@evident-ai/cli 3.0.1-dev.4b2470a → 3.0.1-dev.4b2c312

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.js CHANGED
@@ -1,6 +1,7 @@
1
1
  #!/usr/bin/env node
2
2
 
3
3
  // src/index.ts
4
+ import { createRequire } from "module";
4
5
  import { Command } from "commander";
5
6
 
6
7
  // src/commands/login.ts
@@ -284,14 +285,14 @@ function blank() {
284
285
  console.log();
285
286
  }
286
287
  function waitForEnter(prompt = "Press Enter to continue...") {
287
- return new Promise((resolve) => {
288
+ return new Promise((resolve2) => {
288
289
  process.stdout.write(chalk.dim(prompt));
289
290
  const handler = () => {
290
291
  process.stdin.removeListener("data", handler);
291
292
  process.stdin.setRawMode?.(false);
292
293
  process.stdin.pause();
293
294
  console.log();
294
- resolve();
295
+ resolve2();
295
296
  };
296
297
  if (process.stdin.isTTY) {
297
298
  process.stdin.setRawMode?.(true);
@@ -301,7 +302,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
301
302
  });
302
303
  }
303
304
  function sleep(ms) {
304
- return new Promise((resolve) => setTimeout(resolve, ms));
305
+ return new Promise((resolve2) => setTimeout(resolve2, ms));
305
306
  }
306
307
 
307
308
  // src/commands/login.ts
@@ -375,19 +376,19 @@ async function tokenLogin() {
375
376
  console.log("Visit your Evident dashboard to generate a CLI token.");
376
377
  blank();
377
378
  process.stdout.write("Paste token: ");
378
- const token = await new Promise((resolve) => {
379
+ const token = await new Promise((resolve2) => {
379
380
  let data = "";
380
381
  process.stdin.setEncoding("utf8");
381
382
  process.stdin.on("data", (chunk) => {
382
383
  data += chunk;
383
384
  });
384
385
  process.stdin.on("end", () => {
385
- resolve(data.trim());
386
+ resolve2(data.trim());
386
387
  });
387
388
  if (process.stdin.isTTY) {
388
389
  process.stdin.once("data", (chunk) => {
389
390
  process.stdin.pause();
390
- resolve(chunk.toString().trim());
391
+ resolve2(chunk.toString().trim());
391
392
  });
392
393
  process.stdin.resume();
393
394
  }
@@ -484,8 +485,34 @@ var TelemetryEventTypes = {
484
485
  var MAX_FRAME_BYTES = 256 * 1024;
485
486
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
486
487
 
488
+ // ../../packages/types/src/logging/index.ts
489
+ var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
+ function log(level, event, fields) {
491
+ const method = level === "debug" ? "log" : level;
492
+ try {
493
+ console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
494
+ } catch (err) {
495
+ console.error(
496
+ "[evident] log_serialize_failed",
497
+ event,
498
+ err instanceof Error ? err.message : String(err)
499
+ );
500
+ }
501
+ }
502
+ function stripQuery(url) {
503
+ try {
504
+ return new URL(url).pathname;
505
+ } catch {
506
+ const q = url.indexOf("?");
507
+ return q === -1 ? url : url.slice(0, q);
508
+ }
509
+ }
510
+
487
511
  // src/lib/telemetry.ts
488
- var CLI_VERSION = process.env.npm_package_version || "unknown";
512
+ var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
513
+ function getCliVersion() {
514
+ return CLI_VERSION;
515
+ }
489
516
  var eventBuffer = [];
490
517
  var flushTimeout = null;
491
518
  var isShuttingDown = false;
@@ -679,20 +706,20 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
679
706
  if (health.healthy) {
680
707
  return health;
681
708
  }
682
- await new Promise((resolve) => setTimeout(resolve, 1e3));
709
+ await new Promise((resolve2) => setTimeout(resolve2, 1e3));
683
710
  }
684
711
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
685
712
  }
686
713
 
687
714
  // src/lib/opencode/opencode-version-gate.ts
688
- var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
689
- function isQueueValidatedVersion(version) {
690
- if (!version) return false;
691
- return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version);
692
- }
693
- function buildOpenCodeVersionWarning(version) {
694
- if (isQueueValidatedVersion(version)) return null;
695
- const detected = version ? `v${version}` : "unknown";
715
+ var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
716
+ function isQueueValidatedVersion(version2) {
717
+ if (!version2) return false;
718
+ return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
719
+ }
720
+ function buildOpenCodeVersionWarning(version2) {
721
+ if (isQueueValidatedVersion(version2)) return null;
722
+ const detected = version2 ? `v${version2}` : "unknown";
696
723
  const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
697
724
  return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
698
725
  }
@@ -1009,7 +1036,11 @@ function roleOf(m) {
1009
1036
  }
1010
1037
  function completedOf(m) {
1011
1038
  if (!m || typeof m !== "object") return void 0;
1012
- return m.info?.time?.completed;
1039
+ return m.info?.time?.completed ?? m.time?.completed;
1040
+ }
1041
+ function createdOf(m) {
1042
+ if (!m || typeof m !== "object") return void 0;
1043
+ return m.info?.time?.created ?? m.time?.created;
1013
1044
  }
1014
1045
  function idOf(m) {
1015
1046
  if (!m || typeof m !== "object") return void 0;
@@ -1029,6 +1060,66 @@ function finishOf(m) {
1029
1060
  const infoFinish = m.info?.finish;
1030
1061
  return typeof infoFinish === "string" ? infoFinish : void 0;
1031
1062
  }
1063
+ function errorOf(m) {
1064
+ if (!m || typeof m !== "object") return void 0;
1065
+ return m.info?.error ?? m.error;
1066
+ }
1067
+ function isAssistantInFlight(m) {
1068
+ if (completedOf(m) == null) return true;
1069
+ return finishOf(m) === "tool-calls";
1070
+ }
1071
+ async function getSessionMessages(port, sessionId) {
1072
+ try {
1073
+ const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
1074
+ if (!res.ok) return null;
1075
+ const body = await res.json();
1076
+ return Array.isArray(body) ? body : null;
1077
+ } catch {
1078
+ return null;
1079
+ }
1080
+ }
1081
+ function sessionLastActivityMs(session) {
1082
+ const candidates = [
1083
+ session.time?.updated,
1084
+ session.time?.created,
1085
+ session.time_updated,
1086
+ session.time_created,
1087
+ session.updated,
1088
+ session.created
1089
+ ];
1090
+ for (const c of candidates) {
1091
+ if (typeof c === "number" && Number.isFinite(c)) return c;
1092
+ }
1093
+ return null;
1094
+ }
1095
+ async function listSessions(port) {
1096
+ try {
1097
+ const res = await fetch(`${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
+ }
1122
+ }
1032
1123
  async function createOpenCodeSession(port, directory) {
1033
1124
  const url = new URL(`${opencodeBase(port)}/session`);
1034
1125
  if (directory && directory.trim()) {
@@ -1046,9 +1137,16 @@ async function createOpenCodeSession(port, directory) {
1046
1137
  const data = await response.json();
1047
1138
  return data.id;
1048
1139
  }
1049
- async function sendPromptAsync(port, sessionId, content, options, messageId) {
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
+ );
1050
1149
  const body = {
1051
- messageID: messageId,
1052
1150
  parts: [{ type: "text", text: content }]
1053
1151
  };
1054
1152
  if (options?.agent) {
@@ -1072,6 +1170,29 @@ async function sendPromptAsync(port, sessionId, content, options, messageId) {
1072
1170
  const text = await res.text().catch(() => "");
1073
1171
  throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1074
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 };
1187
+ }
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;
1075
1196
  }
1076
1197
  function findAssistantReplyAfter(messages, userMessageId) {
1077
1198
  if (!messages || messages.length === 0) return null;
@@ -1088,19 +1209,31 @@ function findAssistantReplyAfter(messages, userMessageId) {
1088
1209
  }
1089
1210
  function findLastAssistantReplyFor(messages, userMessageId) {
1090
1211
  if (!messages || messages.length === 0) return null;
1212
+ let lastCorrelated = null;
1213
+ let lastNonErrored = null;
1091
1214
  for (let i = messages.length - 1; i >= 0; i--) {
1092
1215
  const m = messages[i];
1093
- if (roleOf(m) === "assistant" && parentIdOf(m) === userMessageId) return m;
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
+ }
1094
1222
  }
1223
+ if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
1095
1224
  const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1096
1225
  if (userIndex === -1) return null;
1097
1226
  let last = null;
1227
+ let lastOk = null;
1098
1228
  for (let i = userIndex + 1; i < messages.length; i++) {
1099
1229
  const role = roleOf(messages[i]);
1100
1230
  if (role === "user") break;
1101
- if (role === "assistant") last = messages[i];
1231
+ if (role === "assistant") {
1232
+ last = messages[i];
1233
+ if (errorOf(messages[i]) == null) lastOk = messages[i];
1234
+ }
1102
1235
  }
1103
- return last;
1236
+ return lastOk ?? last;
1104
1237
  }
1105
1238
  function messageRunState(messages, userMessageId) {
1106
1239
  if (!messages || messages.length === 0) return "unknown";
@@ -1110,13 +1243,131 @@ function messageRunState(messages, userMessageId) {
1110
1243
  if (!reply) return "unknown";
1111
1244
  }
1112
1245
  if (!reply) return "queued";
1113
- if (completedOf(reply) == null) return "running";
1114
- if (finishOf(reply) === "tool-calls") return "running";
1115
- return "done";
1246
+ if (isAssistantInFlight(reply)) return "running";
1247
+ return errorOf(reply) != null ? "failed" : "done";
1116
1248
  }
1117
- function opencodeMessageIdFor(queuedMessageId) {
1118
- const sanitized = queuedMessageId.replace(/[^a-zA-Z0-9]/g, "_");
1119
- return `msg_${sanitized}`;
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 "${input}": must be a positive value.`);
1287
+ }
1288
+ return value * DURATION_UNIT_MS[match[2]];
1289
+ }
1290
+ function selectSessionsToDelete(sessions, opts) {
1291
+ const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
1292
+ if (maxAgeMs === void 0 && maxCount === void 0) return [];
1293
+ const ageEligible = (s) => {
1294
+ if (maxAgeMs === void 0) return false;
1295
+ if (s.lastActivityMs === null) return true;
1296
+ return nowMs - s.lastActivityMs > maxAgeMs;
1297
+ };
1298
+ const countEligibleIds = /* @__PURE__ */ new Set();
1299
+ if (maxCount !== void 0) {
1300
+ const byActivityDesc = [...sessions].sort(
1301
+ (a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
1302
+ );
1303
+ for (const s of byActivityDesc.slice(maxCount)) {
1304
+ countEligibleIds.add(s.id);
1305
+ }
1306
+ }
1307
+ const toDelete = [];
1308
+ for (const s of sessions) {
1309
+ if (protectedIds.has(s.id)) continue;
1310
+ if (ageEligible(s) || countEligibleIds.has(s.id)) {
1311
+ toDelete.push(s.id);
1312
+ }
1313
+ }
1314
+ return toDelete;
1315
+ }
1316
+ var DEFAULT_INTERVAL = "1h";
1317
+ function resolve(flag, envValue, fallback) {
1318
+ return flag ?? envValue ?? fallback;
1319
+ }
1320
+ function parseMaxCount(input) {
1321
+ const trimmed = input.trim();
1322
+ if (!/^\d+$/.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) {
1342
+ try {
1343
+ maxAgeMs = parseDurationMs(maxAgeRaw);
1344
+ } catch (err) {
1345
+ warnings.push(
1346
+ `Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
1347
+ );
1348
+ }
1349
+ }
1350
+ let maxCount;
1351
+ if (maxCountRaw !== void 0) {
1352
+ try {
1353
+ maxCount = parseMaxCount(maxCountRaw);
1354
+ } catch (err) {
1355
+ warnings.push(
1356
+ `Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
1357
+ );
1358
+ }
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 };
1120
1371
  }
1121
1372
 
1122
1373
  // src/lib/tunnel/connection.ts
@@ -1187,24 +1438,34 @@ var StreamForwarder = class {
1187
1438
  }
1188
1439
  async handleOpen(frame) {
1189
1440
  const { sid, method, path, headers, has_body } = frame;
1441
+ const correlationId = headers?.[CORRELATION_ID_HEADER];
1442
+ const startedAt = Date.now();
1190
1443
  if (path === TUNNEL_DRAIN_PING_PATH) {
1191
1444
  this.callbacks.onDrainPing?.();
1192
1445
  this.send({ type: "head", sid, status: 204, headers: {} });
1193
1446
  this.send({ type: "res_end", sid });
1194
1447
  return;
1195
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
+ }
1196
1457
  const ac = new AbortController();
1197
1458
  let bodyPromise;
1198
1459
  let pushBody;
1199
1460
  let endBody;
1200
1461
  if (has_body) {
1201
1462
  const chunks = [];
1202
- bodyPromise = new Promise((resolve) => {
1463
+ bodyPromise = new Promise((resolve2) => {
1203
1464
  pushBody = (buf) => {
1204
1465
  chunks.push(buf);
1205
1466
  };
1206
1467
  endBody = () => {
1207
- resolve(Buffer.concat(chunks));
1468
+ resolve2(Buffer.concat(chunks));
1208
1469
  };
1209
1470
  });
1210
1471
  }
@@ -1239,6 +1500,14 @@ var StreamForwarder = class {
1239
1500
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1240
1501
  });
1241
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
+ }
1242
1511
  this.callbacks.onHead?.(sid, upstream.status);
1243
1512
  try {
1244
1513
  if (upstream.body) {
@@ -1314,7 +1583,7 @@ function connectTunnel(options) {
1314
1583
  } = options;
1315
1584
  const tunnelUrl = getTunnelUrlConfig();
1316
1585
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1317
- return new Promise((resolve, reject) => {
1586
+ return new Promise((resolve2, reject) => {
1318
1587
  const ws = new WebSocket2(url, {
1319
1588
  headers: {
1320
1589
  Authorization: authHeader
@@ -1379,7 +1648,7 @@ function connectTunnel(options) {
1379
1648
  clearTimeout(connectionTimeout);
1380
1649
  const connectedAgentId = message.agent_id ?? agentId;
1381
1650
  onConnected?.(connectedAgentId);
1382
- resolve({
1651
+ resolve2({
1383
1652
  ws,
1384
1653
  close: () => ws.close(1e3, "CLI shutdown")
1385
1654
  });
@@ -1508,7 +1777,7 @@ var DEFAULT_RETRY_POLICY = {
1508
1777
  };
1509
1778
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1510
1779
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1511
- var DEFAULT_DISPATCH_CONFIRM_MS = 6e3;
1780
+ var DEFAULT_STUCK_QUEUED_MS = 6e4;
1512
1781
  var ChannelAuthError = class extends Error {
1513
1782
  constructor(message) {
1514
1783
  super(message);
@@ -1543,10 +1812,18 @@ var ChannelDriver = class {
1543
1812
  sleep;
1544
1813
  pausedPollIntervalMs;
1545
1814
  pausedMaxWaitMs;
1546
- dispatchConfirmMs;
1815
+ stuckQueuedMs;
1547
1816
  now;
1548
1817
  /** Cache of conversationId → opencode sessionId. */
1549
1818
  sessions = /* @__PURE__ */ new Map();
1819
+ /**
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();
1550
1827
  /**
1551
1828
  * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1552
1829
  * session: one polling loop services all of that session's in-flight messages.
@@ -1563,6 +1840,54 @@ var ChannelDriver = class {
1563
1840
  * a steady-state-poll re-dispatch will not double-run the message.
1564
1841
  */
1565
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();
1566
1891
  /**
1567
1892
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1568
1893
  * first session creation so drain-created sessions are rooted at the project
@@ -1570,8 +1895,33 @@ var ChannelDriver = class {
1570
1895
  * not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
1571
1896
  */
1572
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();
1573
1908
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
1574
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;
1575
1925
  constructor(config2) {
1576
1926
  this.agentId = config2.agentId;
1577
1927
  this.port = config2.port;
@@ -1585,7 +1935,7 @@ var ChannelDriver = class {
1585
1935
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1586
1936
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1587
1937
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
1588
- this.dispatchConfirmMs = config2.dispatchConfirmMs ?? DEFAULT_DISPATCH_CONFIRM_MS;
1938
+ this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
1589
1939
  this.now = config2.now ?? (() => Date.now());
1590
1940
  }
1591
1941
  /** The IPv4-loopback base URL for the local `opencode serve`. */
@@ -1603,8 +1953,21 @@ var ChannelDriver = class {
1603
1953
  * @returns the number of messages NEWLY dispatched to opencode's native queue.
1604
1954
  */
1605
1955
  async drainPending() {
1956
+ if (this.stopped) return 0;
1606
1957
  if (this.draining) return 0;
1607
1958
  this.draining = true;
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() {
1608
1971
  let dispatched = 0;
1609
1972
  try {
1610
1973
  const conversations = await this.getPendingConversations();
@@ -1616,8 +1979,10 @@ var ChannelDriver = class {
1616
1979
  });
1617
1980
  }
1618
1981
  for (const conv of conversations) {
1982
+ if (this.stopped) break;
1619
1983
  dispatched += await this.processConversation(conv);
1620
1984
  }
1985
+ await this.readoptProcessing();
1621
1986
  } finally {
1622
1987
  this.draining = false;
1623
1988
  }
@@ -1635,6 +2000,73 @@ var ChannelDriver = class {
1635
2000
  }
1636
2001
  return false;
1637
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;
2069
+ }
1638
2070
  /**
1639
2071
  * Await all outstanding per-session watchers (WI-3).
1640
2072
  *
@@ -1669,15 +2101,18 @@ var ChannelDriver = class {
1669
2101
  const sessionId = await this.ensureSession(conv);
1670
2102
  const messages = await this.getPendingMessages(conv.id);
1671
2103
  let dispatched = 0;
2104
+ let skippedAlreadyDispatched = 0;
1672
2105
  for (const message of messages) {
2106
+ if (this.stopped) break;
1673
2107
  if (this.dispatched.has(message.id)) {
2108
+ skippedAlreadyDispatched += 1;
1674
2109
  continue;
1675
2110
  }
1676
- const opencodeMessageId = opencodeMessageIdFor(message.id);
1677
2111
  const options = {
1678
2112
  agent: message.opencode_agent ?? void 0,
1679
2113
  model: message.opencode_model ?? void 0
1680
2114
  };
2115
+ let opencodeMessageId;
1681
2116
  try {
1682
2117
  this.log({
1683
2118
  level: "info",
@@ -1685,10 +2120,23 @@ var ChannelDriver = class {
1685
2120
  conversation_id: conv.id,
1686
2121
  message_id: message.id
1687
2122
  });
1688
- await sendPromptAsync(this.port, sessionId, message.content, options, opencodeMessageId);
2123
+ opencodeMessageId = await this.dispatchLocked(
2124
+ sessionId,
2125
+ () => sendPromptAsync(this.port, sessionId, message.content, options)
2126
+ );
1689
2127
  } catch (err) {
1690
2128
  if (err instanceof ChannelAuthError) throw err;
1691
2129
  this.dispatched.delete(message.id);
2130
+ if (await sessionExists(this.port, sessionId) === false) {
2131
+ this.sessions.delete(conv.id);
2132
+ this.log({
2133
+ level: "info",
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.`,
2135
+ conversation_id: conv.id,
2136
+ message_id: message.id
2137
+ });
2138
+ break;
2139
+ }
1692
2140
  await this.markFailed(conv.id, message.id).catch(() => {
1693
2141
  });
1694
2142
  this.log({
@@ -1699,24 +2147,58 @@ var ChannelDriver = class {
1699
2147
  });
1700
2148
  continue;
1701
2149
  }
2150
+ if (opencodeMessageId === null) {
2151
+ this.log({
2152
+ level: "error",
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`,
2154
+ conversation_id: conv.id,
2155
+ message_id: message.id
2156
+ });
2157
+ continue;
2158
+ }
1702
2159
  this.dispatched.add(message.id);
1703
2160
  this.registerInFlight(conv, sessionId, message, opencodeMessageId);
1704
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
+ });
1705
2170
  }
1706
2171
  this.ensureWatcherRunning(sessionId);
1707
2172
  return dispatched;
1708
2173
  }
1709
2174
  async ensureSession(conv) {
1710
- const cached = this.sessions.get(conv.id);
1711
- if (cached) return cached;
1712
- if (conv.opencode_session_id) {
1713
- this.sessions.set(conv.id, conv.opencode_session_id);
1714
- return conv.opencode_session_id;
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;
1715
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) {
1716
2198
  const directory = await this.resolveOpenCodeDirectory();
1717
2199
  const sessionId = await createOpenCodeSession(this.port, directory);
1718
- this.sessions.set(conv.id, sessionId);
1719
- await this.persistSession(conv.id, sessionId).catch(() => {
2200
+ this.sessions.set(conversationId, sessionId);
2201
+ await this.persistSession(conversationId, sessionId).catch(() => {
1720
2202
  });
1721
2203
  return sessionId;
1722
2204
  }
@@ -1739,6 +2221,25 @@ var ChannelDriver = class {
1739
2221
  // -------------------------------------------------------------------------
1740
2222
  // Per-session watcher (WI-3)
1741
2223
  // -------------------------------------------------------------------------
2224
+ /**
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).
2230
+ */
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
+ }
1742
2243
  /** Register a freshly-dispatched message with its session's watcher state. */
1743
2244
  registerInFlight(conv, sessionId, message, opencodeMessageId) {
1744
2245
  let watcher = this.watchers.get(sessionId);
@@ -1760,7 +2261,54 @@ var ChannelDriver = class {
1760
2261
  dispatchedAt: now,
1761
2262
  deadline: now + this.pausedMaxWaitMs,
1762
2263
  started: false,
1763
- done: false
2264
+ done: false,
2265
+ stuckReported: false
2266
+ });
2267
+ }
2268
+ /**
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).
2276
+ *
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.
2284
+ */
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
1764
2312
  });
1765
2313
  }
1766
2314
  /**
@@ -1826,6 +2374,7 @@ var ChannelDriver = class {
1826
2374
  conversation_id: watcher.conv.id
1827
2375
  });
1828
2376
  for (const evidentMessageId of [...watcher.inFlight.keys()]) {
2377
+ this.readopted.delete(evidentMessageId);
1829
2378
  this.removeInFlight(watcher, evidentMessageId);
1830
2379
  }
1831
2380
  return;
@@ -1846,10 +2395,15 @@ var ChannelDriver = class {
1846
2395
  async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
1847
2396
  const conv = watcher.conv;
1848
2397
  const state = messageRunState(messages, inFlight.opencodeMessageId);
1849
- if ((state === "running" || state === "done") && !inFlight.started) {
2398
+ if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
1850
2399
  let claimed;
1851
2400
  try {
1852
- claimed = await this.markProcessing(conv.id, inFlight.evidentMessageId, sessionId);
2401
+ claimed = await this.markProcessing(
2402
+ conv.id,
2403
+ inFlight.evidentMessageId,
2404
+ sessionId,
2405
+ inFlight.opencodeMessageId
2406
+ );
1853
2407
  } catch (err) {
1854
2408
  if (err instanceof ChannelAuthError) throw err;
1855
2409
  this.log({
@@ -1879,7 +2433,12 @@ var ChannelDriver = class {
1879
2433
  message_id: inFlight.evidentMessageId
1880
2434
  });
1881
2435
  try {
1882
- await this.markDone(conv.id, inFlight.evidentMessageId, sessionId);
2436
+ await this.markDone(
2437
+ conv.id,
2438
+ inFlight.evidentMessageId,
2439
+ sessionId,
2440
+ inFlight.opencodeMessageId
2441
+ );
1883
2442
  } catch (err) {
1884
2443
  if (err instanceof ChannelAuthError) throw err;
1885
2444
  if (err instanceof ChannelTerminalError) {
@@ -1915,10 +2474,59 @@ var ChannelDriver = class {
1915
2474
  this.removeInFlight(watcher, inFlight.evidentMessageId);
1916
2475
  return;
1917
2476
  }
1918
- if (state === "unknown") {
1919
- if (this.now() - inFlight.dispatchedAt >= this.dispatchConfirmMs) {
1920
- await this.redispatchInFlight(sessionId, inFlight);
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;
1921
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
+ });
1922
2530
  }
1923
2531
  if (this.now() >= inFlight.deadline) {
1924
2532
  this.log({
@@ -1927,48 +2535,395 @@ var ChannelDriver = class {
1927
2535
  conversation_id: conv.id,
1928
2536
  message_id: inFlight.evidentMessageId
1929
2537
  });
2538
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2539
+ watched_for_ms: this.now() - inFlight.dispatchedAt
2540
+ });
1930
2541
  this.removeInFlight(watcher, inFlight.evidentMessageId);
1931
2542
  }
1932
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
+ }
1933
2627
  /**
1934
- * Re-dispatch a message whose user row never appeared (idle-path guard). Safe:
1935
- * opencode treats a duplicate caller-supplied `messageID` as idempotent (PoC
1936
- * fact 9) — one user message + one reply even if the original DID land. Resets
1937
- * the dispatch timestamp so the guard doesn't immediately fire again.
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.
1938
2644
  */
1939
- async redispatchInFlight(sessionId, inFlight) {
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
+ }
1940
2807
  const options = {
1941
- agent: inFlight.message.opencode_agent ?? void 0,
1942
- model: inFlight.message.opencode_model ?? void 0
2808
+ agent: row.opencode_agent ?? void 0,
2809
+ model: row.opencode_model ?? void 0
1943
2810
  };
1944
2811
  this.log({
1945
2812
  level: "info",
1946
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} not observed after dispatch \u2014 re-dispatching (idle-path guard)`,
1947
- message_id: inFlight.evidentMessageId
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
1948
2816
  });
2817
+ this.awaitingReadopt.add(row.id);
2818
+ let ocId;
1949
2819
  try {
1950
- await sendPromptAsync(
1951
- this.port,
2820
+ ocId = await this.dispatchLocked(
1952
2821
  sessionId,
1953
- inFlight.message.content,
1954
- options,
1955
- inFlight.opencodeMessageId
2822
+ () => sendPromptAsync(this.port, sessionId, row.content, options)
1956
2823
  );
1957
2824
  } catch (err) {
2825
+ this.awaitingReadopt.delete(row.id);
2826
+ if (err instanceof ChannelAuthError) throw err;
1958
2827
  this.log({
1959
2828
  level: "error",
1960
- message: `Re-dispatch failed for message ${inFlight.evidentMessageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
1961
- message_id: inFlight.evidentMessageId
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
1962
2842
  });
2843
+ return;
1963
2844
  }
1964
- inFlight.dispatchedAt = this.now();
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
+ };
1965
2902
  }
1966
2903
  /**
1967
2904
  * Remove a message from the in-flight set AND the authoritative dispatched
1968
2905
  * set. Once the in-flight set empties, the watcher loop's `while` guard exits
1969
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).
1970
2915
  */
1971
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
+ }
1972
2927
  watcher.inFlight.delete(evidentMessageId);
1973
2928
  this.dispatched.delete(evidentMessageId);
1974
2929
  }
@@ -1997,8 +2952,8 @@ var ChannelDriver = class {
1997
2952
  } catch {
1998
2953
  }
1999
2954
  for (const q of questions) {
2000
- if (q.sessionID !== sessionId) continue;
2001
2955
  if (watcher.reportedQuestions.has(q.id)) continue;
2956
+ if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
2002
2957
  const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
2003
2958
  const reported = await this.reportInteraction(
2004
2959
  watcher.conv.id,
@@ -2018,8 +2973,8 @@ var ChannelDriver = class {
2018
2973
  } catch {
2019
2974
  }
2020
2975
  for (const p of permissions) {
2021
- if (p.sessionID !== sessionId) continue;
2022
2976
  if (watcher.reportedPermissions.has(p.id)) continue;
2977
+ if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
2023
2978
  const paused = this.attributeInteraction(watcher, p.messageID, messages);
2024
2979
  const reported = await this.reportInteraction(
2025
2980
  watcher.conv.id,
@@ -2030,6 +2985,50 @@ var ChannelDriver = class {
2030
2985
  if (reported) watcher.reportedPermissions.add(p.id);
2031
2986
  }
2032
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/${sessionId}`);
3022
+ if (res.ok) {
3023
+ const body = await res.json();
3024
+ parent = body && typeof body.parentID === "string" ? body.parentID : null;
3025
+ }
3026
+ } catch {
3027
+ parent = void 0;
3028
+ }
3029
+ if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3030
+ return parent;
3031
+ }
2033
3032
  /**
2034
3033
  * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
2035
3034
  *
@@ -2112,6 +3111,35 @@ var ChannelDriver = class {
2112
3111
  }
2113
3112
  return await res.json();
2114
3113
  }
3114
+ /**
3115
+ * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
3116
+ * The pending path (`getPendingConversations`/`getPendingMessages`) only
3117
+ * surfaces `pending` rows, so a message already `processing` when the runner
3118
+ * died is invisible to it — this dedicated endpoint returns exactly those rows
3119
+ * with the fields the re-adopt path needs (`processed_at`,
3120
+ * `opencode_session_id`, routing).
3121
+ *
3122
+ * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
3123
+ * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
3124
+ * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
3125
+ * next tick retries.
3126
+ */
3127
+ async getProcessingMessages() {
3128
+ const res = await this.fetchImpl(
3129
+ `${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 ${res.status}`);
3135
+ }
3136
+ const data = await res.json();
3137
+ let messages = data.messages ?? [];
3138
+ if (this.conversationFilter) {
3139
+ messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
3140
+ }
3141
+ return messages;
3142
+ }
2115
3143
  /**
2116
3144
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
2117
3145
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -2133,13 +3161,17 @@ var ChannelDriver = class {
2133
3161
  * A single attempt (no internal retry): the watcher's per-tick loop is the
2134
3162
  * retry vehicle for the swap-to-running.
2135
3163
  */
2136
- async markProcessing(conversationId, messageId, sessionId) {
3164
+ async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
2137
3165
  const res = await this.fetchImpl(
2138
3166
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
2139
3167
  {
2140
3168
  method: "PATCH",
2141
3169
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2142
- body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
3170
+ body: JSON.stringify({
3171
+ status: "processing",
3172
+ opencode_session_id: sessionId,
3173
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
3174
+ })
2143
3175
  }
2144
3176
  );
2145
3177
  this.assertAuth(res, "marking message as processing");
@@ -2177,13 +3209,17 @@ var ChannelDriver = class {
2177
3209
  * watcher retries next tick within the
2178
3210
  * deadline, Finding 4).
2179
3211
  */
2180
- async markDone(conversationId, messageId, sessionId) {
3212
+ async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
2181
3213
  const res = await this.fetchImpl(
2182
3214
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
2183
3215
  {
2184
3216
  method: "PATCH",
2185
3217
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2186
- body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
3218
+ body: JSON.stringify({
3219
+ status: "done",
3220
+ opencode_session_id: sessionId,
3221
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
3222
+ })
2187
3223
  }
2188
3224
  );
2189
3225
  this.assertAuth(res, "marking message as done");
@@ -2193,7 +3229,17 @@ var ChannelDriver = class {
2193
3229
  }
2194
3230
  throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
2195
3231
  }
2196
- async markFailed(conversationId, messageId) {
3232
+ /**
3233
+ * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3234
+ * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3235
+ * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3236
+ * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3237
+ * failure reason reaches the channel.
3238
+ */
3239
+ async markFailed(conversationId, messageId, sessionId, error2) {
3240
+ const body = { status: "failed" };
3241
+ if (sessionId !== void 0) body.opencode_session_id = sessionId;
3242
+ if (error2 !== void 0) body.error = error2;
2197
3243
  await this.callWithRetry(
2198
3244
  "marking message as failed",
2199
3245
  () => this.fetchImpl(
@@ -2201,11 +3247,47 @@ var ChannelDriver = class {
2201
3247
  {
2202
3248
  method: "PATCH",
2203
3249
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2204
- body: JSON.stringify({ status: "failed" })
3250
+ body: JSON.stringify(body)
2205
3251
  }
2206
3252
  )
2207
3253
  );
2208
3254
  }
3255
+ /**
3256
+ * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3257
+ * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
3258
+ * — the server records it via `log()` (no DB write, no notification). This is
3259
+ * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
3260
+ * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
3261
+ * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
3262
+ * context (no silent catch, per development-workflow).
3263
+ */
3264
+ async postSignal(conversationId, messageId, signal, extra) {
3265
+ try {
3266
+ const res = await this.fetchImpl(
3267
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
3268
+ {
3269
+ method: "POST",
3270
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3271
+ body: JSON.stringify({ signal, ...extra })
3272
+ }
3273
+ );
3274
+ if (!res.ok) {
3275
+ this.log({
3276
+ level: "error",
3277
+ message: `Signal '${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): ${err instanceof Error ? err.message : String(err)}`,
3286
+ conversation_id: conversationId,
3287
+ message_id: messageId
3288
+ });
3289
+ }
3290
+ }
2209
3291
  async persistSession(conversationId, sessionId) {
2210
3292
  const res = await this.fetchImpl(
2211
3293
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}`,
@@ -2476,6 +3558,25 @@ async function resolveAgentIdFromKey(authHeader) {
2476
3558
  return { error: `Failed to resolve agent from key: ${message}` };
2477
3559
  }
2478
3560
  }
3561
+ async function notifyAgentDisconnected(agentId, authHeader) {
3562
+ const apiUrl = getApiUrlConfig();
3563
+ try {
3564
+ const response = await fetch(`${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
+ }
2479
3580
  async function getAgentInfo(agentId, authHeader) {
2480
3581
  const apiUrl = getApiUrlConfig();
2481
3582
  try {
@@ -2521,7 +3622,9 @@ async function getAgentInfo(agentId, authHeader) {
2521
3622
  // src/commands/run.ts
2522
3623
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
2523
3624
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
2524
- 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) {
2525
3628
  if (state.json) {
2526
3629
  console.log(
2527
3630
  JSON.stringify({
@@ -2546,9 +3649,9 @@ function logActivity(state, entry) {
2546
3649
  }
2547
3650
  if (!state.interactive) {
2548
3651
  if (entry.type === "error") {
2549
- log(state, entry.error ?? "Unknown error", true);
3652
+ log2(state, entry.error ?? "Unknown error", true);
2550
3653
  } else if (entry.type === "info" && entry.message) {
2551
- log(state, entry.message);
3654
+ log2(state, entry.message);
2552
3655
  }
2553
3656
  }
2554
3657
  }
@@ -2634,6 +3737,7 @@ async function handleAuthError(state, error2) {
2634
3737
  }
2635
3738
  async function driveChannels(state, driver) {
2636
3739
  let idlePolls = 0;
3740
+ let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
2637
3741
  while (state.running) {
2638
3742
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
2639
3743
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2643,7 +3747,9 @@ async function driveChannels(state, driver) {
2643
3747
  try {
2644
3748
  const processed = await driver.drainPending();
2645
3749
  state.messageCount += processed;
2646
- if (processed > 0 || driver.hasInFlightWatchers()) {
3750
+ const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
3751
+ lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
3752
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
2647
3753
  idlePolls = 0;
2648
3754
  if (processed > 0 && state.interactive) displayStatus(state);
2649
3755
  } else if (state.idleTimeout !== null) {
@@ -2672,7 +3778,7 @@ async function driveChannels(state, driver) {
2672
3778
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
2673
3779
  if (state.interactive) displayStatus(state);
2674
3780
  }
2675
- await new Promise((resolve) => setTimeout(resolve, CHANNEL_POLL_INTERVAL_MS));
3781
+ await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
2676
3782
  if (state.idleTimeout !== null && idlePolls >= 2) {
2677
3783
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
2678
3784
  if (idleMs > state.idleTimeout * 1e3) {
@@ -2683,8 +3789,122 @@ async function driveChannels(state, driver) {
2683
3789
  }
2684
3790
  }
2685
3791
  }
2686
- async function cleanup(state) {
3792
+ var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
3793
+ async function runSweep(state, driver, config2) {
3794
+ const mode = `age=${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 = {}) {
2687
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);
2688
3908
  if (state.connection) {
2689
3909
  state.connection.close();
2690
3910
  state.connection = null;
@@ -2695,7 +3915,7 @@ async function cleanup(state) {
2695
3915
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
2696
3916
  displayStatus(state);
2697
3917
  } else {
2698
- log(state, "Stopped OpenCode process");
3918
+ log2(state, "Stopped OpenCode process");
2699
3919
  }
2700
3920
  state.opencodeProcess = null;
2701
3921
  }
@@ -2715,26 +3935,32 @@ async function run(options) {
2715
3935
  opencodeVersion: null,
2716
3936
  opencodeProcess: null,
2717
3937
  connection: null,
3938
+ channelDriver: null,
2718
3939
  running: true,
3940
+ shuttingDown: false,
2719
3941
  activityLog: [],
2720
3942
  messageCount: 0,
3943
+ lastProxiedActivityAt: null,
3944
+ sessionCleanupTimers: [],
2721
3945
  authHeader: ""
2722
3946
  };
2723
3947
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
2724
- log(
3948
+ log2(
2725
3949
  state,
2726
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.",
2727
3951
  false
2728
3952
  );
2729
3953
  }
2730
3954
  const handleSignal = async () => {
3955
+ if (state.shuttingDown) return;
3956
+ state.shuttingDown = true;
2731
3957
  if (state.interactive) {
2732
3958
  logActivity(state, { type: "info", message: "Shutting down..." });
2733
3959
  displayStatus(state);
2734
3960
  } else {
2735
- log(state, "Shutting down...");
3961
+ log2(state, "Shutting down...");
2736
3962
  }
2737
- await cleanup(state);
3963
+ await cleanup(state, { graceful: true });
2738
3964
  await shutdownTelemetry();
2739
3965
  process.exit(0);
2740
3966
  };
@@ -2765,7 +3991,7 @@ async function run(options) {
2765
3991
  const resolved = await resolveAgentIdFromKey(state.authHeader);
2766
3992
  if (resolved.agent_id) {
2767
3993
  state.agentId = resolved.agent_id;
2768
- log(state, `Resolved agent ID from key: ${state.agentId}`);
3994
+ log2(state, `Resolved agent ID from key: ${state.agentId}`);
2769
3995
  if (state.interactive && !state.json) {
2770
3996
  logActivity(state, {
2771
3997
  type: "info",
@@ -2828,17 +4054,17 @@ async function run(options) {
2828
4054
  port: state.port,
2829
4055
  interactive: state.interactive,
2830
4056
  agentId: state.agentId,
2831
- log: (message) => log(state, message)
4057
+ log: (message) => log2(state, message)
2832
4058
  });
2833
4059
  state.port = oc.port;
2834
4060
  state.opencodeProcess = oc.process;
2835
4061
  state.opencodeVersion = oc.version;
2836
4062
  state.opencodeConnected = oc.process !== null || oc.version !== null;
2837
- const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
2838
- ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
4063
+ const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
4064
+ ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
2839
4065
  const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
2840
4066
  if (versionWarning) {
2841
- log(state, versionWarning, false);
4067
+ log2(state, versionWarning, false);
2842
4068
  if (state.interactive && !state.json) {
2843
4069
  logActivity(state, { type: "info", message: versionWarning });
2844
4070
  }
@@ -2854,12 +4080,14 @@ async function run(options) {
2854
4080
  apiUrl: getApiUrlConfig(),
2855
4081
  getAuthHeader: () => state.authHeader,
2856
4082
  conversationFilter: state.conversationFilter,
4083
+ stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
2857
4084
  log: (entry) => logActivity(state, {
2858
4085
  type: entry.level === "error" ? "error" : "info",
2859
4086
  message: entry.message,
2860
4087
  error: entry.level === "error" ? entry.message : void 0
2861
4088
  })
2862
4089
  });
4090
+ state.channelDriver = channelDriver;
2863
4091
  const connection = new RunnerConnection({
2864
4092
  agentId: state.agentId,
2865
4093
  getAuthHeader: () => state.authHeader,
@@ -2873,7 +4101,11 @@ async function run(options) {
2873
4101
  type: "info",
2874
4102
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
2875
4103
  });
2876
- emitAgentConnected(state.agentId, { port: state.port });
4104
+ emitAgentConnected(state.agentId, {
4105
+ port: state.port,
4106
+ cli_version: getCliVersion(),
4107
+ opencode_version: state.opencodeVersion
4108
+ });
2877
4109
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
2878
4110
  if (state.interactive) displayStatus(state);
2879
4111
  channelDriver.drainPending().then((processed) => {
@@ -2907,9 +4139,14 @@ async function run(options) {
2907
4139
  logActivity(state, { type: "error", error: error2 });
2908
4140
  if (state.interactive) displayStatus(state);
2909
4141
  },
2910
- // 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.
2911
4147
  onResponse: () => {
2912
4148
  state.opencodeConnected = true;
4149
+ state.lastProxiedActivityAt = Date.now();
2913
4150
  },
2914
4151
  // A channel message was queued and the api-worker pinged us over the
2915
4152
  // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2947,10 +4184,12 @@ async function run(options) {
2947
4184
  if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
2948
4185
  throw error2;
2949
4186
  }
4187
+ scheduleSessionCleanup(state, channelDriver, options);
2950
4188
  if (!interactive || state.json) {
2951
- log(state, "Driving channel messages...");
4189
+ log2(state, "Driving channel messages...");
2952
4190
  }
2953
4191
  await driveChannels(state, channelDriver);
4192
+ if (state.shuttingDown) return;
2954
4193
  await cleanup(state);
2955
4194
  if (state.json) {
2956
4195
  console.log(
@@ -2960,11 +4199,12 @@ async function run(options) {
2960
4199
  })
2961
4200
  );
2962
4201
  } else if (!interactive) {
2963
- log(state, `Completed. Processed ${state.messageCount} message(s).`);
4202
+ log2(state, `Completed. Processed ${state.messageCount} message(s).`);
2964
4203
  }
2965
4204
  await shutdownTelemetry();
2966
4205
  process.exit(0);
2967
4206
  } catch (error2) {
4207
+ if (state.shuttingDown) return;
2968
4208
  await cleanup(state);
2969
4209
  const message = error2 instanceof Error ? error2.message : String(error2);
2970
4210
  if (state.json) {
@@ -2982,8 +4222,9 @@ async function run(options) {
2982
4222
  }
2983
4223
 
2984
4224
  // src/index.ts
4225
+ var { version } = createRequire(import.meta.url)("../package.json");
2985
4226
  var program = new Command();
2986
- 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(
2987
4228
  "--endpoint <url>",
2988
4229
  "Evident API base URL (default: production; e.g. http://localhost:3001)"
2989
4230
  ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {
@@ -2998,7 +4239,16 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
2998
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);
2999
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 }));
3000
4241
  program.command("whoami").description("Show the currently logged in user").action(whoami);
3001
- program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Agent ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option("-v, --verbose", "Show detailed request/response information").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").action(
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(
3002
4252
  (options) => {
3003
4253
  run({
3004
4254
  agent: options.agent,
@@ -3006,7 +4256,11 @@ program.command("run").description("Connect to Evident and process messages").op
3006
4256
  verbose: options.verbose,
3007
4257
  conversation: options.conversation,
3008
4258
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
3009
- 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
3010
4264
  });
3011
4265
  }
3012
4266
  );