2158,10 @@ var ChannelDriver = class {
1616
2158
  });
1617
2159
  }
1618
2160
  for (const conv of conversations) {
2161
+ if (this.stopped) break;
1619
2162
  dispatched += await this.processConversation(conv);
1620
2163
  }
2164
+ await this.readoptProcessing();
1621
2165
  } finally {
1622
2166
  this.draining = false;
1623
2167
  }
@@ -1635,6 +2179,73 @@ var ChannelDriver = class {
1635
2179
  }
1636
2180
  return false;
1637
2181
  }
2182
+ /**
2183
+ * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2184
+ * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
2185
+ * `watchers` entry whose `inFlight` set is non-empty — the same predicate
2186
+ * `hasInFlightWatchers()` uses, lifted to return the ids.
2187
+ *
2188
+ * Deliberately does NOT include `this.sessions` (the permanent, never-pruned
2189
+ * conversation→session cache). Protecting every bound-but-idle session there
2190
+ * would shield nearly every session and defeat cleanup — AND it is unnecessary:
2191
+ * `ensureSession` is self-healing (it recreates a session whose id no longer
2192
+ * exists), so deleting an idle bound session is harmless — the conversation's
2193
+ * next turn transparently rebinds a fresh one. The only thing worth protecting
2194
+ * is a session with a turn ACTIVELY in flight right now: tearing that down
2195
+ * mid-turn would strand the running `prompt_async`. Idle sessions are fair game.
2196
+ */
2197
+ protectedSessionIds() {
2198
+ const ids = /* @__PURE__ */ new Set();
2199
+ for (const [sessionId, watcher] of this.watchers) {
2200
+ if (watcher.inFlight.size > 0) ids.add(sessionId);
2201
+ }
2202
+ return ids;
2203
+ }
2204
+ /**
2205
+ * Begin a graceful stop: stop accepting NEW channel work. Idempotent. After
2206
+ * this, `drainPending()` is a no-op (returns 0), so no new message is dispatched
2207
+ * — but the watcher loops already tracking in-flight turns keep running, so a
2208
+ * turn that has finished (or is about to) still fires `markDone` and delivers
2209
+ * its reply. Pair with `waitForInFlight()` to bound how long shutdown waits.
2210
+ */
2211
+ stop() {
2212
+ this.stopped = true;
2213
+ }
2214
+ /**
2215
+ * Wait (up to `timeoutMs`) for in-flight watcher work to settle during a
2216
+ * graceful shutdown, so a turn whose reply is ready — or completes within the
2217
+ * window — is delivered before the process exits, instead of being cut off and
2218
+ * left for the ADR-0046 restart-recovery path.
2219
+ *
2220
+ * Bounded on purpose: the watcher's own give-up deadline is up to 10 minutes,
2221
+ * far longer than a shutdown grace period (e.g. Fargate's SIGTERM→SIGKILL
2222
+ * window). We poll `hasInFlightWatchers()` and return as soon as the in-flight
2223
+ * set empties OR the timeout elapses. Anything still in flight at the timeout is
2224
+ * safe to abandon — it stays `processing` server-side and is re-adopted on the
2225
+ * next runner start (ADR-0046).
2226
+ *
2227
+ * @returns true if all in-flight work settled within the window; false if the
2228
+ * timeout elapsed with work still in flight.
2229
+ */
2230
+ async waitForInFlight(timeoutMs) {
2231
+ const deadline = this.now() + timeoutMs;
2232
+ const step = Math.min(this.pausedPollIntervalMs, 250);
2233
+ if (this.activeDrain) {
2234
+ let drainSettled = false;
2235
+ void this.activeDrain.then(() => {
2236
+ drainSettled = true;
2237
+ });
2238
+ while (!drainSettled) {
2239
+ if (this.now() >= deadline) return false;
2240
+ await this.sleep(step);
2241
+ }
2242
+ }
2243
+ while (this.hasInFlightWatchers()) {
2244
+ if (this.now() >= deadline) return false;
2245
+ await this.sleep(step);
2246
+ }
2247
+ return true;
2248
+ }
1638
2249
  /**
1639
2250
  * Await all outstanding per-session watchers (WI-3).
1640
2251
  *
@@ -1654,9 +2265,7 @@ var ChannelDriver = class {
1654
2265
  if (!stillLive) return;
1655
2266
  }
1656
2267
  }
1657
- // -------------------------------------------------------------------------
1658
2268
  // Conversation processing (WI-3 — async dispatch)
1659
- // -------------------------------------------------------------------------
1660
2269
  /**
1661
2270
  * Dispatch each pending message for a conversation to opencode's native queue
1662
2271
  * via `prompt_async` (Task 3.2) and register it with the conversation's
@@ -1669,15 +2278,18 @@ var ChannelDriver = class {
1669
2278
  const sessionId = await this.ensureSession(conv);
1670
2279
  const messages = await this.getPendingMessages(conv.id);
1671
2280
  let dispatched = 0;
2281
+ let skippedAlreadyDispatched = 0;
1672
2282
  for (const message of messages) {
2283
+ if (this.stopped) break;
1673
2284
  if (this.dispatched.has(message.id)) {
2285
+ skippedAlreadyDispatched += 1;
1674
2286
  continue;
1675
2287
  }
1676
- const opencodeMessageId = opencodeMessageIdFor(message.id);
1677
2288
  const options = {
1678
2289
  agent: message.opencode_agent ?? void 0,
1679
2290
  model: message.opencode_model ?? void 0
1680
2291
  };
2292
+ let opencodeMessageId;
1681
2293
  try {
1682
2294
  this.log({
1683
2295
  level: "info",
@@ -1685,10 +2297,24 @@ var ChannelDriver = class {
1685
2297
  conversation_id: conv.id,
1686
2298
  message_id: message.id
1687
2299
  });
1688
- await sendPromptAsync(this.port, sessionId, message.content, options, opencodeMessageId);
2300
+ const sendAttachments = this.buildSendAttachments(conv, message);
2301
+ opencodeMessageId = await this.dispatchLocked(
2302
+ sessionId,
2303
+ () => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
2304
+ );
1689
2305
  } catch (err) {
1690
2306
  if (err instanceof ChannelAuthError) throw err;
1691
2307
  this.dispatched.delete(message.id);
2308
+ if (await sessionExists(this.port, sessionId) === false) {
2309
+ this.sessions.delete(conv.id);
2310
+ this.log({
2311
+ level: "warn",
2312
+ message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
2313
+ conversation_id: conv.id,
2314
+ message_id: message.id
2315
+ });
2316
+ break;
2317
+ }
1692
2318
  await this.markFailed(conv.id, message.id).catch(() => {
1693
2319
  });
1694
2320
  this.log({
@@ -1699,24 +2325,58 @@ var ChannelDriver = class {
1699
2325
  });
1700
2326
  continue;
1701
2327
  }
2328
+ if (opencodeMessageId === null) {
2329
+ this.log({
2330
+ level: "warn",
2331
+ message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
2332
+ conversation_id: conv.id,
2333
+ message_id: message.id
2334
+ });
2335
+ continue;
2336
+ }
1702
2337
  this.dispatched.add(message.id);
1703
2338
  this.registerInFlight(conv, sessionId, message, opencodeMessageId);
1704
2339
  dispatched += 1;
2340
+ void this.postSignal(conv.id, message.id, "dispatched");
2341
+ }
2342
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
2343
+ this.log({
2344
+ level: "warn",
2345
+ message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
2346
+ conversation_id: conv.id
2347
+ });
1705
2348
  }
1706
2349
  this.ensureWatcherRunning(sessionId);
1707
2350
  return dispatched;
1708
2351
  }
1709
2352
  async ensureSession(conv) {
1710
- const cached = this.sessions.get(conv.id);
1711
- if (cached) return cached;
1712
- if (conv.opencode_session_id) {
1713
- this.sessions.set(conv.id, conv.opencode_session_id);
1714
- return conv.opencode_session_id;
2353
+ const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
2354
+ if (bound) {
2355
+ const exists = await sessionExists(this.port, bound);
2356
+ if (exists === false) {
2357
+ this.log({
2358
+ level: "debug",
2359
+ message: `OpenCode session ${bound} for conversation ${conv.id.slice(0, 8)} no longer exists (deleted or DB reset) \u2014 creating a fresh session and rebinding.`,
2360
+ conversation_id: conv.id
2361
+ });
2362
+ this.sessions.delete(conv.id);
2363
+ return this.createAndBindSession(conv.id);
2364
+ }
2365
+ this.sessions.set(conv.id, bound);
2366
+ return bound;
1715
2367
  }
2368
+ return this.createAndBindSession(conv.id);
2369
+ }
2370
+ /**
2371
+ * Create a fresh OpenCode session for a conversation, cache the binding, and
2372
+ * best-effort persist it server-side. Shared by the first-ever bind and the
2373
+ * self-heal recreate path in `ensureSession`.
2374
+ */
2375
+ async createAndBindSession(conversationId) {
1716
2376
  const directory = await this.resolveOpenCodeDirectory();
1717
2377
  const sessionId = await createOpenCodeSession(this.port, directory);
1718
- this.sessions.set(conv.id, sessionId);
1719
- await this.persistSession(conv.id, sessionId).catch(() => {
2378
+ this.sessions.set(conversationId, sessionId);
2379
+ await this.persistSession(conversationId, sessionId).catch(() => {
1720
2380
  });
1721
2381
  return sessionId;
1722
2382
  }
@@ -1730,15 +2390,129 @@ var ChannelDriver = class {
1730
2390
  this.opencodeDirectory = await getOpenCodeDirectory(this.port);
1731
2391
  if (!this.opencodeDirectory) {
1732
2392
  this.log({
1733
- level: "info",
2393
+ level: "warn",
1734
2394
  message: "Could not determine opencode directory (GET /path) \u2014 new sessions may not appear in opencode web"
1735
2395
  });
1736
2396
  }
1737
2397
  return this.opencodeDirectory;
1738
2398
  }
1739
- // -------------------------------------------------------------------------
1740
2399
  // Per-session watcher (WI-3)
1741
- // -------------------------------------------------------------------------
2400
+ /**
2401
+ * Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
2402
+ * opencode session (Task 2.1a), so two dispatches into the SAME session can
2403
+ * never interleave and mis-correlate their read-backs. Distinct sessions run
2404
+ * concurrently. The chained tail intentionally ignores the prior result/error
2405
+ * (each dispatch reports its own outcome to its caller).
2406
+ */
2407
+ dispatchLocked(sessionId, fn) {
2408
+ const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
2409
+ const run2 = prior.then(fn, fn);
2410
+ this.sessionDispatchLocks.set(
2411
+ sessionId,
2412
+ run2.then(
2413
+ () => void 0,
2414
+ () => void 0
2415
+ )
2416
+ );
2417
+ return run2;
2418
+ }
2419
+ // Inbound image attachments (#255, WI-8)
2420
+ /**
2421
+ * Build the `SendAttachmentsInput` for a message's inbound images, or
2422
+ * `undefined` when the message has none (so a text-only turn is unchanged).
2423
+ *
2424
+ * The driver OWNS the two channel-facing concerns the session module cannot:
2425
+ * - the AUTHENTICATED byte fetch through Evident's WI-6 endpoint
2426
+ * (`fetchAttachmentDataUrl`), using the SAME `getAuthHeader()` as every
2427
+ * other combinedAuth callback — the CLI NEVER talks to Slack directly;
2428
+ * - the in-thread SKIP NOTE (`signalAttachmentsSkipped`) posted over the
2429
+ * existing callback surface when any image was skipped/failed.
2430
+ * `sendPromptAsync` applies the capability gate + appends the `file` parts and
2431
+ * reports outcomes back via `onOutcomes`.
2432
+ */
2433
+ buildSendAttachments(conv, message) {
2434
+ const refs = message.attachments;
2435
+ if (!refs || refs.length === 0) return void 0;
2436
+ return {
2437
+ inputs: refs.map((a, index) => ({
2438
+ index,
2439
+ mime: a.mime,
2440
+ ...a.filename ? { filename: a.filename } : {}
2441
+ })),
2442
+ fetchDataUrl: (index) => this.fetchAttachmentDataUrl(message.id, index, refs[index].mime),
2443
+ onOutcomes: ({ outcomes, capabilityUnknown }) => this.signalAttachmentsSkipped(conv.id, message.id, outcomes, capabilityUnknown)
2444
+ };
2445
+ }
2446
+ /**
2447
+ * Fetch ONE inbound image's bytes through Evident's WI-6 endpoint
2448
+ * (`GET {apiUrl}/agents/{agentId}/attachments/{messageId}/{index}`) using the
2449
+ * existing authenticated fetch, and base64-encode into a
2450
+ * `data:<mime>;base64,<…>` URL for the opencode `file` part's `url`.
2451
+ *
2452
+ * The endpoint streams the source bytes verbatim (200), or returns 404
2453
+ * (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
2454
+ * (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
2455
+ * OMITS that one image and the text turn still sends — NEVER throws the turn.
2456
+ * Failures are logged with context (no silent swallow).
2457
+ */
2458
+ async fetchAttachmentDataUrl(messageId, index, mime) {
2459
+ try {
2460
+ const res = await this.fetchImpl(
2461
+ `${this.apiUrl}/agents/${this.agentId}/attachments/${messageId}/${index}`,
2462
+ { headers: { Authorization: this.getAuthHeader() } }
2463
+ );
2464
+ if (!res.ok) {
2465
+ this.log({
2466
+ level: "error",
2467
+ message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
2468
+ message_id: messageId
2469
+ });
2470
+ return null;
2471
+ }
2472
+ const buf = await res.arrayBuffer();
2473
+ const base64 = Buffer.from(buf).toString("base64");
2474
+ const dataMime = cleanImageMime(res.headers.get("content-type")) || mime;
2475
+ return `data:${dataMime};base64,${base64}`;
2476
+ } catch (err) {
2477
+ this.log({
2478
+ level: "error",
2479
+ message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} failed \u2014 omitting this image (text turn proceeds): ${err instanceof Error ? err.message : String(err)}`,
2480
+ message_id: messageId
2481
+ });
2482
+ return null;
2483
+ }
2484
+ }
2485
+ /**
2486
+ * On any skipped/failed image, post an in-thread note to Evident over the
2487
+ * EXISTING combinedAuth callback surface — the CLI NEVER posts to Slack directly.
2488
+ * Evident routes the note to source via `conversation.deliver`.
2489
+ *
2490
+ * The `POST .../messages/:id/signal` route accepts `attachments_skipped` (in
2491
+ * `messageSignalSchema`) and turns it into an in-thread note delivered through
2492
+ * `conversation.deliver` (e.g. "N image(s) couldn't be forwarded"), so the note
2493
+ * reaches the channel.
2494
+ *
2495
+ * Fire-and-forget: never throws into the send/tick (logs its own failure).
2496
+ */
2497
+ signalAttachmentsSkipped(conversationId, messageId, outcomes, capabilityUnknown) {
2498
+ const skipped = outcomes.filter((o) => o.status === "skipped").length;
2499
+ const failed = outcomes.filter((o) => o.status === "failed").length;
2500
+ if (skipped === 0 && failed === 0) return;
2501
+ if (this.attachmentsSkippedSignalled.has(messageId)) return;
2502
+ this.attachmentsSkippedSignalled.add(messageId);
2503
+ const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
2504
+ this.log({
2505
+ level: "info",
2506
+ 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`,
2507
+ conversation_id: conversationId,
2508
+ message_id: messageId
2509
+ });
2510
+ void this.postSignal(conversationId, messageId, "attachments_skipped", {
2511
+ skipped,
2512
+ failed,
2513
+ ...skipped > 0 ? { skipped_reason: skippedReason } : {}
2514
+ });
2515
+ }
1742
2516
  /** Register a freshly-dispatched message with its session's watcher state. */
1743
2517
  registerInFlight(conv, sessionId, message, opencodeMessageId) {
1744
2518
  let watcher = this.watchers.get(sessionId);
@@ -1748,7 +2522,9 @@ var ChannelDriver = class {
1748
2522
  inFlight: /* @__PURE__ */ new Map(),
1749
2523
  loop: null,
1750
2524
  reportedQuestions: /* @__PURE__ */ new Set(),
1751
- reportedPermissions: /* @__PURE__ */ new Set()
2525
+ reportedPermissions: /* @__PURE__ */ new Set(),
2526
+ lastGoodPollAt: this.now(),
2527
+ hadUsablePoll: false
1752
2528
  };
1753
2529
  this.watchers.set(sessionId, watcher);
1754
2530
  }
@@ -1758,18 +2534,101 @@ var ChannelDriver = class {
1758
2534
  opencodeMessageId,
1759
2535
  message,
1760
2536
  dispatchedAt: now,
2537
+ processingAnchorMs: now,
1761
2538
  deadline: now + this.pausedMaxWaitMs,
1762
2539
  started: false,
1763
- done: false
2540
+ done: false,
2541
+ stuckReported: false,
2542
+ lastAliveAt: 0,
2543
+ aliveInFlight: false,
2544
+ awaitingHumanLatched: false,
2545
+ pausedOnQuestion: false,
2546
+ pausedOnPermission: false,
2547
+ pausedClearConfirmed: false,
2548
+ pausedInFlight: false,
2549
+ deliveryDeadlineAnchored: false
1764
2550
  });
1765
2551
  }
1766
2552
  /**
1767
- * Start (but do NOT await) the per-session watcher loop if it has in-flight
1768
- * work and is not already running. Single-flight per session. The loop is
1769
- * tracked on the watcher and cleared when it settles; it never rejects (fully
1770
- * guarded), so a failed poll/callback can never crash the run loop — the cron
1771
- * stays as the safety net.
1772
- */
2553
+ * Register a RE-ADOPTED `processing` message with its session watcher
2554
+ * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
2555
+ * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
2556
+ * `now`, so the paused/queued/unreachable cases settle on the same wall-clock a
2557
+ * fresh dispatch would (10 min after `processed_at`, not 10 min from now).
2558
+ *
2559
+ * This re-attaches into the SAME watcher, so the ADR-0047 progressing-vs-paused
2560
+ * give-up (`serviceInFlightMessage`) applies unchanged: a re-adopted turn
2561
+ * opencode reports ACTIVELY `running` is watched to completion (its liveness
2562
+ * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
2563
+ * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
2564
+ * handed to the cron. The old "the `deadline` must settle before the ~15-min
2565
+ * cron or they double-drive" reasoning is superseded: liveness now settles the
2566
+ * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
2567
+ * (only the appear-guard uses it).
2568
+ *
2569
+ * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
2570
+ * `opencodeMessageId` is the id the watcher polls for a reply — for the orphan
2571
+ * fresh-run path these differ (a fresh opencode id under the same server row).
2572
+ *
2573
+ * `started` is set true so the watcher does NOT re-`markProcessing` a row the
2574
+ * server already flipped to `processing`; the running/done transitions still
2575
+ * fire from the watcher's normal branches.
2576
+ */
2577
+ registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
2578
+ let watcher = this.watchers.get(sessionId);
2579
+ if (!watcher) {
2580
+ watcher = {
2581
+ conv,
2582
+ inFlight: /* @__PURE__ */ new Map(),
2583
+ loop: null,
2584
+ reportedQuestions: /* @__PURE__ */ new Set(),
2585
+ reportedPermissions: /* @__PURE__ */ new Set(),
2586
+ lastGoodPollAt: this.now(),
2587
+ hadUsablePoll: false
2588
+ };
2589
+ this.watchers.set(sessionId, watcher);
2590
+ }
2591
+ watcher.inFlight.set(message.id, {
2592
+ evidentMessageId: message.id,
2593
+ opencodeMessageId,
2594
+ message,
2595
+ dispatchedAt: this.now(),
2596
+ // Anchor the absolute-age ceiling to the SERVER-SIDE `processed_at` (the same
2597
+ // value seeding `deadline`), NOT `dispatchedAt` — so a re-adopted zombie's age
2598
+ // reflects the real turn duration and the ceiling fires on the ORIGINAL turn.
2599
+ processingAnchorMs: processedAtMs,
2600
+ deadline: processedAtMs + this.pausedMaxWaitMs,
2601
+ // The server row is ALREADY `processing`; do not re-fire markProcessing.
2602
+ started: true,
2603
+ done: false,
2604
+ // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
2605
+ // AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
2606
+ // on `state === 'queued'` (turn produced no reply), not on `started`, so a
2607
+ // re-adopted row left wedged in `queued` still emits the signal once
2608
+ // (#210/#220 observability).
2609
+ stuckReported: false,
2610
+ // Task 5.2: a re-adopted actively-running row re-attaches into the SAME
2611
+ // watcher and so hits the SAME actively-running heartbeat branch in
2612
+ // `serviceInFlightMessage` as a fresh dispatch — monitoring observes "runner
2613
+ // re-adopted and is confirming this row alive" via that `alive` heartbeat,
2614
+ // with no extra `re_adopted` signal needed (folds old WI-6).
2615
+ lastAliveAt: 0,
2616
+ aliveInFlight: false,
2617
+ awaitingHumanLatched: false,
2618
+ pausedOnQuestion: false,
2619
+ pausedOnPermission: false,
2620
+ pausedClearConfirmed: false,
2621
+ pausedInFlight: false,
2622
+ deliveryDeadlineAnchored: false
2623
+ });
2624
+ }
2625
+ /**
2626
+ * Start (but do NOT await) the per-session watcher loop if it has in-flight
2627
+ * work and is not already running. Single-flight per session. The loop is
2628
+ * tracked on the watcher and cleared when it settles; it never rejects (fully
2629
+ * guarded), so a failed poll/callback can never crash the run loop — the cron
2630
+ * stays as the safety net.
2631
+ */
1773
2632
  ensureWatcherRunning(sessionId) {
1774
2633
  const watcher = this.watchers.get(sessionId);
1775
2634
  if (!watcher) return;
@@ -1811,12 +2670,30 @@ var ChannelDriver = class {
1811
2670
  messages = Array.isArray(body) ? body : null;
1812
2671
  }
1813
2672
  } catch {
1814
- continue;
1815
2673
  }
2674
+ if (messages != null && messages.length > 0) {
2675
+ watcher.lastGoodPollAt = this.now();
2676
+ watcher.hadUsablePoll = true;
2677
+ } else {
2678
+ const emptyButReachable = messages != null;
2679
+ const graceApplies = !emptyButReachable || watcher.hadUsablePoll;
2680
+ if (graceApplies && this.now() - watcher.lastGoodPollAt < POLL_MISS_GRACE_MS) {
2681
+ continue;
2682
+ }
2683
+ }
2684
+ const { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk } = await this.pollInteractions(sessionId, watcher, messages);
1816
2685
  for (const inFlight of [...watcher.inFlight.values()]) {
1817
- await this.serviceInFlightMessage(sessionId, watcher, inFlight, messages);
2686
+ await this.serviceInFlightMessage(
2687
+ sessionId,
2688
+ watcher,
2689
+ inFlight,
2690
+ messages,
2691
+ openQuestions,
2692
+ openPermissions,
2693
+ questionsPolledOk,
2694
+ permissionsPolledOk
2695
+ );
1818
2696
  }
1819
- await this.pollInteractions(sessionId, watcher, messages);
1820
2697
  }
1821
2698
  } catch (err) {
1822
2699
  if (err instanceof ChannelAuthError) {
@@ -1826,6 +2703,7 @@ var ChannelDriver = class {
1826
2703
  conversation_id: watcher.conv.id
1827
2704
  });
1828
2705
  for (const evidentMessageId of [...watcher.inFlight.keys()]) {
2706
+ this.readopted.delete(evidentMessageId);
1829
2707
  this.removeInFlight(watcher, evidentMessageId);
1830
2708
  }
1831
2709
  return;
@@ -1837,23 +2715,55 @@ var ChannelDriver = class {
1837
2715
  });
1838
2716
  }
1839
2717
  }
2718
+ /**
2719
+ * On FIRST observing a terminal (done/failed) state, ensure the delivery
2720
+ * (markDone/markFailed) transient-retry path has a real window. A long
2721
+ * ACTIVELY-running turn is kept past its original `deadline`, so by completion
2722
+ * `now >= deadline` already holds and the retry bound below would fire on the
2723
+ * first transient PATCH failure — dropping the message before its reply lands
2724
+ * (Bugbot "Stale deadline aborts long-turn delivery"). Re-anchor once (latched)
2725
+ * to a fresh `pausedMaxWaitMs` window; only extend if the current deadline is at
2726
+ * or past now, so a still-ample window is left untouched.
2727
+ */
2728
+ anchorDeliveryDeadline(inFlight) {
2729
+ if (inFlight.deliveryDeadlineAnchored) return;
2730
+ inFlight.deliveryDeadlineAnchored = true;
2731
+ if (this.now() >= inFlight.deadline) {
2732
+ inFlight.deadline = this.now() + this.pausedMaxWaitMs;
2733
+ }
2734
+ }
1840
2735
  /**
1841
2736
  * Drive ONE in-flight message's lifecycle from the tick's message snapshot.
1842
2737
  * Fires markProcessing on queued→running and markDone on done (each once),
1843
2738
  * applies the idle-path re-dispatch guard, and removes the message from the
1844
2739
  * in-flight set on completion or timeout.
1845
2740
  */
1846
- async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
2741
+ async serviceInFlightMessage(sessionId, watcher, inFlight, messages, openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk) {
1847
2742
  const conv = watcher.conv;
1848
2743
  const state = messageRunState(messages, inFlight.opencodeMessageId);
1849
- if ((state === "running" || state === "done") && !inFlight.started) {
2744
+ const id = inFlight.evidentMessageId;
2745
+ if (openQuestions.has(id)) inFlight.pausedOnQuestion = true;
2746
+ else if (questionsPolledOk) inFlight.pausedOnQuestion = false;
2747
+ if (openPermissions.has(id)) inFlight.pausedOnPermission = true;
2748
+ else if (permissionsPolledOk) inFlight.pausedOnPermission = false;
2749
+ const observedOpen = openQuestions.has(id) || openPermissions.has(id);
2750
+ const latchedPaused = inFlight.pausedOnQuestion || inFlight.pausedOnPermission;
2751
+ const awaitingHuman = observedOpen || latchedPaused;
2752
+ if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
2753
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
1850
2754
  let claimed;
1851
2755
  try {
1852
- claimed = await this.markProcessing(conv.id, inFlight.evidentMessageId, sessionId);
2756
+ claimed = await this.markProcessing(
2757
+ conv.id,
2758
+ inFlight.evidentMessageId,
2759
+ sessionId,
2760
+ inFlight.opencodeMessageId,
2761
+ title
2762
+ );
1853
2763
  } catch (err) {
1854
2764
  if (err instanceof ChannelAuthError) throw err;
1855
2765
  this.log({
1856
- level: "error",
2766
+ level: "warn",
1857
2767
  message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
1858
2768
  conversation_id: conv.id,
1859
2769
  message_id: inFlight.evidentMessageId
@@ -1863,7 +2773,7 @@ var ChannelDriver = class {
1863
2773
  inFlight.started = true;
1864
2774
  if (!claimed) {
1865
2775
  this.log({
1866
- level: "info",
2776
+ level: "debug",
1867
2777
  message: `Message ${inFlight.evidentMessageId.slice(0, 8)} already marked processing \u2014 continuing`,
1868
2778
  conversation_id: conv.id,
1869
2779
  message_id: inFlight.evidentMessageId
@@ -1871,6 +2781,7 @@ var ChannelDriver = class {
1871
2781
  }
1872
2782
  }
1873
2783
  if (state === "done") {
2784
+ this.anchorDeliveryDeadline(inFlight);
1874
2785
  if (!inFlight.done) {
1875
2786
  this.log({
1876
2787
  level: "info",
@@ -1878,13 +2789,20 @@ var ChannelDriver = class {
1878
2789
  conversation_id: conv.id,
1879
2790
  message_id: inFlight.evidentMessageId
1880
2791
  });
2792
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
1881
2793
  try {
1882
- await this.markDone(conv.id, inFlight.evidentMessageId, sessionId);
2794
+ await this.markDone(
2795
+ conv.id,
2796
+ inFlight.evidentMessageId,
2797
+ sessionId,
2798
+ inFlight.opencodeMessageId,
2799
+ title
2800
+ );
1883
2801
  } catch (err) {
1884
2802
  if (err instanceof ChannelAuthError) throw err;
1885
2803
  if (err instanceof ChannelTerminalError) {
1886
2804
  this.log({
1887
- level: "error",
2805
+ level: "warn",
1888
2806
  message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
1889
2807
  conversation_id: conv.id,
1890
2808
  message_id: inFlight.evidentMessageId
@@ -1894,7 +2812,7 @@ var ChannelDriver = class {
1894
2812
  }
1895
2813
  if (this.now() >= inFlight.deadline) {
1896
2814
  this.log({
1897
- level: "error",
2815
+ level: "warn",
1898
2816
  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)}`,
1899
2817
  conversation_id: conv.id,
1900
2818
  message_id: inFlight.evidentMessageId
@@ -1903,7 +2821,7 @@ var ChannelDriver = class {
1903
2821
  return;
1904
2822
  }
1905
2823
  this.log({
1906
- level: "error",
2824
+ level: "warn",
1907
2825
  message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
1908
2826
  conversation_id: conv.id,
1909
2827
  message_id: inFlight.evidentMessageId
@@ -1915,60 +2833,583 @@ var ChannelDriver = class {
1915
2833
  this.removeInFlight(watcher, inFlight.evidentMessageId);
1916
2834
  return;
1917
2835
  }
1918
- if (state === "unknown") {
1919
- if (this.now() - inFlight.dispatchedAt >= this.dispatchConfirmMs) {
1920
- await this.redispatchInFlight(sessionId, inFlight);
2836
+ if (state === "failed") {
2837
+ this.anchorDeliveryDeadline(inFlight);
2838
+ if (!inFlight.done) {
2839
+ const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
2840
+ this.log({
2841
+ level: "error",
2842
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2843
+ conversation_id: conv.id,
2844
+ message_id: inFlight.evidentMessageId
2845
+ });
2846
+ try {
2847
+ await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
2848
+ } catch (err) {
2849
+ if (err instanceof ChannelAuthError) throw err;
2850
+ if (err instanceof ChannelTerminalError) {
2851
+ this.log({
2852
+ level: "warn",
2853
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2854
+ conversation_id: conv.id,
2855
+ message_id: inFlight.evidentMessageId
2856
+ });
2857
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2858
+ return;
2859
+ }
2860
+ if (this.now() >= inFlight.deadline) {
2861
+ this.log({
2862
+ level: "warn",
2863
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
2864
+ conversation_id: conv.id,
2865
+ message_id: inFlight.evidentMessageId
2866
+ });
2867
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2868
+ return;
2869
+ }
2870
+ this.log({
2871
+ level: "warn",
2872
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2873
+ conversation_id: conv.id,
2874
+ message_id: inFlight.evidentMessageId
2875
+ });
2876
+ return;
2877
+ }
2878
+ inFlight.done = true;
1921
2879
  }
2880
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2881
+ return;
1922
2882
  }
1923
- if (this.now() >= inFlight.deadline) {
2883
+ const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
2884
+ const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
2885
+ if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
2886
+ inFlight.stuckReported = true;
2887
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2888
+ stuck_for_ms: this.now() - inFlight.dispatchedAt
2889
+ });
2890
+ }
2891
+ const activelyRunning = state === "running" && !awaitingHuman;
2892
+ if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
1924
2893
  this.log({
1925
- level: "info",
2894
+ level: "warn",
2895
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} exceeded the absolute processing ceiling (${Math.round((this.now() - inFlight.processingAnchorMs) / 6e4)}min, session ${sessionId}) while still actively running \u2014 releasing so the cron can reclaim it`,
2896
+ conversation_id: conv.id,
2897
+ message_id: inFlight.evidentMessageId
2898
+ });
2899
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2900
+ watched_for_ms: this.now() - inFlight.processingAnchorMs
2901
+ });
2902
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2903
+ return;
2904
+ }
2905
+ if (activelyRunning && !inFlight.awaitingHumanLatched && !inFlight.aliveInFlight && this.now() - inFlight.lastAliveAt >= HEARTBEAT_MS) {
2906
+ inFlight.aliveInFlight = true;
2907
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "alive").then((ok) => {
2908
+ inFlight.aliveInFlight = false;
2909
+ if (ok) inFlight.lastAliveAt = this.now();
2910
+ });
2911
+ }
2912
+ if (awaitingHuman) {
2913
+ if (!inFlight.awaitingHumanLatched) {
2914
+ inFlight.deadline = this.now() + this.pausedMaxWaitMs;
2915
+ inFlight.awaitingHumanLatched = true;
2916
+ }
2917
+ if (!inFlight.pausedClearConfirmed && !inFlight.pausedInFlight) {
2918
+ inFlight.pausedInFlight = true;
2919
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "paused").then((ok) => {
2920
+ inFlight.pausedInFlight = false;
2921
+ if (ok && inFlight.awaitingHumanLatched) inFlight.pausedClearConfirmed = true;
2922
+ });
2923
+ }
2924
+ } else if (inFlight.awaitingHumanLatched) {
2925
+ inFlight.awaitingHumanLatched = false;
2926
+ inFlight.pausedOnQuestion = false;
2927
+ inFlight.pausedOnPermission = false;
2928
+ inFlight.pausedClearConfirmed = false;
2929
+ }
2930
+ const siblingPaused = (sib) => openQuestions.has(sib.evidentMessageId) || openPermissions.has(sib.evidentMessageId) || sib.awaitingHumanLatched || sib.pausedOnQuestion || sib.pausedOnPermission;
2931
+ const hasActivelyRunningSibling = [...watcher.inFlight.values()].some(
2932
+ (sib) => sib.evidentMessageId !== inFlight.evidentMessageId && messageRunState(messages, sib.opencodeMessageId) === "running" && !siblingPaused(sib)
2933
+ );
2934
+ const queuedBehindRunningSibling = state === "queued" && hasActivelyRunningSibling;
2935
+ if (!activelyRunning && !queuedBehindRunningSibling && this.now() >= inFlight.deadline) {
2936
+ this.log({
2937
+ level: "debug",
1926
2938
  message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
1927
2939
  conversation_id: conv.id,
1928
2940
  message_id: inFlight.evidentMessageId
1929
2941
  });
2942
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2943
+ watched_for_ms: this.now() - inFlight.dispatchedAt
2944
+ });
1930
2945
  this.removeInFlight(watcher, inFlight.evidentMessageId);
1931
2946
  }
1932
2947
  }
2948
+ // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
2949
+ /**
2950
+ * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
2951
+ *
2952
+ * The pending drain only re-drives `pending` rows; a message already flipped to
2953
+ * `processing` before the runner died is watched by nobody until the 15-min
2954
+ * cron resets it. Here we fetch those rows, and per row resolve its correlated
2955
+ * reply against opencode's OWN session store — completing, re-attaching, or
2956
+ * (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
2957
+ * is idempotent per message (Invariant 2): a row a watcher already tracks is
2958
+ * skipped in `readoptOne` — one driver, no double-drive.
2959
+ *
2960
+ * Only `ChannelAuthError` propagates (to `drainPending`, like the pending
2961
+ * path); every other early return LOGS a reason with context — no silent drop.
2962
+ */
2963
+ async readoptProcessing() {
2964
+ const rows = await this.getProcessingMessages();
2965
+ if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0 || this.readoptPollUnresolvedSignalled.size > 0) {
2966
+ const stillProcessing = new Set(rows.map((r) => r.id));
2967
+ for (const id of [
2968
+ ...this.dontRedispatch,
2969
+ ...this.doneUndeliverable,
2970
+ ...this.readoptPollUnresolvedSignalled
2971
+ ]) {
2972
+ if (!stillProcessing.has(id)) {
2973
+ const cleared = this.dontRedispatch.delete(id);
2974
+ const clearedUndeliverable = this.doneUndeliverable.delete(id);
2975
+ this.readoptPollUnresolvedSignalled.delete(id);
2976
+ if (cleared || clearedUndeliverable) {
2977
+ this.log({
2978
+ level: "debug",
2979
+ message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
2980
+ message_id: id
2981
+ });
2982
+ }
2983
+ }
2984
+ }
2985
+ }
2986
+ if (rows.length === 0) return;
2987
+ const bySession = /* @__PURE__ */ new Map();
2988
+ for (const row of rows) {
2989
+ if (!row.opencode_session_id) {
2990
+ this.log({
2991
+ level: "warn",
2992
+ message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
2993
+ conversation_id: row.conversation_id,
2994
+ message_id: row.id
2995
+ });
2996
+ continue;
2997
+ }
2998
+ const list = bySession.get(row.opencode_session_id) ?? [];
2999
+ list.push(row);
3000
+ bySession.set(row.opencode_session_id, list);
3001
+ }
3002
+ for (const [sessionId, sessionRows] of bySession) {
3003
+ let messages;
3004
+ try {
3005
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
3006
+ if (!res.ok) {
3007
+ this.log({
3008
+ level: "warn",
3009
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
3010
+ });
3011
+ continue;
3012
+ }
3013
+ const body = await res.json();
3014
+ if (!Array.isArray(body)) {
3015
+ this.log({
3016
+ level: "warn",
3017
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
3018
+ });
3019
+ continue;
3020
+ }
3021
+ messages = body;
3022
+ } catch (err) {
3023
+ this.log({
3024
+ level: "warn",
3025
+ message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
3026
+ });
3027
+ continue;
3028
+ }
3029
+ const anyUntracked = sessionRows.some((row) => !this.isTracked(sessionId, row.id));
3030
+ const sessionOngoing = anyUntracked ? await isSessionOngoing(this.port, sessionId) : null;
3031
+ for (const row of sessionRows) {
3032
+ await this.readoptOne(sessionId, row, messages, sessionOngoing);
3033
+ }
3034
+ }
3035
+ }
1933
3036
  /**
1934
- * Re-dispatch a message whose user row never appeared (idle-path guard). Safe:
1935
- * opencode treats a duplicate caller-supplied `messageID` as idempotent (PoC
1936
- * fact 9) — one user message + one reply even if the original DID land. Resets
1937
- * the dispatch timestamp so the guard doesn't immediately fire again.
3037
+ * Re-adopt ONE `processing` row against the tick's session message snapshot
3038
+ * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
3039
+ *
3040
+ * Branches on `messageRunState(messages, row.opencode_message_id)` the
3041
+ * opencode-assigned user-message id persisted on the first `processing` PATCH
3042
+ * (#218). A row with a NULL stored id (dispatched but the read-back never landed
3043
+ * before the restart) has no id to correlate → treated as an orphan and
3044
+ * re-dispatched (at most once, see `forceReadoptRun`):
3045
+ * - `done` → `markDone` now (guarded like the watcher's done branch);
3046
+ * - `failed` → `markFailed` with the surfaced error (issue #182), so an
3047
+ * errored turn is reported failed on restart, NOT re-dispatched;
3048
+ * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
3049
+ * tracking the stored id so the reply correlates by it;
3050
+ * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
3051
+ *
3052
+ * Only `ChannelAuthError` propagates.
1938
3053
  */
1939
- async redispatchInFlight(sessionId, inFlight) {
3054
+ async readoptOne(sessionId, row, messages, sessionOngoing) {
3055
+ if (this.isTracked(sessionId, row.id)) {
3056
+ this.log({
3057
+ level: "debug",
3058
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
3059
+ conversation_id: row.conversation_id,
3060
+ message_id: row.id
3061
+ });
3062
+ return;
3063
+ }
3064
+ const ocId = row.opencode_message_id;
3065
+ const state = messageRunState(messages, ocId ?? "");
3066
+ if (state === "done") {
3067
+ if (this.doneUndeliverable.has(row.id)) {
3068
+ this.log({
3069
+ level: "debug",
3070
+ message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
3071
+ conversation_id: row.conversation_id,
3072
+ message_id: row.id
3073
+ });
3074
+ return;
3075
+ }
3076
+ this.log({
3077
+ level: "info",
3078
+ message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
3079
+ conversation_id: row.conversation_id,
3080
+ message_id: row.id
3081
+ });
3082
+ try {
3083
+ const title = await this.resolveSessionTitle(sessionId, row.conversation_id);
3084
+ await this.markDone(row.conversation_id, row.id, sessionId, ocId, title);
3085
+ } catch (err) {
3086
+ if (err instanceof ChannelAuthError) throw err;
3087
+ if (err instanceof ChannelTerminalError) {
3088
+ this.doneUndeliverable.add(row.id);
3089
+ this.log({
3090
+ level: "warn",
3091
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 parking until it leaves processing; leaving for the cron safety net: ${err.message}`,
3092
+ conversation_id: row.conversation_id,
3093
+ message_id: row.id
3094
+ });
3095
+ void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
3096
+ return;
3097
+ }
3098
+ this.log({
3099
+ level: "warn",
3100
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3101
+ conversation_id: row.conversation_id,
3102
+ message_id: row.id
3103
+ });
3104
+ return;
3105
+ }
3106
+ this.dontRedispatch.delete(row.id);
3107
+ void this.postSignal(row.conversation_id, row.id, "readopt_done");
3108
+ return;
3109
+ }
3110
+ if (state === "failed") {
3111
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
3112
+ this.log({
3113
+ level: "error",
3114
+ message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
3115
+ conversation_id: row.conversation_id,
3116
+ message_id: row.id
3117
+ });
3118
+ try {
3119
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2);
3120
+ } catch (err) {
3121
+ if (err instanceof ChannelAuthError) throw err;
3122
+ if (err instanceof ChannelTerminalError) {
3123
+ this.doneUndeliverable.add(row.id);
3124
+ this.log({
3125
+ level: "warn",
3126
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 parking until it leaves processing; leaving for the cron safety net: ${err.message}`,
3127
+ conversation_id: row.conversation_id,
3128
+ message_id: row.id
3129
+ });
3130
+ void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
3131
+ return;
3132
+ }
3133
+ this.log({
3134
+ level: "warn",
3135
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3136
+ conversation_id: row.conversation_id,
3137
+ message_id: row.id
3138
+ });
3139
+ return;
3140
+ }
3141
+ this.dontRedispatch.delete(row.id);
3142
+ void this.postSignal(row.conversation_id, row.id, "readopt_failed");
3143
+ return;
3144
+ }
3145
+ if (this.dontRedispatch.has(row.id)) {
3146
+ this.log({
3147
+ level: "debug",
3148
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
3149
+ conversation_id: row.conversation_id,
3150
+ message_id: row.id
3151
+ });
3152
+ return;
3153
+ }
3154
+ let statusReadableOngoing = null;
3155
+ if (state === "running" && ocId) {
3156
+ const reply = findLastAssistantReplyFor(messages, ocId);
3157
+ const shape = this.replyCompletionShape(reply);
3158
+ const ongoing = sessionOngoing;
3159
+ statusReadableOngoing = ongoing;
3160
+ if (ongoing === false) {
3161
+ this.log({
3162
+ level: "info",
3163
+ message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} but session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status (absent/idle) \u2014 re-dispatching from scratch (status-gated recovery)`,
3164
+ conversation_id: row.conversation_id,
3165
+ message_id: row.id
3166
+ });
3167
+ await this.forceReadoptRun(sessionId, row);
3168
+ return;
3169
+ }
3170
+ if (ongoing === true) {
3171
+ this.log({
3172
+ level: "debug",
3173
+ message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} and session ${sessionId.slice(0, 8)} is ongoing per GET /session/status (busy/retry) \u2014 re-attaching watcher (no re-dispatch)`,
3174
+ conversation_id: row.conversation_id,
3175
+ message_id: row.id
3176
+ });
3177
+ } else {
3178
+ if (shape === "b1") {
3179
+ this.log({
3180
+ level: "debug",
3181
+ message: `Re-adopt: message ${row.id.slice(0, 8)} running/b1 but GET /session/status was unreadable (null) for session ${sessionId.slice(0, 8)} \u2014 NOT latching a b1 row on a transient status blip; leaving it un-tracked to re-evaluate on the next drain`,
3182
+ conversation_id: row.conversation_id,
3183
+ message_id: row.id
3184
+ });
3185
+ if (!this.readoptPollUnresolvedSignalled.has(row.id)) {
3186
+ this.readoptPollUnresolvedSignalled.add(row.id);
3187
+ void this.postSignal(row.conversation_id, row.id, "readopt_poll_unresolved");
3188
+ }
3189
+ return;
3190
+ }
3191
+ this.log({
3192
+ level: "debug",
3193
+ message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} but GET /session/status was unreadable (null) for session ${sessionId.slice(0, 8)} \u2014 falling back to the #253 preamble + descendant cross-check`,
3194
+ conversation_id: row.conversation_id,
3195
+ message_id: row.id
3196
+ });
3197
+ }
3198
+ }
3199
+ if (statusReadableOngoing === null && state === "running" && ocId && isPreamblePinnedRunning(messages, ocId)) {
3200
+ const descendantAlive = await this.isAnyDescendantSessionAlive(sessionId);
3201
+ if (descendantAlive === true) {
3202
+ this.log({
3203
+ level: "debug",
3204
+ message: `Re-adopt: message ${row.id.slice(0, 8)} preamble-pinned running (root ${sessionId.slice(0, 8)}) but a live descendant sub-agent session was found \u2014 treating as still running, re-attaching watcher (no re-dispatch)`,
3205
+ conversation_id: row.conversation_id,
3206
+ message_id: row.id
3207
+ });
3208
+ } else {
3209
+ this.log({
3210
+ level: "info",
3211
+ message: `Re-adopt: message ${row.id.slice(0, 8)} preamble-pinned on recovery (root ${sessionId.slice(0, 8)}), no live descendant runner \u2014 re-dispatching from scratch${descendantAlive === null ? " (descendant liveness indeterminate; a restart guarantees no live runner, so this does NOT block the re-dispatch)" : ""}`,
3212
+ conversation_id: row.conversation_id,
3213
+ message_id: row.id
3214
+ });
3215
+ await this.forceReadoptRun(sessionId, row);
3216
+ return;
3217
+ }
3218
+ }
3219
+ if ((state === "running" || state === "queued") && ocId) {
3220
+ const conv = this.convForRow(sessionId, row);
3221
+ const message = this.queuedMessageForRow(row);
3222
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
3223
+ this.dispatched.add(row.id);
3224
+ this.readopted.add(row.id);
3225
+ this.ensureWatcherRunning(sessionId);
3226
+ this.log({
3227
+ level: "debug",
3228
+ message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
3229
+ conversation_id: row.conversation_id,
3230
+ message_id: row.id
3231
+ });
3232
+ void this.postSignal(row.conversation_id, row.id, "readopt_reattached");
3233
+ return;
3234
+ }
3235
+ await this.forceReadoptRun(sessionId, row);
3236
+ }
3237
+ /**
3238
+ * Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
3239
+ *
3240
+ * #218/WI-5: the row's user message is absent (never kept, or a null stored id),
3241
+ * so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
3242
+ * read it back, and register the watcher under the assigned id so the reply
3243
+ * correlates server-side.
3244
+ *
3245
+ * ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
3246
+ * id). Without a guard, if this dispatches on tick N but the read-back+persist
3247
+ * hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
3248
+ * tick N+1 would dispatch AGAIN → duplicate user turns. The `awaitingReadopt`
3249
+ * latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
3250
+ * short-circuit while the row is latched; clear it on a successful dispatch (the
3251
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
3252
+ * re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
3253
+ * may retry exactly once more).
3254
+ *
3255
+ * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
3256
+ * `processed_at` (Invariant 1).
3257
+ */
3258
+ async forceReadoptRun(sessionId, row) {
3259
+ if (this.stopped) {
3260
+ this.log({
3261
+ level: "debug",
3262
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned but the runner is stopping \u2014 not starting a fresh turn; leaving for restart recovery`,
3263
+ conversation_id: row.conversation_id,
3264
+ message_id: row.id
3265
+ });
3266
+ return;
3267
+ }
3268
+ if (this.awaitingReadopt.has(row.id)) {
3269
+ this.log({
3270
+ level: "debug",
3271
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
3272
+ conversation_id: row.conversation_id,
3273
+ message_id: row.id
3274
+ });
3275
+ return;
3276
+ }
3277
+ if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
3278
+ this.dontRedispatch.add(row.id);
3279
+ this.log({
3280
+ level: "debug",
3281
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned but its re-adopt window has already elapsed \u2014 not dispatching an unwatchable turn; parking until it leaves processing (cron will reset it)`,
3282
+ conversation_id: row.conversation_id,
3283
+ message_id: row.id
3284
+ });
3285
+ void this.postSignal(row.conversation_id, row.id, "readopt_window_elapsed");
3286
+ return;
3287
+ }
1940
3288
  const options = {
1941
- agent: inFlight.message.opencode_agent ?? void 0,
1942
- model: inFlight.message.opencode_model ?? void 0
3289
+ agent: row.opencode_agent ?? void 0,
3290
+ model: row.opencode_model ?? void 0
1943
3291
  };
1944
3292
  this.log({
1945
3293
  level: "info",
1946
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} not observed after dispatch \u2014 re-dispatching (idle-path guard)`,
1947
- message_id: inFlight.evidentMessageId
3294
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
3295
+ conversation_id: row.conversation_id,
3296
+ message_id: row.id
1948
3297
  });
3298
+ this.awaitingReadopt.add(row.id);
3299
+ const readoptConv = this.convForRow(sessionId, row);
3300
+ const readoptMessage = this.queuedMessageForRow(row);
3301
+ const sendAttachments = this.buildSendAttachments(readoptConv, readoptMessage);
3302
+ let ocId;
1949
3303
  try {
1950
- await sendPromptAsync(
1951
- this.port,
3304
+ ocId = await this.dispatchLocked(
1952
3305
  sessionId,
1953
- inFlight.message.content,
1954
- options,
1955
- inFlight.opencodeMessageId
3306
+ () => sendPromptAsync(this.port, sessionId, row.content, options, sendAttachments)
1956
3307
  );
1957
3308
  } catch (err) {
3309
+ this.awaitingReadopt.delete(row.id);
3310
+ if (err instanceof ChannelAuthError) throw err;
1958
3311
  this.log({
1959
- level: "error",
1960
- message: `Re-dispatch failed for message ${inFlight.evidentMessageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
1961
- message_id: inFlight.evidentMessageId
3312
+ level: "warn",
3313
+ message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3314
+ conversation_id: row.conversation_id,
3315
+ message_id: row.id
1962
3316
  });
3317
+ void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3318
+ return;
1963
3319
  }
1964
- inFlight.dispatchedAt = this.now();
3320
+ if (ocId === null) {
3321
+ this.awaitingReadopt.delete(row.id);
3322
+ this.log({
3323
+ level: "warn",
3324
+ message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next drain`,
3325
+ conversation_id: row.conversation_id,
3326
+ message_id: row.id
3327
+ });
3328
+ void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3329
+ return;
3330
+ }
3331
+ this.registerReadopted(readoptConv, sessionId, readoptMessage, ocId, this.processedAtMs(row));
3332
+ this.dispatched.add(row.id);
3333
+ this.readopted.add(row.id);
3334
+ this.awaitingReadopt.delete(row.id);
3335
+ this.ensureWatcherRunning(sessionId);
3336
+ void this.postSignal(row.conversation_id, row.id, "readopt_redispatched");
3337
+ }
3338
+ /**
3339
+ * True if `evidentMessageId` is already being driven — either in the
3340
+ * authoritative `dispatched` set or a live watcher's in-flight set for this
3341
+ * session (Invariant 2, WI-5). Either signal means a watcher owns the row.
3342
+ */
3343
+ isTracked(sessionId, evidentMessageId) {
3344
+ if (this.dispatched.has(evidentMessageId)) return true;
3345
+ const watcher = this.watchers.get(sessionId);
3346
+ return watcher?.inFlight.has(evidentMessageId) ?? false;
3347
+ }
3348
+ /**
3349
+ * Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
3350
+ * deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
3351
+ * for `processing` rows, but if it is somehow null/unparseable fall back to
3352
+ * `now` (defensive) AND log — a fallback means the anchor is weaker than
3353
+ * intended, which is worth surfacing.
3354
+ */
3355
+ processedAtMs(row) {
3356
+ const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
3357
+ if (!Number.isNaN(parsed)) return parsed;
3358
+ this.log({
3359
+ level: "error",
3360
+ message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
3361
+ conversation_id: row.conversation_id,
3362
+ message_id: row.id
3363
+ });
3364
+ return this.now();
3365
+ }
3366
+ /** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
3367
+ convForRow(sessionId, row) {
3368
+ return {
3369
+ id: row.conversation_id,
3370
+ agent_id: this.agentId,
3371
+ opencode_session_id: sessionId,
3372
+ pending_message_count: 0,
3373
+ oldest_pending_at: row.processed_at
3374
+ };
3375
+ }
3376
+ /** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
3377
+ queuedMessageForRow(row) {
3378
+ return {
3379
+ id: row.id,
3380
+ content: row.content,
3381
+ status: "processing",
3382
+ opencode_agent: row.opencode_agent,
3383
+ opencode_model: row.opencode_model,
3384
+ source_message_id: row.source_message_id,
3385
+ slack_user_id: row.slack_user_id,
3386
+ attachments: row.attachments ?? null
3387
+ };
1965
3388
  }
1966
3389
  /**
1967
3390
  * Remove a message from the in-flight set AND the authoritative dispatched
1968
3391
  * set. Once the in-flight set empties, the watcher loop's `while` guard exits
1969
3392
  * and its `.finally` removes the session entry from `this.watchers`.
3393
+ *
3394
+ * Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
3395
+ * (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
3396
+ * cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
3397
+ * re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
3398
+ * re-adopted message that completed (`done`) needs no marker — it's leaving
3399
+ * `processing`. This suppresses only re-dispatch: if its reply later completes,
3400
+ * the done branch still delivers it (Bugbot #202).
1970
3401
  */
1971
3402
  removeInFlight(watcher, evidentMessageId) {
3403
+ const inFlight = watcher.inFlight.get(evidentMessageId);
3404
+ if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
3405
+ this.dontRedispatch.add(evidentMessageId);
3406
+ this.log({
3407
+ level: "debug",
3408
+ message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
3409
+ conversation_id: watcher.conv.id,
3410
+ message_id: evidentMessageId
3411
+ });
3412
+ }
1972
3413
  watcher.inFlight.delete(evidentMessageId);
1973
3414
  this.dispatched.delete(evidentMessageId);
1974
3415
  }
@@ -1985,21 +3426,41 @@ var ChannelDriver = class {
1985
3426
  * RUNNING (not done) is the one that paused. With one running message that is
1986
3427
  * unambiguous; with several we prefer an explicit messageID match, else the
1987
3428
  * oldest running message.
3429
+ *
3430
+ * Returns the set of in-flight Evident message ids that are paused awaiting a
3431
+ * human — an outstanding (still-open) question/permission is attributed to them.
3432
+ * `serviceInFlightMessage` uses this to keep an actively-running turn watched
3433
+ * forever (ADR-0047) while still bounding a turn merely blocked on a person who
3434
+ * may never answer. Attribution here covers ALL open interactions, not just
3435
+ * NEW (un-deduped) ones — a question stays "awaiting a human" until answered,
3436
+ * even after it was already surfaced to the channel.
1988
3437
  */
1989
3438
  async pollInteractions(sessionId, watcher, messages) {
3439
+ const openQuestions = /* @__PURE__ */ new Set();
3440
+ const openPermissions = /* @__PURE__ */ new Set();
3441
+ let questionsPolledOk = true;
3442
+ let permissionsPolledOk = true;
1990
3443
  let questions = [];
1991
3444
  try {
1992
3445
  const res = await this.fetchImpl(`${this.opencodeBase}/question`);
1993
3446
  if (res.ok) {
1994
3447
  const body = await res.json();
1995
- questions = Array.isArray(body) ? body : [];
3448
+ if (Array.isArray(body)) {
3449
+ questions = body;
3450
+ } else {
3451
+ questionsPolledOk = false;
3452
+ }
3453
+ } else {
3454
+ questionsPolledOk = false;
1996
3455
  }
1997
3456
  } catch {
3457
+ questionsPolledOk = false;
1998
3458
  }
1999
3459
  for (const q of questions) {
2000
- if (q.sessionID !== sessionId) continue;
2001
- if (watcher.reportedQuestions.has(q.id)) continue;
3460
+ if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
2002
3461
  const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
3462
+ if (paused) openQuestions.add(paused.evidentMessageId);
3463
+ if (watcher.reportedQuestions.has(q.id)) continue;
2003
3464
  const reported = await this.reportInteraction(
2004
3465
  watcher.conv.id,
2005
3466
  "question",
@@ -2013,14 +3474,22 @@ var ChannelDriver = class {
2013
3474
  const res = await this.fetchImpl(`${this.opencodeBase}/permission`);
2014
3475
  if (res.ok) {
2015
3476
  const body = await res.json();
2016
- permissions = Array.isArray(body) ? body : [];
3477
+ if (Array.isArray(body)) {
3478
+ permissions = body;
3479
+ } else {
3480
+ permissionsPolledOk = false;
3481
+ }
3482
+ } else {
3483
+ permissionsPolledOk = false;
2017
3484
  }
2018
3485
  } catch {
3486
+ permissionsPolledOk = false;
2019
3487
  }
2020
3488
  for (const p of permissions) {
2021
- if (p.sessionID !== sessionId) continue;
2022
- if (watcher.reportedPermissions.has(p.id)) continue;
3489
+ if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
2023
3490
  const paused = this.attributeInteraction(watcher, p.messageID, messages);
3491
+ if (paused) openPermissions.add(paused.evidentMessageId);
3492
+ if (watcher.reportedPermissions.has(p.id)) continue;
2024
3493
  const reported = await this.reportInteraction(
2025
3494
  watcher.conv.id,
2026
3495
  "permission",
@@ -2029,6 +3498,173 @@ var ChannelDriver = class {
2029
3498
  );
2030
3499
  if (reported) watcher.reportedPermissions.add(p.id);
2031
3500
  }
3501
+ return { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk };
3502
+ }
3503
+ /**
3504
+ * True when `sessionId` is the `rootSessionId` itself OR a descendant of it —
3505
+ * i.e. its `parentID` chain (resolved via `GET /session/:id`) reaches the
3506
+ * watched root. Sub-agents spawned via the `task` tool run in child sessions,
3507
+ * so their questions/permissions live under a different `sessionID` that must
3508
+ * still be attributed to the root conversation the watcher owns.
3509
+ *
3510
+ * Parents are cached in `sessionParents` so we walk each session at most once;
3511
+ * a bounded depth cap guards against a cycle or a pathological chain, and any
3512
+ * fetch failure is treated as "not a descendant" (best-effort — the interaction
3513
+ * simply isn't surfaced this tick and is retried next tick once resolvable).
3514
+ */
3515
+ async sessionBelongsTo(sessionId, rootSessionId) {
3516
+ let current = sessionId;
3517
+ for (let depth = 0; current && depth < 32; depth++) {
3518
+ if (current === rootSessionId) return true;
3519
+ const parent = await this.resolveSessionParent(current);
3520
+ if (parent === null || parent === void 0) return false;
3521
+ current = parent;
3522
+ }
3523
+ return false;
3524
+ }
3525
+ /**
3526
+ * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3527
+ * `null` for a root session (no parent) and `undefined` when opencode is
3528
+ * unreachable / the session can't be read (so the caller stops walking without
3529
+ * caching a wrong answer — the next tick retries).
3530
+ */
3531
+ async resolveSessionParent(sessionId) {
3532
+ const cached = this.sessionParents.get(sessionId);
3533
+ if (cached !== void 0) return cached;
3534
+ let parent = void 0;
3535
+ try {
3536
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
3537
+ if (res.ok) {
3538
+ const body = await res.json();
3539
+ parent = body && typeof body.parentID === "string" ? body.parentID : null;
3540
+ }
3541
+ } catch {
3542
+ parent = void 0;
3543
+ }
3544
+ if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3545
+ return parent;
3546
+ }
3547
+ /**
3548
+ * Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
3549
+ * status PATCH can carry it into the "Live sessions" list. Driver-level cache so
3550
+ * BOTH the watcher completion path and the restart-recovery re-adopt path (which
3551
+ * has no watcher) can use it. `conversationId` is passed only for log context.
3552
+ * Best-effort:
3553
+ * - a resolved NON-EMPTY title is cached and terminal (a real session name
3554
+ * won't later un-name), so we do NOT re-GET `/session/:id` every tick;
3555
+ * - while the title is still absent/empty we do NOT latch it — OpenCode names
3556
+ * sessions asynchronously mid-turn, so an early call (e.g. at `processing`)
3557
+ * must leave the cache unresolved and re-fetch on the next need so a later
3558
+ * call (e.g. at `done`) picks up the name assigned in the meantime. Such a
3559
+ * call returns `null` (omit the title on THIS PATCH) without caching;
3560
+ * - a failed request likewise leaves the cache unresolved (retry next need)
3561
+ * and returns `null` — it must NEVER throw or block completion.
3562
+ * A failure is logged with agent/session context (no silent catch).
3563
+ */
3564
+ async resolveSessionTitle(sessionId, conversationId) {
3565
+ const cached = this.sessionTitles.get(sessionId);
3566
+ if (cached != null) return cached;
3567
+ try {
3568
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
3569
+ if (res.ok) {
3570
+ const body = await res.json();
3571
+ const title = body && typeof body.title === "string" ? body.title.trim() : "";
3572
+ if (title.length > 0) {
3573
+ this.sessionTitles.set(sessionId, title);
3574
+ return title;
3575
+ }
3576
+ return null;
3577
+ }
3578
+ this.log({
3579
+ level: "debug",
3580
+ message: `Session title fetch for session ${sessionId.slice(0, 8)} (agent ${this.agentId.slice(0, 8)}) returned HTTP ${res.status} \u2014 omitting title`,
3581
+ conversation_id: conversationId
3582
+ });
3583
+ } catch (err) {
3584
+ this.log({
3585
+ level: "debug",
3586
+ message: `Best-effort session title fetch failed for session ${sessionId.slice(0, 8)} (agent ${this.agentId.slice(0, 8)}) \u2014 omitting title: ${err instanceof Error ? err.message : String(err)}`,
3587
+ conversation_id: conversationId
3588
+ });
3589
+ }
3590
+ return null;
3591
+ }
3592
+ /**
3593
+ * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3594
+ * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
3595
+ *
3596
+ * The PRIMARY recovery trigger is "preamble-pinned on recovery ⇒ idle" — a
3597
+ * runner restart wipes OpenCode's in-memory `SessionStatus`/`Runner`, so a
3598
+ * completed `finish: "tool-calls"` root reply encountered during re-adoption is
3599
+ * idle by OpenCode's own definition and is re-dispatched. This method exists only
3600
+ * so the WI-3 caller can VETO that re-dispatch in the rare case a descendant is
3601
+ * provably in flight at the exact moment of recovery.
3602
+ *
3603
+ * "Alive" criterion (TIGHTENED): a descendant is alive only when it is PROVABLY,
3604
+ * ACTIVELY generating — its LAST message is an assistant still mid-generation
3605
+ * (`completed == null`, via `isSessionActivelyGenerating`). An
3606
+ * INCOMPLETE-BUT-NOT-GENERATING child — last message a user message, or a
3607
+ * completed `finish: "tool-calls"` step — is NOT alive after a restart (nothing
3608
+ * is generating once the runner is gone), so it does NOT veto. (This is
3609
+ * deliberately NOT `!isTurnComplete`, which also matches those dead-but-non-terminal
3610
+ * shapes and would falsely veto — re-hanging the very turn this path recovers.)
3611
+ *
3612
+ * Return contract (encoded so WI-3 need not re-derive it):
3613
+ * - `true` → a descendant is provably, actively generating (veto re-dispatch).
3614
+ * - `false` → descendants exist but none is actively generating (the restart
3615
+ * case), OR no descendant is found at all.
3616
+ * - `null` → liveness is INDETERMINATE (enumeration via `listSessions` failed).
3617
+ *
3618
+ * ⚠️ `null` (UNKNOWN) MUST NOT be treated as "alive": WI-3 treats `null` the same
3619
+ * as `false` and does NOT veto — a restart guarantees no live runner, so an
3620
+ * indeterminate cross-check almost always means "couldn't reach a child that no
3621
+ * longer exists". The inversion lives in the caller; this method just reports
3622
+ * true/false/null faithfully.
3623
+ *
3624
+ * VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
3625
+ * (already proven by the existing child-session interaction tests, via
3626
+ * `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
3627
+ * terminal state. We do NOT depend on any session-level `busy`/`idle` field —
3628
+ * there is none on `GET /session/:id`; OpenCode's busy state is in-memory
3629
+ * `SessionStatus` only.
3630
+ */
3631
+ async isAnyDescendantSessionAlive(rootSessionId) {
3632
+ const sessions = await listSessions(this.port);
3633
+ if (!sessions) {
3634
+ this.log({
3635
+ level: "warn",
3636
+ message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
3637
+ });
3638
+ return null;
3639
+ }
3640
+ for (const candidate of sessions) {
3641
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
3642
+ if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
3643
+ const childMsgs = await getSessionMessages(this.port, candidate.id);
3644
+ if (isSessionActivelyGenerating(childMsgs)) {
3645
+ return true;
3646
+ }
3647
+ }
3648
+ return false;
3649
+ }
3650
+ /**
3651
+ * Cheap decision-telemetry label for a running row's LAST correlated reply
3652
+ * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
3653
+ * - `b1` — the reply itself is still in flight (`time.completed == null`) —
3654
+ * the aborted-in-flight production bug after a restart.
3655
+ * - `b2` — a COMPLETED reply pinned running only by `finish === "tool-calls"`
3656
+ * (the sub-agent preamble — #253's shape).
3657
+ * - `other` — any other shape (defensive; a running row is normally b1 or b2).
3658
+ * Reads `info.time.completed` / `info.finish` (tolerating the legacy top-level
3659
+ * shape) directly rather than re-importing the module-private `completedOf`/
3660
+ * `finishOf` — this is a display label only, not a correctness predicate.
3661
+ */
3662
+ replyCompletionShape(reply) {
3663
+ if (!reply) return "other";
3664
+ const completed = reply.info?.time?.completed ?? reply.time?.completed;
3665
+ if (completed == null) return "b1";
3666
+ const finish = reply.info?.finish ?? reply.finish;
3667
+ return finish === "tool-calls" ? "b2" : "other";
2032
3668
  }
2033
3669
  /**
2034
3670
  * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
@@ -2080,9 +3716,7 @@ var ChannelDriver = class {
2080
3716
  }
2081
3717
  return inFlight.sort(byOldest)[0];
2082
3718
  }
2083
- // -------------------------------------------------------------------------
2084
3719
  // Evident API calls (combinedAuth thread routes)
2085
- // -------------------------------------------------------------------------
2086
3720
  async getPendingConversations() {
2087
3721
  const res = await this.fetchImpl(
2088
3722
  `${this.apiUrl}/agents/${this.agentId}/conversations/pending`,
@@ -2112,6 +3746,35 @@ var ChannelDriver = class {
2112
3746
  }
2113
3747
  return await res.json();
2114
3748
  }
3749
+ /**
3750
+ * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
3751
+ * The pending path (`getPendingConversations`/`getPendingMessages`) only
3752
+ * surfaces `pending` rows, so a message already `processing` when the runner
3753
+ * died is invisible to it — this dedicated endpoint returns exactly those rows
3754
+ * with the fields the re-adopt path needs (`processed_at`,
3755
+ * `opencode_session_id`, routing).
3756
+ *
3757
+ * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
3758
+ * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
3759
+ * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
3760
+ * next tick retries.
3761
+ */
3762
+ async getProcessingMessages() {
3763
+ const res = await this.fetchImpl(
3764
+ `${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
3765
+ { headers: { Authorization: this.getAuthHeader() } }
3766
+ );
3767
+ this.assertAuth(res, "fetching processing messages");
3768
+ if (!res.ok) {
3769
+ throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
3770
+ }
3771
+ const data = await res.json();
3772
+ let messages = data.messages ?? [];
3773
+ if (this.conversationFilter) {
3774
+ messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
3775
+ }
3776
+ return messages;
3777
+ }
2115
3778
  /**
2116
3779
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
2117
3780
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -2133,13 +3796,18 @@ var ChannelDriver = class {
2133
3796
  * A single attempt (no internal retry): the watcher's per-tick loop is the
2134
3797
  * retry vehicle for the swap-to-running.
2135
3798
  */
2136
- async markProcessing(conversationId, messageId, sessionId) {
3799
+ async markProcessing(conversationId, messageId, sessionId, opencodeMessageId, title) {
2137
3800
  const res = await this.fetchImpl(
2138
3801
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
2139
3802
  {
2140
3803
  method: "PATCH",
2141
3804
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2142
- body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
3805
+ body: JSON.stringify({
3806
+ status: "processing",
3807
+ opencode_session_id: sessionId,
3808
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3809
+ ...title ? { title } : {}
3810
+ })
2143
3811
  }
2144
3812
  );
2145
3813
  this.assertAuth(res, "marking message as processing");
@@ -2177,13 +3845,18 @@ var ChannelDriver = class {
2177
3845
  * watcher retries next tick within the
2178
3846
  * deadline, Finding 4).
2179
3847
  */
2180
- async markDone(conversationId, messageId, sessionId) {
3848
+ async markDone(conversationId, messageId, sessionId, opencodeMessageId, title) {
2181
3849
  const res = await this.fetchImpl(
2182
3850
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
2183
3851
  {
2184
3852
  method: "PATCH",
2185
3853
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2186
- body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
3854
+ body: JSON.stringify({
3855
+ status: "done",
3856
+ opencode_session_id: sessionId,
3857
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3858
+ ...title ? { title } : {}
3859
+ })
2187
3860
  }
2188
3861
  );
2189
3862
  this.assertAuth(res, "marking message as done");
@@ -2193,7 +3866,17 @@ var ChannelDriver = class {
2193
3866
  }
2194
3867
  throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
2195
3868
  }
2196
- async markFailed(conversationId, messageId) {
3869
+ /**
3870
+ * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3871
+ * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3872
+ * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3873
+ * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3874
+ * failure reason reaches the channel.
3875
+ */
3876
+ async markFailed(conversationId, messageId, sessionId, error2) {
3877
+ const body = { status: "failed" };
3878
+ if (sessionId !== void 0) body.opencode_session_id = sessionId;
3879
+ if (error2 !== void 0) body.error = error2;
2197
3880
  await this.callWithRetry(
2198
3881
  "marking message as failed",
2199
3882
  () => this.fetchImpl(
@@ -2201,11 +3884,56 @@ var ChannelDriver = class {
2201
3884
  {
2202
3885
  method: "PATCH",
2203
3886
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2204
- body: JSON.stringify({ status: "failed" })
3887
+ body: JSON.stringify(body)
2205
3888
  }
2206
3889
  )
2207
3890
  );
2208
3891
  }
3892
+ /**
3893
+ * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3894
+ * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
3895
+ * — the server records it via `log()` (no DB write, no notification). This is
3896
+ * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
3897
+ * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
3898
+ * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
3899
+ * context (no silent catch, per development-workflow).
3900
+ *
3901
+ * Returns whether the POST SUCCEEDED (2xx). Most callers ignore this (pure
3902
+ * telemetry), but the `paused` liveness-clear uses it to know whether to
3903
+ * RE-ASSERT on a later tick — a single dropped `paused` POST must not leave a
3904
+ * stale `last_seen_alive_at` on a still-paused row (Bugbot "Failed paused signal
3905
+ * leaves liveness").
3906
+ */
3907
+ async postSignal(conversationId, messageId, signal, extra) {
3908
+ try {
3909
+ const res = await this.fetchImpl(
3910
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
3911
+ {
3912
+ method: "POST",
3913
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3914
+ body: JSON.stringify({ signal, ...extra })
3915
+ }
3916
+ );
3917
+ if (!res.ok) {
3918
+ this.log({
3919
+ level: "warn",
3920
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
3921
+ conversation_id: conversationId,
3922
+ message_id: messageId
3923
+ });
3924
+ return false;
3925
+ }
3926
+ return true;
3927
+ } catch (err) {
3928
+ this.log({
3929
+ level: "warn",
3930
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
3931
+ conversation_id: conversationId,
3932
+ message_id: messageId
3933
+ });
3934
+ return false;
3935
+ }
3936
+ }
2209
3937
  async persistSession(conversationId, sessionId) {
2210
3938
  const res = await this.fetchImpl(
2211
3939
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}`,
@@ -2260,9 +3988,7 @@ var ChannelDriver = class {
2260
3988
  return false;
2261
3989
  }
2262
3990
  }
2263
- // -------------------------------------------------------------------------
2264
3991
  // Retry wrapper
2265
- // -------------------------------------------------------------------------
2266
3992
  /**
2267
3993
  * Invoke an Evident API call, retrying on transient failures (5xx / 429 /
2268
3994
  * network errors) with exponential backoff + jitter (capped). Auth failures
@@ -2461,7 +4187,7 @@ async function resolveAgentIdFromKey(authHeader) {
2461
4187
  if (!response.ok) {
2462
4188
  const serverMessage = await readErrorMessage(response);
2463
4189
  return {
2464
- error: `Failed to resolve agent from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
4190
+ error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
2465
4191
  };
2466
4192
  }
2467
4193
  const data = await response.json();
@@ -2469,11 +4195,30 @@ async function resolveAgentIdFromKey(authHeader) {
2469
4195
  return { agent_id: data.agent_id };
2470
4196
  }
2471
4197
  return {
2472
- error: "Cannot resolve agent ID: auth type is not agent_key. Please provide --agent explicitly."
4198
+ error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
2473
4199
  };
2474
4200
  } catch (error2) {
2475
4201
  const message = error2 instanceof Error ? error2.message : "Unknown error";
2476
- return { error: `Failed to resolve agent from key: ${message}` };
4202
+ return { error: `Failed to resolve runner from key: ${message}` };
4203
+ }
4204
+ }
4205
+ async function notifyAgentDisconnected(agentId, authHeader) {
4206
+ const apiUrl = getApiUrlConfig();
4207
+ try {
4208
+ const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
4209
+ method: "POST",
4210
+ headers: { Authorization: authHeader }
4211
+ });
4212
+ if (!response.ok) {
4213
+ const serverMessage = await readErrorMessage(response);
4214
+ return {
4215
+ ok: false,
4216
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
4217
+ };
4218
+ }
4219
+ return { ok: true };
4220
+ } catch (error2) {
4221
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
2477
4222
  }
2478
4223
  }
2479
4224
  async function getAgentInfo(agentId, authHeader) {
@@ -2490,12 +4235,12 @@ async function getAgentInfo(agentId, authHeader) {
2490
4235
  const serverMessage = await readErrorMessage(response);
2491
4236
  return {
2492
4237
  valid: false,
2493
- error: serverMessage ?? "You do not have access to this agent (it may belong to a different team or organization)."
4238
+ error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
2494
4239
  };
2495
4240
  }
2496
4241
  if (response.status === 404) {
2497
4242
  const serverMessage = await readErrorMessage(response);
2498
- return { valid: false, error: serverMessage ?? `Agent ${agentId} not found` };
4243
+ return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
2499
4244
  }
2500
4245
  if (!response.ok) {
2501
4246
  const serverMessage = await readErrorMessage(response);
@@ -2508,36 +4253,68 @@ async function getAgentInfo(agentId, authHeader) {
2508
4253
  if (agent.agent_type !== "local") {
2509
4254
  return {
2510
4255
  valid: false,
2511
- error: `Agent is type '${agent.agent_type}', must be 'local' for CLI connection`
4256
+ error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
2512
4257
  };
2513
4258
  }
2514
4259
  return { valid: true, agent };
2515
4260
  } catch (error2) {
2516
4261
  const message = error2 instanceof Error ? error2.message : "Unknown error";
2517
- return { valid: false, error: `Failed to validate agent: ${message}` };
4262
+ return { valid: false, error: `Failed to validate runner: ${message}` };
2518
4263
  }
2519
4264
  }
2520
4265
 
2521
4266
  // src/commands/run.ts
2522
4267
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
2523
4268
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
2524
- function log(state, message, isError = false) {
4269
+ var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4270
+ var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
4271
+ function resolveLogLevel(options) {
4272
+ const accepted = Object.keys(LOG_LEVELS);
4273
+ const validate = (value, source) => {
4274
+ const normalized = value.trim().toLowerCase();
4275
+ if (!accepted.includes(normalized)) {
4276
+ throw new Error(
4277
+ `Invalid log level "${value}"${source}; expected one of ${accepted.join(", ")}`
4278
+ );
4279
+ }
4280
+ return normalized;
4281
+ };
4282
+ if (options.logLevel !== void 0) {
4283
+ return validate(options.logLevel, " (--log-level)");
4284
+ }
4285
+ if (options.verbose) {
4286
+ return "debug";
4287
+ }
4288
+ const env = process.env.EVIDENT_LOG_LEVEL;
4289
+ if (env !== void 0 && env !== "") {
4290
+ return validate(env, " (EVIDENT_LOG_LEVEL)");
4291
+ }
4292
+ return "info";
4293
+ }
4294
+ function meetsThreshold(state, level) {
4295
+ return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
4296
+ }
4297
+ function log2(state, message, level = "info") {
4298
+ if (!meetsThreshold(state, level)) return;
2525
4299
  if (state.json) {
2526
4300
  console.log(
2527
4301
  JSON.stringify({
2528
4302
  timestamp: (/* @__PURE__ */ new Date()).toISOString(),
2529
- level: isError ? "error" : "info",
4303
+ level,
2530
4304
  message
2531
4305
  })
2532
4306
  );
2533
4307
  } else if (!state.interactive) {
2534
- const prefix = isError ? chalk6.red("\u2717") : chalk6.green("\u2022");
4308
+ const prefix = level === "error" ? chalk6.red("\u2717") : level === "warn" ? chalk6.yellow("!") : level === "debug" ? chalk6.dim("\xB7") : chalk6.green("\u2022");
2535
4309
  console.log(`${prefix} ${message}`);
2536
4310
  }
2537
4311
  }
2538
4312
  function logActivity(state, entry) {
4313
+ const level = entry.level ?? (entry.type === "error" ? "error" : "info");
4314
+ if (!meetsThreshold(state, level)) return;
2539
4315
  const fullEntry = {
2540
4316
  ...entry,
4317
+ level,
2541
4318
  timestamp: /* @__PURE__ */ new Date()
2542
4319
  };
2543
4320
  state.activityLog.push(fullEntry);
@@ -2546,9 +4323,9 @@ function logActivity(state, entry) {
2546
4323
  }
2547
4324
  if (!state.interactive) {
2548
4325
  if (entry.type === "error") {
2549
- log(state, entry.error ?? "Unknown error", true);
2550
- } else if (entry.type === "info" && entry.message) {
2551
- log(state, entry.message);
4326
+ log2(state, entry.error ?? "Unknown error", level);
4327
+ } else if (entry.message) {
4328
+ log2(state, entry.message, level);
2552
4329
  }
2553
4330
  }
2554
4331
  }
@@ -2634,6 +4411,7 @@ async function handleAuthError(state, error2) {
2634
4411
  }
2635
4412
  async function driveChannels(state, driver) {
2636
4413
  let idlePolls = 0;
4414
+ let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
2637
4415
  while (state.running) {
2638
4416
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
2639
4417
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2643,7 +4421,9 @@ async function driveChannels(state, driver) {
2643
4421
  try {
2644
4422
  const processed = await driver.drainPending();
2645
4423
  state.messageCount += processed;
2646
- if (processed > 0 || driver.hasInFlightWatchers()) {
4424
+ const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4425
+ lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4426
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
2647
4427
  idlePolls = 0;
2648
4428
  if (processed > 0 && state.interactive) displayStatus(state);
2649
4429
  } else if (state.idleTimeout !== null) {
@@ -2672,7 +4452,7 @@ async function driveChannels(state, driver) {
2672
4452
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
2673
4453
  if (state.interactive) displayStatus(state);
2674
4454
  }
2675
- await new Promise((resolve) => setTimeout(resolve, CHANNEL_POLL_INTERVAL_MS));
4455
+ await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
2676
4456
  if (state.idleTimeout !== null && idlePolls >= 2) {
2677
4457
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
2678
4458
  if (idleMs > state.idleTimeout * 1e3) {
@@ -2683,8 +4463,122 @@ async function driveChannels(state, driver) {
2683
4463
  }
2684
4464
  }
2685
4465
  }
2686
- async function cleanup(state) {
4466
+ var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
4467
+ async function runSweep(state, driver, config2) {
4468
+ const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
4469
+ try {
4470
+ const sessions = await listSessions(state.port);
4471
+ if (sessions === null) {
4472
+ logActivity(state, {
4473
+ type: "info",
4474
+ message: `Session cleanup: could not list sessions (opencode unreachable); skipping this sweep (${mode})`
4475
+ });
4476
+ return;
4477
+ }
4478
+ const toDelete = selectSessionsToDelete(
4479
+ sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
4480
+ {
4481
+ maxAgeMs: config2.maxAgeMs,
4482
+ maxCount: config2.maxCount,
4483
+ nowMs: Date.now(),
4484
+ protectedIds: driver.protectedSessionIds()
4485
+ }
4486
+ );
4487
+ const protectedNow = driver.protectedSessionIds();
4488
+ let deleted = 0;
4489
+ let failed = 0;
4490
+ let skippedNewlyActive = 0;
4491
+ for (const id of toDelete) {
4492
+ if (protectedNow.has(id)) {
4493
+ skippedNewlyActive++;
4494
+ logActivity(state, {
4495
+ type: "info",
4496
+ message: `Session cleanup: skipping ${id} \u2014 became active/bound after selection (${mode})`
4497
+ });
4498
+ continue;
4499
+ }
4500
+ if (await deleteSession(state.port, id)) deleted++;
4501
+ else failed++;
4502
+ }
4503
+ const failedNote = failed > 0 ? `, failed ${failed}` : "";
4504
+ const skippedNote = skippedNewlyActive > 0 ? `, skipped ${skippedNewlyActive} newly-active` : "";
4505
+ logActivity(state, {
4506
+ type: "info",
4507
+ message: `Session cleanup: inspected ${sessions.length}, deleted ${deleted}${failedNote}${skippedNote} (${mode})`
4508
+ });
4509
+ } catch (error2) {
4510
+ const message = error2 instanceof Error ? error2.message : String(error2);
4511
+ logActivity(state, {
4512
+ type: "error",
4513
+ error: `Session cleanup sweep failed (non-fatal, ${mode}): ${message}`
4514
+ });
4515
+ }
4516
+ }
4517
+ function scheduleSessionCleanup(state, driver, options) {
4518
+ const config2 = resolveSessionCleanupConfig(
4519
+ {
4520
+ maxAge: options.sessionCleanupMaxAge,
4521
+ maxCount: options.sessionCleanupMaxCount,
4522
+ interval: options.sessionCleanupInterval
4523
+ },
4524
+ process.env
4525
+ );
4526
+ for (const warning2 of config2.warnings) {
4527
+ logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
4528
+ }
4529
+ if (!config2.enabled) return;
4530
+ logActivity(state, {
4531
+ type: "info",
4532
+ message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
4533
+ });
4534
+ const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
4535
+ const firstSweep = setTimeout(
4536
+ () => void runSweep(state, driver, config2),
4537
+ SESSION_CLEANUP_FIRST_SWEEP_MS
4538
+ );
4539
+ state.sessionCleanupTimers.push(interval, firstSweep);
4540
+ }
4541
+ async function notifyOffline(state) {
4542
+ if (!state.agentId || !state.authHeader) return;
4543
+ if (!state.connected) {
4544
+ log2(state, "Skipping offline signal \u2014 this runner does not hold the live tunnel");
4545
+ return;
4546
+ }
4547
+ const result = await notifyAgentDisconnected(state.agentId, state.authHeader);
4548
+ if (result.ok) {
4549
+ log2(state, "Notified Evident the runner is going offline");
4550
+ } else {
4551
+ logActivity(state, {
4552
+ type: "error",
4553
+ error: `Could not notify Evident of offline status (relay will still report it): ${result.error}`
4554
+ });
4555
+ if (state.interactive) displayStatus(state);
4556
+ }
4557
+ }
4558
+ async function cleanup(state, opts = {}) {
2687
4559
  state.running = false;
4560
+ for (const timer of state.sessionCleanupTimers) {
4561
+ clearInterval(timer);
4562
+ clearTimeout(timer);
4563
+ }
4564
+ state.sessionCleanupTimers = [];
4565
+ if (opts.graceful && state.channelDriver) {
4566
+ state.channelDriver.stop();
4567
+ log2(state, "Draining in-flight channel work before shutdown...");
4568
+ if (state.interactive) {
4569
+ logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4570
+ displayStatus(state);
4571
+ }
4572
+ const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
4573
+ if (!settled) {
4574
+ logActivity(state, {
4575
+ type: "info",
4576
+ message: "Shutdown drain timed out with work still in flight \u2014 leaving it for restart recovery"
4577
+ });
4578
+ if (state.interactive) displayStatus(state);
4579
+ }
4580
+ }
4581
+ await notifyOffline(state);
2688
4582
  if (state.connection) {
2689
4583
  state.connection.close();
2690
4584
  state.connection = null;
@@ -2695,13 +4589,27 @@ async function cleanup(state) {
2695
4589
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
2696
4590
  displayStatus(state);
2697
4591
  } else {
2698
- log(state, "Stopped OpenCode process");
4592
+ log2(state, "Stopped OpenCode process");
2699
4593
  }
2700
4594
  state.opencodeProcess = null;
2701
4595
  }
2702
4596
  }
2703
4597
  async function run(options) {
2704
4598
  const interactive = isInteractive(options.json);
4599
+ let logLevel;
4600
+ try {
4601
+ logLevel = resolveLogLevel(options);
4602
+ } catch (error2) {
4603
+ const message = error2 instanceof Error ? error2.message : String(error2);
4604
+ if (options.json) {
4605
+ console.log(JSON.stringify({ status: "error", error: message }));
4606
+ } else {
4607
+ printError(message);
4608
+ }
4609
+ await shutdownTelemetry();
4610
+ process.exit(1);
4611
+ return;
4612
+ }
2705
4613
  const state = {
2706
4614
  agentId: options.agent || "",
2707
4615
  agentName: null,
@@ -2710,31 +4618,38 @@ async function run(options) {
2710
4618
  idleTimeout: options.idleTimeout ?? null,
2711
4619
  json: options.json ?? false,
2712
4620
  interactive,
4621
+ logLevel,
2713
4622
  connected: false,
2714
4623
  opencodeConnected: false,
2715
4624
  opencodeVersion: null,
2716
4625
  opencodeProcess: null,
2717
4626
  connection: null,
4627
+ channelDriver: null,
2718
4628
  running: true,
4629
+ shuttingDown: false,
2719
4630
  activityLog: [],
2720
4631
  messageCount: 0,
4632
+ lastProxiedActivityAt: null,
4633
+ sessionCleanupTimers: [],
2721
4634
  authHeader: ""
2722
4635
  };
2723
4636
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
2724
- log(
4637
+ log2(
2725
4638
  state,
2726
- "Warning: No --idle-timeout set in CI environment. The runner will poll indefinitely until the job times out. Consider adding --idle-timeout 30 to avoid wasting runner minutes.",
2727
- false
4639
+ "No --idle-timeout set in CI environment. The runner will poll indefinitely until the job times out. Consider adding --idle-timeout 30 to avoid wasting runner minutes.",
4640
+ "warn"
2728
4641
  );
2729
4642
  }
2730
4643
  const handleSignal = async () => {
4644
+ if (state.shuttingDown) return;
4645
+ state.shuttingDown = true;
2731
4646
  if (state.interactive) {
2732
4647
  logActivity(state, { type: "info", message: "Shutting down..." });
2733
4648
  displayStatus(state);
2734
4649
  } else {
2735
- log(state, "Shutting down...");
4650
+ log2(state, "Shutting down...");
2736
4651
  }
2737
- await cleanup(state);
4652
+ await cleanup(state, { graceful: true });
2738
4653
  await shutdownTelemetry();
2739
4654
  process.exit(0);
2740
4655
  };
@@ -2765,15 +4680,15 @@ async function run(options) {
2765
4680
  const resolved = await resolveAgentIdFromKey(state.authHeader);
2766
4681
  if (resolved.agent_id) {
2767
4682
  state.agentId = resolved.agent_id;
2768
- log(state, `Resolved agent ID from key: ${state.agentId}`);
4683
+ log2(state, `Resolved runner ID from key: ${state.agentId}`);
2769
4684
  if (state.interactive && !state.json) {
2770
4685
  logActivity(state, {
2771
4686
  type: "info",
2772
- message: `Agent ID resolved from key: ${state.agentId}`
4687
+ message: `Runner ID resolved from key: ${state.agentId}`
2773
4688
  });
2774
4689
  }
2775
4690
  } else {
2776
- printError(resolved.error || "Failed to resolve agent ID from key");
4691
+ printError(resolved.error || "Failed to resolve runner ID from key");
2777
4692
  process.exit(1);
2778
4693
  }
2779
4694
  } else {
@@ -2801,7 +4716,7 @@ async function run(options) {
2801
4716
  console.log(chalk6.bold("Evident Run"));
2802
4717
  console.log(chalk6.dim("-".repeat(40)));
2803
4718
  }
2804
- const spinner = interactive && !state.json ? ora3("Validating agent...").start() : null;
4719
+ const spinner = interactive && !state.json ? ora3("Validating runner...").start() : null;
2805
4720
  let validation = await getAgentInfo(state.agentId, state.authHeader);
2806
4721
  if (!validation.valid && validation.authFailed && interactive) {
2807
4722
  spinner?.fail("Authentication failed");
@@ -2813,14 +4728,14 @@ async function run(options) {
2813
4728
  "Login successful! Retrying..."
2814
4729
  );
2815
4730
  state.authHeader = getAuthHeader(credentials2);
2816
- spinner?.start("Validating agent...");
4731
+ spinner?.start("Validating runner...");
2817
4732
  validation = await getAgentInfo(state.agentId, state.authHeader);
2818
4733
  }
2819
4734
  if (!validation.valid) {
2820
- spinner?.fail(`Agent validation failed: ${validation.error}`);
4735
+ spinner?.fail(`Runner validation failed: ${validation.error}`);
2821
4736
  throw new Error(validation.error);
2822
4737
  }
2823
- spinner?.succeed(`Agent: ${validation.agent.name || state.agentId}`);
4738
+ spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
2824
4739
  state.agentName = validation.agent.name;
2825
4740
  const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
2826
4741
  try {
@@ -2828,19 +4743,19 @@ async function run(options) {
2828
4743
  port: state.port,
2829
4744
  interactive: state.interactive,
2830
4745
  agentId: state.agentId,
2831
- log: (message) => log(state, message)
4746
+ log: (message) => log2(state, message)
2832
4747
  });
2833
4748
  state.port = oc.port;
2834
4749
  state.opencodeProcess = oc.process;
2835
4750
  state.opencodeVersion = oc.version;
2836
4751
  state.opencodeConnected = oc.process !== null || oc.version !== null;
2837
- const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
2838
- ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
4752
+ const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
4753
+ ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
2839
4754
  const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
2840
4755
  if (versionWarning) {
2841
- log(state, versionWarning, false);
4756
+ log2(state, versionWarning, "warn");
2842
4757
  if (state.interactive && !state.json) {
2843
- logActivity(state, { type: "info", message: versionWarning });
4758
+ logActivity(state, { type: "info", level: "warn", message: versionWarning });
2844
4759
  }
2845
4760
  }
2846
4761
  } catch (error2) {
@@ -2854,12 +4769,20 @@ async function run(options) {
2854
4769
  apiUrl: getApiUrlConfig(),
2855
4770
  getAuthHeader: () => state.authHeader,
2856
4771
  conversationFilter: state.conversationFilter,
2857
- log: (entry) => logActivity(state, {
2858
- type: entry.level === "error" ? "error" : "info",
2859
- message: entry.message,
2860
- error: entry.level === "error" ? entry.message : void 0
2861
- })
4772
+ stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
4773
+ log: (entry) => (
4774
+ // Thread the driver's real level straight through so `debug`/`warn`
4775
+ // survive the sink filter (they no longer collapse to info). `type`
4776
+ // stays the coarse error/non-error split the activity log renders with.
4777
+ logActivity(state, {
4778
+ type: entry.level === "error" ? "error" : "info",
4779
+ level: entry.level,
4780
+ message: entry.message,
4781
+ error: entry.level === "error" ? entry.message : void 0
4782
+ })
4783
+ )
2862
4784
  });
4785
+ state.channelDriver = channelDriver;
2863
4786
  const connection = new RunnerConnection({
2864
4787
  agentId: state.agentId,
2865
4788
  getAuthHeader: () => state.authHeader,
@@ -2871,9 +4794,13 @@ async function run(options) {
2871
4794
  state.agentId = agentId;
2872
4795
  logActivity(state, {
2873
4796
  type: "info",
2874
- message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
4797
+ message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
4798
+ });
4799
+ emitAgentConnected(state.agentId, {
4800
+ port: state.port,
4801
+ cli_version: getCliVersion(),
4802
+ opencode_version: state.opencodeVersion
2875
4803
  });
2876
- emitAgentConnected(state.agentId, { port: state.port });
2877
4804
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
2878
4805
  if (state.interactive) displayStatus(state);
2879
4806
  channelDriver.drainPending().then((processed) => {
@@ -2907,9 +4834,14 @@ async function run(options) {
2907
4834
  logActivity(state, { type: "error", error: error2 });
2908
4835
  if (state.interactive) displayStatus(state);
2909
4836
  },
2910
- // Web traffic is proxied transparently; only note opencode is live.
4837
+ // Web traffic is proxied transparently; note opencode is live and stamp
4838
+ // proxied activity so the idle loop treats interactive proxy use as work.
4839
+ // Fires per forwarded response head (incl. every SSE open) and excludes
4840
+ // the internal drain-ping, so an actively-used proxy keeps the timer
4841
+ // fresh while a lone idle SSE with no follow-up requests still ages out.
2911
4842
  onResponse: () => {
2912
4843
  state.opencodeConnected = true;
4844
+ state.lastProxiedActivityAt = Date.now();
2913
4845
  },
2914
4846
  // A channel message was queued and the api-worker pinged us over the
2915
4847
  // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2947,10 +4879,12 @@ async function run(options) {
2947
4879
  if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
2948
4880
  throw error2;
2949
4881
  }
4882
+ scheduleSessionCleanup(state, channelDriver, options);
2950
4883
  if (!interactive || state.json) {
2951
- log(state, "Driving channel messages...");
4884
+ log2(state, "Driving channel messages...");
2952
4885
  }
2953
4886
  await driveChannels(state, channelDriver);
4887
+ if (state.shuttingDown) return;
2954
4888
  await cleanup(state);
2955
4889
  if (state.json) {
2956
4890
  console.log(
@@ -2960,11 +4894,12 @@ async function run(options) {
2960
4894
  })
2961
4895
  );
2962
4896
  } else if (!interactive) {
2963
- log(state, `Completed. Processed ${state.messageCount} message(s).`);
4897
+ log2(state, `Completed. Processed ${state.messageCount} message(s).`);
2964
4898
  }
2965
4899
  await shutdownTelemetry();
2966
4900
  process.exit(0);
2967
4901
  } catch (error2) {
4902
+ if (state.shuttingDown) return;
2968
4903
  await cleanup(state);
2969
4904
  const message = error2 instanceof Error ? error2.message : String(error2);
2970
4905
  if (state.json) {
@@ -2982,8 +4917,9 @@ async function run(options) {
2982
4917
  }
2983
4918
 
2984
4919
  // src/index.ts
4920
+ var { version } = createRequire(import.meta.url)("../package.json");
2985
4921
  var program = new Command();
2986
- program.name("evident").description("Run OpenCode locally and connect it to Evident").version("0.1.0").option(
4922
+ program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
2987
4923
  "--endpoint <url>",
2988
4924
  "Evident API base URL (default: production; e.g. http://localhost:3001)"
2989
4925
  ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {
@@ -2998,15 +4934,34 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
2998
4934
  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);
2999
4935
  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 }));
3000
4936
  program.command("whoami").description("Show the currently logged in user").action(whoami);
3001
- program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Agent ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option("-v, --verbose", "Show detailed request/response information").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").action(
4937
+ 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("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
4938
+ "--log-level <level>",
4939
+ "Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
4940
+ ).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(
4941
+ "--session-cleanup-max-age <duration>",
4942
+ "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
4943
+ ).option(
4944
+ "--session-cleanup-max-count <n>",
4945
+ "Keep only the newest N OpenCode sessions. Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_COUNT"
4946
+ ).option(
4947
+ "--session-cleanup-interval <duration>",
4948
+ "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
4949
+ ).action(
3002
4950
  (options) => {
3003
4951
  run({
3004
4952
  agent: options.agent,
3005
4953
  port: parseInt(options.port, 10),
4954
+ // Raw string — validation/precedence is single-sourced in run.ts's
4955
+ // resolveLogLevel (flag > -v > EVIDENT_LOG_LEVEL > info).
4956
+ logLevel: options.logLevel,
3006
4957
  verbose: options.verbose,
3007
4958
  conversation: options.conversation,
3008
4959
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
3009
- json: options.json
4960
+ json: options.json,
4961
+ // Raw strings — the resolver in run.ts single-sources parsing (M1).
4962
+ sessionCleanupMaxAge: options.sessionCleanupMaxAge,
4963
+ sessionCleanupMaxCount: options.sessionCleanupMaxCount,
4964
+ sessionCleanupInterval: options.sessionCleanupInterval
3010
4965
  });
3011
4966
  }
3012
4967
  );