@evident-ai/cli 3.0.1-dev.90655e3 → 3.0.1-dev.95a86c0

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;
@@ -679,11 +706,24 @@ 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
 
714
+ // src/lib/opencode/opencode-version-gate.ts
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${version2}` : "unknown";
723
+ const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
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.`;
725
+ }
726
+
687
727
  // src/lib/opencode/process.ts
688
728
  import { execSync, spawn } from "child_process";
689
729
  var OPENCODE_PORT_RANGE = [4096, 4097, 4098, 4099, 4100];
@@ -996,7 +1036,37 @@ function roleOf(m) {
996
1036
  }
997
1037
  function completedOf(m) {
998
1038
  if (!m || typeof m !== "object") return void 0;
999
- 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;
1044
+ }
1045
+ function idOf(m) {
1046
+ if (!m || typeof m !== "object") return void 0;
1047
+ if (typeof m.id === "string") return m.id;
1048
+ const infoId = m.info?.id;
1049
+ return typeof infoId === "string" ? infoId : void 0;
1050
+ }
1051
+ function parentIdOf(m) {
1052
+ if (!m || typeof m !== "object") return void 0;
1053
+ if (typeof m.parentID === "string") return m.parentID;
1054
+ const infoParent = m.info?.parentID;
1055
+ return typeof infoParent === "string" ? infoParent : void 0;
1056
+ }
1057
+ function finishOf(m) {
1058
+ if (!m || typeof m !== "object") return void 0;
1059
+ if (typeof m.finish === "string") return m.finish;
1060
+ const infoFinish = m.info?.finish;
1061
+ return typeof infoFinish === "string" ? infoFinish : void 0;
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";
1000
1070
  }
1001
1071
  async function getSessionMessages(port, sessionId) {
1002
1072
  try {
@@ -1008,11 +1078,83 @@ async function getSessionMessages(port, sessionId) {
1008
1078
  return null;
1009
1079
  }
1010
1080
  }
1011
- function isTurnComplete(messages) {
1081
+ function isSessionActivelyGenerating(messages) {
1012
1082
  if (!messages || messages.length === 0) return false;
1013
1083
  const last = messages[messages.length - 1];
1014
1084
  if (roleOf(last) !== "assistant") return false;
1015
- return completedOf(last) != null;
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}): ${err instanceof Error ? err.message : String(err)}`
1149
+ );
1150
+ return null;
1151
+ }
1152
+ }
1153
+ async function isSessionOngoing(port, id) {
1154
+ const map = await getSessionStatuses(port);
1155
+ if (map == null) return null;
1156
+ const entry = map[id];
1157
+ return entry != null && entry.type !== "idle";
1016
1158
  }
1017
1159
  async function createOpenCodeSession(port, directory) {
1018
1160
  const url = new URL(`${opencodeBase(port)}/session`);
@@ -1031,7 +1173,15 @@ async function createOpenCodeSession(port, directory) {
1031
1173
  const data = await response.json();
1032
1174
  return data.id;
1033
1175
  }
1034
- async function sendMessageToOpenCode(port, sessionId, content, options, hooks, maxWaitMs = 10 * 60 * 1e3) {
1176
+ function messageText(m) {
1177
+ if (!m || !Array.isArray(m.parts)) return "";
1178
+ return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
1179
+ }
1180
+ async function sendPromptAsync(port, sessionId, content, options) {
1181
+ const before = await getSessionMessages(port, sessionId);
1182
+ const knownUserIds = new Set(
1183
+ (before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
1184
+ );
1035
1185
  const body = {
1036
1186
  parts: [{ type: "text", text: content }]
1037
1187
  };
@@ -1047,79 +1197,218 @@ async function sendMessageToOpenCode(port, sessionId, content, options, hooks, m
1047
1197
  };
1048
1198
  }
1049
1199
  }
1050
- let pollDone = false;
1051
- const reportedQuestions = /* @__PURE__ */ new Set();
1052
- const reportedPermissions = /* @__PURE__ */ new Set();
1053
- const pollInteractive = async () => {
1054
- while (!pollDone) {
1055
- await new Promise((resolve) => setTimeout(resolve, 1e3));
1056
- if (pollDone) break;
1057
- if (hooks?.onQuestion) {
1058
- try {
1059
- const res = await fetch(`${opencodeBase(port)}/question`);
1060
- if (res.ok) {
1061
- const questions = await res.json();
1062
- for (const q of questions) {
1063
- if (q.sessionID === sessionId && !reportedQuestions.has(q.id)) {
1064
- reportedQuestions.add(q.id);
1065
- await hooks.onQuestion(q);
1066
- }
1067
- }
1068
- }
1069
- } catch {
1070
- }
1071
- }
1072
- if (hooks?.onPermission) {
1073
- try {
1074
- const res = await fetch(`${opencodeBase(port)}/permission`);
1075
- if (res.ok) {
1076
- const permissions = await res.json();
1077
- for (const p of permissions) {
1078
- if (p.sessionID === sessionId && !reportedPermissions.has(p.id)) {
1079
- reportedPermissions.add(p.id);
1080
- await hooks.onPermission(p);
1081
- }
1082
- }
1083
- }
1084
- } catch {
1200
+ const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/prompt_async`, {
1201
+ method: "POST",
1202
+ headers: { "Content-Type": "application/json" },
1203
+ body: JSON.stringify(body)
1204
+ });
1205
+ if (res.status < 200 || res.status >= 300) {
1206
+ const text = await res.text().catch(() => "");
1207
+ throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1208
+ }
1209
+ const READ_BACK_ATTEMPTS = 5;
1210
+ const READ_BACK_DELAY_MS = 150;
1211
+ for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
1212
+ const after = await getSessionMessages(port, sessionId);
1213
+ if (after) {
1214
+ let best = null;
1215
+ for (const m of after) {
1216
+ if (roleOf(m) !== "user") continue;
1217
+ const id = idOf(m);
1218
+ if (typeof id !== "string" || knownUserIds.has(id)) continue;
1219
+ if (messageText(m) !== content) continue;
1220
+ const created = createdOf(m) ?? 0;
1221
+ if (best === null || created > best.created) {
1222
+ best = { id, created };
1085
1223
  }
1086
1224
  }
1225
+ if (best) return best.id;
1087
1226
  }
1227
+ if (attempt < READ_BACK_ATTEMPTS - 1) {
1228
+ await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1229
+ }
1230
+ }
1231
+ return null;
1232
+ }
1233
+ function findAssistantReplyAfter(messages, userMessageId) {
1234
+ if (!messages || messages.length === 0) return null;
1235
+ const byParent = messages.find(
1236
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) === userMessageId
1237
+ );
1238
+ if (byParent) return byParent;
1239
+ const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1240
+ if (userIndex === -1) return null;
1241
+ for (let i = userIndex + 1; i < messages.length; i++) {
1242
+ if (roleOf(messages[i]) === "assistant") return messages[i];
1243
+ }
1244
+ return null;
1245
+ }
1246
+ function findLastAssistantReplyFor(messages, userMessageId) {
1247
+ if (!messages || messages.length === 0) return null;
1248
+ let lastCorrelated = null;
1249
+ let lastNonErrored = null;
1250
+ for (let i = messages.length - 1; i >= 0; i--) {
1251
+ const m = messages[i];
1252
+ if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
1253
+ if (lastCorrelated === null) lastCorrelated = m;
1254
+ if (errorOf(m) == null) {
1255
+ lastNonErrored = m;
1256
+ break;
1257
+ }
1258
+ }
1259
+ if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
1260
+ const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1261
+ if (userIndex === -1) return null;
1262
+ let last = null;
1263
+ let lastOk = null;
1264
+ for (let i = userIndex + 1; i < messages.length; i++) {
1265
+ const role = roleOf(messages[i]);
1266
+ if (role === "user") break;
1267
+ if (role === "assistant") {
1268
+ last = messages[i];
1269
+ if (errorOf(messages[i]) == null) lastOk = messages[i];
1270
+ }
1271
+ }
1272
+ return lastOk ?? last;
1273
+ }
1274
+ function messageRunState(messages, userMessageId) {
1275
+ if (!messages || messages.length === 0) return "unknown";
1276
+ const hasUser = messages.some((m) => idOf(m) === userMessageId);
1277
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1278
+ if (!hasUser) {
1279
+ if (!reply) return "unknown";
1280
+ }
1281
+ if (!reply) return "queued";
1282
+ if (isAssistantInFlight(reply)) return "running";
1283
+ return errorOf(reply) != null ? "failed" : "done";
1284
+ }
1285
+ function isPreamblePinnedRunning(messages, userMessageId) {
1286
+ if (messageRunState(messages, userMessageId) !== "running") return false;
1287
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1288
+ return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1289
+ }
1290
+ function messageError(messages, userMessageId) {
1291
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1292
+ const error2 = errorOf(reply);
1293
+ if (error2 == null) return null;
1294
+ if (typeof error2 === "string") return error2;
1295
+ if (typeof error2 === "object") {
1296
+ const e = error2;
1297
+ const dataMessage = e.data?.message;
1298
+ if (typeof dataMessage === "string") return dataMessage;
1299
+ if (typeof e.message === "string") return e.message;
1300
+ }
1301
+ return "The agent run failed.";
1302
+ }
1303
+ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1304
+ if (!messages || messages.length === 0) return false;
1305
+ return messages.some(
1306
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1307
+ );
1308
+ }
1309
+
1310
+ // src/lib/opencode/session-cleanup.ts
1311
+ var DURATION_UNIT_MS = {
1312
+ s: 1e3,
1313
+ m: 60 * 1e3,
1314
+ h: 60 * 60 * 1e3,
1315
+ d: 24 * 60 * 60 * 1e3
1316
+ };
1317
+ function parseDurationMs(input) {
1318
+ const trimmed = input.trim();
1319
+ const match = /^(\d+)([smhd])$/.exec(trimmed);
1320
+ if (!match) {
1321
+ throw new Error(
1322
+ `Invalid duration "${input}": expected <number><unit> where unit is one of s, m, h, d (e.g. "7d", "24h", "30m", "90s").`
1323
+ );
1324
+ }
1325
+ const value = Number(match[1]);
1326
+ if (value <= 0) {
1327
+ throw new Error(`Invalid duration "${input}": must be a positive value.`);
1328
+ }
1329
+ return value * DURATION_UNIT_MS[match[2]];
1330
+ }
1331
+ function selectSessionsToDelete(sessions, opts) {
1332
+ const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
1333
+ if (maxAgeMs === void 0 && maxCount === void 0) return [];
1334
+ const ageEligible = (s) => {
1335
+ if (maxAgeMs === void 0) return false;
1336
+ if (s.lastActivityMs === null) return true;
1337
+ return nowMs - s.lastActivityMs > maxAgeMs;
1088
1338
  };
1089
- const sendMessage = async () => {
1090
- const controller = new AbortController();
1091
- const timer = setTimeout(() => controller.abort(), maxWaitMs);
1339
+ const countEligibleIds = /* @__PURE__ */ new Set();
1340
+ if (maxCount !== void 0) {
1341
+ const byActivityDesc = [...sessions].sort(
1342
+ (a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
1343
+ );
1344
+ for (const s of byActivityDesc.slice(maxCount)) {
1345
+ countEligibleIds.add(s.id);
1346
+ }
1347
+ }
1348
+ const toDelete = [];
1349
+ for (const s of sessions) {
1350
+ if (protectedIds.has(s.id)) continue;
1351
+ if (ageEligible(s) || countEligibleIds.has(s.id)) {
1352
+ toDelete.push(s.id);
1353
+ }
1354
+ }
1355
+ return toDelete;
1356
+ }
1357
+ var DEFAULT_INTERVAL = "1h";
1358
+ function resolve(flag, envValue, fallback) {
1359
+ return flag ?? envValue ?? fallback;
1360
+ }
1361
+ function parseMaxCount(input) {
1362
+ const trimmed = input.trim();
1363
+ if (!/^\d+$/.test(trimmed)) {
1364
+ throw new Error(`Invalid max-count "${input}": expected a positive integer.`);
1365
+ }
1366
+ const value = Number(trimmed);
1367
+ if (value <= 0) {
1368
+ throw new Error(`Invalid max-count "${input}": must be greater than 0.`);
1369
+ }
1370
+ return value;
1371
+ }
1372
+ function resolveSessionCleanupConfig(flags, env = process.env) {
1373
+ const warnings = [];
1374
+ const maxAgeRaw = resolve(flags.maxAge, env.EVIDENT_SESSION_CLEANUP_MAX_AGE);
1375
+ const maxCountRaw = resolve(flags.maxCount, env.EVIDENT_SESSION_CLEANUP_MAX_COUNT);
1376
+ const intervalRaw = resolve(
1377
+ flags.interval,
1378
+ env.EVIDENT_SESSION_CLEANUP_INTERVAL,
1379
+ DEFAULT_INTERVAL
1380
+ );
1381
+ let maxAgeMs;
1382
+ if (maxAgeRaw !== void 0) {
1092
1383
  try {
1093
- const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`, {
1094
- method: "POST",
1095
- headers: { "Content-Type": "application/json" },
1096
- body: JSON.stringify(body),
1097
- signal: controller.signal
1098
- });
1099
- if (!res.ok) {
1100
- const text = await res.text().catch(() => "");
1101
- throw new Error(`OpenCode message failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1102
- }
1103
- const sessionRes = await fetch(`${opencodeBase(port)}/session/${sessionId}`).catch(
1104
- () => null
1384
+ maxAgeMs = parseDurationMs(maxAgeRaw);
1385
+ } catch (err) {
1386
+ warnings.push(
1387
+ `Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
1105
1388
  );
1106
- const session = sessionRes?.ok ? await sessionRes.json() : null;
1107
- const reportedInteraction = reportedQuestions.size > 0 || reportedPermissions.size > 0;
1108
- const turnComplete = isTurnComplete(await getSessionMessages(port, sessionId));
1109
- const awaitingInteraction = reportedInteraction && !turnComplete;
1110
- return { title: session?.title, awaitingInteraction };
1389
+ }
1390
+ }
1391
+ let maxCount;
1392
+ if (maxCountRaw !== void 0) {
1393
+ try {
1394
+ maxCount = parseMaxCount(maxCountRaw);
1111
1395
  } catch (err) {
1112
- if (err instanceof Error && err.name === "AbortError") {
1113
- throw new Error("Message processing timed out");
1114
- }
1115
- throw err;
1116
- } finally {
1117
- clearTimeout(timer);
1118
- pollDone = true;
1396
+ warnings.push(
1397
+ `Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
1398
+ );
1119
1399
  }
1120
- };
1121
- const [result] = await Promise.all([sendMessage(), pollInteractive()]);
1122
- return result;
1400
+ }
1401
+ let intervalMs;
1402
+ try {
1403
+ intervalMs = parseDurationMs(intervalRaw ?? DEFAULT_INTERVAL);
1404
+ } catch (err) {
1405
+ warnings.push(
1406
+ `Ignoring invalid --session-cleanup-interval, using default ${DEFAULT_INTERVAL}: ${err instanceof Error ? err.message : String(err)}`
1407
+ );
1408
+ intervalMs = parseDurationMs(DEFAULT_INTERVAL);
1409
+ }
1410
+ const enabled = maxAgeMs !== void 0 || maxCount !== void 0;
1411
+ return { enabled, maxAgeMs, maxCount, intervalMs, warnings };
1123
1412
  }
1124
1413
 
1125
1414
  // src/lib/tunnel/connection.ts
@@ -1190,24 +1479,34 @@ var StreamForwarder = class {
1190
1479
  }
1191
1480
  async handleOpen(frame) {
1192
1481
  const { sid, method, path, headers, has_body } = frame;
1482
+ const correlationId = headers?.[CORRELATION_ID_HEADER];
1483
+ const startedAt = Date.now();
1193
1484
  if (path === TUNNEL_DRAIN_PING_PATH) {
1194
1485
  this.callbacks.onDrainPing?.();
1195
1486
  this.send({ type: "head", sid, status: 204, headers: {} });
1196
1487
  this.send({ type: "res_end", sid });
1197
1488
  return;
1198
1489
  }
1490
+ if (process.env.DEBUG) {
1491
+ log("debug", "agent_request", {
1492
+ correlation_id: correlationId,
1493
+ sid,
1494
+ method,
1495
+ path: stripQuery(path)
1496
+ });
1497
+ }
1199
1498
  const ac = new AbortController();
1200
1499
  let bodyPromise;
1201
1500
  let pushBody;
1202
1501
  let endBody;
1203
1502
  if (has_body) {
1204
1503
  const chunks = [];
1205
- bodyPromise = new Promise((resolve) => {
1504
+ bodyPromise = new Promise((resolve2) => {
1206
1505
  pushBody = (buf) => {
1207
1506
  chunks.push(buf);
1208
1507
  };
1209
1508
  endBody = () => {
1210
- resolve(Buffer.concat(chunks));
1509
+ resolve2(Buffer.concat(chunks));
1211
1510
  };
1212
1511
  });
1213
1512
  }
@@ -1242,6 +1541,14 @@ var StreamForwarder = class {
1242
1541
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1243
1542
  });
1244
1543
  this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
1544
+ if (process.env.DEBUG) {
1545
+ log("debug", "agent_response", {
1546
+ correlation_id: correlationId,
1547
+ sid,
1548
+ status: upstream.status,
1549
+ duration_ms: Date.now() - startedAt
1550
+ });
1551
+ }
1245
1552
  this.callbacks.onHead?.(sid, upstream.status);
1246
1553
  try {
1247
1554
  if (upstream.body) {
@@ -1317,7 +1624,7 @@ function connectTunnel(options) {
1317
1624
  } = options;
1318
1625
  const tunnelUrl = getTunnelUrlConfig();
1319
1626
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1320
- return new Promise((resolve, reject) => {
1627
+ return new Promise((resolve2, reject) => {
1321
1628
  const ws = new WebSocket2(url, {
1322
1629
  headers: {
1323
1630
  Authorization: authHeader
@@ -1382,7 +1689,7 @@ function connectTunnel(options) {
1382
1689
  clearTimeout(connectionTimeout);
1383
1690
  const connectedAgentId = message.agent_id ?? agentId;
1384
1691
  onConnected?.(connectedAgentId);
1385
- resolve({
1692
+ resolve2({
1386
1693
  ws,
1387
1694
  close: () => ws.close(1e3, "CLI shutdown")
1388
1695
  });
@@ -1498,6 +1805,12 @@ var RunnerConnection = class {
1498
1805
  };
1499
1806
 
1500
1807
  // src/lib/channels/driver.ts
1808
+ function messageIdOf(m) {
1809
+ if (!m || typeof m !== "object") return void 0;
1810
+ if (typeof m.id === "string") return m.id;
1811
+ const infoId = m.info?.id;
1812
+ return typeof infoId === "string" ? infoId : void 0;
1813
+ }
1501
1814
  var DEFAULT_RETRY_POLICY = {
1502
1815
  maxAttempts: 6,
1503
1816
  baseDelayMs: 500,
@@ -1505,12 +1818,24 @@ var DEFAULT_RETRY_POLICY = {
1505
1818
  };
1506
1819
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1507
1820
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1821
+ var DEFAULT_STUCK_QUEUED_MS = 6e4;
1822
+ var HEARTBEAT_MS = 6e4;
1823
+ var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
1824
+ var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
1508
1825
  var ChannelAuthError = class extends Error {
1509
1826
  constructor(message) {
1510
1827
  super(message);
1511
1828
  this.name = "ChannelAuthError";
1512
1829
  }
1513
1830
  };
1831
+ var ChannelTerminalError = class extends Error {
1832
+ status;
1833
+ constructor(message, status) {
1834
+ super(message);
1835
+ this.name = "ChannelTerminalError";
1836
+ this.status = status;
1837
+ }
1838
+ };
1514
1839
  function backoffDelay(attempt, policy) {
1515
1840
  const exp = policy.baseDelayMs * Math.pow(2, attempt);
1516
1841
  const capped = Math.min(policy.maxDelayMs, exp);
@@ -1531,15 +1856,91 @@ var ChannelDriver = class {
1531
1856
  sleep;
1532
1857
  pausedPollIntervalMs;
1533
1858
  pausedMaxWaitMs;
1859
+ stuckQueuedMs;
1860
+ now;
1534
1861
  /** Cache of conversationId → opencode sessionId. */
1535
1862
  sessions = /* @__PURE__ */ new Map();
1536
1863
  /**
1537
- * Outstanding paused-session watchers, keyed by `message.id` (WI-2-CLI).
1538
- * Single-flight per message: while a watcher is live for a message we never
1539
- * start a second one. The message stays `processing` for the watcher's lifetime
1540
- * so the `?status=pending` drain cannot double-claim it.
1864
+ * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
1865
+ * longer idempotent (no caller-supplied `messageID`), and its read-back picks
1866
+ * "the one new user row" which is only unambiguous if no OTHER dispatch into
1867
+ * the SAME session interleaves its snapshot→POST→read-back. This map chains each
1868
+ * session's dispatches so they run serially; distinct sessions stay concurrent.
1869
+ */
1870
+ sessionDispatchLocks = /* @__PURE__ */ new Map();
1871
+ /**
1872
+ * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1873
+ * session: one polling loop services all of that session's in-flight messages.
1874
+ * A session entry exists while it has any in-flight (dispatched-but-not-done)
1875
+ * message; it is removed once its in-flight set empties.
1541
1876
  */
1542
1877
  watchers = /* @__PURE__ */ new Map();
1878
+ /**
1879
+ * AUTHORITATIVE local dedup (WI-3): Evident message ids that have been
1880
+ * dispatched and are still in-flight. A message in this set is never
1881
+ * re-`prompt_async`-ed by a subsequent poll tick while it is queued/running.
1882
+ * Backed by a stable minted opencode `messageID` whose duplicate re-enqueue is
1883
+ * idempotent on opencode (PoC fact 9) — so even if this set is lost on restart,
1884
+ * a steady-state-poll re-dispatch will not double-run the message.
1885
+ */
1886
+ dispatched = /* @__PURE__ */ new Set();
1887
+ /**
1888
+ * Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
1889
+ * Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
1890
+ * so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
1891
+ * it is re-adopted and removed when its watcher settles or it is observed off
1892
+ * the processing list.
1893
+ */
1894
+ readopted = /* @__PURE__ */ new Set();
1895
+ /**
1896
+ * "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
1897
+ * re-adopted running/orphan row's watcher hit its `processed_at`-anchored
1898
+ * deadline (or an orphan whose window already elapsed): the still-`processing`
1899
+ * server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
1900
+ * drain until the 15-min cron resets it — spamming new turns.
1901
+ *
1902
+ * CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
1903
+ * does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
1904
+ * in opencode must still be delivered via `markDone` on the next drain — so
1905
+ * `readoptOne` computes `state` FIRST and this set is checked only on the
1906
+ * non-done path. It is cleared once the row leaves the processing list (cron
1907
+ * reset → it drains normally as `pending`), so it can never leak.
1908
+ */
1909
+ dontRedispatch = /* @__PURE__ */ new Set();
1910
+ /**
1911
+ * "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
1912
+ * re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
1913
+ * never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
1914
+ * that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
1915
+ * markDone failure must NOT land here (it must still retry next drain). Separate
1916
+ * from `dontRedispatch` because the two concerns are independent: a row can need
1917
+ * "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
1918
+ * the row leaves the processing list, exactly like `dontRedispatch`.
1919
+ */
1920
+ doneUndeliverable = /* @__PURE__ */ new Set();
1921
+ /**
1922
+ * "Already emitted `readopt_poll_unresolved` for this row" (#229). The b1 /
1923
+ * unreadable-status re-evaluate leaf leaves the row UN-tracked so it is re-read
1924
+ * every ~2s drain until the status map becomes readable — but the server-visible
1925
+ * signal is an OUTCOME, so it must fire at most ONCE per row, not once per drain
1926
+ * (Bugbot "Re-adopt signals flood every drain"). Cleared when the row leaves the
1927
+ * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
1928
+ */
1929
+ readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
1930
+ /**
1931
+ * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
1932
+ * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
1933
+ * is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
1934
+ * persist hasn't landed before tick N+1 re-reads the still-null
1935
+ * `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
1936
+ * A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
1937
+ * short-circuits while it is present, so a null-id row is re-dispatched AT MOST
1938
+ * ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
1939
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
1940
+ * re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
1941
+ * so the NEXT tick may retry exactly once more).
1942
+ */
1943
+ awaitingReadopt = /* @__PURE__ */ new Set();
1543
1944
  /**
1544
1945
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1545
1946
  * first session creation so drain-created sessions are rooted at the project
@@ -1547,8 +1948,43 @@ var ChannelDriver = class {
1547
1948
  * not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
1548
1949
  */
1549
1950
  opencodeDirectory = void 0;
1951
+ /**
1952
+ * Cache of opencode `sessionId → parentID` (its parent session, or `null` when
1953
+ * the session is a root with no parent). Sub-agents spawned via the `task` tool
1954
+ * run in CHILD sessions whose `parentID` chains up to the Evident-created
1955
+ * (watched) session; we resolve this once per session so a child-session
1956
+ * question/permission can be attributed to the watched session's subtree
1957
+ * (`sessionBelongsTo`) instead of being dropped by an exact-id filter. A missing
1958
+ * entry = not yet resolved; `null` = resolved root (stop walking).
1959
+ */
1960
+ sessionParents = /* @__PURE__ */ new Map();
1961
+ /**
1962
+ * Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
1963
+ * NON-EMPTY name is stored (terminal — a real session name won't later un-name),
1964
+ * so we do NOT re-GET `/session/:id` every tick. A missing entry = not yet
1965
+ * resolved OR resolved-but-still-empty → re-fetch on next need, since OpenCode
1966
+ * names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
1967
+ * the watcher completion path AND the restart-recovery re-adopt path (which has
1968
+ * no watcher) can resolve the title.
1969
+ */
1970
+ sessionTitles = /* @__PURE__ */ new Map();
1550
1971
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
1551
1972
  draining = false;
1973
+ /**
1974
+ * The currently-executing `drainPending()` promise, or null when idle. Lets a
1975
+ * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
1976
+ * is about to dispatch is not missed by the `hasInFlightWatchers()` check (a
1977
+ * drain that entered before `stop()` still registers its watcher).
1978
+ */
1979
+ activeDrain = null;
1980
+ /**
1981
+ * Set by `stop()` on graceful shutdown. Once stopped, `drainPending` no longer
1982
+ * dispatches NEW work (it returns 0 immediately) — but the per-session watcher
1983
+ * loops already running keep going so in-flight turns can finish and deliver
1984
+ * their reply. `run.ts` awaits `waitForInFlight()` before it closes the tunnel
1985
+ * and stops opencode.
1986
+ */
1987
+ stopped = false;
1552
1988
  constructor(config2) {
1553
1989
  this.agentId = config2.agentId;
1554
1990
  this.port = config2.port;
@@ -1562,6 +1998,8 @@ var ChannelDriver = class {
1562
1998
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1563
1999
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1564
2000
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
2001
+ this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
2002
+ this.now = config2.now ?? (() => Date.now());
1565
2003
  }
1566
2004
  /** The IPv4-loopback base URL for the local `opencode serve`. */
1567
2005
  get opencodeBase() {
@@ -1571,16 +2009,29 @@ var ChannelDriver = class {
1571
2009
  // Public API
1572
2010
  // -------------------------------------------------------------------------
1573
2011
  /**
1574
- * Drain all pending channel conversations once: poll → processcallback.
2012
+ * Drain all pending channel conversations once: poll → dispatchregister.
1575
2013
  * Called on tunnel `connected` (WI-CHAN-4) and on each poll tick by `run.ts`.
1576
2014
  * Re-entrant calls while a drain is in flight are skipped (return 0).
1577
2015
  *
1578
- * @returns the number of messages processed.
2016
+ * @returns the number of messages NEWLY dispatched to opencode's native queue.
1579
2017
  */
1580
2018
  async drainPending() {
2019
+ if (this.stopped) return 0;
1581
2020
  if (this.draining) return 0;
1582
2021
  this.draining = true;
1583
- let processed = 0;
2022
+ const run2 = this.runDrain();
2023
+ this.activeDrain = run2.then(
2024
+ () => {
2025
+ this.activeDrain = null;
2026
+ },
2027
+ () => {
2028
+ this.activeDrain = null;
2029
+ }
2030
+ );
2031
+ return run2;
2032
+ }
2033
+ async runDrain() {
2034
+ let dispatched = 0;
1584
2035
  try {
1585
2036
  const conversations = await this.getPendingConversations();
1586
2037
  if (conversations.length > 0) {
@@ -1591,107 +2042,226 @@ var ChannelDriver = class {
1591
2042
  });
1592
2043
  }
1593
2044
  for (const conv of conversations) {
1594
- processed += await this.processConversation(conv);
2045
+ if (this.stopped) break;
2046
+ dispatched += await this.processConversation(conv);
1595
2047
  }
2048
+ await this.readoptProcessing();
1596
2049
  } finally {
1597
2050
  this.draining = false;
1598
2051
  }
1599
- return processed;
2052
+ return dispatched;
2053
+ }
2054
+ /**
2055
+ * True while any per-session watcher has a non-empty in-flight dispatched set
2056
+ * (Task 3.7). `run.ts` treats this as NON-idle so `--idle-timeout` cannot exit
2057
+ * the process while a dispatched message is still queued/running — which would
2058
+ * kill the turn and orphan its reply.
2059
+ */
2060
+ hasInFlightWatchers() {
2061
+ for (const watcher of this.watchers.values()) {
2062
+ if (watcher.inFlight.size > 0) return true;
2063
+ }
2064
+ return false;
2065
+ }
2066
+ /**
2067
+ * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2068
+ * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
2069
+ * `watchers` entry whose `inFlight` set is non-empty — the same predicate
2070
+ * `hasInFlightWatchers()` uses, lifted to return the ids.
2071
+ *
2072
+ * Deliberately does NOT include `this.sessions` (the permanent, never-pruned
2073
+ * conversation→session cache). Protecting every bound-but-idle session there
2074
+ * would shield nearly every session and defeat cleanup — AND it is unnecessary:
2075
+ * `ensureSession` is self-healing (it recreates a session whose id no longer
2076
+ * exists), so deleting an idle bound session is harmless — the conversation's
2077
+ * next turn transparently rebinds a fresh one. The only thing worth protecting
2078
+ * is a session with a turn ACTIVELY in flight right now: tearing that down
2079
+ * mid-turn would strand the running `prompt_async`. Idle sessions are fair game.
2080
+ */
2081
+ protectedSessionIds() {
2082
+ const ids = /* @__PURE__ */ new Set();
2083
+ for (const [sessionId, watcher] of this.watchers) {
2084
+ if (watcher.inFlight.size > 0) ids.add(sessionId);
2085
+ }
2086
+ return ids;
2087
+ }
2088
+ /**
2089
+ * Begin a graceful stop: stop accepting NEW channel work. Idempotent. After
2090
+ * this, `drainPending()` is a no-op (returns 0), so no new message is dispatched
2091
+ * — but the watcher loops already tracking in-flight turns keep running, so a
2092
+ * turn that has finished (or is about to) still fires `markDone` and delivers
2093
+ * its reply. Pair with `waitForInFlight()` to bound how long shutdown waits.
2094
+ */
2095
+ stop() {
2096
+ this.stopped = true;
2097
+ }
2098
+ /**
2099
+ * Wait (up to `timeoutMs`) for in-flight watcher work to settle during a
2100
+ * graceful shutdown, so a turn whose reply is ready — or completes within the
2101
+ * window — is delivered before the process exits, instead of being cut off and
2102
+ * left for the ADR-0046 restart-recovery path.
2103
+ *
2104
+ * Bounded on purpose: the watcher's own give-up deadline is up to 10 minutes,
2105
+ * far longer than a shutdown grace period (e.g. Fargate's SIGTERM→SIGKILL
2106
+ * window). We poll `hasInFlightWatchers()` and return as soon as the in-flight
2107
+ * set empties OR the timeout elapses. Anything still in flight at the timeout is
2108
+ * safe to abandon — it stays `processing` server-side and is re-adopted on the
2109
+ * next runner start (ADR-0046).
2110
+ *
2111
+ * @returns true if all in-flight work settled within the window; false if the
2112
+ * timeout elapsed with work still in flight.
2113
+ */
2114
+ async waitForInFlight(timeoutMs) {
2115
+ const deadline = this.now() + timeoutMs;
2116
+ const step = Math.min(this.pausedPollIntervalMs, 250);
2117
+ if (this.activeDrain) {
2118
+ let drainSettled = false;
2119
+ void this.activeDrain.then(() => {
2120
+ drainSettled = true;
2121
+ });
2122
+ while (!drainSettled) {
2123
+ if (this.now() >= deadline) return false;
2124
+ await this.sleep(step);
2125
+ }
2126
+ }
2127
+ while (this.hasInFlightWatchers()) {
2128
+ if (this.now() >= deadline) return false;
2129
+ await this.sleep(step);
2130
+ }
2131
+ return true;
1600
2132
  }
1601
2133
  /**
1602
- * Await all outstanding paused-session watchers (WI-2-CLI).
2134
+ * Await all outstanding per-session watchers (WI-3).
1603
2135
  *
1604
- * In production the watchers are deliberately started-not-awaited so the drain
1605
- * loop never blocks on them and process exit is not held up (the cron recovers
1606
- * any abandoned ones). This helper exists primarily for deterministic tests
1607
- * that need to observe the watcher's effect (the `done` PATCH or its giving up)
1608
- * after a non-blocking `drainPending`. Watchers never reject, so this resolves.
2136
+ * In production the watcher loops are deliberately started-not-awaited so the
2137
+ * drain loop never blocks on them and process exit is not held up (the cron
2138
+ * recovers any abandoned ones). This helper exists primarily for deterministic
2139
+ * tests that need to observe a watcher's effect (the `processing`/`done` PATCH
2140
+ * or its giving up) after a non-blocking `drainPending`. Watcher loops never
2141
+ * reject, so this resolves.
1609
2142
  */
1610
2143
  async flushPausedWatchers() {
1611
- await Promise.all([...this.watchers.values()]);
2144
+ while (true) {
2145
+ const loops = [...this.watchers.values()].map((w) => w.loop).filter((l) => l != null);
2146
+ if (loops.length === 0) return;
2147
+ await Promise.all(loops);
2148
+ const stillLive = [...this.watchers.values()].some((w) => w.loop != null);
2149
+ if (!stillLive) return;
2150
+ }
1612
2151
  }
1613
2152
  // -------------------------------------------------------------------------
1614
- // Conversation processing
2153
+ // Conversation processing (WI-3 — async dispatch)
1615
2154
  // -------------------------------------------------------------------------
2155
+ /**
2156
+ * Dispatch each pending message for a conversation to opencode's native queue
2157
+ * via `prompt_async` (Task 3.2) and register it with the conversation's
2158
+ * per-session watcher. Does NOT block on the turn and does NOT call
2159
+ * `markProcessing` here — that fires from the watcher on running-start.
2160
+ *
2161
+ * @returns the count of messages NEWLY dispatched (not already in-flight).
2162
+ */
1616
2163
  async processConversation(conv) {
1617
2164
  const sessionId = await this.ensureSession(conv);
1618
2165
  const messages = await this.getPendingMessages(conv.id);
1619
- let processed = 0;
2166
+ let dispatched = 0;
2167
+ let skippedAlreadyDispatched = 0;
1620
2168
  for (const message of messages) {
1621
- const claimed = await this.markProcessing(conv.id, message.id);
1622
- if (!claimed) {
1623
- this.log({
1624
- level: "info",
1625
- message: `Message ${message.id.slice(0, 8)} already claimed \u2014 skipping`,
1626
- conversation_id: conv.id,
1627
- message_id: message.id
1628
- });
2169
+ if (this.stopped) break;
2170
+ if (this.dispatched.has(message.id)) {
2171
+ skippedAlreadyDispatched += 1;
1629
2172
  continue;
1630
2173
  }
2174
+ const options = {
2175
+ agent: message.opencode_agent ?? void 0,
2176
+ model: message.opencode_model ?? void 0
2177
+ };
2178
+ let opencodeMessageId;
1631
2179
  try {
1632
2180
  this.log({
1633
2181
  level: "info",
1634
- message: `Sending queued message ${message.id.slice(0, 8)} to OpenCode (session ${sessionId.slice(0, 8)})`,
2182
+ message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
1635
2183
  conversation_id: conv.id,
1636
2184
  message_id: message.id
1637
2185
  });
1638
- const result = await sendMessageToOpenCode(
1639
- this.port,
2186
+ opencodeMessageId = await this.dispatchLocked(
1640
2187
  sessionId,
1641
- message.content,
1642
- {
1643
- agent: message.opencode_agent ?? void 0,
1644
- model: message.opencode_model ?? void 0
1645
- },
1646
- {
1647
- onQuestion: (question) => this.reportInteraction(conv.id, "question", question),
1648
- onPermission: (permission) => this.reportInteraction(conv.id, "permission", permission)
1649
- }
2188
+ () => sendPromptAsync(this.port, sessionId, message.content, options)
1650
2189
  );
1651
- if (result.awaitingInteraction) {
2190
+ } catch (err) {
2191
+ if (err instanceof ChannelAuthError) throw err;
2192
+ this.dispatched.delete(message.id);
2193
+ if (await sessionExists(this.port, sessionId) === false) {
2194
+ this.sessions.delete(conv.id);
1652
2195
  this.log({
1653
2196
  level: "info",
1654
- message: `Message ${message.id.slice(0, 8)} paused awaiting interaction \u2014 watching session for completion`,
2197
+ 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.`,
1655
2198
  conversation_id: conv.id,
1656
2199
  message_id: message.id
1657
2200
  });
1658
- this.startPausedWatcher(conv, message, sessionId);
1659
- continue;
2201
+ break;
1660
2202
  }
1661
- await this.confirmCompletion(sessionId);
1662
- await this.markDone(conv.id, message.id, sessionId);
1663
- processed += 1;
2203
+ await this.markFailed(conv.id, message.id).catch(() => {
2204
+ });
1664
2205
  this.log({
1665
- level: "info",
1666
- message: `Message ${message.id.slice(0, 8)} processed`,
2206
+ level: "error",
2207
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
1667
2208
  conversation_id: conv.id,
1668
2209
  message_id: message.id
1669
2210
  });
1670
- } catch (err) {
1671
- if (err instanceof ChannelAuthError) throw err;
1672
- await this.markFailed(conv.id, message.id).catch(() => {
1673
- });
2211
+ continue;
2212
+ }
2213
+ if (opencodeMessageId === null) {
1674
2214
  this.log({
1675
2215
  level: "error",
1676
- message: `Message ${message.id.slice(0, 8)} failed: ${err instanceof Error ? err.message : String(err)}`,
2216
+ 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`,
1677
2217
  conversation_id: conv.id,
1678
2218
  message_id: message.id
1679
2219
  });
2220
+ continue;
1680
2221
  }
2222
+ this.dispatched.add(message.id);
2223
+ this.registerInFlight(conv, sessionId, message, opencodeMessageId);
2224
+ dispatched += 1;
2225
+ void this.postSignal(conv.id, message.id, "dispatched");
2226
+ }
2227
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
2228
+ this.log({
2229
+ level: "error",
2230
+ 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).`,
2231
+ conversation_id: conv.id
2232
+ });
1681
2233
  }
1682
- return processed;
2234
+ this.ensureWatcherRunning(sessionId);
2235
+ return dispatched;
1683
2236
  }
1684
2237
  async ensureSession(conv) {
1685
- const cached = this.sessions.get(conv.id);
1686
- if (cached) return cached;
1687
- if (conv.opencode_session_id) {
1688
- this.sessions.set(conv.id, conv.opencode_session_id);
1689
- return conv.opencode_session_id;
2238
+ const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
2239
+ if (bound) {
2240
+ const exists = await sessionExists(this.port, bound);
2241
+ if (exists === false) {
2242
+ this.log({
2243
+ level: "info",
2244
+ 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.`,
2245
+ conversation_id: conv.id
2246
+ });
2247
+ this.sessions.delete(conv.id);
2248
+ return this.createAndBindSession(conv.id);
2249
+ }
2250
+ this.sessions.set(conv.id, bound);
2251
+ return bound;
1690
2252
  }
2253
+ return this.createAndBindSession(conv.id);
2254
+ }
2255
+ /**
2256
+ * Create a fresh OpenCode session for a conversation, cache the binding, and
2257
+ * best-effort persist it server-side. Shared by the first-ever bind and the
2258
+ * self-heal recreate path in `ensureSession`.
2259
+ */
2260
+ async createAndBindSession(conversationId) {
1691
2261
  const directory = await this.resolveOpenCodeDirectory();
1692
2262
  const sessionId = await createOpenCodeSession(this.port, directory);
1693
- this.sessions.set(conv.id, sessionId);
1694
- await this.persistSession(conv.id, sessionId).catch(() => {
2263
+ this.sessions.set(conversationId, sessionId);
2264
+ await this.persistSession(conversationId, sessionId).catch(() => {
1695
2265
  });
1696
2266
  return sessionId;
1697
2267
  }
@@ -1711,111 +2281,1230 @@ var ChannelDriver = class {
1711
2281
  }
1712
2282
  return this.opencodeDirectory;
1713
2283
  }
1714
- /**
1715
- * Local reconcile: re-query `GET /session/:id/message` and check whether the
1716
- * last assistant message is message-level complete (`isTurnComplete`).
1717
- *
1718
- * This is a PURE OBSERVABILITY probe on the normal (non-paused) path: the
1719
- * blocking POST already returned, and `markDone` causes the server to re-fetch
1720
- * the messages itself (via `extractTextFromMessages`) when delivering the
1721
- * reply — so this round-trip never gates delivery. We keep it only to surface a
1722
- * truthful diagnostic when opencode hasn't yet recorded a completed assistant
1723
- * turn at reconcile time, then proceed to `markDone` regardless. Uses the
1724
- * injected `fetchImpl` and reuses only the pure `isTurnComplete` predicate.
1725
- */
1726
- async confirmCompletion(sessionId) {
1727
- try {
1728
- const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
1729
- if (!res.ok) return;
1730
- const body = await res.json();
1731
- const messages = Array.isArray(body) ? body : null;
1732
- if (!isTurnComplete(messages)) {
1733
- this.log({
1734
- level: "info",
1735
- message: `Session ${sessionId.slice(0, 8)} messages do not yet show a completed assistant turn on reconcile \u2014 delivering anyway`
1736
- });
1737
- }
1738
- } catch {
1739
- }
1740
- }
1741
2284
  // -------------------------------------------------------------------------
1742
- // Paused-session watcher (WI-2-CLI)
2285
+ // Per-session watcher (WI-3)
1743
2286
  // -------------------------------------------------------------------------
1744
2287
  /**
1745
- * Start (but do NOT await) a watcher that resumes a paused turn to completion.
1746
- *
1747
- * Single-flight per message: if a watcher is already live for this message we
1748
- * skip. The returned watcher promise is tracked in `this.watchers` and removed
1749
- * when it settles; it never rejects (the body is fully guarded), so a failed
1750
- * poll/markDone can never crash the run loop — the cron stays as the safety net.
2288
+ * Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
2289
+ * opencode session (Task 2.1a), so two dispatches into the SAME session can
2290
+ * never interleave and mis-correlate their read-backs. Distinct sessions run
2291
+ * concurrently. The chained tail intentionally ignores the prior result/error
2292
+ * (each dispatch reports its own outcome to its caller).
1751
2293
  */
1752
- startPausedWatcher(conv, message, sessionId) {
1753
- if (this.watchers.has(message.id)) return;
1754
- const watcher = this.watchPausedSession(conv, message, sessionId).finally(() => {
1755
- this.watchers.delete(message.id);
2294
+ dispatchLocked(sessionId, fn) {
2295
+ const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
2296
+ const run2 = prior.then(fn, fn);
2297
+ this.sessionDispatchLocks.set(
2298
+ sessionId,
2299
+ run2.then(
2300
+ () => void 0,
2301
+ () => void 0
2302
+ )
2303
+ );
2304
+ return run2;
2305
+ }
2306
+ /** Register a freshly-dispatched message with its session's watcher state. */
2307
+ registerInFlight(conv, sessionId, message, opencodeMessageId) {
2308
+ let watcher = this.watchers.get(sessionId);
2309
+ if (!watcher) {
2310
+ watcher = {
2311
+ conv,
2312
+ inFlight: /* @__PURE__ */ new Map(),
2313
+ loop: null,
2314
+ reportedQuestions: /* @__PURE__ */ new Set(),
2315
+ reportedPermissions: /* @__PURE__ */ new Set(),
2316
+ lastGoodPollAt: this.now(),
2317
+ hadUsablePoll: false
2318
+ };
2319
+ this.watchers.set(sessionId, watcher);
2320
+ }
2321
+ const now = this.now();
2322
+ watcher.inFlight.set(message.id, {
2323
+ evidentMessageId: message.id,
2324
+ opencodeMessageId,
2325
+ message,
2326
+ dispatchedAt: now,
2327
+ processingAnchorMs: now,
2328
+ deadline: now + this.pausedMaxWaitMs,
2329
+ started: false,
2330
+ done: false,
2331
+ stuckReported: false,
2332
+ lastAliveAt: 0,
2333
+ aliveInFlight: false,
2334
+ awaitingHumanLatched: false,
2335
+ pausedOnQuestion: false,
2336
+ pausedOnPermission: false,
2337
+ pausedClearConfirmed: false,
2338
+ pausedInFlight: false,
2339
+ deliveryDeadlineAnchored: false
1756
2340
  });
1757
- this.watchers.set(message.id, watcher);
1758
2341
  }
1759
2342
  /**
1760
- * Poll `GET /session/:id/message` until the SAME paused turn is message-level
1761
- * complete the last message is an ASSISTANT message whose
1762
- * `info.time.completed` is set (the user answered the question/permission in
1763
- * opencode web and opencode finished the turn) then complete it via the
1764
- * EXISTING `markDone` PATCH NO re-send of the original message. The server
1765
- * re-fetches the assistant messages on completion, so the watcher does not
1766
- * pass any reply text.
2343
+ * Register a RE-ADOPTED `processing` message with its session watcher
2344
+ * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
2345
+ * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
2346
+ * `now`, so the paused/queued/unreachable cases settle on the same wall-clock a
2347
+ * fresh dispatch would (10 min after `processed_at`, not 10 min from now).
1767
2348
  *
1768
- * Fetch seam: the poll uses the injected `this.fetchImpl` (preserving the
1769
- * tests' injection) and reuses only the pure `isTurnComplete` predicate from
1770
- * `session.ts` we do NOT call `getSessionMessages` (which uses the global
1771
- * `fetch`) here.
2349
+ * This re-attaches into the SAME watcher, so the ADR-0047 progressing-vs-paused
2350
+ * give-up (`serviceInFlightMessage`) applies unchanged: a re-adopted turn
2351
+ * opencode reports ACTIVELY `running` is watched to completion (its liveness
2352
+ * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
2353
+ * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
2354
+ * handed to the cron. The old "the `deadline` must settle before the ~15-min
2355
+ * cron or they double-drive" reasoning is superseded: liveness now settles the
2356
+ * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
2357
+ * (only the appear-guard uses it).
1772
2358
  *
1773
- * Bounded by `pausedMaxWaitMs` (default 10 min, strictly < the 15-min cron
1774
- * reset): on timeout we STOP and leave the message `processing` so the cron
1775
- * remains the last-resort safety net. The whole body is wrapped so any
1776
- * poll/markDone failure is logged and swallowed — a watcher MUST NEVER throw
1777
- * out of the run loop.
2359
+ * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
2360
+ * `opencodeMessageId` is the id the watcher polls for a reply for the orphan
2361
+ * fresh-run path these differ (a fresh opencode id under the same server row).
2362
+ *
2363
+ * `started` is set true so the watcher does NOT re-`markProcessing` a row the
2364
+ * server already flipped to `processing`; the running/done transitions still
2365
+ * fire from the watcher's normal branches.
2366
+ */
2367
+ registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
2368
+ let watcher = this.watchers.get(sessionId);
2369
+ if (!watcher) {
2370
+ watcher = {
2371
+ conv,
2372
+ inFlight: /* @__PURE__ */ new Map(),
2373
+ loop: null,
2374
+ reportedQuestions: /* @__PURE__ */ new Set(),
2375
+ reportedPermissions: /* @__PURE__ */ new Set(),
2376
+ lastGoodPollAt: this.now(),
2377
+ hadUsablePoll: false
2378
+ };
2379
+ this.watchers.set(sessionId, watcher);
2380
+ }
2381
+ watcher.inFlight.set(message.id, {
2382
+ evidentMessageId: message.id,
2383
+ opencodeMessageId,
2384
+ message,
2385
+ dispatchedAt: this.now(),
2386
+ // Anchor the absolute-age ceiling to the SERVER-SIDE `processed_at` (the same
2387
+ // value seeding `deadline`), NOT `dispatchedAt` — so a re-adopted zombie's age
2388
+ // reflects the real turn duration and the ceiling fires on the ORIGINAL turn.
2389
+ processingAnchorMs: processedAtMs,
2390
+ deadline: processedAtMs + this.pausedMaxWaitMs,
2391
+ // The server row is ALREADY `processing`; do not re-fire markProcessing.
2392
+ started: true,
2393
+ done: false,
2394
+ // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
2395
+ // AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
2396
+ // on `state === 'queued'` (turn produced no reply), not on `started`, so a
2397
+ // re-adopted row left wedged in `queued` still emits the signal once
2398
+ // (#210/#220 observability).
2399
+ stuckReported: false,
2400
+ // Task 5.2: a re-adopted actively-running row re-attaches into the SAME
2401
+ // watcher and so hits the SAME actively-running heartbeat branch in
2402
+ // `serviceInFlightMessage` as a fresh dispatch — monitoring observes "runner
2403
+ // re-adopted and is confirming this row alive" via that `alive` heartbeat,
2404
+ // with no extra `re_adopted` signal needed (folds old WI-6).
2405
+ lastAliveAt: 0,
2406
+ aliveInFlight: false,
2407
+ awaitingHumanLatched: false,
2408
+ pausedOnQuestion: false,
2409
+ pausedOnPermission: false,
2410
+ pausedClearConfirmed: false,
2411
+ pausedInFlight: false,
2412
+ deliveryDeadlineAnchored: false
2413
+ });
2414
+ }
2415
+ /**
2416
+ * Start (but do NOT await) the per-session watcher loop if it has in-flight
2417
+ * work and is not already running. Single-flight per session. The loop is
2418
+ * tracked on the watcher and cleared when it settles; it never rejects (fully
2419
+ * guarded), so a failed poll/callback can never crash the run loop — the cron
2420
+ * stays as the safety net.
1778
2421
  */
1779
- async watchPausedSession(conv, message, sessionId) {
1780
- const deadline = Date.now() + this.pausedMaxWaitMs;
2422
+ ensureWatcherRunning(sessionId) {
2423
+ const watcher = this.watchers.get(sessionId);
2424
+ if (!watcher) return;
2425
+ if (watcher.loop) return;
2426
+ if (watcher.inFlight.size === 0) {
2427
+ this.watchers.delete(sessionId);
2428
+ return;
2429
+ }
2430
+ const loop = this.runWatcherLoop(sessionId, watcher).finally(() => {
2431
+ watcher.loop = null;
2432
+ if (watcher.inFlight.size === 0) {
2433
+ this.watchers.delete(sessionId);
2434
+ }
2435
+ });
2436
+ watcher.loop = loop;
2437
+ }
2438
+ /**
2439
+ * The per-session polling loop (WI-3). Once per tick it:
2440
+ * 1. polls `GET /session/:id/message` once and, per in-flight message,
2441
+ * computes `messageRunState` and fires markProcessing (queued→running) /
2442
+ * markDone (done) exactly once per transition;
2443
+ * 2. applies the idle-path re-dispatch guard (a dispatched message that never
2444
+ * APPEARS → re-dispatch — D1 obligation 2);
2445
+ * 3. polls `/question` + `/permission` (scoped to the session) and surfaces
2446
+ * NEW ones via `reportInteraction`, carrying the PAUSED message's own
2447
+ * `source_message_id`;
2448
+ * 4. drops messages that completed or timed out from the in-flight set.
2449
+ * Exits when the in-flight set empties. Never throws.
2450
+ */
2451
+ async runWatcherLoop(sessionId, watcher) {
1781
2452
  try {
1782
- while (Date.now() < deadline) {
2453
+ while (watcher.inFlight.size > 0) {
1783
2454
  await this.sleep(this.pausedPollIntervalMs);
1784
- let completed = false;
2455
+ let messages = null;
1785
2456
  try {
1786
2457
  const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
1787
2458
  if (res.ok) {
1788
2459
  const body = await res.json();
1789
- const messages = Array.isArray(body) ? body : null;
1790
- completed = isTurnComplete(messages);
2460
+ messages = Array.isArray(body) ? body : null;
1791
2461
  }
1792
2462
  } catch {
1793
- continue;
1794
2463
  }
1795
- if (!completed) continue;
2464
+ if (messages != null && messages.length > 0) {
2465
+ watcher.lastGoodPollAt = this.now();
2466
+ watcher.hadUsablePoll = true;
2467
+ } else {
2468
+ const emptyButReachable = messages != null;
2469
+ const graceApplies = !emptyButReachable || watcher.hadUsablePoll;
2470
+ if (graceApplies && this.now() - watcher.lastGoodPollAt < POLL_MISS_GRACE_MS) {
2471
+ continue;
2472
+ }
2473
+ }
2474
+ const { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk } = await this.pollInteractions(sessionId, watcher, messages);
2475
+ for (const inFlight of [...watcher.inFlight.values()]) {
2476
+ await this.serviceInFlightMessage(
2477
+ sessionId,
2478
+ watcher,
2479
+ inFlight,
2480
+ messages,
2481
+ openQuestions,
2482
+ openPermissions,
2483
+ questionsPolledOk,
2484
+ permissionsPolledOk
2485
+ );
2486
+ }
2487
+ }
2488
+ } catch (err) {
2489
+ if (err instanceof ChannelAuthError) {
1796
2490
  this.log({
1797
- level: "info",
1798
- message: `Paused session ${sessionId.slice(0, 8)} completed \u2014 marking message ${message.id.slice(0, 8)} done`,
2491
+ level: "error",
2492
+ message: `Session watcher aborted on auth failure for session ${sessionId.slice(0, 8)} \u2014 clearing in-flight state for re-drive after re-auth: ${err.message}`,
2493
+ conversation_id: watcher.conv.id
2494
+ });
2495
+ for (const evidentMessageId of [...watcher.inFlight.keys()]) {
2496
+ this.readopted.delete(evidentMessageId);
2497
+ this.removeInFlight(watcher, evidentMessageId);
2498
+ }
2499
+ return;
2500
+ }
2501
+ this.log({
2502
+ level: "error",
2503
+ message: `Session watcher failed for session ${sessionId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
2504
+ conversation_id: watcher.conv.id
2505
+ });
2506
+ }
2507
+ }
2508
+ /**
2509
+ * On FIRST observing a terminal (done/failed) state, ensure the delivery
2510
+ * (markDone/markFailed) transient-retry path has a real window. A long
2511
+ * ACTIVELY-running turn is kept past its original `deadline`, so by completion
2512
+ * `now >= deadline` already holds and the retry bound below would fire on the
2513
+ * first transient PATCH failure — dropping the message before its reply lands
2514
+ * (Bugbot "Stale deadline aborts long-turn delivery"). Re-anchor once (latched)
2515
+ * to a fresh `pausedMaxWaitMs` window; only extend if the current deadline is at
2516
+ * or past now, so a still-ample window is left untouched.
2517
+ */
2518
+ anchorDeliveryDeadline(inFlight) {
2519
+ if (inFlight.deliveryDeadlineAnchored) return;
2520
+ inFlight.deliveryDeadlineAnchored = true;
2521
+ if (this.now() >= inFlight.deadline) {
2522
+ inFlight.deadline = this.now() + this.pausedMaxWaitMs;
2523
+ }
2524
+ }
2525
+ /**
2526
+ * Drive ONE in-flight message's lifecycle from the tick's message snapshot.
2527
+ * Fires markProcessing on queued→running and markDone on done (each once),
2528
+ * applies the idle-path re-dispatch guard, and removes the message from the
2529
+ * in-flight set on completion or timeout.
2530
+ */
2531
+ async serviceInFlightMessage(sessionId, watcher, inFlight, messages, openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk) {
2532
+ const conv = watcher.conv;
2533
+ const state = messageRunState(messages, inFlight.opencodeMessageId);
2534
+ const id = inFlight.evidentMessageId;
2535
+ if (openQuestions.has(id)) inFlight.pausedOnQuestion = true;
2536
+ else if (questionsPolledOk) inFlight.pausedOnQuestion = false;
2537
+ if (openPermissions.has(id)) inFlight.pausedOnPermission = true;
2538
+ else if (permissionsPolledOk) inFlight.pausedOnPermission = false;
2539
+ const observedOpen = openQuestions.has(id) || openPermissions.has(id);
2540
+ const latchedPaused = inFlight.pausedOnQuestion || inFlight.pausedOnPermission;
2541
+ const awaitingHuman = observedOpen || latchedPaused;
2542
+ if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
2543
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2544
+ let claimed;
2545
+ try {
2546
+ claimed = await this.markProcessing(
2547
+ conv.id,
2548
+ inFlight.evidentMessageId,
2549
+ sessionId,
2550
+ inFlight.opencodeMessageId,
2551
+ title
2552
+ );
2553
+ } catch (err) {
2554
+ if (err instanceof ChannelAuthError) throw err;
2555
+ this.log({
2556
+ level: "error",
2557
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
1799
2558
  conversation_id: conv.id,
1800
- message_id: message.id
2559
+ message_id: inFlight.evidentMessageId
1801
2560
  });
1802
- await this.markDone(conv.id, message.id, sessionId);
1803
2561
  return;
1804
2562
  }
2563
+ inFlight.started = true;
2564
+ if (!claimed) {
2565
+ this.log({
2566
+ level: "info",
2567
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} already marked processing \u2014 continuing`,
2568
+ conversation_id: conv.id,
2569
+ message_id: inFlight.evidentMessageId
2570
+ });
2571
+ }
2572
+ }
2573
+ if (state === "done") {
2574
+ this.anchorDeliveryDeadline(inFlight);
2575
+ if (!inFlight.done) {
2576
+ this.log({
2577
+ level: "info",
2578
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2579
+ conversation_id: conv.id,
2580
+ message_id: inFlight.evidentMessageId
2581
+ });
2582
+ const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
2583
+ try {
2584
+ await this.markDone(
2585
+ conv.id,
2586
+ inFlight.evidentMessageId,
2587
+ sessionId,
2588
+ inFlight.opencodeMessageId,
2589
+ title
2590
+ );
2591
+ } catch (err) {
2592
+ if (err instanceof ChannelAuthError) throw err;
2593
+ if (err instanceof ChannelTerminalError) {
2594
+ this.log({
2595
+ level: "error",
2596
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2597
+ conversation_id: conv.id,
2598
+ message_id: inFlight.evidentMessageId
2599
+ });
2600
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2601
+ return;
2602
+ }
2603
+ if (this.now() >= inFlight.deadline) {
2604
+ this.log({
2605
+ level: "error",
2606
+ 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)}`,
2607
+ conversation_id: conv.id,
2608
+ message_id: inFlight.evidentMessageId
2609
+ });
2610
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2611
+ return;
2612
+ }
2613
+ this.log({
2614
+ level: "error",
2615
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2616
+ conversation_id: conv.id,
2617
+ message_id: inFlight.evidentMessageId
2618
+ });
2619
+ return;
2620
+ }
2621
+ inFlight.done = true;
2622
+ }
2623
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2624
+ return;
2625
+ }
2626
+ if (state === "failed") {
2627
+ this.anchorDeliveryDeadline(inFlight);
2628
+ if (!inFlight.done) {
2629
+ const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
2630
+ this.log({
2631
+ level: "error",
2632
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2633
+ conversation_id: conv.id,
2634
+ message_id: inFlight.evidentMessageId
2635
+ });
2636
+ try {
2637
+ await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
2638
+ } catch (err) {
2639
+ if (err instanceof ChannelAuthError) throw err;
2640
+ if (err instanceof ChannelTerminalError) {
2641
+ this.log({
2642
+ level: "error",
2643
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2644
+ conversation_id: conv.id,
2645
+ message_id: inFlight.evidentMessageId
2646
+ });
2647
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2648
+ return;
2649
+ }
2650
+ if (this.now() >= inFlight.deadline) {
2651
+ this.log({
2652
+ level: "error",
2653
+ 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)}`,
2654
+ conversation_id: conv.id,
2655
+ message_id: inFlight.evidentMessageId
2656
+ });
2657
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2658
+ return;
2659
+ }
2660
+ this.log({
2661
+ level: "error",
2662
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2663
+ conversation_id: conv.id,
2664
+ message_id: inFlight.evidentMessageId
2665
+ });
2666
+ return;
2667
+ }
2668
+ inFlight.done = true;
2669
+ }
2670
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2671
+ return;
2672
+ }
2673
+ const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
2674
+ const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
2675
+ if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
2676
+ inFlight.stuckReported = true;
2677
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2678
+ stuck_for_ms: this.now() - inFlight.dispatchedAt
2679
+ });
2680
+ }
2681
+ const activelyRunning = state === "running" && !awaitingHuman;
2682
+ if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
2683
+ this.log({
2684
+ level: "error",
2685
+ 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`,
2686
+ conversation_id: conv.id,
2687
+ message_id: inFlight.evidentMessageId
2688
+ });
2689
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2690
+ watched_for_ms: this.now() - inFlight.processingAnchorMs
2691
+ });
2692
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2693
+ return;
2694
+ }
2695
+ if (activelyRunning && !inFlight.awaitingHumanLatched && !inFlight.aliveInFlight && this.now() - inFlight.lastAliveAt >= HEARTBEAT_MS) {
2696
+ inFlight.aliveInFlight = true;
2697
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "alive").then((ok) => {
2698
+ inFlight.aliveInFlight = false;
2699
+ if (ok) inFlight.lastAliveAt = this.now();
2700
+ });
2701
+ }
2702
+ if (awaitingHuman) {
2703
+ if (!inFlight.awaitingHumanLatched) {
2704
+ inFlight.deadline = this.now() + this.pausedMaxWaitMs;
2705
+ inFlight.awaitingHumanLatched = true;
2706
+ }
2707
+ if (!inFlight.pausedClearConfirmed && !inFlight.pausedInFlight) {
2708
+ inFlight.pausedInFlight = true;
2709
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "paused").then((ok) => {
2710
+ inFlight.pausedInFlight = false;
2711
+ if (ok && inFlight.awaitingHumanLatched) inFlight.pausedClearConfirmed = true;
2712
+ });
2713
+ }
2714
+ } else if (inFlight.awaitingHumanLatched) {
2715
+ inFlight.awaitingHumanLatched = false;
2716
+ inFlight.pausedOnQuestion = false;
2717
+ inFlight.pausedOnPermission = false;
2718
+ inFlight.pausedClearConfirmed = false;
2719
+ }
2720
+ const siblingPaused = (sib) => openQuestions.has(sib.evidentMessageId) || openPermissions.has(sib.evidentMessageId) || sib.awaitingHumanLatched || sib.pausedOnQuestion || sib.pausedOnPermission;
2721
+ const hasActivelyRunningSibling = [...watcher.inFlight.values()].some(
2722
+ (sib) => sib.evidentMessageId !== inFlight.evidentMessageId && messageRunState(messages, sib.opencodeMessageId) === "running" && !siblingPaused(sib)
2723
+ );
2724
+ const queuedBehindRunningSibling = state === "queued" && hasActivelyRunningSibling;
2725
+ if (!activelyRunning && !queuedBehindRunningSibling && this.now() >= inFlight.deadline) {
1805
2726
  this.log({
1806
2727
  level: "info",
1807
- message: `Paused session ${sessionId.slice(0, 8)} did not complete within the watch window \u2014 leaving message ${message.id.slice(0, 8)} for the cron safety net`,
2728
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
1808
2729
  conversation_id: conv.id,
1809
- message_id: message.id
2730
+ message_id: inFlight.evidentMessageId
2731
+ });
2732
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2733
+ watched_for_ms: this.now() - inFlight.dispatchedAt
2734
+ });
2735
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2736
+ }
2737
+ }
2738
+ // -------------------------------------------------------------------------
2739
+ // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
2740
+ // -------------------------------------------------------------------------
2741
+ /**
2742
+ * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
2743
+ *
2744
+ * The pending drain only re-drives `pending` rows; a message already flipped to
2745
+ * `processing` before the runner died is watched by nobody until the 15-min
2746
+ * cron resets it. Here we fetch those rows, and per row resolve its correlated
2747
+ * reply against opencode's OWN session store — completing, re-attaching, or
2748
+ * (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
2749
+ * is idempotent per message (Invariant 2): a row a watcher already tracks is
2750
+ * skipped in `readoptOne` — one driver, no double-drive.
2751
+ *
2752
+ * Only `ChannelAuthError` propagates (to `drainPending`, like the pending
2753
+ * path); every other early return LOGS a reason with context — no silent drop.
2754
+ */
2755
+ async readoptProcessing() {
2756
+ const rows = await this.getProcessingMessages();
2757
+ if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0 || this.readoptPollUnresolvedSignalled.size > 0) {
2758
+ const stillProcessing = new Set(rows.map((r) => r.id));
2759
+ for (const id of [
2760
+ ...this.dontRedispatch,
2761
+ ...this.doneUndeliverable,
2762
+ ...this.readoptPollUnresolvedSignalled
2763
+ ]) {
2764
+ if (!stillProcessing.has(id)) {
2765
+ const cleared = this.dontRedispatch.delete(id);
2766
+ const clearedUndeliverable = this.doneUndeliverable.delete(id);
2767
+ this.readoptPollUnresolvedSignalled.delete(id);
2768
+ if (cleared || clearedUndeliverable) {
2769
+ this.log({
2770
+ level: "info",
2771
+ message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
2772
+ message_id: id
2773
+ });
2774
+ }
2775
+ }
2776
+ }
2777
+ }
2778
+ if (rows.length === 0) return;
2779
+ const bySession = /* @__PURE__ */ new Map();
2780
+ for (const row of rows) {
2781
+ if (!row.opencode_session_id) {
2782
+ this.log({
2783
+ level: "error",
2784
+ message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
2785
+ conversation_id: row.conversation_id,
2786
+ message_id: row.id
2787
+ });
2788
+ continue;
2789
+ }
2790
+ const list = bySession.get(row.opencode_session_id) ?? [];
2791
+ list.push(row);
2792
+ bySession.set(row.opencode_session_id, list);
2793
+ }
2794
+ for (const [sessionId, sessionRows] of bySession) {
2795
+ let messages;
2796
+ try {
2797
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
2798
+ if (!res.ok) {
2799
+ this.log({
2800
+ level: "error",
2801
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
2802
+ });
2803
+ continue;
2804
+ }
2805
+ const body = await res.json();
2806
+ if (!Array.isArray(body)) {
2807
+ this.log({
2808
+ level: "error",
2809
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
2810
+ });
2811
+ continue;
2812
+ }
2813
+ messages = body;
2814
+ } catch (err) {
2815
+ this.log({
2816
+ level: "error",
2817
+ message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
2818
+ });
2819
+ continue;
2820
+ }
2821
+ const anyUntracked = sessionRows.some((row) => !this.isTracked(sessionId, row.id));
2822
+ const sessionOngoing = anyUntracked ? await isSessionOngoing(this.port, sessionId) : null;
2823
+ for (const row of sessionRows) {
2824
+ await this.readoptOne(sessionId, row, messages, sessionOngoing);
2825
+ }
2826
+ }
2827
+ }
2828
+ /**
2829
+ * Re-adopt ONE `processing` row against the tick's session message snapshot
2830
+ * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
2831
+ *
2832
+ * Branches on `messageRunState(messages, row.opencode_message_id)` — the
2833
+ * opencode-assigned user-message id persisted on the first `processing` PATCH
2834
+ * (#218). A row with a NULL stored id (dispatched but the read-back never landed
2835
+ * before the restart) has no id to correlate → treated as an orphan and
2836
+ * re-dispatched (at most once, see `forceReadoptRun`):
2837
+ * - `done` → `markDone` now (guarded like the watcher's done branch);
2838
+ * - `failed` → `markFailed` with the surfaced error (issue #182), so an
2839
+ * errored turn is reported failed on restart, NOT re-dispatched;
2840
+ * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
2841
+ * tracking the stored id so the reply correlates by it;
2842
+ * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
2843
+ *
2844
+ * Only `ChannelAuthError` propagates.
2845
+ */
2846
+ async readoptOne(sessionId, row, messages, sessionOngoing) {
2847
+ if (this.isTracked(sessionId, row.id)) {
2848
+ this.log({
2849
+ level: "info",
2850
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
2851
+ conversation_id: row.conversation_id,
2852
+ message_id: row.id
2853
+ });
2854
+ return;
2855
+ }
2856
+ const ocId = row.opencode_message_id;
2857
+ const state = messageRunState(messages, ocId ?? "");
2858
+ if (state === "done") {
2859
+ if (this.doneUndeliverable.has(row.id)) {
2860
+ this.log({
2861
+ level: "info",
2862
+ message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
2863
+ conversation_id: row.conversation_id,
2864
+ message_id: row.id
2865
+ });
2866
+ return;
2867
+ }
2868
+ this.log({
2869
+ level: "info",
2870
+ message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
2871
+ conversation_id: row.conversation_id,
2872
+ message_id: row.id
2873
+ });
2874
+ try {
2875
+ const title = await this.resolveSessionTitle(sessionId, row.conversation_id);
2876
+ await this.markDone(row.conversation_id, row.id, sessionId, ocId, title);
2877
+ } catch (err) {
2878
+ if (err instanceof ChannelAuthError) throw err;
2879
+ if (err instanceof ChannelTerminalError) {
2880
+ this.doneUndeliverable.add(row.id);
2881
+ this.log({
2882
+ level: "error",
2883
+ 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}`,
2884
+ conversation_id: row.conversation_id,
2885
+ message_id: row.id
2886
+ });
2887
+ void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
2888
+ return;
2889
+ }
2890
+ this.log({
2891
+ level: "error",
2892
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2893
+ conversation_id: row.conversation_id,
2894
+ message_id: row.id
2895
+ });
2896
+ return;
2897
+ }
2898
+ this.dontRedispatch.delete(row.id);
2899
+ void this.postSignal(row.conversation_id, row.id, "readopt_done");
2900
+ return;
2901
+ }
2902
+ if (state === "failed") {
2903
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
2904
+ this.log({
2905
+ level: "error",
2906
+ message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2907
+ conversation_id: row.conversation_id,
2908
+ message_id: row.id
2909
+ });
2910
+ try {
2911
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2);
2912
+ } catch (err) {
2913
+ if (err instanceof ChannelAuthError) throw err;
2914
+ if (err instanceof ChannelTerminalError) {
2915
+ this.doneUndeliverable.add(row.id);
2916
+ this.log({
2917
+ level: "error",
2918
+ 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}`,
2919
+ conversation_id: row.conversation_id,
2920
+ message_id: row.id
2921
+ });
2922
+ void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
2923
+ return;
2924
+ }
2925
+ this.log({
2926
+ level: "error",
2927
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2928
+ conversation_id: row.conversation_id,
2929
+ message_id: row.id
2930
+ });
2931
+ return;
2932
+ }
2933
+ this.dontRedispatch.delete(row.id);
2934
+ void this.postSignal(row.conversation_id, row.id, "readopt_failed");
2935
+ return;
2936
+ }
2937
+ if (this.dontRedispatch.has(row.id)) {
2938
+ this.log({
2939
+ level: "info",
2940
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
2941
+ conversation_id: row.conversation_id,
2942
+ message_id: row.id
2943
+ });
2944
+ return;
2945
+ }
2946
+ let statusReadableOngoing = null;
2947
+ if (state === "running" && ocId) {
2948
+ const reply = findLastAssistantReplyFor(messages, ocId);
2949
+ const shape = this.replyCompletionShape(reply);
2950
+ const ongoing = sessionOngoing;
2951
+ statusReadableOngoing = ongoing;
2952
+ if (ongoing === false) {
2953
+ this.log({
2954
+ level: "info",
2955
+ 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)`,
2956
+ conversation_id: row.conversation_id,
2957
+ message_id: row.id
2958
+ });
2959
+ await this.forceReadoptRun(sessionId, row);
2960
+ return;
2961
+ }
2962
+ if (ongoing === true) {
2963
+ this.log({
2964
+ level: "info",
2965
+ 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)`,
2966
+ conversation_id: row.conversation_id,
2967
+ message_id: row.id
2968
+ });
2969
+ } else {
2970
+ if (shape === "b1") {
2971
+ this.log({
2972
+ level: "info",
2973
+ 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`,
2974
+ conversation_id: row.conversation_id,
2975
+ message_id: row.id
2976
+ });
2977
+ if (!this.readoptPollUnresolvedSignalled.has(row.id)) {
2978
+ this.readoptPollUnresolvedSignalled.add(row.id);
2979
+ void this.postSignal(row.conversation_id, row.id, "readopt_poll_unresolved");
2980
+ }
2981
+ return;
2982
+ }
2983
+ this.log({
2984
+ level: "info",
2985
+ 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`,
2986
+ conversation_id: row.conversation_id,
2987
+ message_id: row.id
2988
+ });
2989
+ }
2990
+ }
2991
+ if (statusReadableOngoing === null && state === "running" && ocId && isPreamblePinnedRunning(messages, ocId)) {
2992
+ const descendantAlive = await this.isAnyDescendantSessionAlive(sessionId);
2993
+ if (descendantAlive === true) {
2994
+ this.log({
2995
+ level: "info",
2996
+ 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)`,
2997
+ conversation_id: row.conversation_id,
2998
+ message_id: row.id
2999
+ });
3000
+ } else {
3001
+ this.log({
3002
+ level: "info",
3003
+ 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)" : ""}`,
3004
+ conversation_id: row.conversation_id,
3005
+ message_id: row.id
3006
+ });
3007
+ await this.forceReadoptRun(sessionId, row);
3008
+ return;
3009
+ }
3010
+ }
3011
+ if ((state === "running" || state === "queued") && ocId) {
3012
+ const conv = this.convForRow(sessionId, row);
3013
+ const message = this.queuedMessageForRow(row);
3014
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
3015
+ this.dispatched.add(row.id);
3016
+ this.readopted.add(row.id);
3017
+ this.ensureWatcherRunning(sessionId);
3018
+ this.log({
3019
+ level: "info",
3020
+ message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
3021
+ conversation_id: row.conversation_id,
3022
+ message_id: row.id
3023
+ });
3024
+ void this.postSignal(row.conversation_id, row.id, "readopt_reattached");
3025
+ return;
3026
+ }
3027
+ await this.forceReadoptRun(sessionId, row);
3028
+ }
3029
+ /**
3030
+ * Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
3031
+ *
3032
+ * #218/WI-5: the row's user message is absent (never kept, or a null stored id),
3033
+ * so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
3034
+ * read it back, and register the watcher under the assigned id so the reply
3035
+ * correlates server-side.
3036
+ *
3037
+ * ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
3038
+ * id). Without a guard, if this dispatches on tick N but the read-back+persist
3039
+ * hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
3040
+ * tick N+1 would dispatch AGAIN → duplicate user turns. The `awaitingReadopt`
3041
+ * latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
3042
+ * short-circuit while the row is latched; clear it on a successful dispatch (the
3043
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
3044
+ * re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
3045
+ * may retry exactly once more).
3046
+ *
3047
+ * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
3048
+ * `processed_at` (Invariant 1).
3049
+ */
3050
+ async forceReadoptRun(sessionId, row) {
3051
+ if (this.stopped) {
3052
+ this.log({
3053
+ level: "info",
3054
+ 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`,
3055
+ conversation_id: row.conversation_id,
3056
+ message_id: row.id
3057
+ });
3058
+ return;
3059
+ }
3060
+ if (this.awaitingReadopt.has(row.id)) {
3061
+ this.log({
3062
+ level: "info",
3063
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
3064
+ conversation_id: row.conversation_id,
3065
+ message_id: row.id
1810
3066
  });
3067
+ return;
3068
+ }
3069
+ if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
3070
+ this.dontRedispatch.add(row.id);
3071
+ this.log({
3072
+ level: "info",
3073
+ 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)`,
3074
+ conversation_id: row.conversation_id,
3075
+ message_id: row.id
3076
+ });
3077
+ void this.postSignal(row.conversation_id, row.id, "readopt_window_elapsed");
3078
+ return;
3079
+ }
3080
+ const options = {
3081
+ agent: row.opencode_agent ?? void 0,
3082
+ model: row.opencode_model ?? void 0
3083
+ };
3084
+ this.log({
3085
+ level: "info",
3086
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
3087
+ conversation_id: row.conversation_id,
3088
+ message_id: row.id
3089
+ });
3090
+ this.awaitingReadopt.add(row.id);
3091
+ let ocId;
3092
+ try {
3093
+ ocId = await this.dispatchLocked(
3094
+ sessionId,
3095
+ () => sendPromptAsync(this.port, sessionId, row.content, options)
3096
+ );
1811
3097
  } catch (err) {
3098
+ this.awaitingReadopt.delete(row.id);
3099
+ if (err instanceof ChannelAuthError) throw err;
1812
3100
  this.log({
1813
3101
  level: "error",
1814
- message: `Paused-session watcher failed for message ${message.id.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
1815
- conversation_id: conv.id,
1816
- message_id: message.id
3102
+ message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
3103
+ conversation_id: row.conversation_id,
3104
+ message_id: row.id
3105
+ });
3106
+ void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3107
+ return;
3108
+ }
3109
+ if (ocId === null) {
3110
+ this.awaitingReadopt.delete(row.id);
3111
+ this.log({
3112
+ level: "error",
3113
+ 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`,
3114
+ conversation_id: row.conversation_id,
3115
+ message_id: row.id
3116
+ });
3117
+ void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3118
+ return;
3119
+ }
3120
+ const conv = this.convForRow(sessionId, row);
3121
+ const message = this.queuedMessageForRow(row);
3122
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
3123
+ this.dispatched.add(row.id);
3124
+ this.readopted.add(row.id);
3125
+ this.awaitingReadopt.delete(row.id);
3126
+ this.ensureWatcherRunning(sessionId);
3127
+ void this.postSignal(row.conversation_id, row.id, "readopt_redispatched");
3128
+ }
3129
+ /**
3130
+ * True if `evidentMessageId` is already being driven — either in the
3131
+ * authoritative `dispatched` set or a live watcher's in-flight set for this
3132
+ * session (Invariant 2, WI-5). Either signal means a watcher owns the row.
3133
+ */
3134
+ isTracked(sessionId, evidentMessageId) {
3135
+ if (this.dispatched.has(evidentMessageId)) return true;
3136
+ const watcher = this.watchers.get(sessionId);
3137
+ return watcher?.inFlight.has(evidentMessageId) ?? false;
3138
+ }
3139
+ /**
3140
+ * Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
3141
+ * deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
3142
+ * for `processing` rows, but if it is somehow null/unparseable fall back to
3143
+ * `now` (defensive) AND log — a fallback means the anchor is weaker than
3144
+ * intended, which is worth surfacing.
3145
+ */
3146
+ processedAtMs(row) {
3147
+ const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
3148
+ if (!Number.isNaN(parsed)) return parsed;
3149
+ this.log({
3150
+ level: "error",
3151
+ message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
3152
+ conversation_id: row.conversation_id,
3153
+ message_id: row.id
3154
+ });
3155
+ return this.now();
3156
+ }
3157
+ /** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
3158
+ convForRow(sessionId, row) {
3159
+ return {
3160
+ id: row.conversation_id,
3161
+ agent_id: this.agentId,
3162
+ opencode_session_id: sessionId,
3163
+ pending_message_count: 0,
3164
+ oldest_pending_at: row.processed_at
3165
+ };
3166
+ }
3167
+ /** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
3168
+ queuedMessageForRow(row) {
3169
+ return {
3170
+ id: row.id,
3171
+ content: row.content,
3172
+ status: "processing",
3173
+ opencode_agent: row.opencode_agent,
3174
+ opencode_model: row.opencode_model,
3175
+ source_message_id: row.source_message_id,
3176
+ slack_user_id: row.slack_user_id
3177
+ };
3178
+ }
3179
+ /**
3180
+ * Remove a message from the in-flight set AND the authoritative dispatched
3181
+ * set. Once the in-flight set empties, the watcher loop's `while` guard exits
3182
+ * and its `.finally` removes the session entry from `this.watchers`.
3183
+ *
3184
+ * Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
3185
+ * (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
3186
+ * cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
3187
+ * re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
3188
+ * re-adopted message that completed (`done`) needs no marker — it's leaving
3189
+ * `processing`. This suppresses only re-dispatch: if its reply later completes,
3190
+ * the done branch still delivers it (Bugbot #202).
3191
+ */
3192
+ removeInFlight(watcher, evidentMessageId) {
3193
+ const inFlight = watcher.inFlight.get(evidentMessageId);
3194
+ if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
3195
+ this.dontRedispatch.add(evidentMessageId);
3196
+ this.log({
3197
+ level: "info",
3198
+ message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
3199
+ conversation_id: watcher.conv.id,
3200
+ message_id: evidentMessageId
1817
3201
  });
1818
3202
  }
3203
+ watcher.inFlight.delete(evidentMessageId);
3204
+ this.dispatched.delete(evidentMessageId);
3205
+ }
3206
+ /**
3207
+ * Poll `/question` + `/permission` (scoped to the session) and surface NEW ones
3208
+ * via `reportInteraction` (Task 3.5), carrying the PAUSED message's own
3209
+ * `source_message_id` so the server @mentions the correct person under
3210
+ * concurrency. Dedups by interaction id across ticks (reused per-session sets).
3211
+ *
3212
+ * The interaction is attributed to the in-flight message it paused on. opencode
3213
+ * stamps a `messageID` on a permission (and `tool.messageID` on a question) =
3214
+ * the assistant message id, whose `parentID` is the user message id — but the
3215
+ * simplest robust attribution here is: the single in-flight message that is
3216
+ * RUNNING (not done) is the one that paused. With one running message that is
3217
+ * unambiguous; with several we prefer an explicit messageID match, else the
3218
+ * oldest running message.
3219
+ *
3220
+ * Returns the set of in-flight Evident message ids that are paused awaiting a
3221
+ * human — an outstanding (still-open) question/permission is attributed to them.
3222
+ * `serviceInFlightMessage` uses this to keep an actively-running turn watched
3223
+ * forever (ADR-0047) while still bounding a turn merely blocked on a person who
3224
+ * may never answer. Attribution here covers ALL open interactions, not just
3225
+ * NEW (un-deduped) ones — a question stays "awaiting a human" until answered,
3226
+ * even after it was already surfaced to the channel.
3227
+ */
3228
+ async pollInteractions(sessionId, watcher, messages) {
3229
+ const openQuestions = /* @__PURE__ */ new Set();
3230
+ const openPermissions = /* @__PURE__ */ new Set();
3231
+ let questionsPolledOk = true;
3232
+ let permissionsPolledOk = true;
3233
+ let questions = [];
3234
+ try {
3235
+ const res = await this.fetchImpl(`${this.opencodeBase}/question`);
3236
+ if (res.ok) {
3237
+ const body = await res.json();
3238
+ if (Array.isArray(body)) {
3239
+ questions = body;
3240
+ } else {
3241
+ questionsPolledOk = false;
3242
+ }
3243
+ } else {
3244
+ questionsPolledOk = false;
3245
+ }
3246
+ } catch {
3247
+ questionsPolledOk = false;
3248
+ }
3249
+ for (const q of questions) {
3250
+ if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
3251
+ const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
3252
+ if (paused) openQuestions.add(paused.evidentMessageId);
3253
+ if (watcher.reportedQuestions.has(q.id)) continue;
3254
+ const reported = await this.reportInteraction(
3255
+ watcher.conv.id,
3256
+ "question",
3257
+ q,
3258
+ paused?.message.source_message_id ?? void 0
3259
+ );
3260
+ if (reported) watcher.reportedQuestions.add(q.id);
3261
+ }
3262
+ let permissions = [];
3263
+ try {
3264
+ const res = await this.fetchImpl(`${this.opencodeBase}/permission`);
3265
+ if (res.ok) {
3266
+ const body = await res.json();
3267
+ if (Array.isArray(body)) {
3268
+ permissions = body;
3269
+ } else {
3270
+ permissionsPolledOk = false;
3271
+ }
3272
+ } else {
3273
+ permissionsPolledOk = false;
3274
+ }
3275
+ } catch {
3276
+ permissionsPolledOk = false;
3277
+ }
3278
+ for (const p of permissions) {
3279
+ if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
3280
+ const paused = this.attributeInteraction(watcher, p.messageID, messages);
3281
+ if (paused) openPermissions.add(paused.evidentMessageId);
3282
+ if (watcher.reportedPermissions.has(p.id)) continue;
3283
+ const reported = await this.reportInteraction(
3284
+ watcher.conv.id,
3285
+ "permission",
3286
+ p,
3287
+ paused?.message.source_message_id ?? void 0
3288
+ );
3289
+ if (reported) watcher.reportedPermissions.add(p.id);
3290
+ }
3291
+ return { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk };
3292
+ }
3293
+ /**
3294
+ * True when `sessionId` is the `rootSessionId` itself OR a descendant of it —
3295
+ * i.e. its `parentID` chain (resolved via `GET /session/:id`) reaches the
3296
+ * watched root. Sub-agents spawned via the `task` tool run in child sessions,
3297
+ * so their questions/permissions live under a different `sessionID` that must
3298
+ * still be attributed to the root conversation the watcher owns.
3299
+ *
3300
+ * Parents are cached in `sessionParents` so we walk each session at most once;
3301
+ * a bounded depth cap guards against a cycle or a pathological chain, and any
3302
+ * fetch failure is treated as "not a descendant" (best-effort — the interaction
3303
+ * simply isn't surfaced this tick and is retried next tick once resolvable).
3304
+ */
3305
+ async sessionBelongsTo(sessionId, rootSessionId) {
3306
+ let current = sessionId;
3307
+ for (let depth = 0; current && depth < 32; depth++) {
3308
+ if (current === rootSessionId) return true;
3309
+ const parent = await this.resolveSessionParent(current);
3310
+ if (parent === null || parent === void 0) return false;
3311
+ current = parent;
3312
+ }
3313
+ return false;
3314
+ }
3315
+ /**
3316
+ * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3317
+ * `null` for a root session (no parent) and `undefined` when opencode is
3318
+ * unreachable / the session can't be read (so the caller stops walking without
3319
+ * caching a wrong answer — the next tick retries).
3320
+ */
3321
+ async resolveSessionParent(sessionId) {
3322
+ const cached = this.sessionParents.get(sessionId);
3323
+ if (cached !== void 0) return cached;
3324
+ let parent = void 0;
3325
+ try {
3326
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
3327
+ if (res.ok) {
3328
+ const body = await res.json();
3329
+ parent = body && typeof body.parentID === "string" ? body.parentID : null;
3330
+ }
3331
+ } catch {
3332
+ parent = void 0;
3333
+ }
3334
+ if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3335
+ return parent;
3336
+ }
3337
+ /**
3338
+ * Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
3339
+ * status PATCH can carry it into the "Live sessions" list. Driver-level cache so
3340
+ * BOTH the watcher completion path and the restart-recovery re-adopt path (which
3341
+ * has no watcher) can use it. `conversationId` is passed only for log context.
3342
+ * Best-effort:
3343
+ * - a resolved NON-EMPTY title is cached and terminal (a real session name
3344
+ * won't later un-name), so we do NOT re-GET `/session/:id` every tick;
3345
+ * - while the title is still absent/empty we do NOT latch it — OpenCode names
3346
+ * sessions asynchronously mid-turn, so an early call (e.g. at `processing`)
3347
+ * must leave the cache unresolved and re-fetch on the next need so a later
3348
+ * call (e.g. at `done`) picks up the name assigned in the meantime. Such a
3349
+ * call returns `null` (omit the title on THIS PATCH) without caching;
3350
+ * - a failed request likewise leaves the cache unresolved (retry next need)
3351
+ * and returns `null` — it must NEVER throw or block completion.
3352
+ * A failure is logged with agent/session context (no silent catch).
3353
+ */
3354
+ async resolveSessionTitle(sessionId, conversationId) {
3355
+ const cached = this.sessionTitles.get(sessionId);
3356
+ if (cached != null) return cached;
3357
+ try {
3358
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
3359
+ if (res.ok) {
3360
+ const body = await res.json();
3361
+ const title = body && typeof body.title === "string" ? body.title.trim() : "";
3362
+ if (title.length > 0) {
3363
+ this.sessionTitles.set(sessionId, title);
3364
+ return title;
3365
+ }
3366
+ return null;
3367
+ }
3368
+ this.log({
3369
+ level: "info",
3370
+ message: `Session title fetch for session ${sessionId.slice(0, 8)} (agent ${this.agentId.slice(0, 8)}) returned HTTP ${res.status} \u2014 omitting title`,
3371
+ conversation_id: conversationId
3372
+ });
3373
+ } catch (err) {
3374
+ this.log({
3375
+ level: "info",
3376
+ 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)}`,
3377
+ conversation_id: conversationId
3378
+ });
3379
+ }
3380
+ return null;
3381
+ }
3382
+ /**
3383
+ * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3384
+ * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
3385
+ *
3386
+ * The PRIMARY recovery trigger is "preamble-pinned on recovery ⇒ idle" — a
3387
+ * runner restart wipes OpenCode's in-memory `SessionStatus`/`Runner`, so a
3388
+ * completed `finish: "tool-calls"` root reply encountered during re-adoption is
3389
+ * idle by OpenCode's own definition and is re-dispatched. This method exists only
3390
+ * so the WI-3 caller can VETO that re-dispatch in the rare case a descendant is
3391
+ * provably in flight at the exact moment of recovery.
3392
+ *
3393
+ * "Alive" criterion (TIGHTENED): a descendant is alive only when it is PROVABLY,
3394
+ * ACTIVELY generating — its LAST message is an assistant still mid-generation
3395
+ * (`completed == null`, via `isSessionActivelyGenerating`). An
3396
+ * INCOMPLETE-BUT-NOT-GENERATING child — last message a user message, or a
3397
+ * completed `finish: "tool-calls"` step — is NOT alive after a restart (nothing
3398
+ * is generating once the runner is gone), so it does NOT veto. (This is
3399
+ * deliberately NOT `!isTurnComplete`, which also matches those dead-but-non-terminal
3400
+ * shapes and would falsely veto — re-hanging the very turn this path recovers.)
3401
+ *
3402
+ * Return contract (encoded so WI-3 need not re-derive it):
3403
+ * - `true` → a descendant is provably, actively generating (veto re-dispatch).
3404
+ * - `false` → descendants exist but none is actively generating (the restart
3405
+ * case), OR no descendant is found at all.
3406
+ * - `null` → liveness is INDETERMINATE (enumeration via `listSessions` failed).
3407
+ *
3408
+ * ⚠️ `null` (UNKNOWN) MUST NOT be treated as "alive": WI-3 treats `null` the same
3409
+ * as `false` and does NOT veto — a restart guarantees no live runner, so an
3410
+ * indeterminate cross-check almost always means "couldn't reach a child that no
3411
+ * longer exists". The inversion lives in the caller; this method just reports
3412
+ * true/false/null faithfully.
3413
+ *
3414
+ * VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
3415
+ * (already proven by the existing child-session interaction tests, via
3416
+ * `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
3417
+ * terminal state. We do NOT depend on any session-level `busy`/`idle` field —
3418
+ * there is none on `GET /session/:id`; OpenCode's busy state is in-memory
3419
+ * `SessionStatus` only.
3420
+ */
3421
+ async isAnyDescendantSessionAlive(rootSessionId) {
3422
+ const sessions = await listSessions(this.port);
3423
+ if (!sessions) {
3424
+ this.log({
3425
+ level: "error",
3426
+ message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
3427
+ });
3428
+ return null;
3429
+ }
3430
+ for (const candidate of sessions) {
3431
+ if (!candidate?.id || candidate.id === rootSessionId) continue;
3432
+ if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
3433
+ const childMsgs = await getSessionMessages(this.port, candidate.id);
3434
+ if (isSessionActivelyGenerating(childMsgs)) {
3435
+ return true;
3436
+ }
3437
+ }
3438
+ return false;
3439
+ }
3440
+ /**
3441
+ * Cheap decision-telemetry label for a running row's LAST correlated reply
3442
+ * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
3443
+ * - `b1` — the reply itself is still in flight (`time.completed == null`) —
3444
+ * the aborted-in-flight production bug after a restart.
3445
+ * - `b2` — a COMPLETED reply pinned running only by `finish === "tool-calls"`
3446
+ * (the sub-agent preamble — #253's shape).
3447
+ * - `other` — any other shape (defensive; a running row is normally b1 or b2).
3448
+ * Reads `info.time.completed` / `info.finish` (tolerating the legacy top-level
3449
+ * shape) directly rather than re-importing the module-private `completedOf`/
3450
+ * `finishOf` — this is a display label only, not a correctness predicate.
3451
+ */
3452
+ replyCompletionShape(reply) {
3453
+ if (!reply) return "other";
3454
+ const completed = reply.info?.time?.completed ?? reply.time?.completed;
3455
+ if (completed == null) return "b1";
3456
+ const finish = reply.info?.finish ?? reply.finish;
3457
+ return finish === "tool-calls" ? "b2" : "other";
3458
+ }
3459
+ /**
3460
+ * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
3461
+ *
3462
+ * The interaction carries `interactionMessageId` — the ASSISTANT message id
3463
+ * that raised it (a question's `tool.messageID` / a permission's `messageID`).
3464
+ * That assistant message is the reply to ONE of our minted user messages
3465
+ * (correlated by `parentID`, GATE-B). So when we have the tick's message
3466
+ * snapshot, we resolve each running in-flight message's correlated assistant
3467
+ * reply (`findAssistantReplyAfter`) and match its id against
3468
+ * `interactionMessageId` — giving an EXACT attribution even with several
3469
+ * messages in flight concurrently in one session.
3470
+ *
3471
+ * We fall back to the oldest running message ONLY when no exact match is
3472
+ * possible (the id is absent, the snapshot is missing, or the reply has not yet
3473
+ * been correlated). With a single running message either path is exact. Never
3474
+ * throws.
3475
+ *
3476
+ * Attribution must NOT depend on our own `started` PATCH flag: opencode can
3477
+ * START a turn AND raise a question/permission BEFORE our next tick fires
3478
+ * `markProcessing` (which sets `started`). Relying on `started` would leave the
3479
+ * running set empty in that window and let the server fall back to "newest
3480
+ * processing/pending" — possibly @mentioning a FOLLOW-UP author rather than the
3481
+ * person whose active turn actually paused. So we derive "running" from the
3482
+ * tick's `messages` snapshot via `messageRunState` instead.
3483
+ */
3484
+ attributeInteraction(watcher, interactionMessageId, messages) {
3485
+ const inFlight = [...watcher.inFlight.values()].filter((m) => !m.done);
3486
+ if (inFlight.length === 0) return void 0;
3487
+ if (interactionMessageId && messages) {
3488
+ const exact = inFlight.find((m) => {
3489
+ const reply = findAssistantReplyAfter(messages, m.opencodeMessageId);
3490
+ return reply != null && messageIdOf(reply) === interactionMessageId;
3491
+ });
3492
+ if (exact) return exact;
3493
+ }
3494
+ const byOldest = (a, b) => a.dispatchedAt - b.dispatchedAt;
3495
+ if (messages) {
3496
+ const runningPerSnapshot = inFlight.filter(
3497
+ (m) => messageRunState(messages, m.opencodeMessageId) === "running"
3498
+ );
3499
+ if (runningPerSnapshot.length > 0) {
3500
+ return runningPerSnapshot.sort(byOldest)[0];
3501
+ }
3502
+ }
3503
+ const startedRunning = inFlight.filter((m) => m.started);
3504
+ if (startedRunning.length > 0) {
3505
+ return startedRunning.sort(byOldest)[0];
3506
+ }
3507
+ return inFlight.sort(byOldest)[0];
1819
3508
  }
1820
3509
  // -------------------------------------------------------------------------
1821
3510
  // Evident API calls (combinedAuth thread routes)
@@ -1849,49 +3538,193 @@ var ChannelDriver = class {
1849
3538
  }
1850
3539
  return await res.json();
1851
3540
  }
1852
- async markProcessing(conversationId, messageId) {
3541
+ /**
3542
+ * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
3543
+ * The pending path (`getPendingConversations`/`getPendingMessages`) only
3544
+ * surfaces `pending` rows, so a message already `processing` when the runner
3545
+ * died is invisible to it — this dedicated endpoint returns exactly those rows
3546
+ * with the fields the re-adopt path needs (`processed_at`,
3547
+ * `opencode_session_id`, routing).
3548
+ *
3549
+ * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
3550
+ * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
3551
+ * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
3552
+ * next tick retries.
3553
+ */
3554
+ async getProcessingMessages() {
3555
+ const res = await this.fetchImpl(
3556
+ `${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
3557
+ { headers: { Authorization: this.getAuthHeader() } }
3558
+ );
3559
+ this.assertAuth(res, "fetching processing messages");
3560
+ if (!res.ok) {
3561
+ throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
3562
+ }
3563
+ const data = await res.json();
3564
+ let messages = data.messages ?? [];
3565
+ if (this.conversationFilter) {
3566
+ messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
3567
+ }
3568
+ return messages;
3569
+ }
3570
+ /**
3571
+ * EXISTING combinedAuth route — now fired by the watcher on queued→running
3572
+ * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
3573
+ * opencode_session_id}` → `notifyMessageStarted` (hourglass→runner swap +
3574
+ * deep-linked "View in Evident" notice).
3575
+ *
3576
+ * Return/throw contract (consumed by the watcher's swap-to-running guard):
3577
+ * - returns `true` → the server transitioned the row to processing;
3578
+ * - returns `false` → the server gave a DEFINITIVE "already-processing"
3579
+ * answer (a non-retryable, non-auth status — e.g. a
3580
+ * conflict because a duplicate already transitioned it),
3581
+ * so the caller treats it as already-started and does NOT
3582
+ * retry;
3583
+ * - throws `ChannelAuthError` on 401/403 (terminal auth failure);
3584
+ * - throws on a TRANSIENT failure (retryable 5xx/429 status, or a
3585
+ * network-level error from `fetch`) — i.e. NO definitive server response —
3586
+ * so the caller leaves the message un-started and retries the swap on the
3587
+ * next tick.
3588
+ * A single attempt (no internal retry): the watcher's per-tick loop is the
3589
+ * retry vehicle for the swap-to-running.
3590
+ */
3591
+ async markProcessing(conversationId, messageId, sessionId, opencodeMessageId, title) {
1853
3592
  const res = await this.fetchImpl(
1854
3593
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
1855
3594
  {
1856
3595
  method: "PATCH",
1857
3596
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1858
- body: JSON.stringify({ status: "processing" })
3597
+ body: JSON.stringify({
3598
+ status: "processing",
3599
+ opencode_session_id: sessionId,
3600
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3601
+ ...title ? { title } : {}
3602
+ })
1859
3603
  }
1860
3604
  );
1861
3605
  this.assertAuth(res, "marking message as processing");
1862
- return res.ok;
3606
+ if (res.ok) return true;
3607
+ if (isRetryableStatus(res.status)) {
3608
+ throw new Error(`marking message as processing: HTTP ${res.status}`);
3609
+ }
3610
+ return false;
1863
3611
  }
1864
3612
  /**
1865
- * EXISTING combinedAuth completion route — idempotent + retried (WI-CHAN-2).
1866
- * `PATCH .../messages/:id {status:'done', opencode_session_id}`. The server's
3613
+ * EXISTING combinedAuth completion route — idempotent (WI-CHAN-2). `PATCH
3614
+ * .../messages/:id {status:'done', opencode_session_id}`. The server's
1867
3615
  * `queued_conversation_messages.status`/`processed_at` gate makes a re-call
1868
- * for an already-`done` message a no-op (no double Slack post).
3616
+ * for an already-`done` message a no-op (no double Slack post). Fired by the
3617
+ * watcher on per-message completion (Task 3.4) — no `confirmCompletion`
3618
+ * round-trip (we already observed completion via the message list).
3619
+ *
3620
+ * SINGLE ATTEMPT (no in-call `callWithRetry` backoff). The per-session watcher
3621
+ * services its in-flight messages SEQUENTIALLY within a tick
3622
+ * (`runWatcherLoop` → `serviceInFlightMessage`), so a long multi-attempt
3623
+ * backoff here would BLOCK sibling messages in the SAME session/tick: while
3624
+ * message A's done PATCH burned its internal retries, message B could not be
3625
+ * swapped to running even though opencode had already started it. Instead this
3626
+ * does ONE PATCH and surfaces the SAME outcome contract the watcher's markDone
3627
+ * handler already relies on, leaning on the per-tick retry across ticks
3628
+ * (bounded by `inFlight.deadline`) rather than an in-call retry:
3629
+ * - resolves (`void`) → the server transitioned the row to done
3630
+ * (or idempotently confirmed already-done);
3631
+ * - throws `ChannelAuthError` → 401/403 (terminal auth failure → loop
3632
+ * cleanup, Finding 1);
3633
+ * - throws `ChannelTerminalError`→ non-retryable, non-auth 4xx (will never
3634
+ * succeed → straight to the cron, Finding 4);
3635
+ * - throws a plain `Error` → TRANSIENT 5xx/429 or a network-level error
3636
+ * (no definitive server response → the
3637
+ * watcher retries next tick within the
3638
+ * deadline, Finding 4).
1869
3639
  */
1870
- async markDone(conversationId, messageId, sessionId) {
3640
+ async markDone(conversationId, messageId, sessionId, opencodeMessageId, title) {
3641
+ const res = await this.fetchImpl(
3642
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
3643
+ {
3644
+ method: "PATCH",
3645
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3646
+ body: JSON.stringify({
3647
+ status: "done",
3648
+ opencode_session_id: sessionId,
3649
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
3650
+ ...title ? { title } : {}
3651
+ })
3652
+ }
3653
+ );
3654
+ this.assertAuth(res, "marking message as done");
3655
+ if (res.ok) return;
3656
+ if (isRetryableStatus(res.status)) {
3657
+ throw new Error(`marking message as done: HTTP ${res.status}`);
3658
+ }
3659
+ throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
3660
+ }
3661
+ /**
3662
+ * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3663
+ * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3664
+ * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3665
+ * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3666
+ * failure reason reaches the channel.
3667
+ */
3668
+ async markFailed(conversationId, messageId, sessionId, error2) {
3669
+ const body = { status: "failed" };
3670
+ if (sessionId !== void 0) body.opencode_session_id = sessionId;
3671
+ if (error2 !== void 0) body.error = error2;
1871
3672
  await this.callWithRetry(
1872
- "marking message as done",
3673
+ "marking message as failed",
1873
3674
  () => this.fetchImpl(
1874
3675
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
1875
3676
  {
1876
3677
  method: "PATCH",
1877
3678
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1878
- body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
3679
+ body: JSON.stringify(body)
1879
3680
  }
1880
3681
  )
1881
3682
  );
1882
3683
  }
1883
- async markFailed(conversationId, messageId) {
1884
- await this.callWithRetry(
1885
- "marking message as failed",
1886
- () => this.fetchImpl(
1887
- `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
3684
+ /**
3685
+ * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3686
+ * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
3687
+ * — the server records it via `log()` (no DB write, no notification). This is
3688
+ * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
3689
+ * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
3690
+ * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
3691
+ * context (no silent catch, per development-workflow).
3692
+ *
3693
+ * Returns whether the POST SUCCEEDED (2xx). Most callers ignore this (pure
3694
+ * telemetry), but the `paused` liveness-clear uses it to know whether to
3695
+ * RE-ASSERT on a later tick — a single dropped `paused` POST must not leave a
3696
+ * stale `last_seen_alive_at` on a still-paused row (Bugbot "Failed paused signal
3697
+ * leaves liveness").
3698
+ */
3699
+ async postSignal(conversationId, messageId, signal, extra) {
3700
+ try {
3701
+ const res = await this.fetchImpl(
3702
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
1888
3703
  {
1889
- method: "PATCH",
3704
+ method: "POST",
1890
3705
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1891
- body: JSON.stringify({ status: "failed" })
3706
+ body: JSON.stringify({ signal, ...extra })
1892
3707
  }
1893
- )
1894
- );
3708
+ );
3709
+ if (!res.ok) {
3710
+ this.log({
3711
+ level: "error",
3712
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
3713
+ conversation_id: conversationId,
3714
+ message_id: messageId
3715
+ });
3716
+ return false;
3717
+ }
3718
+ return true;
3719
+ } catch (err) {
3720
+ this.log({
3721
+ level: "error",
3722
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
3723
+ conversation_id: conversationId,
3724
+ message_id: messageId
3725
+ });
3726
+ return false;
3727
+ }
1895
3728
  }
1896
3729
  async persistSession(conversationId, sessionId) {
1897
3730
  const res = await this.fetchImpl(
@@ -1906,10 +3739,17 @@ var ChannelDriver = class {
1906
3739
  }
1907
3740
  /**
1908
3741
  * EXISTING combinedAuth interaction route (WI-CHAN-3) — idempotent + retried.
1909
- * `POST .../interactive-event {type, data}`. The server persists the
1910
- * interaction and posts a link to the proxied opencode-web conversation.
3742
+ * `POST .../interactive-event {type, data, source_message_id?}`. The server
3743
+ * persists the interaction and posts a link to the proxied opencode-web
3744
+ * conversation, @mentioning the user who triggered THIS message's turn.
3745
+ *
3746
+ * WI-3 / WI-4 contract: `source_message_id` is the PAUSED message's own Slack
3747
+ * ts (`message.source_message_id`). The server resolves the @mention from that
3748
+ * message's user FIRST (falling back to the old "newest processing" precedence
3749
+ * only when absent), so the correct person is mentioned under concurrency. It
3750
+ * is OPTIONAL for back-compat with older clients / legacy rows.
1911
3751
  */
1912
- async reportInteraction(conversationId, type, data) {
3752
+ async reportInteraction(conversationId, type, data, sourceMessageId) {
1913
3753
  try {
1914
3754
  await this.callWithRetry(
1915
3755
  "reporting interactive event",
@@ -1918,7 +3758,9 @@ var ChannelDriver = class {
1918
3758
  {
1919
3759
  method: "POST",
1920
3760
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1921
- body: JSON.stringify({ type, data })
3761
+ body: JSON.stringify(
3762
+ sourceMessageId ? { type, data, source_message_id: sourceMessageId } : { type, data }
3763
+ )
1922
3764
  }
1923
3765
  )
1924
3766
  );
@@ -1927,6 +3769,7 @@ var ChannelDriver = class {
1927
3769
  message: `${type} surfaced to channel (id: ${data.id.slice(0, 8)})`,
1928
3770
  conversation_id: conversationId
1929
3771
  });
3772
+ return true;
1930
3773
  } catch (err) {
1931
3774
  if (err instanceof ChannelAuthError) throw err;
1932
3775
  this.log({
@@ -1934,6 +3777,7 @@ var ChannelDriver = class {
1934
3777
  message: `Failed to surface ${type}: ${err instanceof Error ? err.message : String(err)}`,
1935
3778
  conversation_id: conversationId
1936
3779
  });
3780
+ return false;
1937
3781
  }
1938
3782
  }
1939
3783
  // -------------------------------------------------------------------------
@@ -1972,8 +3816,9 @@ var ChannelDriver = class {
1972
3816
  await this.sleep(backoffDelay(attempt, this.retry));
1973
3817
  continue;
1974
3818
  }
3819
+ break;
1975
3820
  }
1976
- throw new Error(`${context}: HTTP ${res.status}`);
3821
+ throw new ChannelTerminalError(`${context}: HTTP ${res.status}`, res.status);
1977
3822
  }
1978
3823
  throw lastError instanceof Error ? lastError : new Error(`${context}: exhausted retries`);
1979
3824
  }
@@ -2151,6 +3996,25 @@ async function resolveAgentIdFromKey(authHeader) {
2151
3996
  return { error: `Failed to resolve agent from key: ${message}` };
2152
3997
  }
2153
3998
  }
3999
+ async function notifyAgentDisconnected(agentId, authHeader) {
4000
+ const apiUrl = getApiUrlConfig();
4001
+ try {
4002
+ const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
4003
+ method: "POST",
4004
+ headers: { Authorization: authHeader }
4005
+ });
4006
+ if (!response.ok) {
4007
+ const serverMessage = await readErrorMessage(response);
4008
+ return {
4009
+ ok: false,
4010
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
4011
+ };
4012
+ }
4013
+ return { ok: true };
4014
+ } catch (error2) {
4015
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
4016
+ }
4017
+ }
2154
4018
  async function getAgentInfo(agentId, authHeader) {
2155
4019
  const apiUrl = getApiUrlConfig();
2156
4020
  try {
@@ -2196,7 +4060,9 @@ async function getAgentInfo(agentId, authHeader) {
2196
4060
  // src/commands/run.ts
2197
4061
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
2198
4062
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
2199
- function log(state, message, isError = false) {
4063
+ var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4064
+ var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
4065
+ function log2(state, message, isError = false) {
2200
4066
  if (state.json) {
2201
4067
  console.log(
2202
4068
  JSON.stringify({
@@ -2221,9 +4087,9 @@ function logActivity(state, entry) {
2221
4087
  }
2222
4088
  if (!state.interactive) {
2223
4089
  if (entry.type === "error") {
2224
- log(state, entry.error ?? "Unknown error", true);
4090
+ log2(state, entry.error ?? "Unknown error", true);
2225
4091
  } else if (entry.type === "info" && entry.message) {
2226
- log(state, entry.message);
4092
+ log2(state, entry.message);
2227
4093
  }
2228
4094
  }
2229
4095
  }
@@ -2309,6 +4175,7 @@ async function handleAuthError(state, error2) {
2309
4175
  }
2310
4176
  async function driveChannels(state, driver) {
2311
4177
  let idlePolls = 0;
4178
+ let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
2312
4179
  while (state.running) {
2313
4180
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
2314
4181
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2318,9 +4185,11 @@ async function driveChannels(state, driver) {
2318
4185
  try {
2319
4186
  const processed = await driver.drainPending();
2320
4187
  state.messageCount += processed;
2321
- if (processed > 0) {
4188
+ const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
4189
+ lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
4190
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
2322
4191
  idlePolls = 0;
2323
- if (state.interactive) displayStatus(state);
4192
+ if (processed > 0 && state.interactive) displayStatus(state);
2324
4193
  } else if (state.idleTimeout !== null) {
2325
4194
  idlePolls++;
2326
4195
  if (idlePolls === 1) {
@@ -2347,7 +4216,7 @@ async function driveChannels(state, driver) {
2347
4216
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
2348
4217
  if (state.interactive) displayStatus(state);
2349
4218
  }
2350
- await new Promise((resolve) => setTimeout(resolve, CHANNEL_POLL_INTERVAL_MS));
4219
+ await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
2351
4220
  if (state.idleTimeout !== null && idlePolls >= 2) {
2352
4221
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
2353
4222
  if (idleMs > state.idleTimeout * 1e3) {
@@ -2358,8 +4227,122 @@ async function driveChannels(state, driver) {
2358
4227
  }
2359
4228
  }
2360
4229
  }
2361
- async function cleanup(state) {
4230
+ var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
4231
+ async function runSweep(state, driver, config2) {
4232
+ const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
4233
+ try {
4234
+ const sessions = await listSessions(state.port);
4235
+ if (sessions === null) {
4236
+ logActivity(state, {
4237
+ type: "info",
4238
+ message: `Session cleanup: could not list sessions (opencode unreachable); skipping this sweep (${mode})`
4239
+ });
4240
+ return;
4241
+ }
4242
+ const toDelete = selectSessionsToDelete(
4243
+ sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
4244
+ {
4245
+ maxAgeMs: config2.maxAgeMs,
4246
+ maxCount: config2.maxCount,
4247
+ nowMs: Date.now(),
4248
+ protectedIds: driver.protectedSessionIds()
4249
+ }
4250
+ );
4251
+ const protectedNow = driver.protectedSessionIds();
4252
+ let deleted = 0;
4253
+ let failed = 0;
4254
+ let skippedNewlyActive = 0;
4255
+ for (const id of toDelete) {
4256
+ if (protectedNow.has(id)) {
4257
+ skippedNewlyActive++;
4258
+ logActivity(state, {
4259
+ type: "info",
4260
+ message: `Session cleanup: skipping ${id} \u2014 became active/bound after selection (${mode})`
4261
+ });
4262
+ continue;
4263
+ }
4264
+ if (await deleteSession(state.port, id)) deleted++;
4265
+ else failed++;
4266
+ }
4267
+ const failedNote = failed > 0 ? `, failed ${failed}` : "";
4268
+ const skippedNote = skippedNewlyActive > 0 ? `, skipped ${skippedNewlyActive} newly-active` : "";
4269
+ logActivity(state, {
4270
+ type: "info",
4271
+ message: `Session cleanup: inspected ${sessions.length}, deleted ${deleted}${failedNote}${skippedNote} (${mode})`
4272
+ });
4273
+ } catch (error2) {
4274
+ const message = error2 instanceof Error ? error2.message : String(error2);
4275
+ logActivity(state, {
4276
+ type: "error",
4277
+ error: `Session cleanup sweep failed (non-fatal, ${mode}): ${message}`
4278
+ });
4279
+ }
4280
+ }
4281
+ function scheduleSessionCleanup(state, driver, options) {
4282
+ const config2 = resolveSessionCleanupConfig(
4283
+ {
4284
+ maxAge: options.sessionCleanupMaxAge,
4285
+ maxCount: options.sessionCleanupMaxCount,
4286
+ interval: options.sessionCleanupInterval
4287
+ },
4288
+ process.env
4289
+ );
4290
+ for (const warning2 of config2.warnings) {
4291
+ logActivity(state, { type: "info", message: `Session cleanup: ${warning2}` });
4292
+ }
4293
+ if (!config2.enabled) return;
4294
+ logActivity(state, {
4295
+ type: "info",
4296
+ message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
4297
+ });
4298
+ const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
4299
+ const firstSweep = setTimeout(
4300
+ () => void runSweep(state, driver, config2),
4301
+ SESSION_CLEANUP_FIRST_SWEEP_MS
4302
+ );
4303
+ state.sessionCleanupTimers.push(interval, firstSweep);
4304
+ }
4305
+ async function notifyOffline(state) {
4306
+ if (!state.agentId || !state.authHeader) return;
4307
+ if (!state.connected) {
4308
+ log2(state, "Skipping offline signal \u2014 this runner does not hold the live tunnel");
4309
+ return;
4310
+ }
4311
+ const result = await notifyAgentDisconnected(state.agentId, state.authHeader);
4312
+ if (result.ok) {
4313
+ log2(state, "Notified Evident the agent is going offline");
4314
+ } else {
4315
+ logActivity(state, {
4316
+ type: "error",
4317
+ error: `Could not notify Evident of offline status (relay will still report it): ${result.error}`
4318
+ });
4319
+ if (state.interactive) displayStatus(state);
4320
+ }
4321
+ }
4322
+ async function cleanup(state, opts = {}) {
2362
4323
  state.running = false;
4324
+ for (const timer of state.sessionCleanupTimers) {
4325
+ clearInterval(timer);
4326
+ clearTimeout(timer);
4327
+ }
4328
+ state.sessionCleanupTimers = [];
4329
+ if (opts.graceful && state.channelDriver) {
4330
+ state.channelDriver.stop();
4331
+ log2(state, "Draining in-flight channel work before shutdown...");
4332
+ if (state.interactive) {
4333
+ logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4334
+ displayStatus(state);
4335
+ }
4336
+ const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
4337
+ if (!settled) {
4338
+ logActivity(state, {
4339
+ type: "info",
4340
+ message: "Shutdown drain timed out with work still in flight \u2014 leaving it for restart recovery"
4341
+ });
4342
+ if (state.interactive) displayStatus(state);
4343
+ }
4344
+ }
4345
+ await notifyOffline(state);
2363
4346
  if (state.connection) {
2364
4347
  state.connection.close();
2365
4348
  state.connection = null;
@@ -2370,7 +4353,7 @@ async function cleanup(state) {
2370
4353
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
2371
4354
  displayStatus(state);
2372
4355
  } else {
2373
- log(state, "Stopped OpenCode process");
4356
+ log2(state, "Stopped OpenCode process");
2374
4357
  }
2375
4358
  state.opencodeProcess = null;
2376
4359
  }
@@ -2390,26 +4373,32 @@ async function run(options) {
2390
4373
  opencodeVersion: null,
2391
4374
  opencodeProcess: null,
2392
4375
  connection: null,
4376
+ channelDriver: null,
2393
4377
  running: true,
4378
+ shuttingDown: false,
2394
4379
  activityLog: [],
2395
4380
  messageCount: 0,
4381
+ lastProxiedActivityAt: null,
4382
+ sessionCleanupTimers: [],
2396
4383
  authHeader: ""
2397
4384
  };
2398
4385
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
2399
- log(
4386
+ log2(
2400
4387
  state,
2401
4388
  "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.",
2402
4389
  false
2403
4390
  );
2404
4391
  }
2405
4392
  const handleSignal = async () => {
4393
+ if (state.shuttingDown) return;
4394
+ state.shuttingDown = true;
2406
4395
  if (state.interactive) {
2407
4396
  logActivity(state, { type: "info", message: "Shutting down..." });
2408
4397
  displayStatus(state);
2409
4398
  } else {
2410
- log(state, "Shutting down...");
4399
+ log2(state, "Shutting down...");
2411
4400
  }
2412
- await cleanup(state);
4401
+ await cleanup(state, { graceful: true });
2413
4402
  await shutdownTelemetry();
2414
4403
  process.exit(0);
2415
4404
  };
@@ -2440,7 +4429,7 @@ async function run(options) {
2440
4429
  const resolved = await resolveAgentIdFromKey(state.authHeader);
2441
4430
  if (resolved.agent_id) {
2442
4431
  state.agentId = resolved.agent_id;
2443
- log(state, `Resolved agent ID from key: ${state.agentId}`);
4432
+ log2(state, `Resolved agent ID from key: ${state.agentId}`);
2444
4433
  if (state.interactive && !state.json) {
2445
4434
  logActivity(state, {
2446
4435
  type: "info",
@@ -2503,14 +4492,21 @@ async function run(options) {
2503
4492
  port: state.port,
2504
4493
  interactive: state.interactive,
2505
4494
  agentId: state.agentId,
2506
- log: (message) => log(state, message)
4495
+ log: (message) => log2(state, message)
2507
4496
  });
2508
4497
  state.port = oc.port;
2509
4498
  state.opencodeProcess = oc.process;
2510
4499
  state.opencodeVersion = oc.version;
2511
4500
  state.opencodeConnected = oc.process !== null || oc.version !== null;
2512
- const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
2513
- ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
4501
+ const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
4502
+ ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
4503
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
4504
+ if (versionWarning) {
4505
+ log2(state, versionWarning, false);
4506
+ if (state.interactive && !state.json) {
4507
+ logActivity(state, { type: "info", message: versionWarning });
4508
+ }
4509
+ }
2514
4510
  } catch (error2) {
2515
4511
  ocSpinner?.fail(error2.message);
2516
4512
  throw error2;
@@ -2522,12 +4518,14 @@ async function run(options) {
2522
4518
  apiUrl: getApiUrlConfig(),
2523
4519
  getAuthHeader: () => state.authHeader,
2524
4520
  conversationFilter: state.conversationFilter,
4521
+ stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
2525
4522
  log: (entry) => logActivity(state, {
2526
4523
  type: entry.level === "error" ? "error" : "info",
2527
4524
  message: entry.message,
2528
4525
  error: entry.level === "error" ? entry.message : void 0
2529
4526
  })
2530
4527
  });
4528
+ state.channelDriver = channelDriver;
2531
4529
  const connection = new RunnerConnection({
2532
4530
  agentId: state.agentId,
2533
4531
  getAuthHeader: () => state.authHeader,
@@ -2541,7 +4539,11 @@ async function run(options) {
2541
4539
  type: "info",
2542
4540
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
2543
4541
  });
2544
- emitAgentConnected(state.agentId, { port: state.port });
4542
+ emitAgentConnected(state.agentId, {
4543
+ port: state.port,
4544
+ cli_version: getCliVersion(),
4545
+ opencode_version: state.opencodeVersion
4546
+ });
2545
4547
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
2546
4548
  if (state.interactive) displayStatus(state);
2547
4549
  channelDriver.drainPending().then((processed) => {
@@ -2575,9 +4577,14 @@ async function run(options) {
2575
4577
  logActivity(state, { type: "error", error: error2 });
2576
4578
  if (state.interactive) displayStatus(state);
2577
4579
  },
2578
- // Web traffic is proxied transparently; only note opencode is live.
4580
+ // Web traffic is proxied transparently; note opencode is live and stamp
4581
+ // proxied activity so the idle loop treats interactive proxy use as work.
4582
+ // Fires per forwarded response head (incl. every SSE open) and excludes
4583
+ // the internal drain-ping, so an actively-used proxy keeps the timer
4584
+ // fresh while a lone idle SSE with no follow-up requests still ages out.
2579
4585
  onResponse: () => {
2580
4586
  state.opencodeConnected = true;
4587
+ state.lastProxiedActivityAt = Date.now();
2581
4588
  },
2582
4589
  // A channel message was queued and the api-worker pinged us over the
2583
4590
  // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2615,10 +4622,12 @@ async function run(options) {
2615
4622
  if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
2616
4623
  throw error2;
2617
4624
  }
4625
+ scheduleSessionCleanup(state, channelDriver, options);
2618
4626
  if (!interactive || state.json) {
2619
- log(state, "Driving channel messages...");
4627
+ log2(state, "Driving channel messages...");
2620
4628
  }
2621
4629
  await driveChannels(state, channelDriver);
4630
+ if (state.shuttingDown) return;
2622
4631
  await cleanup(state);
2623
4632
  if (state.json) {
2624
4633
  console.log(
@@ -2628,11 +4637,12 @@ async function run(options) {
2628
4637
  })
2629
4638
  );
2630
4639
  } else if (!interactive) {
2631
- log(state, `Completed. Processed ${state.messageCount} message(s).`);
4640
+ log2(state, `Completed. Processed ${state.messageCount} message(s).`);
2632
4641
  }
2633
4642
  await shutdownTelemetry();
2634
4643
  process.exit(0);
2635
4644
  } catch (error2) {
4645
+ if (state.shuttingDown) return;
2636
4646
  await cleanup(state);
2637
4647
  const message = error2 instanceof Error ? error2.message : String(error2);
2638
4648
  if (state.json) {
@@ -2650,8 +4660,9 @@ async function run(options) {
2650
4660
  }
2651
4661
 
2652
4662
  // src/index.ts
4663
+ var { version } = createRequire(import.meta.url)("../package.json");
2653
4664
  var program = new Command();
2654
- program.name("evident").description("Run OpenCode locally and connect it to Evident").version("0.1.0").option(
4665
+ program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
2655
4666
  "--endpoint <url>",
2656
4667
  "Evident API base URL (default: production; e.g. http://localhost:3001)"
2657
4668
  ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {
@@ -2666,7 +4677,16 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
2666
4677
  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);
2667
4678
  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 }));
2668
4679
  program.command("whoami").description("Show the currently logged in user").action(whoami);
2669
- 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(
4680
+ 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").option(
4681
+ "--session-cleanup-max-age <duration>",
4682
+ "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
4683
+ ).option(
4684
+ "--session-cleanup-max-count <n>",
4685
+ "Keep only the newest N OpenCode sessions. Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_COUNT"
4686
+ ).option(
4687
+ "--session-cleanup-interval <duration>",
4688
+ "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
4689
+ ).action(
2670
4690
  (options) => {
2671
4691
  run({
2672
4692
  agent: options.agent,
@@ -2674,7 +4694,11 @@ program.command("run").description("Connect to Evident and process messages").op
2674
4694
  verbose: options.verbose,
2675
4695
  conversation: options.conversation,
2676
4696
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
2677
- json: options.json
4697
+ json: options.json,
4698
+ // Raw strings — the resolver in run.ts single-sources parsing (M1).
4699
+ sessionCleanupMaxAge: options.sessionCleanupMaxAge,
4700
+ sessionCleanupMaxCount: options.sessionCleanupMaxCount,
4701
+ sessionCleanupInterval: options.sessionCleanupInterval
2678
4702
  });
2679
4703
  }
2680
4704
  );