@evident-ai/cli 3.0.1-dev.82a8d1e → 3.0.1-dev.8358cf0

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 +1078,47 @@ async function getSessionMessages(port, sessionId) {
1008
1078
  return null;
1009
1079
  }
1010
1080
  }
1011
- function isTurnComplete(messages) {
1012
- if (!messages || messages.length === 0) return false;
1013
- const last = messages[messages.length - 1];
1014
- if (roleOf(last) !== "assistant") return false;
1015
- return completedOf(last) != null;
1081
+ function sessionLastActivityMs(session) {
1082
+ const candidates = [
1083
+ session.time?.updated,
1084
+ session.time?.created,
1085
+ session.time_updated,
1086
+ session.time_created,
1087
+ session.updated,
1088
+ session.created
1089
+ ];
1090
+ for (const c of candidates) {
1091
+ if (typeof c === "number" && Number.isFinite(c)) return c;
1092
+ }
1093
+ return null;
1094
+ }
1095
+ async function listSessions(port) {
1096
+ try {
1097
+ const res = await fetch(`${opencodeBase(port)}/session`);
1098
+ if (!res.ok) return null;
1099
+ const body = await res.json();
1100
+ return Array.isArray(body) ? body : null;
1101
+ } catch {
1102
+ return null;
1103
+ }
1104
+ }
1105
+ async function deleteSession(port, id) {
1106
+ try {
1107
+ const res = await fetch(`${opencodeBase(port)}/session/${id}`, { method: "DELETE" });
1108
+ return res.status >= 200 && res.status < 300;
1109
+ } catch {
1110
+ return false;
1111
+ }
1112
+ }
1113
+ async function sessionExists(port, id) {
1114
+ try {
1115
+ const res = await fetch(`${opencodeBase(port)}/session/${id}`);
1116
+ if (res.status >= 200 && res.status < 300) return true;
1117
+ if (res.status === 404) return false;
1118
+ return null;
1119
+ } catch {
1120
+ return null;
1121
+ }
1016
1122
  }
1017
1123
  async function createOpenCodeSession(port, directory) {
1018
1124
  const url = new URL(`${opencodeBase(port)}/session`);
@@ -1031,7 +1137,15 @@ async function createOpenCodeSession(port, directory) {
1031
1137
  const data = await response.json();
1032
1138
  return data.id;
1033
1139
  }
1034
- async function sendMessageToOpenCode(port, sessionId, content, options, hooks, maxWaitMs = 10 * 60 * 1e3) {
1140
+ function messageText(m) {
1141
+ if (!m || !Array.isArray(m.parts)) return "";
1142
+ return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
1143
+ }
1144
+ async function sendPromptAsync(port, sessionId, content, options) {
1145
+ const before = await getSessionMessages(port, sessionId);
1146
+ const knownUserIds = new Set(
1147
+ (before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
1148
+ );
1035
1149
  const body = {
1036
1150
  parts: [{ type: "text", text: content }]
1037
1151
  };
@@ -1047,79 +1161,213 @@ async function sendMessageToOpenCode(port, sessionId, content, options, hooks, m
1047
1161
  };
1048
1162
  }
1049
1163
  }
1050
- let pollDone = false;
1051
- const reportedQuestions = /* @__PURE__ */ new Set();
1052
- const reportedPermissions = /* @__PURE__ */ new Set();
1053
- const pollInteractive = async () => {
1054
- while (!pollDone) {
1055
- await new Promise((resolve) => setTimeout(resolve, 1e3));
1056
- if (pollDone) break;
1057
- if (hooks?.onQuestion) {
1058
- try {
1059
- const res = await fetch(`${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(`${opencodeBase(port)}/permission`);
1075
- if (res.ok) {
1076
- const permissions = await res.json();
1077
- for (const p of permissions) {
1078
- if (p.sessionID === sessionId && !reportedPermissions.has(p.id)) {
1079
- reportedPermissions.add(p.id);
1080
- await hooks.onPermission(p);
1081
- }
1082
- }
1083
- }
1084
- } catch {
1164
+ const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/prompt_async`, {
1165
+ method: "POST",
1166
+ headers: { "Content-Type": "application/json" },
1167
+ body: JSON.stringify(body)
1168
+ });
1169
+ if (res.status < 200 || res.status >= 300) {
1170
+ const text = await res.text().catch(() => "");
1171
+ throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1172
+ }
1173
+ const READ_BACK_ATTEMPTS = 5;
1174
+ const READ_BACK_DELAY_MS = 150;
1175
+ for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
1176
+ const after = await getSessionMessages(port, sessionId);
1177
+ if (after) {
1178
+ let best = null;
1179
+ for (const m of after) {
1180
+ if (roleOf(m) !== "user") continue;
1181
+ const id = idOf(m);
1182
+ if (typeof id !== "string" || knownUserIds.has(id)) continue;
1183
+ if (messageText(m) !== content) continue;
1184
+ const created = createdOf(m) ?? 0;
1185
+ if (best === null || created > best.created) {
1186
+ best = { id, created };
1085
1187
  }
1086
1188
  }
1189
+ if (best) return best.id;
1190
+ }
1191
+ if (attempt < READ_BACK_ATTEMPTS - 1) {
1192
+ await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1193
+ }
1194
+ }
1195
+ return null;
1196
+ }
1197
+ function findAssistantReplyAfter(messages, userMessageId) {
1198
+ if (!messages || messages.length === 0) return null;
1199
+ const byParent = messages.find(
1200
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) === userMessageId
1201
+ );
1202
+ if (byParent) return byParent;
1203
+ const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1204
+ if (userIndex === -1) return null;
1205
+ for (let i = userIndex + 1; i < messages.length; i++) {
1206
+ if (roleOf(messages[i]) === "assistant") return messages[i];
1207
+ }
1208
+ return null;
1209
+ }
1210
+ function findLastAssistantReplyFor(messages, userMessageId) {
1211
+ if (!messages || messages.length === 0) return null;
1212
+ let lastCorrelated = null;
1213
+ let lastNonErrored = null;
1214
+ for (let i = messages.length - 1; i >= 0; i--) {
1215
+ const m = messages[i];
1216
+ if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
1217
+ if (lastCorrelated === null) lastCorrelated = m;
1218
+ if (errorOf(m) == null) {
1219
+ lastNonErrored = m;
1220
+ break;
1221
+ }
1222
+ }
1223
+ if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
1224
+ const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1225
+ if (userIndex === -1) return null;
1226
+ let last = null;
1227
+ let lastOk = null;
1228
+ for (let i = userIndex + 1; i < messages.length; i++) {
1229
+ const role = roleOf(messages[i]);
1230
+ if (role === "user") break;
1231
+ if (role === "assistant") {
1232
+ last = messages[i];
1233
+ if (errorOf(messages[i]) == null) lastOk = messages[i];
1087
1234
  }
1235
+ }
1236
+ return lastOk ?? last;
1237
+ }
1238
+ function messageRunState(messages, userMessageId) {
1239
+ if (!messages || messages.length === 0) return "unknown";
1240
+ const hasUser = messages.some((m) => idOf(m) === userMessageId);
1241
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1242
+ if (!hasUser) {
1243
+ if (!reply) return "unknown";
1244
+ }
1245
+ if (!reply) return "queued";
1246
+ if (isAssistantInFlight(reply)) return "running";
1247
+ return errorOf(reply) != null ? "failed" : "done";
1248
+ }
1249
+ function messageError(messages, userMessageId) {
1250
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1251
+ const error2 = errorOf(reply);
1252
+ if (error2 == null) return null;
1253
+ if (typeof error2 === "string") return error2;
1254
+ if (typeof error2 === "object") {
1255
+ const e = error2;
1256
+ const dataMessage = e.data?.message;
1257
+ if (typeof dataMessage === "string") return dataMessage;
1258
+ if (typeof e.message === "string") return e.message;
1259
+ }
1260
+ return "The agent run failed.";
1261
+ }
1262
+ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1263
+ if (!messages || messages.length === 0) return false;
1264
+ return messages.some(
1265
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1266
+ );
1267
+ }
1268
+
1269
+ // src/lib/opencode/session-cleanup.ts
1270
+ var DURATION_UNIT_MS = {
1271
+ s: 1e3,
1272
+ m: 60 * 1e3,
1273
+ h: 60 * 60 * 1e3,
1274
+ d: 24 * 60 * 60 * 1e3
1275
+ };
1276
+ function parseDurationMs(input) {
1277
+ const trimmed = input.trim();
1278
+ const match = /^(\d+)([smhd])$/.exec(trimmed);
1279
+ if (!match) {
1280
+ throw new Error(
1281
+ `Invalid duration "${input}": expected <number><unit> where unit is one of s, m, h, d (e.g. "7d", "24h", "30m", "90s").`
1282
+ );
1283
+ }
1284
+ const value = Number(match[1]);
1285
+ if (value <= 0) {
1286
+ throw new Error(`Invalid duration "${input}": must be a positive value.`);
1287
+ }
1288
+ return value * DURATION_UNIT_MS[match[2]];
1289
+ }
1290
+ function selectSessionsToDelete(sessions, opts) {
1291
+ const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
1292
+ if (maxAgeMs === void 0 && maxCount === void 0) return [];
1293
+ const ageEligible = (s) => {
1294
+ if (maxAgeMs === void 0) return false;
1295
+ if (s.lastActivityMs === null) return true;
1296
+ return nowMs - s.lastActivityMs > maxAgeMs;
1088
1297
  };
1089
- const sendMessage = async () => {
1090
- const controller = new AbortController();
1091
- const timer = setTimeout(() => controller.abort(), maxWaitMs);
1298
+ const countEligibleIds = /* @__PURE__ */ new Set();
1299
+ if (maxCount !== void 0) {
1300
+ const byActivityDesc = [...sessions].sort(
1301
+ (a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
1302
+ );
1303
+ for (const s of byActivityDesc.slice(maxCount)) {
1304
+ countEligibleIds.add(s.id);
1305
+ }
1306
+ }
1307
+ const toDelete = [];
1308
+ for (const s of sessions) {
1309
+ if (protectedIds.has(s.id)) continue;
1310
+ if (ageEligible(s) || countEligibleIds.has(s.id)) {
1311
+ toDelete.push(s.id);
1312
+ }
1313
+ }
1314
+ return toDelete;
1315
+ }
1316
+ var DEFAULT_INTERVAL = "1h";
1317
+ function resolve(flag, envValue, fallback) {
1318
+ return flag ?? envValue ?? fallback;
1319
+ }
1320
+ function parseMaxCount(input) {
1321
+ const trimmed = input.trim();
1322
+ if (!/^\d+$/.test(trimmed)) {
1323
+ throw new Error(`Invalid max-count "${input}": expected a positive integer.`);
1324
+ }
1325
+ const value = Number(trimmed);
1326
+ if (value <= 0) {
1327
+ throw new Error(`Invalid max-count "${input}": must be greater than 0.`);
1328
+ }
1329
+ return value;
1330
+ }
1331
+ function resolveSessionCleanupConfig(flags, env = process.env) {
1332
+ const warnings = [];
1333
+ const maxAgeRaw = resolve(flags.maxAge, env.EVIDENT_SESSION_CLEANUP_MAX_AGE);
1334
+ const maxCountRaw = resolve(flags.maxCount, env.EVIDENT_SESSION_CLEANUP_MAX_COUNT);
1335
+ const intervalRaw = resolve(
1336
+ flags.interval,
1337
+ env.EVIDENT_SESSION_CLEANUP_INTERVAL,
1338
+ DEFAULT_INTERVAL
1339
+ );
1340
+ let maxAgeMs;
1341
+ if (maxAgeRaw !== void 0) {
1092
1342
  try {
1093
- const res = await fetch(`${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
1343
+ maxAgeMs = parseDurationMs(maxAgeRaw);
1344
+ } catch (err) {
1345
+ warnings.push(
1346
+ `Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
1105
1347
  );
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 };
1348
+ }
1349
+ }
1350
+ let maxCount;
1351
+ if (maxCountRaw !== void 0) {
1352
+ try {
1353
+ maxCount = parseMaxCount(maxCountRaw);
1111
1354
  } 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;
1355
+ warnings.push(
1356
+ `Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
1357
+ );
1119
1358
  }
1120
- };
1121
- const [result] = await Promise.all([sendMessage(), pollInteractive()]);
1122
- return result;
1359
+ }
1360
+ let intervalMs;
1361
+ try {
1362
+ intervalMs = parseDurationMs(intervalRaw ?? DEFAULT_INTERVAL);
1363
+ } catch (err) {
1364
+ warnings.push(
1365
+ `Ignoring invalid --session-cleanup-interval, using default ${DEFAULT_INTERVAL}: ${err instanceof Error ? err.message : String(err)}`
1366
+ );
1367
+ intervalMs = parseDurationMs(DEFAULT_INTERVAL);
1368
+ }
1369
+ const enabled = maxAgeMs !== void 0 || maxCount !== void 0;
1370
+ return { enabled, maxAgeMs, maxCount, intervalMs, warnings };
1123
1371
  }
1124
1372
 
1125
1373
  // src/lib/tunnel/connection.ts
@@ -1190,24 +1438,34 @@ var StreamForwarder = class {
1190
1438
  }
1191
1439
  async handleOpen(frame) {
1192
1440
  const { sid, method, path, headers, has_body } = frame;
1441
+ const correlationId = headers?.[CORRELATION_ID_HEADER];
1442
+ const startedAt = Date.now();
1193
1443
  if (path === TUNNEL_DRAIN_PING_PATH) {
1194
1444
  this.callbacks.onDrainPing?.();
1195
1445
  this.send({ type: "head", sid, status: 204, headers: {} });
1196
1446
  this.send({ type: "res_end", sid });
1197
1447
  return;
1198
1448
  }
1449
+ if (process.env.DEBUG) {
1450
+ log("debug", "agent_request", {
1451
+ correlation_id: correlationId,
1452
+ sid,
1453
+ method,
1454
+ path: stripQuery(path)
1455
+ });
1456
+ }
1199
1457
  const ac = new AbortController();
1200
1458
  let bodyPromise;
1201
1459
  let pushBody;
1202
1460
  let endBody;
1203
1461
  if (has_body) {
1204
1462
  const chunks = [];
1205
- bodyPromise = new Promise((resolve) => {
1463
+ bodyPromise = new Promise((resolve2) => {
1206
1464
  pushBody = (buf) => {
1207
1465
  chunks.push(buf);
1208
1466
  };
1209
1467
  endBody = () => {
1210
- resolve(Buffer.concat(chunks));
1468
+ resolve2(Buffer.concat(chunks));
1211
1469
  };
1212
1470
  });
1213
1471
  }
@@ -1242,6 +1500,14 @@ var StreamForwarder = class {
1242
1500
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1243
1501
  });
1244
1502
  this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
1503
+ if (process.env.DEBUG) {
1504
+ log("debug", "agent_response", {
1505
+ correlation_id: correlationId,
1506
+ sid,
1507
+ status: upstream.status,
1508
+ duration_ms: Date.now() - startedAt
1509
+ });
1510
+ }
1245
1511
  this.callbacks.onHead?.(sid, upstream.status);
1246
1512
  try {
1247
1513
  if (upstream.body) {
@@ -1317,7 +1583,7 @@ function connectTunnel(options) {
1317
1583
  } = options;
1318
1584
  const tunnelUrl = getTunnelUrlConfig();
1319
1585
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1320
- return new Promise((resolve, reject) => {
1586
+ return new Promise((resolve2, reject) => {
1321
1587
  const ws = new WebSocket2(url, {
1322
1588
  headers: {
1323
1589
  Authorization: authHeader
@@ -1382,7 +1648,7 @@ function connectTunnel(options) {
1382
1648
  clearTimeout(connectionTimeout);
1383
1649
  const connectedAgentId = message.agent_id ?? agentId;
1384
1650
  onConnected?.(connectedAgentId);
1385
- resolve({
1651
+ resolve2({
1386
1652
  ws,
1387
1653
  close: () => ws.close(1e3, "CLI shutdown")
1388
1654
  });
@@ -1498,6 +1764,12 @@ var RunnerConnection = class {
1498
1764
  };
1499
1765
 
1500
1766
  // src/lib/channels/driver.ts
1767
+ function messageIdOf(m) {
1768
+ if (!m || typeof m !== "object") return void 0;
1769
+ if (typeof m.id === "string") return m.id;
1770
+ const infoId = m.info?.id;
1771
+ return typeof infoId === "string" ? infoId : void 0;
1772
+ }
1501
1773
  var DEFAULT_RETRY_POLICY = {
1502
1774
  maxAttempts: 6,
1503
1775
  baseDelayMs: 500,
@@ -1505,12 +1777,21 @@ var DEFAULT_RETRY_POLICY = {
1505
1777
  };
1506
1778
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1507
1779
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1780
+ var DEFAULT_STUCK_QUEUED_MS = 6e4;
1508
1781
  var ChannelAuthError = class extends Error {
1509
1782
  constructor(message) {
1510
1783
  super(message);
1511
1784
  this.name = "ChannelAuthError";
1512
1785
  }
1513
1786
  };
1787
+ var ChannelTerminalError = class extends Error {
1788
+ status;
1789
+ constructor(message, status) {
1790
+ super(message);
1791
+ this.name = "ChannelTerminalError";
1792
+ this.status = status;
1793
+ }
1794
+ };
1514
1795
  function backoffDelay(attempt, policy) {
1515
1796
  const exp = policy.baseDelayMs * Math.pow(2, attempt);
1516
1797
  const capped = Math.min(policy.maxDelayMs, exp);
@@ -1531,15 +1812,82 @@ var ChannelDriver = class {
1531
1812
  sleep;
1532
1813
  pausedPollIntervalMs;
1533
1814
  pausedMaxWaitMs;
1815
+ stuckQueuedMs;
1816
+ now;
1534
1817
  /** Cache of conversationId → opencode sessionId. */
1535
1818
  sessions = /* @__PURE__ */ new Map();
1536
1819
  /**
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.
1820
+ * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
1821
+ * longer idempotent (no caller-supplied `messageID`), and its read-back picks
1822
+ * "the one new user row" which is only unambiguous if no OTHER dispatch into
1823
+ * the SAME session interleaves its snapshot→POST→read-back. This map chains each
1824
+ * session's dispatches so they run serially; distinct sessions stay concurrent.
1825
+ */
1826
+ sessionDispatchLocks = /* @__PURE__ */ new Map();
1827
+ /**
1828
+ * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1829
+ * session: one polling loop services all of that session's in-flight messages.
1830
+ * A session entry exists while it has any in-flight (dispatched-but-not-done)
1831
+ * message; it is removed once its in-flight set empties.
1541
1832
  */
1542
1833
  watchers = /* @__PURE__ */ new Map();
1834
+ /**
1835
+ * AUTHORITATIVE local dedup (WI-3): Evident message ids that have been
1836
+ * dispatched and are still in-flight. A message in this set is never
1837
+ * re-`prompt_async`-ed by a subsequent poll tick while it is queued/running.
1838
+ * Backed by a stable minted opencode `messageID` whose duplicate re-enqueue is
1839
+ * idempotent on opencode (PoC fact 9) — so even if this set is lost on restart,
1840
+ * a steady-state-poll re-dispatch will not double-run the message.
1841
+ */
1842
+ dispatched = /* @__PURE__ */ new Set();
1843
+ /**
1844
+ * Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
1845
+ * Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
1846
+ * so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
1847
+ * it is re-adopted and removed when its watcher settles or it is observed off
1848
+ * the processing list.
1849
+ */
1850
+ readopted = /* @__PURE__ */ new Set();
1851
+ /**
1852
+ * "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
1853
+ * re-adopted running/orphan row's watcher hit its `processed_at`-anchored
1854
+ * deadline (or an orphan whose window already elapsed): the still-`processing`
1855
+ * server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
1856
+ * drain until the 15-min cron resets it — spamming new turns.
1857
+ *
1858
+ * CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
1859
+ * does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
1860
+ * in opencode must still be delivered via `markDone` on the next drain — so
1861
+ * `readoptOne` computes `state` FIRST and this set is checked only on the
1862
+ * non-done path. It is cleared once the row leaves the processing list (cron
1863
+ * reset → it drains normally as `pending`), so it can never leak.
1864
+ */
1865
+ dontRedispatch = /* @__PURE__ */ new Set();
1866
+ /**
1867
+ * "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
1868
+ * re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
1869
+ * never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
1870
+ * that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
1871
+ * markDone failure must NOT land here (it must still retry next drain). Separate
1872
+ * from `dontRedispatch` because the two concerns are independent: a row can need
1873
+ * "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
1874
+ * the row leaves the processing list, exactly like `dontRedispatch`.
1875
+ */
1876
+ doneUndeliverable = /* @__PURE__ */ new Set();
1877
+ /**
1878
+ * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
1879
+ * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
1880
+ * is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
1881
+ * persist hasn't landed before tick N+1 re-reads the still-null
1882
+ * `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
1883
+ * A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
1884
+ * short-circuits while it is present, so a null-id row is re-dispatched AT MOST
1885
+ * ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
1886
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
1887
+ * re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
1888
+ * so the NEXT tick may retry exactly once more).
1889
+ */
1890
+ awaitingReadopt = /* @__PURE__ */ new Set();
1543
1891
  /**
1544
1892
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1545
1893
  * first session creation so drain-created sessions are rooted at the project
@@ -1547,8 +1895,33 @@ var ChannelDriver = class {
1547
1895
  * not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
1548
1896
  */
1549
1897
  opencodeDirectory = void 0;
1898
+ /**
1899
+ * Cache of opencode `sessionId → parentID` (its parent session, or `null` when
1900
+ * the session is a root with no parent). Sub-agents spawned via the `task` tool
1901
+ * run in CHILD sessions whose `parentID` chains up to the Evident-created
1902
+ * (watched) session; we resolve this once per session so a child-session
1903
+ * question/permission can be attributed to the watched session's subtree
1904
+ * (`sessionBelongsTo`) instead of being dropped by an exact-id filter. A missing
1905
+ * entry = not yet resolved; `null` = resolved root (stop walking).
1906
+ */
1907
+ sessionParents = /* @__PURE__ */ new Map();
1550
1908
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
1551
1909
  draining = false;
1910
+ /**
1911
+ * The currently-executing `drainPending()` promise, or null when idle. Lets a
1912
+ * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
1913
+ * is about to dispatch is not missed by the `hasInFlightWatchers()` check (a
1914
+ * drain that entered before `stop()` still registers its watcher).
1915
+ */
1916
+ activeDrain = null;
1917
+ /**
1918
+ * Set by `stop()` on graceful shutdown. Once stopped, `drainPending` no longer
1919
+ * dispatches NEW work (it returns 0 immediately) — but the per-session watcher
1920
+ * loops already running keep going so in-flight turns can finish and deliver
1921
+ * their reply. `run.ts` awaits `waitForInFlight()` before it closes the tunnel
1922
+ * and stops opencode.
1923
+ */
1924
+ stopped = false;
1552
1925
  constructor(config2) {
1553
1926
  this.agentId = config2.agentId;
1554
1927
  this.port = config2.port;
@@ -1562,6 +1935,8 @@ var ChannelDriver = class {
1562
1935
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1563
1936
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1564
1937
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
1938
+ this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
1939
+ this.now = config2.now ?? (() => Date.now());
1565
1940
  }
1566
1941
  /** The IPv4-loopback base URL for the local `opencode serve`. */
1567
1942
  get opencodeBase() {
@@ -1571,16 +1946,29 @@ var ChannelDriver = class {
1571
1946
  // Public API
1572
1947
  // -------------------------------------------------------------------------
1573
1948
  /**
1574
- * Drain all pending channel conversations once: poll → processcallback.
1949
+ * Drain all pending channel conversations once: poll → dispatchregister.
1575
1950
  * Called on tunnel `connected` (WI-CHAN-4) and on each poll tick by `run.ts`.
1576
1951
  * Re-entrant calls while a drain is in flight are skipped (return 0).
1577
1952
  *
1578
- * @returns the number of messages processed.
1953
+ * @returns the number of messages NEWLY dispatched to opencode's native queue.
1579
1954
  */
1580
1955
  async drainPending() {
1956
+ if (this.stopped) return 0;
1581
1957
  if (this.draining) return 0;
1582
1958
  this.draining = true;
1583
- let processed = 0;
1959
+ const run2 = this.runDrain();
1960
+ this.activeDrain = run2.then(
1961
+ () => {
1962
+ this.activeDrain = null;
1963
+ },
1964
+ () => {
1965
+ this.activeDrain = null;
1966
+ }
1967
+ );
1968
+ return run2;
1969
+ }
1970
+ async runDrain() {
1971
+ let dispatched = 0;
1584
1972
  try {
1585
1973
  const conversations = await this.getPendingConversations();
1586
1974
  if (conversations.length > 0) {
@@ -1591,107 +1979,226 @@ var ChannelDriver = class {
1591
1979
  });
1592
1980
  }
1593
1981
  for (const conv of conversations) {
1594
- processed += await this.processConversation(conv);
1982
+ if (this.stopped) break;
1983
+ dispatched += await this.processConversation(conv);
1595
1984
  }
1985
+ await this.readoptProcessing();
1596
1986
  } finally {
1597
1987
  this.draining = false;
1598
1988
  }
1599
- return processed;
1989
+ return dispatched;
1990
+ }
1991
+ /**
1992
+ * True while any per-session watcher has a non-empty in-flight dispatched set
1993
+ * (Task 3.7). `run.ts` treats this as NON-idle so `--idle-timeout` cannot exit
1994
+ * the process while a dispatched message is still queued/running — which would
1995
+ * kill the turn and orphan its reply.
1996
+ */
1997
+ hasInFlightWatchers() {
1998
+ for (const watcher of this.watchers.values()) {
1999
+ if (watcher.inFlight.size > 0) return true;
2000
+ }
2001
+ return false;
2002
+ }
2003
+ /**
2004
+ * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2005
+ * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
2006
+ * `watchers` entry whose `inFlight` set is non-empty — the same predicate
2007
+ * `hasInFlightWatchers()` uses, lifted to return the ids.
2008
+ *
2009
+ * Deliberately does NOT include `this.sessions` (the permanent, never-pruned
2010
+ * conversation→session cache). Protecting every bound-but-idle session there
2011
+ * would shield nearly every session and defeat cleanup — AND it is unnecessary:
2012
+ * `ensureSession` is self-healing (it recreates a session whose id no longer
2013
+ * exists), so deleting an idle bound session is harmless — the conversation's
2014
+ * next turn transparently rebinds a fresh one. The only thing worth protecting
2015
+ * is a session with a turn ACTIVELY in flight right now: tearing that down
2016
+ * mid-turn would strand the running `prompt_async`. Idle sessions are fair game.
2017
+ */
2018
+ protectedSessionIds() {
2019
+ const ids = /* @__PURE__ */ new Set();
2020
+ for (const [sessionId, watcher] of this.watchers) {
2021
+ if (watcher.inFlight.size > 0) ids.add(sessionId);
2022
+ }
2023
+ return ids;
2024
+ }
2025
+ /**
2026
+ * Begin a graceful stop: stop accepting NEW channel work. Idempotent. After
2027
+ * this, `drainPending()` is a no-op (returns 0), so no new message is dispatched
2028
+ * — but the watcher loops already tracking in-flight turns keep running, so a
2029
+ * turn that has finished (or is about to) still fires `markDone` and delivers
2030
+ * its reply. Pair with `waitForInFlight()` to bound how long shutdown waits.
2031
+ */
2032
+ stop() {
2033
+ this.stopped = true;
2034
+ }
2035
+ /**
2036
+ * Wait (up to `timeoutMs`) for in-flight watcher work to settle during a
2037
+ * graceful shutdown, so a turn whose reply is ready — or completes within the
2038
+ * window — is delivered before the process exits, instead of being cut off and
2039
+ * left for the ADR-0046 restart-recovery path.
2040
+ *
2041
+ * Bounded on purpose: the watcher's own give-up deadline is up to 10 minutes,
2042
+ * far longer than a shutdown grace period (e.g. Fargate's SIGTERM→SIGKILL
2043
+ * window). We poll `hasInFlightWatchers()` and return as soon as the in-flight
2044
+ * set empties OR the timeout elapses. Anything still in flight at the timeout is
2045
+ * safe to abandon — it stays `processing` server-side and is re-adopted on the
2046
+ * next runner start (ADR-0046).
2047
+ *
2048
+ * @returns true if all in-flight work settled within the window; false if the
2049
+ * timeout elapsed with work still in flight.
2050
+ */
2051
+ async waitForInFlight(timeoutMs) {
2052
+ const deadline = this.now() + timeoutMs;
2053
+ const step = Math.min(this.pausedPollIntervalMs, 250);
2054
+ if (this.activeDrain) {
2055
+ let drainSettled = false;
2056
+ void this.activeDrain.then(() => {
2057
+ drainSettled = true;
2058
+ });
2059
+ while (!drainSettled) {
2060
+ if (this.now() >= deadline) return false;
2061
+ await this.sleep(step);
2062
+ }
2063
+ }
2064
+ while (this.hasInFlightWatchers()) {
2065
+ if (this.now() >= deadline) return false;
2066
+ await this.sleep(step);
2067
+ }
2068
+ return true;
1600
2069
  }
1601
2070
  /**
1602
- * Await all outstanding paused-session watchers (WI-2-CLI).
2071
+ * Await all outstanding per-session watchers (WI-3).
1603
2072
  *
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.
2073
+ * In production the watcher loops are deliberately started-not-awaited so the
2074
+ * drain loop never blocks on them and process exit is not held up (the cron
2075
+ * recovers any abandoned ones). This helper exists primarily for deterministic
2076
+ * tests that need to observe a watcher's effect (the `processing`/`done` PATCH
2077
+ * or its giving up) after a non-blocking `drainPending`. Watcher loops never
2078
+ * reject, so this resolves.
1609
2079
  */
1610
2080
  async flushPausedWatchers() {
1611
- await Promise.all([...this.watchers.values()]);
2081
+ while (true) {
2082
+ const loops = [...this.watchers.values()].map((w) => w.loop).filter((l) => l != null);
2083
+ if (loops.length === 0) return;
2084
+ await Promise.all(loops);
2085
+ const stillLive = [...this.watchers.values()].some((w) => w.loop != null);
2086
+ if (!stillLive) return;
2087
+ }
1612
2088
  }
1613
2089
  // -------------------------------------------------------------------------
1614
- // Conversation processing
2090
+ // Conversation processing (WI-3 — async dispatch)
1615
2091
  // -------------------------------------------------------------------------
2092
+ /**
2093
+ * Dispatch each pending message for a conversation to opencode's native queue
2094
+ * via `prompt_async` (Task 3.2) and register it with the conversation's
2095
+ * per-session watcher. Does NOT block on the turn and does NOT call
2096
+ * `markProcessing` here — that fires from the watcher on running-start.
2097
+ *
2098
+ * @returns the count of messages NEWLY dispatched (not already in-flight).
2099
+ */
1616
2100
  async processConversation(conv) {
1617
2101
  const sessionId = await this.ensureSession(conv);
1618
2102
  const messages = await this.getPendingMessages(conv.id);
1619
- let processed = 0;
2103
+ let dispatched = 0;
2104
+ let skippedAlreadyDispatched = 0;
1620
2105
  for (const message of messages) {
1621
- const claimed = await this.markProcessing(conv.id, message.id, sessionId);
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
- });
2106
+ if (this.stopped) break;
2107
+ if (this.dispatched.has(message.id)) {
2108
+ skippedAlreadyDispatched += 1;
1629
2109
  continue;
1630
2110
  }
2111
+ const options = {
2112
+ agent: message.opencode_agent ?? void 0,
2113
+ model: message.opencode_model ?? void 0
2114
+ };
2115
+ let opencodeMessageId;
1631
2116
  try {
1632
2117
  this.log({
1633
2118
  level: "info",
1634
- message: `Sending queued message ${message.id.slice(0, 8)} to OpenCode (session ${sessionId.slice(0, 8)})`,
2119
+ message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
1635
2120
  conversation_id: conv.id,
1636
2121
  message_id: message.id
1637
2122
  });
1638
- const result = await sendMessageToOpenCode(
1639
- this.port,
2123
+ opencodeMessageId = await this.dispatchLocked(
1640
2124
  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
- }
2125
+ () => sendPromptAsync(this.port, sessionId, message.content, options)
1650
2126
  );
1651
- if (result.awaitingInteraction) {
2127
+ } catch (err) {
2128
+ if (err instanceof ChannelAuthError) throw err;
2129
+ this.dispatched.delete(message.id);
2130
+ if (await sessionExists(this.port, sessionId) === false) {
2131
+ this.sessions.delete(conv.id);
1652
2132
  this.log({
1653
2133
  level: "info",
1654
- message: `Message ${message.id.slice(0, 8)} paused awaiting interaction \u2014 watching session for completion`,
2134
+ 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
2135
  conversation_id: conv.id,
1656
2136
  message_id: message.id
1657
2137
  });
1658
- this.startPausedWatcher(conv, message, sessionId);
1659
- continue;
2138
+ break;
1660
2139
  }
1661
- await this.confirmCompletion(sessionId);
1662
- await this.markDone(conv.id, message.id, sessionId);
1663
- processed += 1;
2140
+ await this.markFailed(conv.id, message.id).catch(() => {
2141
+ });
1664
2142
  this.log({
1665
- level: "info",
1666
- message: `Message ${message.id.slice(0, 8)} processed`,
2143
+ level: "error",
2144
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
1667
2145
  conversation_id: conv.id,
1668
2146
  message_id: message.id
1669
2147
  });
1670
- } catch (err) {
1671
- if (err instanceof ChannelAuthError) throw err;
1672
- await this.markFailed(conv.id, message.id).catch(() => {
1673
- });
2148
+ continue;
2149
+ }
2150
+ if (opencodeMessageId === null) {
1674
2151
  this.log({
1675
2152
  level: "error",
1676
- message: `Message ${message.id.slice(0, 8)} failed: ${err instanceof Error ? err.message : String(err)}`,
2153
+ 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
2154
  conversation_id: conv.id,
1678
2155
  message_id: message.id
1679
2156
  });
2157
+ continue;
1680
2158
  }
2159
+ this.dispatched.add(message.id);
2160
+ this.registerInFlight(conv, sessionId, message, opencodeMessageId);
2161
+ dispatched += 1;
2162
+ void this.postSignal(conv.id, message.id, "dispatched");
2163
+ }
2164
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
2165
+ this.log({
2166
+ level: "error",
2167
+ 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).`,
2168
+ conversation_id: conv.id
2169
+ });
1681
2170
  }
1682
- return processed;
2171
+ this.ensureWatcherRunning(sessionId);
2172
+ return dispatched;
1683
2173
  }
1684
2174
  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;
2175
+ const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
2176
+ if (bound) {
2177
+ const exists = await sessionExists(this.port, bound);
2178
+ if (exists === false) {
2179
+ this.log({
2180
+ level: "info",
2181
+ 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.`,
2182
+ conversation_id: conv.id
2183
+ });
2184
+ this.sessions.delete(conv.id);
2185
+ return this.createAndBindSession(conv.id);
2186
+ }
2187
+ this.sessions.set(conv.id, bound);
2188
+ return bound;
1690
2189
  }
2190
+ return this.createAndBindSession(conv.id);
2191
+ }
2192
+ /**
2193
+ * Create a fresh OpenCode session for a conversation, cache the binding, and
2194
+ * best-effort persist it server-side. Shared by the first-ever bind and the
2195
+ * self-heal recreate path in `ensureSession`.
2196
+ */
2197
+ async createAndBindSession(conversationId) {
1691
2198
  const directory = await this.resolveOpenCodeDirectory();
1692
2199
  const sessionId = await createOpenCodeSession(this.port, directory);
1693
- this.sessions.set(conv.id, sessionId);
1694
- await this.persistSession(conv.id, sessionId).catch(() => {
2200
+ this.sessions.set(conversationId, sessionId);
2201
+ await this.persistSession(conversationId, sessionId).catch(() => {
1695
2202
  });
1696
2203
  return sessionId;
1697
2204
  }
@@ -1711,111 +2218,866 @@ var ChannelDriver = class {
1711
2218
  }
1712
2219
  return this.opencodeDirectory;
1713
2220
  }
2221
+ // -------------------------------------------------------------------------
2222
+ // Per-session watcher (WI-3)
2223
+ // -------------------------------------------------------------------------
1714
2224
  /**
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.
2225
+ * Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
2226
+ * opencode session (Task 2.1a), so two dispatches into the SAME session can
2227
+ * never interleave and mis-correlate their read-backs. Distinct sessions run
2228
+ * concurrently. The chained tail intentionally ignores the prior result/error
2229
+ * (each dispatch reports its own outcome to its caller).
1725
2230
  */
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 {
2231
+ dispatchLocked(sessionId, fn) {
2232
+ const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
2233
+ const run2 = prior.then(fn, fn);
2234
+ this.sessionDispatchLocks.set(
2235
+ sessionId,
2236
+ run2.then(
2237
+ () => void 0,
2238
+ () => void 0
2239
+ )
2240
+ );
2241
+ return run2;
2242
+ }
2243
+ /** Register a freshly-dispatched message with its session's watcher state. */
2244
+ registerInFlight(conv, sessionId, message, opencodeMessageId) {
2245
+ let watcher = this.watchers.get(sessionId);
2246
+ if (!watcher) {
2247
+ watcher = {
2248
+ conv,
2249
+ inFlight: /* @__PURE__ */ new Map(),
2250
+ loop: null,
2251
+ reportedQuestions: /* @__PURE__ */ new Set(),
2252
+ reportedPermissions: /* @__PURE__ */ new Set()
2253
+ };
2254
+ this.watchers.set(sessionId, watcher);
1739
2255
  }
2256
+ const now = this.now();
2257
+ watcher.inFlight.set(message.id, {
2258
+ evidentMessageId: message.id,
2259
+ opencodeMessageId,
2260
+ message,
2261
+ dispatchedAt: now,
2262
+ deadline: now + this.pausedMaxWaitMs,
2263
+ started: false,
2264
+ done: false,
2265
+ stuckReported: false
2266
+ });
1740
2267
  }
1741
- // -------------------------------------------------------------------------
1742
- // Paused-session watcher (WI-2-CLI)
1743
- // -------------------------------------------------------------------------
1744
2268
  /**
1745
- * Start (but do NOT await) a watcher that resumes a paused turn to completion.
2269
+ * Register a RE-ADOPTED `processing` message with its session watcher
2270
+ * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
2271
+ * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
2272
+ * `now`: a row already `processing` for e.g. 5 min must give up ~5 min from now
2273
+ * (10 min after `processed_at`), not 10 min from now — otherwise its deadline
2274
+ * lands ~15 min after `processed_at`, coinciding with the cron reset →
2275
+ * double-drive race. `dispatchedAt` stays `now` (only the appear-guard uses it).
1746
2276
  *
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.
2277
+ * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
2278
+ * `opencodeMessageId` is the id the watcher polls for a reply for the orphan
2279
+ * fresh-run path these differ (a fresh opencode id under the same server row).
2280
+ *
2281
+ * `started` is set true so the watcher does NOT re-`markProcessing` a row the
2282
+ * server already flipped to `processing`; the running/done transitions still
2283
+ * fire from the watcher's normal branches.
1751
2284
  */
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);
2285
+ registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
2286
+ let watcher = this.watchers.get(sessionId);
2287
+ if (!watcher) {
2288
+ watcher = {
2289
+ conv,
2290
+ inFlight: /* @__PURE__ */ new Map(),
2291
+ loop: null,
2292
+ reportedQuestions: /* @__PURE__ */ new Set(),
2293
+ reportedPermissions: /* @__PURE__ */ new Set()
2294
+ };
2295
+ this.watchers.set(sessionId, watcher);
2296
+ }
2297
+ watcher.inFlight.set(message.id, {
2298
+ evidentMessageId: message.id,
2299
+ opencodeMessageId,
2300
+ message,
2301
+ dispatchedAt: this.now(),
2302
+ deadline: processedAtMs + this.pausedMaxWaitMs,
2303
+ // The server row is ALREADY `processing`; do not re-fire markProcessing.
2304
+ started: true,
2305
+ done: false,
2306
+ // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
2307
+ // AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
2308
+ // on `state === 'queued'` (turn produced no reply), not on `started`, so a
2309
+ // re-adopted row left wedged in `queued` still emits the signal once
2310
+ // (#210/#220 observability).
2311
+ stuckReported: false
1756
2312
  });
1757
- this.watchers.set(message.id, watcher);
1758
2313
  }
1759
2314
  /**
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.
1767
- *
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.
1772
- *
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.
2315
+ * Start (but do NOT await) the per-session watcher loop if it has in-flight
2316
+ * work and is not already running. Single-flight per session. The loop is
2317
+ * tracked on the watcher and cleared when it settles; it never rejects (fully
2318
+ * guarded), so a failed poll/callback can never crash the run loop the cron
2319
+ * stays as the safety net.
2320
+ */
2321
+ ensureWatcherRunning(sessionId) {
2322
+ const watcher = this.watchers.get(sessionId);
2323
+ if (!watcher) return;
2324
+ if (watcher.loop) return;
2325
+ if (watcher.inFlight.size === 0) {
2326
+ this.watchers.delete(sessionId);
2327
+ return;
2328
+ }
2329
+ const loop = this.runWatcherLoop(sessionId, watcher).finally(() => {
2330
+ watcher.loop = null;
2331
+ if (watcher.inFlight.size === 0) {
2332
+ this.watchers.delete(sessionId);
2333
+ }
2334
+ });
2335
+ watcher.loop = loop;
2336
+ }
2337
+ /**
2338
+ * The per-session polling loop (WI-3). Once per tick it:
2339
+ * 1. polls `GET /session/:id/message` once and, per in-flight message,
2340
+ * computes `messageRunState` and fires markProcessing (queued→running) /
2341
+ * markDone (done) exactly once per transition;
2342
+ * 2. applies the idle-path re-dispatch guard (a dispatched message that never
2343
+ * APPEARS → re-dispatch — D1 obligation 2);
2344
+ * 3. polls `/question` + `/permission` (scoped to the session) and surfaces
2345
+ * NEW ones via `reportInteraction`, carrying the PAUSED message's own
2346
+ * `source_message_id`;
2347
+ * 4. drops messages that completed or timed out from the in-flight set.
2348
+ * Exits when the in-flight set empties. Never throws.
1778
2349
  */
1779
- async watchPausedSession(conv, message, sessionId) {
1780
- const deadline = Date.now() + this.pausedMaxWaitMs;
2350
+ async runWatcherLoop(sessionId, watcher) {
1781
2351
  try {
1782
- while (Date.now() < deadline) {
2352
+ while (watcher.inFlight.size > 0) {
1783
2353
  await this.sleep(this.pausedPollIntervalMs);
1784
- let completed = false;
2354
+ let messages = null;
1785
2355
  try {
1786
2356
  const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
1787
2357
  if (res.ok) {
1788
2358
  const body = await res.json();
1789
- const messages = Array.isArray(body) ? body : null;
1790
- completed = isTurnComplete(messages);
2359
+ messages = Array.isArray(body) ? body : null;
1791
2360
  }
1792
2361
  } catch {
1793
2362
  continue;
1794
2363
  }
1795
- if (!completed) continue;
2364
+ for (const inFlight of [...watcher.inFlight.values()]) {
2365
+ await this.serviceInFlightMessage(sessionId, watcher, inFlight, messages);
2366
+ }
2367
+ await this.pollInteractions(sessionId, watcher, messages);
2368
+ }
2369
+ } catch (err) {
2370
+ if (err instanceof ChannelAuthError) {
1796
2371
  this.log({
1797
- level: "info",
1798
- message: `Paused session ${sessionId.slice(0, 8)} completed \u2014 marking message ${message.id.slice(0, 8)} done`,
1799
- conversation_id: conv.id,
1800
- message_id: message.id
2372
+ level: "error",
2373
+ 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}`,
2374
+ conversation_id: watcher.conv.id
1801
2375
  });
1802
- await this.markDone(conv.id, message.id, sessionId);
2376
+ for (const evidentMessageId of [...watcher.inFlight.keys()]) {
2377
+ this.readopted.delete(evidentMessageId);
2378
+ this.removeInFlight(watcher, evidentMessageId);
2379
+ }
1803
2380
  return;
1804
2381
  }
2382
+ this.log({
2383
+ level: "error",
2384
+ message: `Session watcher failed for session ${sessionId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
2385
+ conversation_id: watcher.conv.id
2386
+ });
2387
+ }
2388
+ }
2389
+ /**
2390
+ * Drive ONE in-flight message's lifecycle from the tick's message snapshot.
2391
+ * Fires markProcessing on queued→running and markDone on done (each once),
2392
+ * applies the idle-path re-dispatch guard, and removes the message from the
2393
+ * in-flight set on completion or timeout.
2394
+ */
2395
+ async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
2396
+ const conv = watcher.conv;
2397
+ const state = messageRunState(messages, inFlight.opencodeMessageId);
2398
+ if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
2399
+ let claimed;
2400
+ try {
2401
+ claimed = await this.markProcessing(
2402
+ conv.id,
2403
+ inFlight.evidentMessageId,
2404
+ sessionId,
2405
+ inFlight.opencodeMessageId
2406
+ );
2407
+ } catch (err) {
2408
+ if (err instanceof ChannelAuthError) throw err;
2409
+ this.log({
2410
+ level: "error",
2411
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2412
+ conversation_id: conv.id,
2413
+ message_id: inFlight.evidentMessageId
2414
+ });
2415
+ return;
2416
+ }
2417
+ inFlight.started = true;
2418
+ if (!claimed) {
2419
+ this.log({
2420
+ level: "info",
2421
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} already marked processing \u2014 continuing`,
2422
+ conversation_id: conv.id,
2423
+ message_id: inFlight.evidentMessageId
2424
+ });
2425
+ }
2426
+ }
2427
+ if (state === "done") {
2428
+ if (!inFlight.done) {
2429
+ this.log({
2430
+ level: "info",
2431
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2432
+ conversation_id: conv.id,
2433
+ message_id: inFlight.evidentMessageId
2434
+ });
2435
+ try {
2436
+ await this.markDone(
2437
+ conv.id,
2438
+ inFlight.evidentMessageId,
2439
+ sessionId,
2440
+ inFlight.opencodeMessageId
2441
+ );
2442
+ } catch (err) {
2443
+ if (err instanceof ChannelAuthError) throw err;
2444
+ if (err instanceof ChannelTerminalError) {
2445
+ this.log({
2446
+ level: "error",
2447
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2448
+ conversation_id: conv.id,
2449
+ message_id: inFlight.evidentMessageId
2450
+ });
2451
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2452
+ return;
2453
+ }
2454
+ if (this.now() >= inFlight.deadline) {
2455
+ this.log({
2456
+ level: "error",
2457
+ 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)}`,
2458
+ conversation_id: conv.id,
2459
+ message_id: inFlight.evidentMessageId
2460
+ });
2461
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2462
+ return;
2463
+ }
2464
+ this.log({
2465
+ level: "error",
2466
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2467
+ conversation_id: conv.id,
2468
+ message_id: inFlight.evidentMessageId
2469
+ });
2470
+ return;
2471
+ }
2472
+ inFlight.done = true;
2473
+ }
2474
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2475
+ return;
2476
+ }
2477
+ if (state === "failed") {
2478
+ if (!inFlight.done) {
2479
+ const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
2480
+ this.log({
2481
+ level: "error",
2482
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2483
+ conversation_id: conv.id,
2484
+ message_id: inFlight.evidentMessageId
2485
+ });
2486
+ try {
2487
+ await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
2488
+ } catch (err) {
2489
+ if (err instanceof ChannelAuthError) throw err;
2490
+ if (err instanceof ChannelTerminalError) {
2491
+ this.log({
2492
+ level: "error",
2493
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2494
+ conversation_id: conv.id,
2495
+ message_id: inFlight.evidentMessageId
2496
+ });
2497
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2498
+ return;
2499
+ }
2500
+ if (this.now() >= inFlight.deadline) {
2501
+ this.log({
2502
+ level: "error",
2503
+ 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)}`,
2504
+ conversation_id: conv.id,
2505
+ message_id: inFlight.evidentMessageId
2506
+ });
2507
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2508
+ return;
2509
+ }
2510
+ this.log({
2511
+ level: "error",
2512
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2513
+ conversation_id: conv.id,
2514
+ message_id: inFlight.evidentMessageId
2515
+ });
2516
+ return;
2517
+ }
2518
+ inFlight.done = true;
2519
+ }
2520
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2521
+ return;
2522
+ }
2523
+ const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
2524
+ const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
2525
+ if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
2526
+ inFlight.stuckReported = true;
2527
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2528
+ stuck_for_ms: this.now() - inFlight.dispatchedAt
2529
+ });
2530
+ }
2531
+ if (this.now() >= inFlight.deadline) {
1805
2532
  this.log({
1806
2533
  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`,
2534
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
1808
2535
  conversation_id: conv.id,
1809
- message_id: message.id
2536
+ message_id: inFlight.evidentMessageId
2537
+ });
2538
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2539
+ watched_for_ms: this.now() - inFlight.dispatchedAt
1810
2540
  });
2541
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2542
+ }
2543
+ }
2544
+ // -------------------------------------------------------------------------
2545
+ // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
2546
+ // -------------------------------------------------------------------------
2547
+ /**
2548
+ * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
2549
+ *
2550
+ * The pending drain only re-drives `pending` rows; a message already flipped to
2551
+ * `processing` before the runner died is watched by nobody until the 15-min
2552
+ * cron resets it. Here we fetch those rows, and per row resolve its correlated
2553
+ * reply against opencode's OWN session store — completing, re-attaching, or
2554
+ * (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
2555
+ * is idempotent per message (Invariant 2): a row a watcher already tracks is
2556
+ * skipped in `readoptOne` — one driver, no double-drive.
2557
+ *
2558
+ * Only `ChannelAuthError` propagates (to `drainPending`, like the pending
2559
+ * path); every other early return LOGS a reason with context — no silent drop.
2560
+ */
2561
+ async readoptProcessing() {
2562
+ const rows = await this.getProcessingMessages();
2563
+ if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0) {
2564
+ const stillProcessing = new Set(rows.map((r) => r.id));
2565
+ for (const id of [...this.dontRedispatch, ...this.doneUndeliverable]) {
2566
+ if (!stillProcessing.has(id)) {
2567
+ const cleared = this.dontRedispatch.delete(id);
2568
+ const clearedUndeliverable = this.doneUndeliverable.delete(id);
2569
+ if (cleared || clearedUndeliverable) {
2570
+ this.log({
2571
+ level: "info",
2572
+ message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
2573
+ message_id: id
2574
+ });
2575
+ }
2576
+ }
2577
+ }
2578
+ }
2579
+ if (rows.length === 0) return;
2580
+ const bySession = /* @__PURE__ */ new Map();
2581
+ for (const row of rows) {
2582
+ if (!row.opencode_session_id) {
2583
+ this.log({
2584
+ level: "error",
2585
+ message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
2586
+ conversation_id: row.conversation_id,
2587
+ message_id: row.id
2588
+ });
2589
+ continue;
2590
+ }
2591
+ const list = bySession.get(row.opencode_session_id) ?? [];
2592
+ list.push(row);
2593
+ bySession.set(row.opencode_session_id, list);
2594
+ }
2595
+ for (const [sessionId, sessionRows] of bySession) {
2596
+ let messages;
2597
+ try {
2598
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
2599
+ if (!res.ok) {
2600
+ this.log({
2601
+ level: "error",
2602
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
2603
+ });
2604
+ continue;
2605
+ }
2606
+ const body = await res.json();
2607
+ if (!Array.isArray(body)) {
2608
+ this.log({
2609
+ level: "error",
2610
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
2611
+ });
2612
+ continue;
2613
+ }
2614
+ messages = body;
2615
+ } catch (err) {
2616
+ this.log({
2617
+ level: "error",
2618
+ message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
2619
+ });
2620
+ continue;
2621
+ }
2622
+ for (const row of sessionRows) {
2623
+ await this.readoptOne(sessionId, row, messages);
2624
+ }
2625
+ }
2626
+ }
2627
+ /**
2628
+ * Re-adopt ONE `processing` row against the tick's session message snapshot
2629
+ * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
2630
+ *
2631
+ * Branches on `messageRunState(messages, row.opencode_message_id)` — the
2632
+ * opencode-assigned user-message id persisted on the first `processing` PATCH
2633
+ * (#218). A row with a NULL stored id (dispatched but the read-back never landed
2634
+ * before the restart) has no id to correlate → treated as an orphan and
2635
+ * re-dispatched (at most once, see `forceReadoptRun`):
2636
+ * - `done` → `markDone` now (guarded like the watcher's done branch);
2637
+ * - `failed` → `markFailed` with the surfaced error (issue #182), so an
2638
+ * errored turn is reported failed on restart, NOT re-dispatched;
2639
+ * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
2640
+ * tracking the stored id so the reply correlates by it;
2641
+ * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
2642
+ *
2643
+ * Only `ChannelAuthError` propagates.
2644
+ */
2645
+ async readoptOne(sessionId, row, messages) {
2646
+ if (this.isTracked(sessionId, row.id)) {
2647
+ this.log({
2648
+ level: "info",
2649
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
2650
+ conversation_id: row.conversation_id,
2651
+ message_id: row.id
2652
+ });
2653
+ return;
2654
+ }
2655
+ const ocId = row.opencode_message_id;
2656
+ const state = messageRunState(messages, ocId ?? "");
2657
+ if (state === "done") {
2658
+ if (this.doneUndeliverable.has(row.id)) {
2659
+ this.log({
2660
+ level: "info",
2661
+ message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
2662
+ conversation_id: row.conversation_id,
2663
+ message_id: row.id
2664
+ });
2665
+ return;
2666
+ }
2667
+ this.log({
2668
+ level: "info",
2669
+ message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
2670
+ conversation_id: row.conversation_id,
2671
+ message_id: row.id
2672
+ });
2673
+ try {
2674
+ await this.markDone(row.conversation_id, row.id, sessionId, ocId);
2675
+ } catch (err) {
2676
+ if (err instanceof ChannelAuthError) throw err;
2677
+ if (err instanceof ChannelTerminalError) {
2678
+ this.doneUndeliverable.add(row.id);
2679
+ this.log({
2680
+ level: "error",
2681
+ 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}`,
2682
+ conversation_id: row.conversation_id,
2683
+ message_id: row.id
2684
+ });
2685
+ return;
2686
+ }
2687
+ this.log({
2688
+ level: "error",
2689
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2690
+ conversation_id: row.conversation_id,
2691
+ message_id: row.id
2692
+ });
2693
+ return;
2694
+ }
2695
+ this.dontRedispatch.delete(row.id);
2696
+ return;
2697
+ }
2698
+ if (state === "failed") {
2699
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
2700
+ this.log({
2701
+ level: "error",
2702
+ message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2703
+ conversation_id: row.conversation_id,
2704
+ message_id: row.id
2705
+ });
2706
+ try {
2707
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2);
2708
+ } catch (err) {
2709
+ if (err instanceof ChannelAuthError) throw err;
2710
+ if (err instanceof ChannelTerminalError) {
2711
+ this.doneUndeliverable.add(row.id);
2712
+ this.log({
2713
+ level: "error",
2714
+ 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}`,
2715
+ conversation_id: row.conversation_id,
2716
+ message_id: row.id
2717
+ });
2718
+ return;
2719
+ }
2720
+ this.log({
2721
+ level: "error",
2722
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2723
+ conversation_id: row.conversation_id,
2724
+ message_id: row.id
2725
+ });
2726
+ return;
2727
+ }
2728
+ this.dontRedispatch.delete(row.id);
2729
+ return;
2730
+ }
2731
+ if (this.dontRedispatch.has(row.id)) {
2732
+ this.log({
2733
+ level: "info",
2734
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
2735
+ conversation_id: row.conversation_id,
2736
+ message_id: row.id
2737
+ });
2738
+ return;
2739
+ }
2740
+ if ((state === "running" || state === "queued") && ocId) {
2741
+ const conv = this.convForRow(sessionId, row);
2742
+ const message = this.queuedMessageForRow(row);
2743
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
2744
+ this.dispatched.add(row.id);
2745
+ this.readopted.add(row.id);
2746
+ this.ensureWatcherRunning(sessionId);
2747
+ this.log({
2748
+ level: "info",
2749
+ message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
2750
+ conversation_id: row.conversation_id,
2751
+ message_id: row.id
2752
+ });
2753
+ return;
2754
+ }
2755
+ await this.forceReadoptRun(sessionId, row);
2756
+ }
2757
+ /**
2758
+ * Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
2759
+ *
2760
+ * #218/WI-5: the row's user message is absent (never kept, or a null stored id),
2761
+ * so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
2762
+ * read it back, and register the watcher under the assigned id so the reply
2763
+ * correlates server-side.
2764
+ *
2765
+ * ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
2766
+ * id). Without a guard, if this dispatches on tick N but the read-back+persist
2767
+ * hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
2768
+ * tick N+1 would dispatch AGAIN → duplicate user turns. The `awaitingReadopt`
2769
+ * latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
2770
+ * short-circuit while the row is latched; clear it on a successful dispatch (the
2771
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
2772
+ * re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
2773
+ * may retry exactly once more).
2774
+ *
2775
+ * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
2776
+ * `processed_at` (Invariant 1).
2777
+ */
2778
+ async forceReadoptRun(sessionId, row) {
2779
+ if (this.stopped) {
2780
+ this.log({
2781
+ level: "info",
2782
+ 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`,
2783
+ conversation_id: row.conversation_id,
2784
+ message_id: row.id
2785
+ });
2786
+ return;
2787
+ }
2788
+ if (this.awaitingReadopt.has(row.id)) {
2789
+ this.log({
2790
+ level: "info",
2791
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
2792
+ conversation_id: row.conversation_id,
2793
+ message_id: row.id
2794
+ });
2795
+ return;
2796
+ }
2797
+ if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
2798
+ this.dontRedispatch.add(row.id);
2799
+ this.log({
2800
+ level: "info",
2801
+ 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)`,
2802
+ conversation_id: row.conversation_id,
2803
+ message_id: row.id
2804
+ });
2805
+ return;
2806
+ }
2807
+ const options = {
2808
+ agent: row.opencode_agent ?? void 0,
2809
+ model: row.opencode_model ?? void 0
2810
+ };
2811
+ this.log({
2812
+ level: "info",
2813
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
2814
+ conversation_id: row.conversation_id,
2815
+ message_id: row.id
2816
+ });
2817
+ this.awaitingReadopt.add(row.id);
2818
+ let ocId;
2819
+ try {
2820
+ ocId = await this.dispatchLocked(
2821
+ sessionId,
2822
+ () => sendPromptAsync(this.port, sessionId, row.content, options)
2823
+ );
1811
2824
  } catch (err) {
2825
+ this.awaitingReadopt.delete(row.id);
2826
+ if (err instanceof ChannelAuthError) throw err;
1812
2827
  this.log({
1813
2828
  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
2829
+ message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2830
+ conversation_id: row.conversation_id,
2831
+ message_id: row.id
2832
+ });
2833
+ return;
2834
+ }
2835
+ if (ocId === null) {
2836
+ this.awaitingReadopt.delete(row.id);
2837
+ this.log({
2838
+ level: "error",
2839
+ 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`,
2840
+ conversation_id: row.conversation_id,
2841
+ message_id: row.id
2842
+ });
2843
+ return;
2844
+ }
2845
+ const conv = this.convForRow(sessionId, row);
2846
+ const message = this.queuedMessageForRow(row);
2847
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
2848
+ this.dispatched.add(row.id);
2849
+ this.readopted.add(row.id);
2850
+ this.awaitingReadopt.delete(row.id);
2851
+ this.ensureWatcherRunning(sessionId);
2852
+ }
2853
+ /**
2854
+ * True if `evidentMessageId` is already being driven — either in the
2855
+ * authoritative `dispatched` set or a live watcher's in-flight set for this
2856
+ * session (Invariant 2, WI-5). Either signal means a watcher owns the row.
2857
+ */
2858
+ isTracked(sessionId, evidentMessageId) {
2859
+ if (this.dispatched.has(evidentMessageId)) return true;
2860
+ const watcher = this.watchers.get(sessionId);
2861
+ return watcher?.inFlight.has(evidentMessageId) ?? false;
2862
+ }
2863
+ /**
2864
+ * Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
2865
+ * deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
2866
+ * for `processing` rows, but if it is somehow null/unparseable fall back to
2867
+ * `now` (defensive) AND log — a fallback means the anchor is weaker than
2868
+ * intended, which is worth surfacing.
2869
+ */
2870
+ processedAtMs(row) {
2871
+ const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
2872
+ if (!Number.isNaN(parsed)) return parsed;
2873
+ this.log({
2874
+ level: "error",
2875
+ message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
2876
+ conversation_id: row.conversation_id,
2877
+ message_id: row.id
2878
+ });
2879
+ return this.now();
2880
+ }
2881
+ /** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
2882
+ convForRow(sessionId, row) {
2883
+ return {
2884
+ id: row.conversation_id,
2885
+ agent_id: this.agentId,
2886
+ opencode_session_id: sessionId,
2887
+ pending_message_count: 0,
2888
+ oldest_pending_at: row.processed_at
2889
+ };
2890
+ }
2891
+ /** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
2892
+ queuedMessageForRow(row) {
2893
+ return {
2894
+ id: row.id,
2895
+ content: row.content,
2896
+ status: "processing",
2897
+ opencode_agent: row.opencode_agent,
2898
+ opencode_model: row.opencode_model,
2899
+ source_message_id: row.source_message_id,
2900
+ slack_user_id: row.slack_user_id
2901
+ };
2902
+ }
2903
+ /**
2904
+ * Remove a message from the in-flight set AND the authoritative dispatched
2905
+ * set. Once the in-flight set empties, the watcher loop's `while` guard exits
2906
+ * and its `.finally` removes the session entry from `this.watchers`.
2907
+ *
2908
+ * Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
2909
+ * (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
2910
+ * cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
2911
+ * re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
2912
+ * re-adopted message that completed (`done`) needs no marker — it's leaving
2913
+ * `processing`. This suppresses only re-dispatch: if its reply later completes,
2914
+ * the done branch still delivers it (Bugbot #202).
2915
+ */
2916
+ removeInFlight(watcher, evidentMessageId) {
2917
+ const inFlight = watcher.inFlight.get(evidentMessageId);
2918
+ if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
2919
+ this.dontRedispatch.add(evidentMessageId);
2920
+ this.log({
2921
+ level: "info",
2922
+ message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
2923
+ conversation_id: watcher.conv.id,
2924
+ message_id: evidentMessageId
2925
+ });
2926
+ }
2927
+ watcher.inFlight.delete(evidentMessageId);
2928
+ this.dispatched.delete(evidentMessageId);
2929
+ }
2930
+ /**
2931
+ * Poll `/question` + `/permission` (scoped to the session) and surface NEW ones
2932
+ * via `reportInteraction` (Task 3.5), carrying the PAUSED message's own
2933
+ * `source_message_id` so the server @mentions the correct person under
2934
+ * concurrency. Dedups by interaction id across ticks (reused per-session sets).
2935
+ *
2936
+ * The interaction is attributed to the in-flight message it paused on. opencode
2937
+ * stamps a `messageID` on a permission (and `tool.messageID` on a question) =
2938
+ * the assistant message id, whose `parentID` is the user message id — but the
2939
+ * simplest robust attribution here is: the single in-flight message that is
2940
+ * RUNNING (not done) is the one that paused. With one running message that is
2941
+ * unambiguous; with several we prefer an explicit messageID match, else the
2942
+ * oldest running message.
2943
+ */
2944
+ async pollInteractions(sessionId, watcher, messages) {
2945
+ let questions = [];
2946
+ try {
2947
+ const res = await this.fetchImpl(`${this.opencodeBase}/question`);
2948
+ if (res.ok) {
2949
+ const body = await res.json();
2950
+ questions = Array.isArray(body) ? body : [];
2951
+ }
2952
+ } catch {
2953
+ }
2954
+ for (const q of questions) {
2955
+ if (watcher.reportedQuestions.has(q.id)) continue;
2956
+ if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
2957
+ const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
2958
+ const reported = await this.reportInteraction(
2959
+ watcher.conv.id,
2960
+ "question",
2961
+ q,
2962
+ paused?.message.source_message_id ?? void 0
2963
+ );
2964
+ if (reported) watcher.reportedQuestions.add(q.id);
2965
+ }
2966
+ let permissions = [];
2967
+ try {
2968
+ const res = await this.fetchImpl(`${this.opencodeBase}/permission`);
2969
+ if (res.ok) {
2970
+ const body = await res.json();
2971
+ permissions = Array.isArray(body) ? body : [];
2972
+ }
2973
+ } catch {
2974
+ }
2975
+ for (const p of permissions) {
2976
+ if (watcher.reportedPermissions.has(p.id)) continue;
2977
+ if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
2978
+ const paused = this.attributeInteraction(watcher, p.messageID, messages);
2979
+ const reported = await this.reportInteraction(
2980
+ watcher.conv.id,
2981
+ "permission",
2982
+ p,
2983
+ paused?.message.source_message_id ?? void 0
2984
+ );
2985
+ if (reported) watcher.reportedPermissions.add(p.id);
2986
+ }
2987
+ }
2988
+ /**
2989
+ * True when `sessionId` is the `rootSessionId` itself OR a descendant of it —
2990
+ * i.e. its `parentID` chain (resolved via `GET /session/:id`) reaches the
2991
+ * watched root. Sub-agents spawned via the `task` tool run in child sessions,
2992
+ * so their questions/permissions live under a different `sessionID` that must
2993
+ * still be attributed to the root conversation the watcher owns.
2994
+ *
2995
+ * Parents are cached in `sessionParents` so we walk each session at most once;
2996
+ * a bounded depth cap guards against a cycle or a pathological chain, and any
2997
+ * fetch failure is treated as "not a descendant" (best-effort — the interaction
2998
+ * simply isn't surfaced this tick and is retried next tick once resolvable).
2999
+ */
3000
+ async sessionBelongsTo(sessionId, rootSessionId) {
3001
+ let current = sessionId;
3002
+ for (let depth = 0; current && depth < 32; depth++) {
3003
+ if (current === rootSessionId) return true;
3004
+ const parent = await this.resolveSessionParent(current);
3005
+ if (parent === null || parent === void 0) return false;
3006
+ current = parent;
3007
+ }
3008
+ return false;
3009
+ }
3010
+ /**
3011
+ * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3012
+ * `null` for a root session (no parent) and `undefined` when opencode is
3013
+ * unreachable / the session can't be read (so the caller stops walking without
3014
+ * caching a wrong answer — the next tick retries).
3015
+ */
3016
+ async resolveSessionParent(sessionId) {
3017
+ const cached = this.sessionParents.get(sessionId);
3018
+ if (cached !== void 0) return cached;
3019
+ let parent = void 0;
3020
+ try {
3021
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
3022
+ if (res.ok) {
3023
+ const body = await res.json();
3024
+ parent = body && typeof body.parentID === "string" ? body.parentID : null;
3025
+ }
3026
+ } catch {
3027
+ parent = void 0;
3028
+ }
3029
+ if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3030
+ return parent;
3031
+ }
3032
+ /**
3033
+ * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
3034
+ *
3035
+ * The interaction carries `interactionMessageId` — the ASSISTANT message id
3036
+ * that raised it (a question's `tool.messageID` / a permission's `messageID`).
3037
+ * That assistant message is the reply to ONE of our minted user messages
3038
+ * (correlated by `parentID`, GATE-B). So when we have the tick's message
3039
+ * snapshot, we resolve each running in-flight message's correlated assistant
3040
+ * reply (`findAssistantReplyAfter`) and match its id against
3041
+ * `interactionMessageId` — giving an EXACT attribution even with several
3042
+ * messages in flight concurrently in one session.
3043
+ *
3044
+ * We fall back to the oldest running message ONLY when no exact match is
3045
+ * possible (the id is absent, the snapshot is missing, or the reply has not yet
3046
+ * been correlated). With a single running message either path is exact. Never
3047
+ * throws.
3048
+ *
3049
+ * Attribution must NOT depend on our own `started` PATCH flag: opencode can
3050
+ * START a turn AND raise a question/permission BEFORE our next tick fires
3051
+ * `markProcessing` (which sets `started`). Relying on `started` would leave the
3052
+ * running set empty in that window and let the server fall back to "newest
3053
+ * processing/pending" — possibly @mentioning a FOLLOW-UP author rather than the
3054
+ * person whose active turn actually paused. So we derive "running" from the
3055
+ * tick's `messages` snapshot via `messageRunState` instead.
3056
+ */
3057
+ attributeInteraction(watcher, interactionMessageId, messages) {
3058
+ const inFlight = [...watcher.inFlight.values()].filter((m) => !m.done);
3059
+ if (inFlight.length === 0) return void 0;
3060
+ if (interactionMessageId && messages) {
3061
+ const exact = inFlight.find((m) => {
3062
+ const reply = findAssistantReplyAfter(messages, m.opencodeMessageId);
3063
+ return reply != null && messageIdOf(reply) === interactionMessageId;
1817
3064
  });
3065
+ if (exact) return exact;
3066
+ }
3067
+ const byOldest = (a, b) => a.dispatchedAt - b.dispatchedAt;
3068
+ if (messages) {
3069
+ const runningPerSnapshot = inFlight.filter(
3070
+ (m) => messageRunState(messages, m.opencodeMessageId) === "running"
3071
+ );
3072
+ if (runningPerSnapshot.length > 0) {
3073
+ return runningPerSnapshot.sort(byOldest)[0];
3074
+ }
1818
3075
  }
3076
+ const startedRunning = inFlight.filter((m) => m.started);
3077
+ if (startedRunning.length > 0) {
3078
+ return startedRunning.sort(byOldest)[0];
3079
+ }
3080
+ return inFlight.sort(byOldest)[0];
1819
3081
  }
1820
3082
  // -------------------------------------------------------------------------
1821
3083
  // Evident API calls (combinedAuth thread routes)
@@ -1849,52 +3111,182 @@ var ChannelDriver = class {
1849
3111
  }
1850
3112
  return await res.json();
1851
3113
  }
1852
- async markProcessing(conversationId, messageId, sessionId) {
3114
+ /**
3115
+ * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
3116
+ * The pending path (`getPendingConversations`/`getPendingMessages`) only
3117
+ * surfaces `pending` rows, so a message already `processing` when the runner
3118
+ * died is invisible to it — this dedicated endpoint returns exactly those rows
3119
+ * with the fields the re-adopt path needs (`processed_at`,
3120
+ * `opencode_session_id`, routing).
3121
+ *
3122
+ * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
3123
+ * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
3124
+ * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
3125
+ * next tick retries.
3126
+ */
3127
+ async getProcessingMessages() {
3128
+ const res = await this.fetchImpl(
3129
+ `${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
3130
+ { headers: { Authorization: this.getAuthHeader() } }
3131
+ );
3132
+ this.assertAuth(res, "fetching processing messages");
3133
+ if (!res.ok) {
3134
+ throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
3135
+ }
3136
+ const data = await res.json();
3137
+ let messages = data.messages ?? [];
3138
+ if (this.conversationFilter) {
3139
+ messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
3140
+ }
3141
+ return messages;
3142
+ }
3143
+ /**
3144
+ * EXISTING combinedAuth route — now fired by the watcher on queued→running
3145
+ * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
3146
+ * opencode_session_id}` → `notifyMessageStarted` (hourglass→runner swap +
3147
+ * deep-linked "View in Evident" notice).
3148
+ *
3149
+ * Return/throw contract (consumed by the watcher's swap-to-running guard):
3150
+ * - returns `true` → the server transitioned the row to processing;
3151
+ * - returns `false` → the server gave a DEFINITIVE "already-processing"
3152
+ * answer (a non-retryable, non-auth status — e.g. a
3153
+ * conflict because a duplicate already transitioned it),
3154
+ * so the caller treats it as already-started and does NOT
3155
+ * retry;
3156
+ * - throws `ChannelAuthError` on 401/403 (terminal auth failure);
3157
+ * - throws on a TRANSIENT failure (retryable 5xx/429 status, or a
3158
+ * network-level error from `fetch`) — i.e. NO definitive server response —
3159
+ * so the caller leaves the message un-started and retries the swap on the
3160
+ * next tick.
3161
+ * A single attempt (no internal retry): the watcher's per-tick loop is the
3162
+ * retry vehicle for the swap-to-running.
3163
+ */
3164
+ async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
1853
3165
  const res = await this.fetchImpl(
1854
3166
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
1855
3167
  {
1856
3168
  method: "PATCH",
1857
3169
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1858
- // Include the opencode session id so the server can deep-link the
1859
- // "View in Evident" notice straight to the conversation (the session
1860
- // already exists by now — ensureSession runs before claiming).
1861
- body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
3170
+ body: JSON.stringify({
3171
+ status: "processing",
3172
+ opencode_session_id: sessionId,
3173
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
3174
+ })
1862
3175
  }
1863
3176
  );
1864
3177
  this.assertAuth(res, "marking message as processing");
1865
- return res.ok;
3178
+ if (res.ok) return true;
3179
+ if (isRetryableStatus(res.status)) {
3180
+ throw new Error(`marking message as processing: HTTP ${res.status}`);
3181
+ }
3182
+ return false;
1866
3183
  }
1867
3184
  /**
1868
- * EXISTING combinedAuth completion route — idempotent + retried (WI-CHAN-2).
1869
- * `PATCH .../messages/:id {status:'done', opencode_session_id}`. The server's
3185
+ * EXISTING combinedAuth completion route — idempotent (WI-CHAN-2). `PATCH
3186
+ * .../messages/:id {status:'done', opencode_session_id}`. The server's
1870
3187
  * `queued_conversation_messages.status`/`processed_at` gate makes a re-call
1871
- * for an already-`done` message a no-op (no double Slack post).
3188
+ * for an already-`done` message a no-op (no double Slack post). Fired by the
3189
+ * watcher on per-message completion (Task 3.4) — no `confirmCompletion`
3190
+ * round-trip (we already observed completion via the message list).
3191
+ *
3192
+ * SINGLE ATTEMPT (no in-call `callWithRetry` backoff). The per-session watcher
3193
+ * services its in-flight messages SEQUENTIALLY within a tick
3194
+ * (`runWatcherLoop` → `serviceInFlightMessage`), so a long multi-attempt
3195
+ * backoff here would BLOCK sibling messages in the SAME session/tick: while
3196
+ * message A's done PATCH burned its internal retries, message B could not be
3197
+ * swapped to running even though opencode had already started it. Instead this
3198
+ * does ONE PATCH and surfaces the SAME outcome contract the watcher's markDone
3199
+ * handler already relies on, leaning on the per-tick retry across ticks
3200
+ * (bounded by `inFlight.deadline`) rather than an in-call retry:
3201
+ * - resolves (`void`) → the server transitioned the row to done
3202
+ * (or idempotently confirmed already-done);
3203
+ * - throws `ChannelAuthError` → 401/403 (terminal auth failure → loop
3204
+ * cleanup, Finding 1);
3205
+ * - throws `ChannelTerminalError`→ non-retryable, non-auth 4xx (will never
3206
+ * succeed → straight to the cron, Finding 4);
3207
+ * - throws a plain `Error` → TRANSIENT 5xx/429 or a network-level error
3208
+ * (no definitive server response → the
3209
+ * watcher retries next tick within the
3210
+ * deadline, Finding 4).
1872
3211
  */
1873
- async markDone(conversationId, messageId, sessionId) {
3212
+ async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
3213
+ const res = await this.fetchImpl(
3214
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
3215
+ {
3216
+ method: "PATCH",
3217
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3218
+ body: JSON.stringify({
3219
+ status: "done",
3220
+ opencode_session_id: sessionId,
3221
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
3222
+ })
3223
+ }
3224
+ );
3225
+ this.assertAuth(res, "marking message as done");
3226
+ if (res.ok) return;
3227
+ if (isRetryableStatus(res.status)) {
3228
+ throw new Error(`marking message as done: HTTP ${res.status}`);
3229
+ }
3230
+ throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
3231
+ }
3232
+ /**
3233
+ * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3234
+ * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3235
+ * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3236
+ * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3237
+ * failure reason reaches the channel.
3238
+ */
3239
+ async markFailed(conversationId, messageId, sessionId, error2) {
3240
+ const body = { status: "failed" };
3241
+ if (sessionId !== void 0) body.opencode_session_id = sessionId;
3242
+ if (error2 !== void 0) body.error = error2;
1874
3243
  await this.callWithRetry(
1875
- "marking message as done",
3244
+ "marking message as failed",
1876
3245
  () => this.fetchImpl(
1877
3246
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
1878
3247
  {
1879
3248
  method: "PATCH",
1880
3249
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1881
- body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
3250
+ body: JSON.stringify(body)
1882
3251
  }
1883
3252
  )
1884
3253
  );
1885
3254
  }
1886
- async markFailed(conversationId, messageId) {
1887
- await this.callWithRetry(
1888
- "marking message as failed",
1889
- () => this.fetchImpl(
1890
- `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
3255
+ /**
3256
+ * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3257
+ * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
3258
+ * — the server records it via `log()` (no DB write, no notification). This is
3259
+ * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
3260
+ * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
3261
+ * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
3262
+ * context (no silent catch, per development-workflow).
3263
+ */
3264
+ async postSignal(conversationId, messageId, signal, extra) {
3265
+ try {
3266
+ const res = await this.fetchImpl(
3267
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
1891
3268
  {
1892
- method: "PATCH",
3269
+ method: "POST",
1893
3270
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1894
- body: JSON.stringify({ status: "failed" })
3271
+ body: JSON.stringify({ signal, ...extra })
1895
3272
  }
1896
- )
1897
- );
3273
+ );
3274
+ if (!res.ok) {
3275
+ this.log({
3276
+ level: "error",
3277
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
3278
+ conversation_id: conversationId,
3279
+ message_id: messageId
3280
+ });
3281
+ }
3282
+ } catch (err) {
3283
+ this.log({
3284
+ level: "error",
3285
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
3286
+ conversation_id: conversationId,
3287
+ message_id: messageId
3288
+ });
3289
+ }
1898
3290
  }
1899
3291
  async persistSession(conversationId, sessionId) {
1900
3292
  const res = await this.fetchImpl(
@@ -1909,10 +3301,17 @@ var ChannelDriver = class {
1909
3301
  }
1910
3302
  /**
1911
3303
  * EXISTING combinedAuth interaction route (WI-CHAN-3) — idempotent + retried.
1912
- * `POST .../interactive-event {type, data}`. The server persists the
1913
- * interaction and posts a link to the proxied opencode-web conversation.
3304
+ * `POST .../interactive-event {type, data, source_message_id?}`. The server
3305
+ * persists the interaction and posts a link to the proxied opencode-web
3306
+ * conversation, @mentioning the user who triggered THIS message's turn.
3307
+ *
3308
+ * WI-3 / WI-4 contract: `source_message_id` is the PAUSED message's own Slack
3309
+ * ts (`message.source_message_id`). The server resolves the @mention from that
3310
+ * message's user FIRST (falling back to the old "newest processing" precedence
3311
+ * only when absent), so the correct person is mentioned under concurrency. It
3312
+ * is OPTIONAL for back-compat with older clients / legacy rows.
1914
3313
  */
1915
- async reportInteraction(conversationId, type, data) {
3314
+ async reportInteraction(conversationId, type, data, sourceMessageId) {
1916
3315
  try {
1917
3316
  await this.callWithRetry(
1918
3317
  "reporting interactive event",
@@ -1921,7 +3320,9 @@ var ChannelDriver = class {
1921
3320
  {
1922
3321
  method: "POST",
1923
3322
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1924
- body: JSON.stringify({ type, data })
3323
+ body: JSON.stringify(
3324
+ sourceMessageId ? { type, data, source_message_id: sourceMessageId } : { type, data }
3325
+ )
1925
3326
  }
1926
3327
  )
1927
3328
  );
@@ -1930,6 +3331,7 @@ var ChannelDriver = class {
1930
3331
  message: `${type} surfaced to channel (id: ${data.id.slice(0, 8)})`,
1931
3332
  conversation_id: conversationId
1932
3333
  });
3334
+ return true;
1933
3335
  } catch (err) {
1934
3336
  if (err instanceof ChannelAuthError) throw err;
1935
3337
  this.log({
@@ -1937,6 +3339,7 @@ var ChannelDriver = class {
1937
3339
  message: `Failed to surface ${type}: ${err instanceof Error ? err.message : String(err)}`,
1938
3340
  conversation_id: conversationId
1939
3341
  });
3342
+ return false;
1940
3343
  }
1941
3344
  }
1942
3345
  // -------------------------------------------------------------------------
@@ -1975,8 +3378,9 @@ var ChannelDriver = class {
1975
3378
  await this.sleep(backoffDelay(attempt, this.retry));
1976
3379
  continue;
1977
3380
  }
3381
+ break;
1978
3382
  }
1979
- throw new Error(`${context}: HTTP ${res.status}`);
3383
+ throw new ChannelTerminalError(`${context}: HTTP ${res.status}`, res.status);
1980
3384
  }
1981
3385
  throw lastError instanceof Error ? lastError : new Error(`${context}: exhausted retries`);
1982
3386
  }
@@ -2154,6 +3558,25 @@ async function resolveAgentIdFromKey(authHeader) {
2154
3558
  return { error: `Failed to resolve agent from key: ${message}` };
2155
3559
  }
2156
3560
  }
3561
+ async function notifyAgentDisconnected(agentId, authHeader) {
3562
+ const apiUrl = getApiUrlConfig();
3563
+ try {
3564
+ const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
3565
+ method: "POST",
3566
+ headers: { Authorization: authHeader }
3567
+ });
3568
+ if (!response.ok) {
3569
+ const serverMessage = await readErrorMessage(response);
3570
+ return {
3571
+ ok: false,
3572
+ error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
3573
+ };
3574
+ }
3575
+ return { ok: true };
3576
+ } catch (error2) {
3577
+ return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
3578
+ }
3579
+ }
2157
3580
  async function getAgentInfo(agentId, authHeader) {
2158
3581
  const apiUrl = getApiUrlConfig();
2159
3582
  try {
@@ -2199,7 +3622,9 @@ async function getAgentInfo(agentId, authHeader) {
2199
3622
  // src/commands/run.ts
2200
3623
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
2201
3624
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
2202
- function log(state, message, isError = false) {
3625
+ var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
3626
+ var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
3627
+ function log2(state, message, isError = false) {
2203
3628
  if (state.json) {
2204
3629
  console.log(
2205
3630
  JSON.stringify({
@@ -2224,9 +3649,9 @@ function logActivity(state, entry) {
2224
3649
  }
2225
3650
  if (!state.interactive) {
2226
3651
  if (entry.type === "error") {
2227
- log(state, entry.error ?? "Unknown error", true);
3652
+ log2(state, entry.error ?? "Unknown error", true);
2228
3653
  } else if (entry.type === "info" && entry.message) {
2229
- log(state, entry.message);
3654
+ log2(state, entry.message);
2230
3655
  }
2231
3656
  }
2232
3657
  }
@@ -2312,6 +3737,7 @@ async function handleAuthError(state, error2) {
2312
3737
  }
2313
3738
  async function driveChannels(state, driver) {
2314
3739
  let idlePolls = 0;
3740
+ let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
2315
3741
  while (state.running) {
2316
3742
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
2317
3743
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2321,9 +3747,11 @@ async function driveChannels(state, driver) {
2321
3747
  try {
2322
3748
  const processed = await driver.drainPending();
2323
3749
  state.messageCount += processed;
2324
- if (processed > 0) {
3750
+ const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
3751
+ lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
3752
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
2325
3753
  idlePolls = 0;
2326
- if (state.interactive) displayStatus(state);
3754
+ if (processed > 0 && state.interactive) displayStatus(state);
2327
3755
  } else if (state.idleTimeout !== null) {
2328
3756
  idlePolls++;
2329
3757
  if (idlePolls === 1) {
@@ -2350,7 +3778,7 @@ async function driveChannels(state, driver) {
2350
3778
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
2351
3779
  if (state.interactive) displayStatus(state);
2352
3780
  }
2353
- await new Promise((resolve) => setTimeout(resolve, CHANNEL_POLL_INTERVAL_MS));
3781
+ await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
2354
3782
  if (state.idleTimeout !== null && idlePolls >= 2) {
2355
3783
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
2356
3784
  if (idleMs > state.idleTimeout * 1e3) {
@@ -2361,8 +3789,122 @@ async function driveChannels(state, driver) {
2361
3789
  }
2362
3790
  }
2363
3791
  }
2364
- async function cleanup(state) {
3792
+ var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
3793
+ async function runSweep(state, driver, config2) {
3794
+ const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
3795
+ try {
3796
+ const sessions = await listSessions(state.port);
3797
+ if (sessions === null) {
3798
+ logActivity(state, {
3799
+ type: "info",
3800
+ message: `Session cleanup: could not list sessions (opencode unreachable); skipping this sweep (${mode})`
3801
+ });
3802
+ return;
3803
+ }
3804
+ const toDelete = selectSessionsToDelete(
3805
+ sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
3806
+ {
3807
+ maxAgeMs: config2.maxAgeMs,
3808
+ maxCount: config2.maxCount,
3809
+ nowMs: Date.now(),
3810
+ protectedIds: driver.protectedSessionIds()
3811
+ }
3812
+ );
3813
+ const protectedNow = driver.protectedSessionIds();
3814
+ let deleted = 0;
3815
+ let failed = 0;
3816
+ let skippedNewlyActive = 0;
3817
+ for (const id of toDelete) {
3818
+ if (protectedNow.has(id)) {
3819
+ skippedNewlyActive++;
3820
+ logActivity(state, {
3821
+ type: "info",
3822
+ message: `Session cleanup: skipping ${id} \u2014 became active/bound after selection (${mode})`
3823
+ });
3824
+ continue;
3825
+ }
3826
+ if (await deleteSession(state.port, id)) deleted++;
3827
+ else failed++;
3828
+ }
3829
+ const failedNote = failed > 0 ? `, failed ${failed}` : "";
3830
+ const skippedNote = skippedNewlyActive > 0 ? `, skipped ${skippedNewlyActive} newly-active` : "";
3831
+ logActivity(state, {
3832
+ type: "info",
3833
+ message: `Session cleanup: inspected ${sessions.length}, deleted ${deleted}${failedNote}${skippedNote} (${mode})`
3834
+ });
3835
+ } catch (error2) {
3836
+ const message = error2 instanceof Error ? error2.message : String(error2);
3837
+ logActivity(state, {
3838
+ type: "error",
3839
+ error: `Session cleanup sweep failed (non-fatal, ${mode}): ${message}`
3840
+ });
3841
+ }
3842
+ }
3843
+ function scheduleSessionCleanup(state, driver, options) {
3844
+ const config2 = resolveSessionCleanupConfig(
3845
+ {
3846
+ maxAge: options.sessionCleanupMaxAge,
3847
+ maxCount: options.sessionCleanupMaxCount,
3848
+ interval: options.sessionCleanupInterval
3849
+ },
3850
+ process.env
3851
+ );
3852
+ for (const warning2 of config2.warnings) {
3853
+ logActivity(state, { type: "info", message: `Session cleanup: ${warning2}` });
3854
+ }
3855
+ if (!config2.enabled) return;
3856
+ logActivity(state, {
3857
+ type: "info",
3858
+ message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
3859
+ });
3860
+ const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
3861
+ const firstSweep = setTimeout(
3862
+ () => void runSweep(state, driver, config2),
3863
+ SESSION_CLEANUP_FIRST_SWEEP_MS
3864
+ );
3865
+ state.sessionCleanupTimers.push(interval, firstSweep);
3866
+ }
3867
+ async function notifyOffline(state) {
3868
+ if (!state.agentId || !state.authHeader) return;
3869
+ if (!state.connected) {
3870
+ log2(state, "Skipping offline signal \u2014 this runner does not hold the live tunnel");
3871
+ return;
3872
+ }
3873
+ const result = await notifyAgentDisconnected(state.agentId, state.authHeader);
3874
+ if (result.ok) {
3875
+ log2(state, "Notified Evident the agent is going offline");
3876
+ } else {
3877
+ logActivity(state, {
3878
+ type: "error",
3879
+ error: `Could not notify Evident of offline status (relay will still report it): ${result.error}`
3880
+ });
3881
+ if (state.interactive) displayStatus(state);
3882
+ }
3883
+ }
3884
+ async function cleanup(state, opts = {}) {
2365
3885
  state.running = false;
3886
+ for (const timer of state.sessionCleanupTimers) {
3887
+ clearInterval(timer);
3888
+ clearTimeout(timer);
3889
+ }
3890
+ state.sessionCleanupTimers = [];
3891
+ if (opts.graceful && state.channelDriver) {
3892
+ state.channelDriver.stop();
3893
+ log2(state, "Draining in-flight channel work before shutdown...");
3894
+ if (state.interactive) {
3895
+ logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
3896
+ displayStatus(state);
3897
+ }
3898
+ const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
3899
+ if (!settled) {
3900
+ logActivity(state, {
3901
+ type: "info",
3902
+ message: "Shutdown drain timed out with work still in flight \u2014 leaving it for restart recovery"
3903
+ });
3904
+ if (state.interactive) displayStatus(state);
3905
+ }
3906
+ }
3907
+ await notifyOffline(state);
2366
3908
  if (state.connection) {
2367
3909
  state.connection.close();
2368
3910
  state.connection = null;
@@ -2373,7 +3915,7 @@ async function cleanup(state) {
2373
3915
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
2374
3916
  displayStatus(state);
2375
3917
  } else {
2376
- log(state, "Stopped OpenCode process");
3918
+ log2(state, "Stopped OpenCode process");
2377
3919
  }
2378
3920
  state.opencodeProcess = null;
2379
3921
  }
@@ -2393,26 +3935,32 @@ async function run(options) {
2393
3935
  opencodeVersion: null,
2394
3936
  opencodeProcess: null,
2395
3937
  connection: null,
3938
+ channelDriver: null,
2396
3939
  running: true,
3940
+ shuttingDown: false,
2397
3941
  activityLog: [],
2398
3942
  messageCount: 0,
3943
+ lastProxiedActivityAt: null,
3944
+ sessionCleanupTimers: [],
2399
3945
  authHeader: ""
2400
3946
  };
2401
3947
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
2402
- log(
3948
+ log2(
2403
3949
  state,
2404
3950
  "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.",
2405
3951
  false
2406
3952
  );
2407
3953
  }
2408
3954
  const handleSignal = async () => {
3955
+ if (state.shuttingDown) return;
3956
+ state.shuttingDown = true;
2409
3957
  if (state.interactive) {
2410
3958
  logActivity(state, { type: "info", message: "Shutting down..." });
2411
3959
  displayStatus(state);
2412
3960
  } else {
2413
- log(state, "Shutting down...");
3961
+ log2(state, "Shutting down...");
2414
3962
  }
2415
- await cleanup(state);
3963
+ await cleanup(state, { graceful: true });
2416
3964
  await shutdownTelemetry();
2417
3965
  process.exit(0);
2418
3966
  };
@@ -2443,7 +3991,7 @@ async function run(options) {
2443
3991
  const resolved = await resolveAgentIdFromKey(state.authHeader);
2444
3992
  if (resolved.agent_id) {
2445
3993
  state.agentId = resolved.agent_id;
2446
- log(state, `Resolved agent ID from key: ${state.agentId}`);
3994
+ log2(state, `Resolved agent ID from key: ${state.agentId}`);
2447
3995
  if (state.interactive && !state.json) {
2448
3996
  logActivity(state, {
2449
3997
  type: "info",
@@ -2506,14 +4054,21 @@ async function run(options) {
2506
4054
  port: state.port,
2507
4055
  interactive: state.interactive,
2508
4056
  agentId: state.agentId,
2509
- log: (message) => log(state, message)
4057
+ log: (message) => log2(state, message)
2510
4058
  });
2511
4059
  state.port = oc.port;
2512
4060
  state.opencodeProcess = oc.process;
2513
4061
  state.opencodeVersion = oc.version;
2514
4062
  state.opencodeConnected = oc.process !== null || oc.version !== null;
2515
- const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
2516
- ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
4063
+ const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
4064
+ ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
4065
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
4066
+ if (versionWarning) {
4067
+ log2(state, versionWarning, false);
4068
+ if (state.interactive && !state.json) {
4069
+ logActivity(state, { type: "info", message: versionWarning });
4070
+ }
4071
+ }
2517
4072
  } catch (error2) {
2518
4073
  ocSpinner?.fail(error2.message);
2519
4074
  throw error2;
@@ -2525,12 +4080,14 @@ async function run(options) {
2525
4080
  apiUrl: getApiUrlConfig(),
2526
4081
  getAuthHeader: () => state.authHeader,
2527
4082
  conversationFilter: state.conversationFilter,
4083
+ stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
2528
4084
  log: (entry) => logActivity(state, {
2529
4085
  type: entry.level === "error" ? "error" : "info",
2530
4086
  message: entry.message,
2531
4087
  error: entry.level === "error" ? entry.message : void 0
2532
4088
  })
2533
4089
  });
4090
+ state.channelDriver = channelDriver;
2534
4091
  const connection = new RunnerConnection({
2535
4092
  agentId: state.agentId,
2536
4093
  getAuthHeader: () => state.authHeader,
@@ -2544,7 +4101,11 @@ async function run(options) {
2544
4101
  type: "info",
2545
4102
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
2546
4103
  });
2547
- emitAgentConnected(state.agentId, { port: state.port });
4104
+ emitAgentConnected(state.agentId, {
4105
+ port: state.port,
4106
+ cli_version: getCliVersion(),
4107
+ opencode_version: state.opencodeVersion
4108
+ });
2548
4109
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
2549
4110
  if (state.interactive) displayStatus(state);
2550
4111
  channelDriver.drainPending().then((processed) => {
@@ -2578,9 +4139,14 @@ async function run(options) {
2578
4139
  logActivity(state, { type: "error", error: error2 });
2579
4140
  if (state.interactive) displayStatus(state);
2580
4141
  },
2581
- // Web traffic is proxied transparently; only note opencode is live.
4142
+ // Web traffic is proxied transparently; note opencode is live and stamp
4143
+ // proxied activity so the idle loop treats interactive proxy use as work.
4144
+ // Fires per forwarded response head (incl. every SSE open) and excludes
4145
+ // the internal drain-ping, so an actively-used proxy keeps the timer
4146
+ // fresh while a lone idle SSE with no follow-up requests still ages out.
2582
4147
  onResponse: () => {
2583
4148
  state.opencodeConnected = true;
4149
+ state.lastProxiedActivityAt = Date.now();
2584
4150
  },
2585
4151
  // A channel message was queued and the api-worker pinged us over the
2586
4152
  // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2618,10 +4184,12 @@ async function run(options) {
2618
4184
  if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
2619
4185
  throw error2;
2620
4186
  }
4187
+ scheduleSessionCleanup(state, channelDriver, options);
2621
4188
  if (!interactive || state.json) {
2622
- log(state, "Driving channel messages...");
4189
+ log2(state, "Driving channel messages...");
2623
4190
  }
2624
4191
  await driveChannels(state, channelDriver);
4192
+ if (state.shuttingDown) return;
2625
4193
  await cleanup(state);
2626
4194
  if (state.json) {
2627
4195
  console.log(
@@ -2631,11 +4199,12 @@ async function run(options) {
2631
4199
  })
2632
4200
  );
2633
4201
  } else if (!interactive) {
2634
- log(state, `Completed. Processed ${state.messageCount} message(s).`);
4202
+ log2(state, `Completed. Processed ${state.messageCount} message(s).`);
2635
4203
  }
2636
4204
  await shutdownTelemetry();
2637
4205
  process.exit(0);
2638
4206
  } catch (error2) {
4207
+ if (state.shuttingDown) return;
2639
4208
  await cleanup(state);
2640
4209
  const message = error2 instanceof Error ? error2.message : String(error2);
2641
4210
  if (state.json) {
@@ -2653,8 +4222,9 @@ async function run(options) {
2653
4222
  }
2654
4223
 
2655
4224
  // src/index.ts
4225
+ var { version } = createRequire(import.meta.url)("../package.json");
2656
4226
  var program = new Command();
2657
- program.name("evident").description("Run OpenCode locally and connect it to Evident").version("0.1.0").option(
4227
+ program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
2658
4228
  "--endpoint <url>",
2659
4229
  "Evident API base URL (default: production; e.g. http://localhost:3001)"
2660
4230
  ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {
@@ -2669,7 +4239,16 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
2669
4239
  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);
2670
4240
  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 }));
2671
4241
  program.command("whoami").description("Show the currently logged in user").action(whoami);
2672
- 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(
4242
+ 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(
4243
+ "--session-cleanup-max-age <duration>",
4244
+ "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
4245
+ ).option(
4246
+ "--session-cleanup-max-count <n>",
4247
+ "Keep only the newest N OpenCode sessions. Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_COUNT"
4248
+ ).option(
4249
+ "--session-cleanup-interval <duration>",
4250
+ "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
4251
+ ).action(
2673
4252
  (options) => {
2674
4253
  run({
2675
4254
  agent: options.agent,
@@ -2677,7 +4256,11 @@ program.command("run").description("Connect to Evident and process messages").op
2677
4256
  verbose: options.verbose,
2678
4257
  conversation: options.conversation,
2679
4258
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
2680
- json: options.json
4259
+ json: options.json,
4260
+ // Raw strings — the resolver in run.ts single-sources parsing (M1).
4261
+ sessionCleanupMaxAge: options.sessionCleanupMaxAge,
4262
+ sessionCleanupMaxCount: options.sessionCleanupMaxCount,
4263
+ sessionCleanupInterval: options.sessionCleanupInterval
2681
4264
  });
2682
4265
  }
2683
4266
  );