@evident-ai/cli 3.0.1-dev.9cb8b53 → 3.0.1-dev.a748d55

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
@@ -285,14 +285,14 @@ function blank() {
285
285
  console.log();
286
286
  }
287
287
  function waitForEnter(prompt = "Press Enter to continue...") {
288
- return new Promise((resolve2) => {
288
+ return new Promise((resolve) => {
289
289
  process.stdout.write(chalk.dim(prompt));
290
290
  const handler = () => {
291
291
  process.stdin.removeListener("data", handler);
292
292
  process.stdin.setRawMode?.(false);
293
293
  process.stdin.pause();
294
294
  console.log();
295
- resolve2();
295
+ resolve();
296
296
  };
297
297
  if (process.stdin.isTTY) {
298
298
  process.stdin.setRawMode?.(true);
@@ -302,7 +302,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
302
302
  });
303
303
  }
304
304
  function sleep(ms) {
305
- return new Promise((resolve2) => setTimeout(resolve2, ms));
305
+ return new Promise((resolve) => setTimeout(resolve, ms));
306
306
  }
307
307
 
308
308
  // src/commands/login.ts
@@ -376,19 +376,19 @@ async function tokenLogin() {
376
376
  console.log("Visit your Evident dashboard to generate a CLI token.");
377
377
  blank();
378
378
  process.stdout.write("Paste token: ");
379
- const token = await new Promise((resolve2) => {
379
+ const token = await new Promise((resolve) => {
380
380
  let data = "";
381
381
  process.stdin.setEncoding("utf8");
382
382
  process.stdin.on("data", (chunk) => {
383
383
  data += chunk;
384
384
  });
385
385
  process.stdin.on("end", () => {
386
- resolve2(data.trim());
386
+ resolve(data.trim());
387
387
  });
388
388
  if (process.stdin.isTTY) {
389
389
  process.stdin.once("data", (chunk) => {
390
390
  process.stdin.pause();
391
- resolve2(chunk.toString().trim());
391
+ resolve(chunk.toString().trim());
392
392
  });
393
393
  process.stdin.resume();
394
394
  }
@@ -471,6 +471,12 @@ import chalk6 from "chalk";
471
471
  import ora3 from "ora";
472
472
  import { select as select3 } from "@inquirer/prompts";
473
473
 
474
+ // ../../packages/types/src/opencode/index.ts
475
+ function opencodeMessageIdFor(queuedMessageId) {
476
+ const sanitized = queuedMessageId.replace(/[^a-zA-Z0-9]/g, "_");
477
+ return `msg_${sanitized}`;
478
+ }
479
+
474
480
  // ../../packages/types/src/telemetry/index.ts
475
481
  var TelemetryEventTypes = {
476
482
  // Agent activity events (shown in web UI activity log)
@@ -509,10 +515,7 @@ function stripQuery(url) {
509
515
  }
510
516
 
511
517
  // src/lib/telemetry.ts
512
- var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
513
- function getCliVersion() {
514
- return CLI_VERSION;
515
- }
518
+ var CLI_VERSION = process.env.npm_package_version || "unknown";
516
519
  var eventBuffer = [];
517
520
  var flushTimeout = null;
518
521
  var isShuttingDown = false;
@@ -706,13 +709,13 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
706
709
  if (health.healthy) {
707
710
  return health;
708
711
  }
709
- await new Promise((resolve2) => setTimeout(resolve2, 1e3));
712
+ await new Promise((resolve) => setTimeout(resolve, 1e3));
710
713
  }
711
714
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
712
715
  }
713
716
 
714
717
  // src/lib/opencode/opencode-version-gate.ts
715
- var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
718
+ var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
716
719
  function isQueueValidatedVersion(version2) {
717
720
  if (!version2) return false;
718
721
  return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
@@ -1068,94 +1071,6 @@ function isAssistantInFlight(m) {
1068
1071
  if (completedOf(m) == null) return true;
1069
1072
  return finishOf(m) === "tool-calls";
1070
1073
  }
1071
- async function getSessionMessages(port, sessionId) {
1072
- try {
1073
- const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
1074
- if (!res.ok) return null;
1075
- const body = await res.json();
1076
- return Array.isArray(body) ? body : null;
1077
- } catch {
1078
- return null;
1079
- }
1080
- }
1081
- function isSessionActivelyGenerating(messages) {
1082
- if (!messages || messages.length === 0) return false;
1083
- const last = messages[messages.length - 1];
1084
- if (roleOf(last) !== "assistant") return false;
1085
- return completedOf(last) == null;
1086
- }
1087
- function sessionLastActivityMs(session) {
1088
- const candidates = [
1089
- session.time?.updated,
1090
- session.time?.created,
1091
- session.time_updated,
1092
- session.time_created,
1093
- session.updated,
1094
- session.created
1095
- ];
1096
- for (const c of candidates) {
1097
- if (typeof c === "number" && Number.isFinite(c)) return c;
1098
- }
1099
- return null;
1100
- }
1101
- async function listSessions(port) {
1102
- try {
1103
- const res = await fetch(`${opencodeBase(port)}/session`);
1104
- if (!res.ok) return null;
1105
- const body = await res.json();
1106
- return Array.isArray(body) ? body : null;
1107
- } catch {
1108
- return null;
1109
- }
1110
- }
1111
- async function deleteSession(port, id) {
1112
- try {
1113
- const res = await fetch(`${opencodeBase(port)}/session/${id}`, { method: "DELETE" });
1114
- return res.status >= 200 && res.status < 300;
1115
- } catch {
1116
- return false;
1117
- }
1118
- }
1119
- async function sessionExists(port, id) {
1120
- try {
1121
- const res = await fetch(`${opencodeBase(port)}/session/${id}`);
1122
- if (res.status >= 200 && res.status < 300) return true;
1123
- if (res.status === 404) return false;
1124
- return null;
1125
- } catch {
1126
- return null;
1127
- }
1128
- }
1129
- async function getSessionStatuses(port) {
1130
- try {
1131
- const res = await fetch(`${opencodeBase(port)}/session/status`);
1132
- if (!res.ok) {
1133
- console.error(
1134
- `[getSessionStatuses] GET /session/status returned HTTP ${res.status} (port ${port})`
1135
- );
1136
- return null;
1137
- }
1138
- const body = await res.json();
1139
- if (body == null || typeof body !== "object" || Array.isArray(body)) {
1140
- console.error(
1141
- `[getSessionStatuses] GET /session/status body was not a plain object (port ${port})`
1142
- );
1143
- return null;
1144
- }
1145
- return body;
1146
- } catch (err) {
1147
- console.error(
1148
- `[getSessionStatuses] GET /session/status failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
1149
- );
1150
- return null;
1151
- }
1152
- }
1153
- async function isSessionOngoing(port, id) {
1154
- const map = await getSessionStatuses(port);
1155
- if (map == null) return null;
1156
- const entry = map[id];
1157
- return entry != null && entry.type !== "idle";
1158
- }
1159
1074
  async function createOpenCodeSession(port, directory) {
1160
1075
  const url = new URL(`${opencodeBase(port)}/session`);
1161
1076
  if (directory && directory.trim()) {
@@ -1173,16 +1088,9 @@ async function createOpenCodeSession(port, directory) {
1173
1088
  const data = await response.json();
1174
1089
  return data.id;
1175
1090
  }
1176
- function messageText(m) {
1177
- if (!m || !Array.isArray(m.parts)) return "";
1178
- return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
1179
- }
1180
- async function sendPromptAsync(port, sessionId, content, options) {
1181
- const before = await getSessionMessages(port, sessionId);
1182
- const knownUserIds = new Set(
1183
- (before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
1184
- );
1091
+ async function sendPromptAsync(port, sessionId, content, options, messageId) {
1185
1092
  const body = {
1093
+ messageID: messageId,
1186
1094
  parts: [{ type: "text", text: content }]
1187
1095
  };
1188
1096
  if (options?.agent) {
@@ -1206,29 +1114,6 @@ async function sendPromptAsync(port, sessionId, content, options) {
1206
1114
  const text = await res.text().catch(() => "");
1207
1115
  throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1208
1116
  }
1209
- const READ_BACK_ATTEMPTS = 5;
1210
- const READ_BACK_DELAY_MS = 150;
1211
- for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
1212
- const after = await getSessionMessages(port, sessionId);
1213
- if (after) {
1214
- let best = null;
1215
- for (const m of after) {
1216
- if (roleOf(m) !== "user") continue;
1217
- const id = idOf(m);
1218
- if (typeof id !== "string" || knownUserIds.has(id)) continue;
1219
- if (messageText(m) !== content) continue;
1220
- const created = createdOf(m) ?? 0;
1221
- if (best === null || created > best.created) {
1222
- best = { id, created };
1223
- }
1224
- }
1225
- if (best) return best.id;
1226
- }
1227
- if (attempt < READ_BACK_ATTEMPTS - 1) {
1228
- await new Promise((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
1229
- }
1230
- }
1231
- return null;
1232
1117
  }
1233
1118
  function findAssistantReplyAfter(messages, userMessageId) {
1234
1119
  if (!messages || messages.length === 0) return null;
@@ -1282,11 +1167,6 @@ function messageRunState(messages, userMessageId) {
1282
1167
  if (isAssistantInFlight(reply)) return "running";
1283
1168
  return errorOf(reply) != null ? "failed" : "done";
1284
1169
  }
1285
- function isPreamblePinnedRunning(messages, userMessageId) {
1286
- if (messageRunState(messages, userMessageId) !== "running") return false;
1287
- const reply = findLastAssistantReplyFor(messages, userMessageId);
1288
- return completedOf(reply) != null && finishOf(reply) === "tool-calls";
1289
- }
1290
1170
  function messageError(messages, userMessageId) {
1291
1171
  const reply = findLastAssistantReplyFor(messages, userMessageId);
1292
1172
  const error2 = errorOf(reply);
@@ -1306,109 +1186,30 @@ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1306
1186
  (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1307
1187
  );
1308
1188
  }
1309
-
1310
- // src/lib/opencode/session-cleanup.ts
1311
- var DURATION_UNIT_MS = {
1312
- s: 1e3,
1313
- m: 60 * 1e3,
1314
- h: 60 * 60 * 1e3,
1315
- d: 24 * 60 * 60 * 1e3
1316
- };
1317
- function parseDurationMs(input) {
1318
- const trimmed = input.trim();
1319
- const match = /^(\d+)([smhd])$/.exec(trimmed);
1320
- if (!match) {
1321
- throw new Error(
1322
- `Invalid duration "${input}": expected <number><unit> where unit is one of s, m, h, d (e.g. "7d", "24h", "30m", "90s").`
1323
- );
1324
- }
1325
- const value = Number(match[1]);
1326
- if (value <= 0) {
1327
- throw new Error(`Invalid duration "${input}": must be a positive value.`);
1328
- }
1329
- return value * DURATION_UNIT_MS[match[2]];
1330
- }
1331
- function selectSessionsToDelete(sessions, opts) {
1332
- const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
1333
- if (maxAgeMs === void 0 && maxCount === void 0) return [];
1334
- const ageEligible = (s) => {
1335
- if (maxAgeMs === void 0) return false;
1336
- if (s.lastActivityMs === null) return true;
1337
- return nowMs - s.lastActivityMs > maxAgeMs;
1338
- };
1339
- const countEligibleIds = /* @__PURE__ */ new Set();
1340
- if (maxCount !== void 0) {
1341
- const byActivityDesc = [...sessions].sort(
1342
- (a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
1343
- );
1344
- for (const s of byActivityDesc.slice(maxCount)) {
1345
- countEligibleIds.add(s.id);
1189
+ function isSessionSettled(messages, now, settleMs) {
1190
+ if (!messages || messages.length === 0) return true;
1191
+ let newestAssistantCreated = null;
1192
+ for (const m of messages) {
1193
+ if (roleOf(m) !== "assistant") continue;
1194
+ if (isAssistantInFlight(m)) return false;
1195
+ const created = createdOf(m);
1196
+ if (typeof created === "number" && (newestAssistantCreated === null || created > newestAssistantCreated)) {
1197
+ newestAssistantCreated = created;
1346
1198
  }
1347
1199
  }
1348
- const toDelete = [];
1349
- for (const s of sessions) {
1350
- if (protectedIds.has(s.id)) continue;
1351
- if (ageEligible(s) || countEligibleIds.has(s.id)) {
1352
- toDelete.push(s.id);
1353
- }
1354
- }
1355
- return toDelete;
1200
+ if (newestAssistantCreated === null) return true;
1201
+ return now - newestAssistantCreated >= settleMs;
1356
1202
  }
1357
- var DEFAULT_INTERVAL = "1h";
1358
- function resolve(flag, envValue, fallback) {
1359
- return flag ?? envValue ?? fallback;
1203
+ function opencodeMessageIdFor2(queuedMessageId) {
1204
+ return opencodeMessageIdFor(queuedMessageId);
1360
1205
  }
1361
- function parseMaxCount(input) {
1362
- const trimmed = input.trim();
1363
- if (!/^\d+$/.test(trimmed)) {
1364
- throw new Error(`Invalid max-count "${input}": expected a positive integer.`);
1206
+ var NATIVE_ID_ALPHABET = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
1207
+ function nativeOpencodeMessageId() {
1208
+ let id = "msg_";
1209
+ for (let i = 0; i < 24; i++) {
1210
+ id += NATIVE_ID_ALPHABET[Math.floor(Math.random() * NATIVE_ID_ALPHABET.length)];
1365
1211
  }
1366
- const value = Number(trimmed);
1367
- if (value <= 0) {
1368
- throw new Error(`Invalid max-count "${input}": must be greater than 0.`);
1369
- }
1370
- return value;
1371
- }
1372
- function resolveSessionCleanupConfig(flags, env = process.env) {
1373
- const warnings = [];
1374
- const maxAgeRaw = resolve(flags.maxAge, env.EVIDENT_SESSION_CLEANUP_MAX_AGE);
1375
- const maxCountRaw = resolve(flags.maxCount, env.EVIDENT_SESSION_CLEANUP_MAX_COUNT);
1376
- const intervalRaw = resolve(
1377
- flags.interval,
1378
- env.EVIDENT_SESSION_CLEANUP_INTERVAL,
1379
- DEFAULT_INTERVAL
1380
- );
1381
- let maxAgeMs;
1382
- if (maxAgeRaw !== void 0) {
1383
- try {
1384
- maxAgeMs = parseDurationMs(maxAgeRaw);
1385
- } catch (err) {
1386
- warnings.push(
1387
- `Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
1388
- );
1389
- }
1390
- }
1391
- let maxCount;
1392
- if (maxCountRaw !== void 0) {
1393
- try {
1394
- maxCount = parseMaxCount(maxCountRaw);
1395
- } catch (err) {
1396
- warnings.push(
1397
- `Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
1398
- );
1399
- }
1400
- }
1401
- let intervalMs;
1402
- try {
1403
- intervalMs = parseDurationMs(intervalRaw ?? DEFAULT_INTERVAL);
1404
- } catch (err) {
1405
- warnings.push(
1406
- `Ignoring invalid --session-cleanup-interval, using default ${DEFAULT_INTERVAL}: ${err instanceof Error ? err.message : String(err)}`
1407
- );
1408
- intervalMs = parseDurationMs(DEFAULT_INTERVAL);
1409
- }
1410
- const enabled = maxAgeMs !== void 0 || maxCount !== void 0;
1411
- return { enabled, maxAgeMs, maxCount, intervalMs, warnings };
1212
+ return id;
1412
1213
  }
1413
1214
 
1414
1215
  // src/lib/tunnel/connection.ts
@@ -1487,26 +1288,24 @@ var StreamForwarder = class {
1487
1288
  this.send({ type: "res_end", sid });
1488
1289
  return;
1489
1290
  }
1490
- if (process.env.DEBUG) {
1491
- log("debug", "agent_request", {
1492
- correlation_id: correlationId,
1493
- sid,
1494
- method,
1495
- path: stripQuery(path)
1496
- });
1497
- }
1291
+ log("info", "agent_request", {
1292
+ correlation_id: correlationId,
1293
+ sid,
1294
+ method,
1295
+ path: stripQuery(path)
1296
+ });
1498
1297
  const ac = new AbortController();
1499
1298
  let bodyPromise;
1500
1299
  let pushBody;
1501
1300
  let endBody;
1502
1301
  if (has_body) {
1503
1302
  const chunks = [];
1504
- bodyPromise = new Promise((resolve2) => {
1303
+ bodyPromise = new Promise((resolve) => {
1505
1304
  pushBody = (buf) => {
1506
1305
  chunks.push(buf);
1507
1306
  };
1508
1307
  endBody = () => {
1509
- resolve2(Buffer.concat(chunks));
1308
+ resolve(Buffer.concat(chunks));
1510
1309
  };
1511
1310
  });
1512
1311
  }
@@ -1541,14 +1340,12 @@ var StreamForwarder = class {
1541
1340
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1542
1341
  });
1543
1342
  this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
1544
- if (process.env.DEBUG) {
1545
- log("debug", "agent_response", {
1546
- correlation_id: correlationId,
1547
- sid,
1548
- status: upstream.status,
1549
- duration_ms: Date.now() - startedAt
1550
- });
1551
- }
1343
+ log("info", "agent_response", {
1344
+ correlation_id: correlationId,
1345
+ sid,
1346
+ status: upstream.status,
1347
+ duration_ms: Date.now() - startedAt
1348
+ });
1552
1349
  this.callbacks.onHead?.(sid, upstream.status);
1553
1350
  try {
1554
1351
  if (upstream.body) {
@@ -1624,7 +1421,7 @@ function connectTunnel(options) {
1624
1421
  } = options;
1625
1422
  const tunnelUrl = getTunnelUrlConfig();
1626
1423
  const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
1627
- return new Promise((resolve2, reject) => {
1424
+ return new Promise((resolve, reject) => {
1628
1425
  const ws = new WebSocket2(url, {
1629
1426
  headers: {
1630
1427
  Authorization: authHeader
@@ -1689,7 +1486,7 @@ function connectTunnel(options) {
1689
1486
  clearTimeout(connectionTimeout);
1690
1487
  const connectedAgentId = message.agent_id ?? agentId;
1691
1488
  onConnected?.(connectedAgentId);
1692
- resolve2({
1489
+ resolve({
1693
1490
  ws,
1694
1491
  close: () => ws.close(1e3, "CLI shutdown")
1695
1492
  });
@@ -1818,10 +1615,10 @@ var DEFAULT_RETRY_POLICY = {
1818
1615
  };
1819
1616
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1820
1617
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1618
+ var DEFAULT_DISPATCH_CONFIRM_MS = 6e3;
1821
1619
  var DEFAULT_STUCK_QUEUED_MS = 6e4;
1822
- var HEARTBEAT_MS = 6e4;
1823
- var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
1824
- var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
1620
+ var DEFAULT_STUCK_QUEUED_REDRIVE_MAX = 3;
1621
+ var DEFAULT_SETTLE_MS = 3500;
1825
1622
  var ChannelAuthError = class extends Error {
1826
1623
  constructor(message) {
1827
1624
  super(message);
@@ -1856,18 +1653,13 @@ var ChannelDriver = class {
1856
1653
  sleep;
1857
1654
  pausedPollIntervalMs;
1858
1655
  pausedMaxWaitMs;
1656
+ dispatchConfirmMs;
1859
1657
  stuckQueuedMs;
1658
+ stuckQueuedRedriveMax;
1659
+ settleMs;
1860
1660
  now;
1861
1661
  /** Cache of conversationId → opencode sessionId. */
1862
1662
  sessions = /* @__PURE__ */ new Map();
1863
- /**
1864
- * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
1865
- * longer idempotent (no caller-supplied `messageID`), and its read-back picks
1866
- * "the one new user row" — which is only unambiguous if no OTHER dispatch into
1867
- * the SAME session interleaves its snapshot→POST→read-back. This map chains each
1868
- * session's dispatches so they run serially; distinct sessions stay concurrent.
1869
- */
1870
- sessionDispatchLocks = /* @__PURE__ */ new Map();
1871
1663
  /**
1872
1664
  * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1873
1665
  * session: one polling loop services all of that session's in-flight messages.
@@ -1918,29 +1710,6 @@ var ChannelDriver = class {
1918
1710
  * the row leaves the processing list, exactly like `dontRedispatch`.
1919
1711
  */
1920
1712
  doneUndeliverable = /* @__PURE__ */ new Set();
1921
- /**
1922
- * "Already emitted `readopt_poll_unresolved` for this row" (#229). The b1 /
1923
- * unreadable-status re-evaluate leaf leaves the row UN-tracked so it is re-read
1924
- * every ~2s drain until the status map becomes readable — but the server-visible
1925
- * signal is an OUTCOME, so it must fire at most ONCE per row, not once per drain
1926
- * (Bugbot "Re-adopt signals flood every drain"). Cleared when the row leaves the
1927
- * processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
1928
- */
1929
- readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
1930
- /**
1931
- * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
1932
- * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
1933
- * is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
1934
- * persist hasn't landed before tick N+1 re-reads the still-null
1935
- * `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
1936
- * A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
1937
- * short-circuits while it is present, so a null-id row is re-dispatched AT MOST
1938
- * ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
1939
- * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
1940
- * re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
1941
- * so the NEXT tick may retry exactly once more).
1942
- */
1943
- awaitingReadopt = /* @__PURE__ */ new Set();
1944
1713
  /**
1945
1714
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1946
1715
  * first session creation so drain-created sessions are rooted at the project
@@ -1948,33 +1717,8 @@ var ChannelDriver = class {
1948
1717
  * not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
1949
1718
  */
1950
1719
  opencodeDirectory = void 0;
1951
- /**
1952
- * Cache of opencode `sessionId → parentID` (its parent session, or `null` when
1953
- * the session is a root with no parent). Sub-agents spawned via the `task` tool
1954
- * run in CHILD sessions whose `parentID` chains up to the Evident-created
1955
- * (watched) session; we resolve this once per session so a child-session
1956
- * question/permission can be attributed to the watched session's subtree
1957
- * (`sessionBelongsTo`) instead of being dropped by an exact-id filter. A missing
1958
- * entry = not yet resolved; `null` = resolved root (stop walking).
1959
- */
1960
- sessionParents = /* @__PURE__ */ new Map();
1961
1720
  /** Serialises drains so a reconnect during a drain doesn't double-process. */
1962
1721
  draining = false;
1963
- /**
1964
- * The currently-executing `drainPending()` promise, or null when idle. Lets a
1965
- * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
1966
- * is about to dispatch is not missed by the `hasInFlightWatchers()` check (a
1967
- * drain that entered before `stop()` still registers its watcher).
1968
- */
1969
- activeDrain = null;
1970
- /**
1971
- * Set by `stop()` on graceful shutdown. Once stopped, `drainPending` no longer
1972
- * dispatches NEW work (it returns 0 immediately) — but the per-session watcher
1973
- * loops already running keep going so in-flight turns can finish and deliver
1974
- * their reply. `run.ts` awaits `waitForInFlight()` before it closes the tunnel
1975
- * and stops opencode.
1976
- */
1977
- stopped = false;
1978
1722
  constructor(config2) {
1979
1723
  this.agentId = config2.agentId;
1980
1724
  this.port = config2.port;
@@ -1988,7 +1732,10 @@ var ChannelDriver = class {
1988
1732
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1989
1733
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1990
1734
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
1735
+ this.dispatchConfirmMs = config2.dispatchConfirmMs ?? DEFAULT_DISPATCH_CONFIRM_MS;
1991
1736
  this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
1737
+ this.stuckQueuedRedriveMax = config2.stuckQueuedRedriveMax ?? DEFAULT_STUCK_QUEUED_REDRIVE_MAX;
1738
+ this.settleMs = config2.settleMs ?? DEFAULT_SETTLE_MS;
1992
1739
  this.now = config2.now ?? (() => Date.now());
1993
1740
  }
1994
1741
  /** The IPv4-loopback base URL for the local `opencode serve`. */
@@ -2006,21 +1753,8 @@ var ChannelDriver = class {
2006
1753
  * @returns the number of messages NEWLY dispatched to opencode's native queue.
2007
1754
  */
2008
1755
  async drainPending() {
2009
- if (this.stopped) return 0;
2010
1756
  if (this.draining) return 0;
2011
1757
  this.draining = true;
2012
- const run2 = this.runDrain();
2013
- this.activeDrain = run2.then(
2014
- () => {
2015
- this.activeDrain = null;
2016
- },
2017
- () => {
2018
- this.activeDrain = null;
2019
- }
2020
- );
2021
- return run2;
2022
- }
2023
- async runDrain() {
2024
1758
  let dispatched = 0;
2025
1759
  try {
2026
1760
  const conversations = await this.getPendingConversations();
@@ -2032,7 +1766,6 @@ var ChannelDriver = class {
2032
1766
  });
2033
1767
  }
2034
1768
  for (const conv of conversations) {
2035
- if (this.stopped) break;
2036
1769
  dispatched += await this.processConversation(conv);
2037
1770
  }
2038
1771
  await this.readoptProcessing();
@@ -2053,73 +1786,6 @@ var ChannelDriver = class {
2053
1786
  }
2054
1787
  return false;
2055
1788
  }
2056
- /**
2057
- * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
2058
- * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
2059
- * `watchers` entry whose `inFlight` set is non-empty — the same predicate
2060
- * `hasInFlightWatchers()` uses, lifted to return the ids.
2061
- *
2062
- * Deliberately does NOT include `this.sessions` (the permanent, never-pruned
2063
- * conversation→session cache). Protecting every bound-but-idle session there
2064
- * would shield nearly every session and defeat cleanup — AND it is unnecessary:
2065
- * `ensureSession` is self-healing (it recreates a session whose id no longer
2066
- * exists), so deleting an idle bound session is harmless — the conversation's
2067
- * next turn transparently rebinds a fresh one. The only thing worth protecting
2068
- * is a session with a turn ACTIVELY in flight right now: tearing that down
2069
- * mid-turn would strand the running `prompt_async`. Idle sessions are fair game.
2070
- */
2071
- protectedSessionIds() {
2072
- const ids = /* @__PURE__ */ new Set();
2073
- for (const [sessionId, watcher] of this.watchers) {
2074
- if (watcher.inFlight.size > 0) ids.add(sessionId);
2075
- }
2076
- return ids;
2077
- }
2078
- /**
2079
- * Begin a graceful stop: stop accepting NEW channel work. Idempotent. After
2080
- * this, `drainPending()` is a no-op (returns 0), so no new message is dispatched
2081
- * — but the watcher loops already tracking in-flight turns keep running, so a
2082
- * turn that has finished (or is about to) still fires `markDone` and delivers
2083
- * its reply. Pair with `waitForInFlight()` to bound how long shutdown waits.
2084
- */
2085
- stop() {
2086
- this.stopped = true;
2087
- }
2088
- /**
2089
- * Wait (up to `timeoutMs`) for in-flight watcher work to settle during a
2090
- * graceful shutdown, so a turn whose reply is ready — or completes within the
2091
- * window — is delivered before the process exits, instead of being cut off and
2092
- * left for the ADR-0046 restart-recovery path.
2093
- *
2094
- * Bounded on purpose: the watcher's own give-up deadline is up to 10 minutes,
2095
- * far longer than a shutdown grace period (e.g. Fargate's SIGTERM→SIGKILL
2096
- * window). We poll `hasInFlightWatchers()` and return as soon as the in-flight
2097
- * set empties OR the timeout elapses. Anything still in flight at the timeout is
2098
- * safe to abandon — it stays `processing` server-side and is re-adopted on the
2099
- * next runner start (ADR-0046).
2100
- *
2101
- * @returns true if all in-flight work settled within the window; false if the
2102
- * timeout elapsed with work still in flight.
2103
- */
2104
- async waitForInFlight(timeoutMs) {
2105
- const deadline = this.now() + timeoutMs;
2106
- const step = Math.min(this.pausedPollIntervalMs, 250);
2107
- if (this.activeDrain) {
2108
- let drainSettled = false;
2109
- void this.activeDrain.then(() => {
2110
- drainSettled = true;
2111
- });
2112
- while (!drainSettled) {
2113
- if (this.now() >= deadline) return false;
2114
- await this.sleep(step);
2115
- }
2116
- }
2117
- while (this.hasInFlightWatchers()) {
2118
- if (this.now() >= deadline) return false;
2119
- await this.sleep(step);
2120
- }
2121
- return true;
2122
- }
2123
1789
  /**
2124
1790
  * Await all outstanding per-session watchers (WI-3).
2125
1791
  *
@@ -2156,16 +1822,15 @@ var ChannelDriver = class {
2156
1822
  let dispatched = 0;
2157
1823
  let skippedAlreadyDispatched = 0;
2158
1824
  for (const message of messages) {
2159
- if (this.stopped) break;
2160
1825
  if (this.dispatched.has(message.id)) {
2161
1826
  skippedAlreadyDispatched += 1;
2162
1827
  continue;
2163
1828
  }
1829
+ const opencodeMessageId = opencodeMessageIdFor2(message.id);
2164
1830
  const options = {
2165
1831
  agent: message.opencode_agent ?? void 0,
2166
1832
  model: message.opencode_model ?? void 0
2167
1833
  };
2168
- let opencodeMessageId;
2169
1834
  try {
2170
1835
  this.log({
2171
1836
  level: "info",
@@ -2173,23 +1838,10 @@ var ChannelDriver = class {
2173
1838
  conversation_id: conv.id,
2174
1839
  message_id: message.id
2175
1840
  });
2176
- opencodeMessageId = await this.dispatchLocked(
2177
- sessionId,
2178
- () => sendPromptAsync(this.port, sessionId, message.content, options)
2179
- );
1841
+ await sendPromptAsync(this.port, sessionId, message.content, options, opencodeMessageId);
2180
1842
  } catch (err) {
2181
1843
  if (err instanceof ChannelAuthError) throw err;
2182
1844
  this.dispatched.delete(message.id);
2183
- if (await sessionExists(this.port, sessionId) === false) {
2184
- this.sessions.delete(conv.id);
2185
- this.log({
2186
- level: "info",
2187
- 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.`,
2188
- conversation_id: conv.id,
2189
- message_id: message.id
2190
- });
2191
- break;
2192
- }
2193
1845
  await this.markFailed(conv.id, message.id).catch(() => {
2194
1846
  });
2195
1847
  this.log({
@@ -2200,15 +1852,6 @@ var ChannelDriver = class {
2200
1852
  });
2201
1853
  continue;
2202
1854
  }
2203
- if (opencodeMessageId === null) {
2204
- this.log({
2205
- level: "error",
2206
- 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`,
2207
- conversation_id: conv.id,
2208
- message_id: message.id
2209
- });
2210
- continue;
2211
- }
2212
1855
  this.dispatched.add(message.id);
2213
1856
  this.registerInFlight(conv, sessionId, message, opencodeMessageId);
2214
1857
  dispatched += 1;
@@ -2225,33 +1868,16 @@ var ChannelDriver = class {
2225
1868
  return dispatched;
2226
1869
  }
2227
1870
  async ensureSession(conv) {
2228
- const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
2229
- if (bound) {
2230
- const exists = await sessionExists(this.port, bound);
2231
- if (exists === false) {
2232
- this.log({
2233
- level: "info",
2234
- 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.`,
2235
- conversation_id: conv.id
2236
- });
2237
- this.sessions.delete(conv.id);
2238
- return this.createAndBindSession(conv.id);
2239
- }
2240
- this.sessions.set(conv.id, bound);
2241
- return bound;
1871
+ const cached = this.sessions.get(conv.id);
1872
+ if (cached) return cached;
1873
+ if (conv.opencode_session_id) {
1874
+ this.sessions.set(conv.id, conv.opencode_session_id);
1875
+ return conv.opencode_session_id;
2242
1876
  }
2243
- return this.createAndBindSession(conv.id);
2244
- }
2245
- /**
2246
- * Create a fresh OpenCode session for a conversation, cache the binding, and
2247
- * best-effort persist it server-side. Shared by the first-ever bind and the
2248
- * self-heal recreate path in `ensureSession`.
2249
- */
2250
- async createAndBindSession(conversationId) {
2251
1877
  const directory = await this.resolveOpenCodeDirectory();
2252
1878
  const sessionId = await createOpenCodeSession(this.port, directory);
2253
- this.sessions.set(conversationId, sessionId);
2254
- await this.persistSession(conversationId, sessionId).catch(() => {
1879
+ this.sessions.set(conv.id, sessionId);
1880
+ await this.persistSession(conv.id, sessionId).catch(() => {
2255
1881
  });
2256
1882
  return sessionId;
2257
1883
  }
@@ -2274,25 +1900,6 @@ var ChannelDriver = class {
2274
1900
  // -------------------------------------------------------------------------
2275
1901
  // Per-session watcher (WI-3)
2276
1902
  // -------------------------------------------------------------------------
2277
- /**
2278
- * Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
2279
- * opencode session (Task 2.1a), so two dispatches into the SAME session can
2280
- * never interleave and mis-correlate their read-backs. Distinct sessions run
2281
- * concurrently. The chained tail intentionally ignores the prior result/error
2282
- * (each dispatch reports its own outcome to its caller).
2283
- */
2284
- dispatchLocked(sessionId, fn) {
2285
- const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
2286
- const run2 = prior.then(fn, fn);
2287
- this.sessionDispatchLocks.set(
2288
- sessionId,
2289
- run2.then(
2290
- () => void 0,
2291
- () => void 0
2292
- )
2293
- );
2294
- return run2;
2295
- }
2296
1903
  /** Register a freshly-dispatched message with its session's watcher state. */
2297
1904
  registerInFlight(conv, sessionId, message, opencodeMessageId) {
2298
1905
  let watcher = this.watchers.get(sessionId);
@@ -2302,9 +1909,7 @@ var ChannelDriver = class {
2302
1909
  inFlight: /* @__PURE__ */ new Map(),
2303
1910
  loop: null,
2304
1911
  reportedQuestions: /* @__PURE__ */ new Set(),
2305
- reportedPermissions: /* @__PURE__ */ new Set(),
2306
- lastGoodPollAt: this.now(),
2307
- hadUsablePoll: false
1912
+ reportedPermissions: /* @__PURE__ */ new Set()
2308
1913
  };
2309
1914
  this.watchers.set(sessionId, watcher);
2310
1915
  }
@@ -2314,37 +1919,24 @@ var ChannelDriver = class {
2314
1919
  opencodeMessageId,
2315
1920
  message,
2316
1921
  dispatchedAt: now,
2317
- processingAnchorMs: now,
2318
1922
  deadline: now + this.pausedMaxWaitMs,
2319
1923
  started: false,
2320
1924
  done: false,
2321
1925
  stuckReported: false,
2322
- lastAliveAt: 0,
2323
- aliveInFlight: false,
2324
- awaitingHumanLatched: false,
2325
- pausedOnQuestion: false,
2326
- pausedOnPermission: false,
2327
- pausedClearConfirmed: false,
2328
- pausedInFlight: false,
2329
- deliveryDeadlineAnchored: false
1926
+ redriveAttempts: 0,
1927
+ lastRedriveAt: null,
1928
+ redriveOpencodeMessageId: null,
1929
+ attemptedOpencodeMessageIds: [opencodeMessageId]
2330
1930
  });
2331
1931
  }
2332
1932
  /**
2333
1933
  * Register a RE-ADOPTED `processing` message with its session watcher
2334
1934
  * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
2335
1935
  * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
2336
- * `now`, so the paused/queued/unreachable cases settle on the same wall-clock a
2337
- * fresh dispatch would (10 min after `processed_at`, not 10 min from now).
2338
- *
2339
- * This re-attaches into the SAME watcher, so the ADR-0047 progressing-vs-paused
2340
- * give-up (`serviceInFlightMessage`) applies unchanged: a re-adopted turn
2341
- * opencode reports ACTIVELY `running` is watched to completion (its liveness
2342
- * heartbeat keeps the cron off its row), while a re-adopted turn that is paused
2343
- * awaiting a human — or queued/unreachable — is still bounded by `deadline` and
2344
- * handed to the cron. The old "the `deadline` must settle before the ~15-min
2345
- * cron or they double-drive" reasoning is superseded: liveness now settles the
2346
- * actively-running case; `deadline` settles the rest. `dispatchedAt` stays `now`
2347
- * (only the appear-guard uses it).
1936
+ * `now`: a row already `processing` for e.g. 5 min must give up ~5 min from now
1937
+ * (10 min after `processed_at`), not 10 min from now — otherwise its deadline
1938
+ * lands ~15 min after `processed_at`, coinciding with the cron reset →
1939
+ * double-drive race. `dispatchedAt` stays `now` (only the appear-guard uses it).
2348
1940
  *
2349
1941
  * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
2350
1942
  * `opencodeMessageId` is the id the watcher polls for a reply — for the orphan
@@ -2354,7 +1946,7 @@ var ChannelDriver = class {
2354
1946
  * server already flipped to `processing`; the running/done transitions still
2355
1947
  * fire from the watcher's normal branches.
2356
1948
  */
2357
- registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
1949
+ registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs, redriveOpencodeMessageId = null) {
2358
1950
  let watcher = this.watchers.get(sessionId);
2359
1951
  if (!watcher) {
2360
1952
  watcher = {
@@ -2362,9 +1954,7 @@ var ChannelDriver = class {
2362
1954
  inFlight: /* @__PURE__ */ new Map(),
2363
1955
  loop: null,
2364
1956
  reportedQuestions: /* @__PURE__ */ new Set(),
2365
- reportedPermissions: /* @__PURE__ */ new Set(),
2366
- lastGoodPollAt: this.now(),
2367
- hadUsablePoll: false
1957
+ reportedPermissions: /* @__PURE__ */ new Set()
2368
1958
  };
2369
1959
  this.watchers.set(sessionId, watcher);
2370
1960
  }
@@ -2373,33 +1963,22 @@ var ChannelDriver = class {
2373
1963
  opencodeMessageId,
2374
1964
  message,
2375
1965
  dispatchedAt: this.now(),
2376
- // Anchor the absolute-age ceiling to the SERVER-SIDE `processed_at` (the same
2377
- // value seeding `deadline`), NOT `dispatchedAt` — so a re-adopted zombie's age
2378
- // reflects the real turn duration and the ceiling fires on the ORIGINAL turn.
2379
- processingAnchorMs: processedAtMs,
2380
1966
  deadline: processedAtMs + this.pausedMaxWaitMs,
2381
1967
  // The server row is ALREADY `processing`; do not re-fire markProcessing.
2382
1968
  started: true,
2383
1969
  done: false,
2384
1970
  // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
2385
- // AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
2386
- // on `state === 'queued'` (turn produced no reply), not on `started`, so a
2387
- // re-adopted row left wedged in `queued` still emits the signal once
2388
- // (#210/#220 observability).
1971
+ // AND the stuck-queued observer now INCLUDES re-adopted queued wedges: it
1972
+ // gates on `state === 'queued'` (turn produced no reply), not on `started`,
1973
+ // so a re-adopted row left wedged in `queued` still emits the signal once
1974
+ // (queued-followup-redrive, #210). A re-adopted `queued` wedge is ALSO
1975
+ // re-driven by the same path (ADR-0046 §c deferred the live-session redrive
1976
+ // to here) — `redriveAttempts` starts fresh so it gets the full budget.
2389
1977
  stuckReported: false,
2390
- // Task 5.2: a re-adopted actively-running row re-attaches into the SAME
2391
- // watcher and so hits the SAME actively-running heartbeat branch in
2392
- // `serviceInFlightMessage` as a fresh dispatch — monitoring observes "runner
2393
- // re-adopted and is confirming this row alive" via that `alive` heartbeat,
2394
- // with no extra `re_adopted` signal needed (folds old WI-6).
2395
- lastAliveAt: 0,
2396
- aliveInFlight: false,
2397
- awaitingHumanLatched: false,
2398
- pausedOnQuestion: false,
2399
- pausedOnPermission: false,
2400
- pausedClearConfirmed: false,
2401
- pausedInFlight: false,
2402
- deliveryDeadlineAnchored: false
1978
+ redriveAttempts: 0,
1979
+ lastRedriveAt: null,
1980
+ redriveOpencodeMessageId,
1981
+ attemptedOpencodeMessageIds: [opencodeMessageId]
2403
1982
  });
2404
1983
  }
2405
1984
  /**
@@ -2450,30 +2029,12 @@ var ChannelDriver = class {
2450
2029
  messages = Array.isArray(body) ? body : null;
2451
2030
  }
2452
2031
  } catch {
2032
+ continue;
2453
2033
  }
2454
- if (messages != null && messages.length > 0) {
2455
- watcher.lastGoodPollAt = this.now();
2456
- watcher.hadUsablePoll = true;
2457
- } else {
2458
- const emptyButReachable = messages != null;
2459
- const graceApplies = !emptyButReachable || watcher.hadUsablePoll;
2460
- if (graceApplies && this.now() - watcher.lastGoodPollAt < POLL_MISS_GRACE_MS) {
2461
- continue;
2462
- }
2463
- }
2464
- const { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk } = await this.pollInteractions(sessionId, watcher, messages);
2465
2034
  for (const inFlight of [...watcher.inFlight.values()]) {
2466
- await this.serviceInFlightMessage(
2467
- sessionId,
2468
- watcher,
2469
- inFlight,
2470
- messages,
2471
- openQuestions,
2472
- openPermissions,
2473
- questionsPolledOk,
2474
- permissionsPolledOk
2475
- );
2035
+ await this.serviceInFlightMessage(sessionId, watcher, inFlight, messages);
2476
2036
  }
2037
+ await this.pollInteractions(sessionId, watcher, messages);
2477
2038
  }
2478
2039
  } catch (err) {
2479
2040
  if (err instanceof ChannelAuthError) {
@@ -2495,40 +2056,15 @@ var ChannelDriver = class {
2495
2056
  });
2496
2057
  }
2497
2058
  }
2498
- /**
2499
- * On FIRST observing a terminal (done/failed) state, ensure the delivery
2500
- * (markDone/markFailed) transient-retry path has a real window. A long
2501
- * ACTIVELY-running turn is kept past its original `deadline`, so by completion
2502
- * `now >= deadline` already holds and the retry bound below would fire on the
2503
- * first transient PATCH failure — dropping the message before its reply lands
2504
- * (Bugbot "Stale deadline aborts long-turn delivery"). Re-anchor once (latched)
2505
- * to a fresh `pausedMaxWaitMs` window; only extend if the current deadline is at
2506
- * or past now, so a still-ample window is left untouched.
2507
- */
2508
- anchorDeliveryDeadline(inFlight) {
2509
- if (inFlight.deliveryDeadlineAnchored) return;
2510
- inFlight.deliveryDeadlineAnchored = true;
2511
- if (this.now() >= inFlight.deadline) {
2512
- inFlight.deadline = this.now() + this.pausedMaxWaitMs;
2513
- }
2514
- }
2515
2059
  /**
2516
2060
  * Drive ONE in-flight message's lifecycle from the tick's message snapshot.
2517
2061
  * Fires markProcessing on queued→running and markDone on done (each once),
2518
2062
  * applies the idle-path re-dispatch guard, and removes the message from the
2519
2063
  * in-flight set on completion or timeout.
2520
2064
  */
2521
- async serviceInFlightMessage(sessionId, watcher, inFlight, messages, openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk) {
2065
+ async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
2522
2066
  const conv = watcher.conv;
2523
2067
  const state = messageRunState(messages, inFlight.opencodeMessageId);
2524
- const id = inFlight.evidentMessageId;
2525
- if (openQuestions.has(id)) inFlight.pausedOnQuestion = true;
2526
- else if (questionsPolledOk) inFlight.pausedOnQuestion = false;
2527
- if (openPermissions.has(id)) inFlight.pausedOnPermission = true;
2528
- else if (permissionsPolledOk) inFlight.pausedOnPermission = false;
2529
- const observedOpen = openQuestions.has(id) || openPermissions.has(id);
2530
- const latchedPaused = inFlight.pausedOnQuestion || inFlight.pausedOnPermission;
2531
- const awaitingHuman = observedOpen || latchedPaused;
2532
2068
  if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
2533
2069
  let claimed;
2534
2070
  try {
@@ -2536,7 +2072,7 @@ var ChannelDriver = class {
2536
2072
  conv.id,
2537
2073
  inFlight.evidentMessageId,
2538
2074
  sessionId,
2539
- inFlight.opencodeMessageId
2075
+ inFlight.redriveOpencodeMessageId
2540
2076
  );
2541
2077
  } catch (err) {
2542
2078
  if (err instanceof ChannelAuthError) throw err;
@@ -2559,7 +2095,6 @@ var ChannelDriver = class {
2559
2095
  }
2560
2096
  }
2561
2097
  if (state === "done") {
2562
- this.anchorDeliveryDeadline(inFlight);
2563
2098
  if (!inFlight.done) {
2564
2099
  this.log({
2565
2100
  level: "info",
@@ -2572,7 +2107,7 @@ var ChannelDriver = class {
2572
2107
  conv.id,
2573
2108
  inFlight.evidentMessageId,
2574
2109
  sessionId,
2575
- inFlight.opencodeMessageId
2110
+ inFlight.redriveOpencodeMessageId
2576
2111
  );
2577
2112
  } catch (err) {
2578
2113
  if (err instanceof ChannelAuthError) throw err;
@@ -2610,7 +2145,6 @@ var ChannelDriver = class {
2610
2145
  return;
2611
2146
  }
2612
2147
  if (state === "failed") {
2613
- this.anchorDeliveryDeadline(inFlight);
2614
2148
  if (!inFlight.done) {
2615
2149
  const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
2616
2150
  this.log({
@@ -2656,70 +2190,211 @@ var ChannelDriver = class {
2656
2190
  this.removeInFlight(watcher, inFlight.evidentMessageId);
2657
2191
  return;
2658
2192
  }
2193
+ if (state === "unknown") {
2194
+ if (this.now() - inFlight.dispatchedAt >= this.dispatchConfirmMs) {
2195
+ await this.redispatchInFlight(sessionId, inFlight);
2196
+ }
2197
+ }
2659
2198
  const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
2660
2199
  const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
2661
- if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
2662
- inFlight.stuckReported = true;
2663
- void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2664
- stuck_for_ms: this.now() - inFlight.dispatchedAt
2665
- });
2200
+ if (state === "queued" && pastStuckBound && sessionIdle) {
2201
+ if (!inFlight.stuckReported) {
2202
+ inFlight.stuckReported = true;
2203
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2204
+ stuck_for_ms: this.now() - inFlight.dispatchedAt
2205
+ });
2206
+ }
2207
+ const dueForRedrive = inFlight.lastRedriveAt == null || this.now() - inFlight.lastRedriveAt >= this.stuckQueuedMs;
2208
+ if (dueForRedrive && isSessionSettled(messages, this.now(), this.settleMs)) {
2209
+ const removed = await this.redriveStuckQueued(sessionId, watcher, inFlight, messages);
2210
+ if (removed) return;
2211
+ }
2666
2212
  }
2667
- const activelyRunning = state === "running" && !awaitingHuman;
2668
- if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
2213
+ if (this.now() >= inFlight.deadline) {
2669
2214
  this.log({
2670
- level: "error",
2671
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} exceeded the absolute processing ceiling (${Math.round((this.now() - inFlight.processingAnchorMs) / 6e4)}min, session ${sessionId}) while still actively running \u2014 releasing so the cron can reclaim it`,
2215
+ level: "info",
2216
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
2672
2217
  conversation_id: conv.id,
2673
2218
  message_id: inFlight.evidentMessageId
2674
2219
  });
2675
- void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2676
- watched_for_ms: this.now() - inFlight.processingAnchorMs
2677
- });
2678
2220
  this.removeInFlight(watcher, inFlight.evidentMessageId);
2679
- return;
2680
2221
  }
2681
- if (activelyRunning && !inFlight.awaitingHumanLatched && !inFlight.aliveInFlight && this.now() - inFlight.lastAliveAt >= HEARTBEAT_MS) {
2682
- inFlight.aliveInFlight = true;
2683
- void this.postSignal(conv.id, inFlight.evidentMessageId, "alive").then((ok) => {
2684
- inFlight.aliveInFlight = false;
2685
- if (ok) inFlight.lastAliveAt = this.now();
2222
+ }
2223
+ /**
2224
+ * Re-dispatch a message whose user row never appeared (idle-path guard). Safe:
2225
+ * opencode treats a duplicate caller-supplied `messageID` as idempotent (PoC
2226
+ * fact 9) one user message + one reply even if the original DID land. Resets
2227
+ * the dispatch timestamp so the guard doesn't immediately fire again.
2228
+ */
2229
+ async redispatchInFlight(sessionId, inFlight) {
2230
+ const options = {
2231
+ agent: inFlight.message.opencode_agent ?? void 0,
2232
+ model: inFlight.message.opencode_model ?? void 0
2233
+ };
2234
+ this.log({
2235
+ level: "info",
2236
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} not observed after dispatch \u2014 re-dispatching (idle-path guard)`,
2237
+ message_id: inFlight.evidentMessageId
2238
+ });
2239
+ try {
2240
+ await sendPromptAsync(
2241
+ this.port,
2242
+ sessionId,
2243
+ inFlight.message.content,
2244
+ options,
2245
+ inFlight.opencodeMessageId
2246
+ );
2247
+ } catch (err) {
2248
+ this.log({
2249
+ level: "error",
2250
+ message: `Re-dispatch failed for message ${inFlight.evidentMessageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
2251
+ message_id: inFlight.evidentMessageId
2686
2252
  });
2687
2253
  }
2688
- if (awaitingHuman) {
2689
- if (!inFlight.awaitingHumanLatched) {
2690
- inFlight.deadline = this.now() + this.pausedMaxWaitMs;
2691
- inFlight.awaitingHumanLatched = true;
2692
- }
2693
- if (!inFlight.pausedClearConfirmed && !inFlight.pausedInFlight) {
2694
- inFlight.pausedInFlight = true;
2695
- void this.postSignal(conv.id, inFlight.evidentMessageId, "paused").then((ok) => {
2696
- inFlight.pausedInFlight = false;
2697
- if (ok && inFlight.awaitingHumanLatched) inFlight.pausedClearConfirmed = true;
2254
+ inFlight.dispatchedAt = this.now();
2255
+ }
2256
+ /**
2257
+ * RE-DRIVE a stuck-`queued` follow-up so opencode actually runs it
2258
+ * (queued-followup-redrive; the fix for the 2026-07-18 dev incident).
2259
+ *
2260
+ * The wedge: a follow-up `prompt_async`'d mid-turn / in the post-turn settling
2261
+ * window is orphaned its user message persists but the turn never runs
2262
+ * (`messageRunState === 'queued'`). Proven against real opencode 1.18.3 by an
2263
+ * isolated, interleaved experiment (`.harness/followup-send-mechanism-finding.md`):
2264
+ * the un-sticking variable is the `messageID` FORMAT. A follow-up carrying the
2265
+ * runner's CUSTOM underscore id (`msg_<sanitized-uuid>`) is NEVER picked up mid-
2266
+ * turn (0/24 trials); the SAME follow-up carrying a NATIVE-format id
2267
+ * (`msg_`+24 base62) runs once the prior turn settles (~50% locally, reliably on
2268
+ * dev). `parts[].id` and `agent`/`model` were proven IRRELEVANT.
2269
+ *
2270
+ * So we re-`prompt_async` into the SAME opencode session (opencode-web does
2271
+ * exactly this and it works — UPDATE 2 in the investigation doc), preserving the
2272
+ * conversation's history/continuity: we do NOT create a fresh session and we do
2273
+ * NOT overwrite the conversation's `opencode_session_id`. Only the opencode
2274
+ * user-message id changes — to a fresh native id per attempt (opencode's
2275
+ * caller-supplied id dedup is global+permanent, so each attempt needs a never-
2276
+ * seen id).
2277
+ *
2278
+ * Reply correlation is PRESERVED: the native id is random and NOT re-derivable
2279
+ * from the row id, so the runner carries it to the server (on markProcessing/
2280
+ * markDone via `redriveOpencodeMessageId`), which persists it on the row and
2281
+ * correlates the reply by THAT id (`conversation-notification.ts`). A normal,
2282
+ * never-re-driven message still correlates by the derived stable id — unchanged.
2283
+ *
2284
+ * Bounded to `stuckQueuedRedriveMax` attempts. On exhaustion the row is marked
2285
+ * FAILED (existing channel failure affordance) so the user is TOLD it could not
2286
+ * be answered rather than left silent, and it is removed from the in-flight set.
2287
+ *
2288
+ * @returns true if the message LEFT this watcher's in-flight set (gave up +
2289
+ * marked failed) — the caller then stops servicing it this tick. Returns false
2290
+ * when the follow-up stays in THIS watcher (re-driven in place, or the re-drive
2291
+ * send failed and the next window retries).
2292
+ */
2293
+ async redriveStuckQueued(sessionId, watcher, inFlight, messages) {
2294
+ if (await this.deliverIfAnyAttemptCompleted(sessionId, watcher, inFlight, messages)) {
2295
+ return true;
2296
+ }
2297
+ if (inFlight.redriveAttempts >= this.stuckQueuedRedriveMax) {
2298
+ this.log({
2299
+ level: "error",
2300
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} stuck queued after ${inFlight.redriveAttempts} re-drive attempt(s) \u2014 marking failed`,
2301
+ conversation_id: watcher.conv.id,
2302
+ message_id: inFlight.evidentMessageId
2303
+ });
2304
+ try {
2305
+ await this.markFailed(watcher.conv.id, inFlight.evidentMessageId);
2306
+ } catch (err) {
2307
+ if (err instanceof ChannelAuthError) throw err;
2308
+ this.log({
2309
+ level: "error",
2310
+ message: `Failed to mark stuck message ${inFlight.evidentMessageId.slice(0, 8)} failed (leaving for the cron safety net): ${err instanceof Error ? err.message : String(err)}`,
2311
+ conversation_id: watcher.conv.id,
2312
+ message_id: inFlight.evidentMessageId
2698
2313
  });
2699
2314
  }
2700
- } else if (inFlight.awaitingHumanLatched) {
2701
- inFlight.awaitingHumanLatched = false;
2702
- inFlight.pausedOnQuestion = false;
2703
- inFlight.pausedOnPermission = false;
2704
- inFlight.pausedClearConfirmed = false;
2315
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2316
+ return true;
2705
2317
  }
2706
- const siblingPaused = (sib) => openQuestions.has(sib.evidentMessageId) || openPermissions.has(sib.evidentMessageId) || sib.awaitingHumanLatched || sib.pausedOnQuestion || sib.pausedOnPermission;
2707
- const hasActivelyRunningSibling = [...watcher.inFlight.values()].some(
2708
- (sib) => sib.evidentMessageId !== inFlight.evidentMessageId && messageRunState(messages, sib.opencodeMessageId) === "running" && !siblingPaused(sib)
2709
- );
2710
- const queuedBehindRunningSibling = state === "queued" && hasActivelyRunningSibling;
2711
- if (!activelyRunning && !queuedBehindRunningSibling && this.now() >= inFlight.deadline) {
2318
+ inFlight.redriveAttempts += 1;
2319
+ inFlight.lastRedriveAt = this.now();
2320
+ this.log({
2321
+ level: "info",
2322
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} stuck queued on an idle session \u2014 re-driving into the SAME session with a native id (attempt ${inFlight.redriveAttempts}/${this.stuckQueuedRedriveMax})`,
2323
+ conversation_id: watcher.conv.id,
2324
+ message_id: inFlight.evidentMessageId
2325
+ });
2326
+ void this.postSignal(watcher.conv.id, inFlight.evidentMessageId, "redriven", {
2327
+ redrive_attempt: inFlight.redriveAttempts
2328
+ });
2329
+ const options = {
2330
+ agent: inFlight.message.opencode_agent ?? void 0,
2331
+ model: inFlight.message.opencode_model ?? void 0
2332
+ };
2333
+ const nativeId = nativeOpencodeMessageId();
2334
+ try {
2335
+ await sendPromptAsync(this.port, sessionId, inFlight.message.content, options, nativeId);
2336
+ } catch (err) {
2337
+ if (err instanceof ChannelAuthError) throw err;
2338
+ inFlight.redriveAttempts -= 1;
2712
2339
  this.log({
2713
- level: "info",
2714
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
2715
- conversation_id: conv.id,
2340
+ level: "error",
2341
+ message: `Re-drive failed for stuck message ${inFlight.evidentMessageId.slice(0, 8)} (will retry next window): ${err instanceof Error ? err.message : String(err)}`,
2342
+ conversation_id: watcher.conv.id,
2716
2343
  message_id: inFlight.evidentMessageId
2717
2344
  });
2718
- void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2719
- watched_for_ms: this.now() - inFlight.dispatchedAt
2345
+ return false;
2346
+ }
2347
+ inFlight.opencodeMessageId = nativeId;
2348
+ inFlight.redriveOpencodeMessageId = nativeId;
2349
+ inFlight.attemptedOpencodeMessageIds.push(nativeId);
2350
+ inFlight.started = false;
2351
+ return false;
2352
+ }
2353
+ /**
2354
+ * Finding 2 (Bugbot #217): scan EVERY opencode id this stuck message has been
2355
+ * driven under (`attemptedOpencodeMessageIds`) for a COMPLETED correlated reply
2356
+ * in the tick's snapshot; if one is found, markDone off THAT id (so the reply an
2357
+ * earlier re-drive attempt eventually produced is delivered) instead of
2358
+ * re-driving again or marking failed. Idempotent + guarded like the watcher's
2359
+ * done branch. `@returns` true when the message left the in-flight set (delivered
2360
+ * or terminally-undeliverable) — the caller then stops servicing it this tick.
2361
+ *
2362
+ * The LATEST id is normally handled by `serviceInFlightMessage`'s own `done`
2363
+ * branch; this covers the ids a re-drive OVERWROTE, which that branch no longer
2364
+ * polls. `messageRunState === 'done'` means a completed, non-errored correlated
2365
+ * reply exists for that id.
2366
+ */
2367
+ async deliverIfAnyAttemptCompleted(sessionId, watcher, inFlight, messages) {
2368
+ const completedId = inFlight.attemptedOpencodeMessageIds.find(
2369
+ (id) => messageRunState(messages, id) === "done"
2370
+ );
2371
+ if (!completedId) return false;
2372
+ if (inFlight.done) {
2373
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2374
+ return true;
2375
+ }
2376
+ this.log({
2377
+ level: "info",
2378
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed under an earlier re-drive attempt's id \u2014 marking done (not re-driving/failing)`,
2379
+ conversation_id: watcher.conv.id,
2380
+ message_id: inFlight.evidentMessageId
2381
+ });
2382
+ try {
2383
+ await this.markDone(watcher.conv.id, inFlight.evidentMessageId, sessionId, completedId);
2384
+ } catch (err) {
2385
+ if (err instanceof ChannelAuthError) throw err;
2386
+ this.log({
2387
+ level: "error",
2388
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done off an earlier re-drive attempt (leaving for the cron safety net): ${err instanceof Error ? err.message : String(err)}`,
2389
+ conversation_id: watcher.conv.id,
2390
+ message_id: inFlight.evidentMessageId
2720
2391
  });
2721
2392
  this.removeInFlight(watcher, inFlight.evidentMessageId);
2393
+ return true;
2722
2394
  }
2395
+ inFlight.done = true;
2396
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2397
+ return true;
2723
2398
  }
2724
2399
  // -------------------------------------------------------------------------
2725
2400
  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
@@ -2740,17 +2415,12 @@ var ChannelDriver = class {
2740
2415
  */
2741
2416
  async readoptProcessing() {
2742
2417
  const rows = await this.getProcessingMessages();
2743
- if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0 || this.readoptPollUnresolvedSignalled.size > 0) {
2418
+ if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0) {
2744
2419
  const stillProcessing = new Set(rows.map((r) => r.id));
2745
- for (const id of [
2746
- ...this.dontRedispatch,
2747
- ...this.doneUndeliverable,
2748
- ...this.readoptPollUnresolvedSignalled
2749
- ]) {
2420
+ for (const id of [...this.dontRedispatch, ...this.doneUndeliverable]) {
2750
2421
  if (!stillProcessing.has(id)) {
2751
2422
  const cleared = this.dontRedispatch.delete(id);
2752
2423
  const clearedUndeliverable = this.doneUndeliverable.delete(id);
2753
- this.readoptPollUnresolvedSignalled.delete(id);
2754
2424
  if (cleared || clearedUndeliverable) {
2755
2425
  this.log({
2756
2426
  level: "info",
@@ -2804,10 +2474,8 @@ var ChannelDriver = class {
2804
2474
  });
2805
2475
  continue;
2806
2476
  }
2807
- const anyUntracked = sessionRows.some((row) => !this.isTracked(sessionId, row.id));
2808
- const sessionOngoing = anyUntracked ? await isSessionOngoing(this.port, sessionId) : null;
2809
2477
  for (const row of sessionRows) {
2810
- await this.readoptOne(sessionId, row, messages, sessionOngoing);
2478
+ await this.readoptOne(sessionId, row, messages);
2811
2479
  }
2812
2480
  }
2813
2481
  }
@@ -2815,21 +2483,23 @@ var ChannelDriver = class {
2815
2483
  * Re-adopt ONE `processing` row against the tick's session message snapshot
2816
2484
  * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
2817
2485
  *
2818
- * Branches on `messageRunState(messages, row.opencode_message_id)` — the
2819
- * opencode-assigned user-message id persisted on the first `processing` PATCH
2820
- * (#218). A row with a NULL stored id (dispatched but the read-back never landed
2821
- * before the restart) has no id to correlate treated as an orphan and
2822
- * re-dispatched (at most once, see `forceReadoptRun`):
2486
+ * Branches on `messageRunState(messages, effectiveId)`, where `effectiveId` is
2487
+ * the NATIVE id a prior lifetime's re-drive ran this row under
2488
+ * (`row.opencode_message_id`) if present, else the STABLE derived id
2489
+ * (`opencodeMessageIdFor(row.id)`). Consulting the native id is what makes a
2490
+ * re-driven row that ALREADY ran/completed resolve correctly on restart instead
2491
+ * of looking `queued` under the (never-run) stable id and being re-driven AGAIN
2492
+ * (Bugbot #217 Finding 3 — the duplicate-turn bug):
2823
2493
  * - `done` → `markDone` now (guarded like the watcher's done branch);
2824
2494
  * - `failed` → `markFailed` with the surfaced error (issue #182), so an
2825
2495
  * errored turn is reported failed on restart, NOT re-dispatched;
2826
2496
  * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
2827
- * tracking the stored id so the reply correlates by it;
2828
- * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
2497
+ * tracking the effective id so the reply correlates by it;
2498
+ * - `unknown` → re-dispatch the STABLE id + attach a watcher (orphan).
2829
2499
  *
2830
2500
  * Only `ChannelAuthError` propagates.
2831
2501
  */
2832
- async readoptOne(sessionId, row, messages, sessionOngoing) {
2502
+ async readoptOne(sessionId, row, messages) {
2833
2503
  if (this.isTracked(sessionId, row.id)) {
2834
2504
  this.log({
2835
2505
  level: "info",
@@ -2839,8 +2509,10 @@ var ChannelDriver = class {
2839
2509
  });
2840
2510
  return;
2841
2511
  }
2842
- const ocId = row.opencode_message_id;
2843
- const state = messageRunState(messages, ocId ?? "");
2512
+ const stableId = opencodeMessageIdFor2(row.id);
2513
+ const nativeRedriveId = row.opencode_message_id;
2514
+ const ocId = nativeRedriveId ?? stableId;
2515
+ const state = messageRunState(messages, ocId);
2844
2516
  if (state === "done") {
2845
2517
  if (this.doneUndeliverable.has(row.id)) {
2846
2518
  this.log({
@@ -2858,7 +2530,7 @@ var ChannelDriver = class {
2858
2530
  message_id: row.id
2859
2531
  });
2860
2532
  try {
2861
- await this.markDone(row.conversation_id, row.id, sessionId, ocId);
2533
+ await this.markDone(row.conversation_id, row.id, sessionId, nativeRedriveId);
2862
2534
  } catch (err) {
2863
2535
  if (err instanceof ChannelAuthError) throw err;
2864
2536
  if (err instanceof ChannelTerminalError) {
@@ -2869,7 +2541,6 @@ var ChannelDriver = class {
2869
2541
  conversation_id: row.conversation_id,
2870
2542
  message_id: row.id
2871
2543
  });
2872
- void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
2873
2544
  return;
2874
2545
  }
2875
2546
  this.log({
@@ -2881,11 +2552,10 @@ var ChannelDriver = class {
2881
2552
  return;
2882
2553
  }
2883
2554
  this.dontRedispatch.delete(row.id);
2884
- void this.postSignal(row.conversation_id, row.id, "readopt_done");
2885
2555
  return;
2886
2556
  }
2887
2557
  if (state === "failed") {
2888
- const error2 = messageError(messages, ocId ?? "") ?? void 0;
2558
+ const error2 = messageError(messages, ocId) ?? void 0;
2889
2559
  this.log({
2890
2560
  level: "error",
2891
2561
  message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
@@ -2904,7 +2574,6 @@ var ChannelDriver = class {
2904
2574
  conversation_id: row.conversation_id,
2905
2575
  message_id: row.id
2906
2576
  });
2907
- void this.postSignal(row.conversation_id, row.id, "readopt_undeliverable");
2908
2577
  return;
2909
2578
  }
2910
2579
  this.log({
@@ -2916,7 +2585,6 @@ var ChannelDriver = class {
2916
2585
  return;
2917
2586
  }
2918
2587
  this.dontRedispatch.delete(row.id);
2919
- void this.postSignal(row.conversation_id, row.id, "readopt_failed");
2920
2588
  return;
2921
2589
  }
2922
2590
  if (this.dontRedispatch.has(row.id)) {
@@ -2928,129 +2596,48 @@ var ChannelDriver = class {
2928
2596
  });
2929
2597
  return;
2930
2598
  }
2931
- let statusReadableOngoing = null;
2932
- if (state === "running" && ocId) {
2933
- const reply = findLastAssistantReplyFor(messages, ocId);
2934
- const shape = this.replyCompletionShape(reply);
2935
- const ongoing = sessionOngoing;
2936
- statusReadableOngoing = ongoing;
2937
- if (ongoing === false) {
2938
- this.log({
2939
- level: "info",
2940
- message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} but session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status (absent/idle) \u2014 re-dispatching from scratch (status-gated recovery)`,
2941
- conversation_id: row.conversation_id,
2942
- message_id: row.id
2943
- });
2944
- await this.forceReadoptRun(sessionId, row);
2945
- return;
2946
- }
2947
- if (ongoing === true) {
2948
- this.log({
2949
- level: "info",
2950
- message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} and session ${sessionId.slice(0, 8)} is ongoing per GET /session/status (busy/retry) \u2014 re-attaching watcher (no re-dispatch)`,
2951
- conversation_id: row.conversation_id,
2952
- message_id: row.id
2953
- });
2954
- } else {
2955
- if (shape === "b1") {
2956
- this.log({
2957
- level: "info",
2958
- message: `Re-adopt: message ${row.id.slice(0, 8)} running/b1 but GET /session/status was unreadable (null) for session ${sessionId.slice(0, 8)} \u2014 NOT latching a b1 row on a transient status blip; leaving it un-tracked to re-evaluate on the next drain`,
2959
- conversation_id: row.conversation_id,
2960
- message_id: row.id
2961
- });
2962
- if (!this.readoptPollUnresolvedSignalled.has(row.id)) {
2963
- this.readoptPollUnresolvedSignalled.add(row.id);
2964
- void this.postSignal(row.conversation_id, row.id, "readopt_poll_unresolved");
2965
- }
2966
- return;
2967
- }
2968
- this.log({
2969
- level: "info",
2970
- message: `Re-adopt: message ${row.id.slice(0, 8)} running/${shape} but GET /session/status was unreadable (null) for session ${sessionId.slice(0, 8)} \u2014 falling back to the #253 preamble + descendant cross-check`,
2971
- conversation_id: row.conversation_id,
2972
- message_id: row.id
2973
- });
2974
- }
2975
- }
2976
- if (statusReadableOngoing === null && state === "running" && ocId && isPreamblePinnedRunning(messages, ocId)) {
2977
- const descendantAlive = await this.isAnyDescendantSessionAlive(sessionId);
2978
- if (descendantAlive === true) {
2979
- this.log({
2980
- level: "info",
2981
- message: `Re-adopt: message ${row.id.slice(0, 8)} preamble-pinned running (root ${sessionId.slice(0, 8)}) but a live descendant sub-agent session was found \u2014 treating as still running, re-attaching watcher (no re-dispatch)`,
2982
- conversation_id: row.conversation_id,
2983
- message_id: row.id
2984
- });
2985
- } else {
2986
- this.log({
2987
- level: "info",
2988
- message: `Re-adopt: message ${row.id.slice(0, 8)} preamble-pinned on recovery (root ${sessionId.slice(0, 8)}), no live descendant runner \u2014 re-dispatching from scratch${descendantAlive === null ? " (descendant liveness indeterminate; a restart guarantees no live runner, so this does NOT block the re-dispatch)" : ""}`,
2989
- conversation_id: row.conversation_id,
2990
- message_id: row.id
2991
- });
2992
- await this.forceReadoptRun(sessionId, row);
2993
- return;
2994
- }
2995
- }
2996
- if ((state === "running" || state === "queued") && ocId) {
2599
+ if (state === "running" || state === "queued") {
2997
2600
  const conv = this.convForRow(sessionId, row);
2998
2601
  const message = this.queuedMessageForRow(row);
2999
- this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
2602
+ this.registerReadopted(
2603
+ conv,
2604
+ sessionId,
2605
+ message,
2606
+ ocId,
2607
+ this.processedAtMs(row),
2608
+ nativeRedriveId
2609
+ );
3000
2610
  this.dispatched.add(row.id);
3001
2611
  this.readopted.add(row.id);
3002
2612
  this.ensureWatcherRunning(sessionId);
3003
2613
  this.log({
3004
2614
  level: "info",
3005
- message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
2615
+ message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (${nativeRedriveId ? "native re-drive id" : "stable id"}, no re-dispatch)`,
3006
2616
  conversation_id: row.conversation_id,
3007
2617
  message_id: row.id
3008
2618
  });
3009
- void this.postSignal(row.conversation_id, row.id, "readopt_reattached");
3010
2619
  return;
3011
2620
  }
3012
2621
  await this.forceReadoptRun(sessionId, row);
3013
2622
  }
3014
2623
  /**
3015
- * Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
3016
- *
3017
- * #218/WI-5: the row's user message is absent (never kept, or a null stored id),
3018
- * so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
3019
- * read it back, and register the watcher under the assigned id so the reply
3020
- * correlates server-side.
2624
+ * Re-dispatch an orphaned (`unknown`) `processing` row (ADR-0046 Decision §2).
3021
2625
  *
3022
- * ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
3023
- * id). Without a guard, if this dispatches on tick N but the read-back+persist
3024
- * hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
3025
- * tick N+1 would dispatch AGAIN duplicate user turns. The `awaitingReadopt`
3026
- * latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
3027
- * short-circuit while the row is latched; clear it on a successful dispatch (the
3028
- * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
3029
- * re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
3030
- * may retry exactly once more).
2626
+ * The stable-id user message is absent from the session, so we (re-)dispatch with
2627
+ * the STABLE id (`opencodeMessageIdFor(row.id)`) NOT a divergent per-attempt id.
2628
+ * This is what keeps the reply correlatable: the server's completion
2629
+ * notification looks for the reply under the stable id, so the fresh turn's reply
2630
+ * (which hangs off the stable id) is found and delivered. The residual
2631
+ * duplicate-incomplete-turn semantics (ADR §2, `.harness/restart-recovery-orphan-finding.md`)
2632
+ * are unchanged and, for an ABSENT id, cannot bite there is no existing turn
2633
+ * to swallow the duplicate.
3031
2634
  *
3032
- * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
2635
+ * `evidentMessageId = row.id` addresses the SERVER row; the stable
2636
+ * `opencodeMessageId` is what the watcher polls. Deadline anchored to
3033
2637
  * `processed_at` (Invariant 1).
3034
2638
  */
3035
2639
  async forceReadoptRun(sessionId, row) {
3036
- if (this.stopped) {
3037
- this.log({
3038
- level: "info",
3039
- 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`,
3040
- conversation_id: row.conversation_id,
3041
- message_id: row.id
3042
- });
3043
- return;
3044
- }
3045
- if (this.awaitingReadopt.has(row.id)) {
3046
- this.log({
3047
- level: "info",
3048
- message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
3049
- conversation_id: row.conversation_id,
3050
- message_id: row.id
3051
- });
3052
- return;
3053
- }
2640
+ const ocId = opencodeMessageIdFor2(row.id);
3054
2641
  if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
3055
2642
  this.dontRedispatch.add(row.id);
3056
2643
  this.log({
@@ -3059,7 +2646,6 @@ var ChannelDriver = class {
3059
2646
  conversation_id: row.conversation_id,
3060
2647
  message_id: row.id
3061
2648
  });
3062
- void this.postSignal(row.conversation_id, row.id, "readopt_window_elapsed");
3063
2649
  return;
3064
2650
  }
3065
2651
  const options = {
@@ -3068,19 +2654,13 @@ var ChannelDriver = class {
3068
2654
  };
3069
2655
  this.log({
3070
2656
  level: "info",
3071
- message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
2657
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching with the stable id`,
3072
2658
  conversation_id: row.conversation_id,
3073
2659
  message_id: row.id
3074
2660
  });
3075
- this.awaitingReadopt.add(row.id);
3076
- let ocId;
3077
2661
  try {
3078
- ocId = await this.dispatchLocked(
3079
- sessionId,
3080
- () => sendPromptAsync(this.port, sessionId, row.content, options)
3081
- );
2662
+ await sendPromptAsync(this.port, sessionId, row.content, options, ocId);
3082
2663
  } catch (err) {
3083
- this.awaitingReadopt.delete(row.id);
3084
2664
  if (err instanceof ChannelAuthError) throw err;
3085
2665
  this.log({
3086
2666
  level: "error",
@@ -3088,18 +2668,6 @@ var ChannelDriver = class {
3088
2668
  conversation_id: row.conversation_id,
3089
2669
  message_id: row.id
3090
2670
  });
3091
- void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3092
- return;
3093
- }
3094
- if (ocId === null) {
3095
- this.awaitingReadopt.delete(row.id);
3096
- this.log({
3097
- level: "error",
3098
- 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`,
3099
- conversation_id: row.conversation_id,
3100
- message_id: row.id
3101
- });
3102
- void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
3103
2671
  return;
3104
2672
  }
3105
2673
  const conv = this.convForRow(sessionId, row);
@@ -3107,9 +2675,7 @@ var ChannelDriver = class {
3107
2675
  this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
3108
2676
  this.dispatched.add(row.id);
3109
2677
  this.readopted.add(row.id);
3110
- this.awaitingReadopt.delete(row.id);
3111
2678
  this.ensureWatcherRunning(sessionId);
3112
- void this.postSignal(row.conversation_id, row.id, "readopt_redispatched");
3113
2679
  }
3114
2680
  /**
3115
2681
  * True if `evidentMessageId` is already being driven — either in the
@@ -3201,41 +2767,21 @@ var ChannelDriver = class {
3201
2767
  * RUNNING (not done) is the one that paused. With one running message that is
3202
2768
  * unambiguous; with several we prefer an explicit messageID match, else the
3203
2769
  * oldest running message.
3204
- *
3205
- * Returns the set of in-flight Evident message ids that are paused awaiting a
3206
- * human — an outstanding (still-open) question/permission is attributed to them.
3207
- * `serviceInFlightMessage` uses this to keep an actively-running turn watched
3208
- * forever (ADR-0047) while still bounding a turn merely blocked on a person who
3209
- * may never answer. Attribution here covers ALL open interactions, not just
3210
- * NEW (un-deduped) ones — a question stays "awaiting a human" until answered,
3211
- * even after it was already surfaced to the channel.
3212
2770
  */
3213
2771
  async pollInteractions(sessionId, watcher, messages) {
3214
- const openQuestions = /* @__PURE__ */ new Set();
3215
- const openPermissions = /* @__PURE__ */ new Set();
3216
- let questionsPolledOk = true;
3217
- let permissionsPolledOk = true;
3218
2772
  let questions = [];
3219
2773
  try {
3220
2774
  const res = await this.fetchImpl(`${this.opencodeBase}/question`);
3221
2775
  if (res.ok) {
3222
2776
  const body = await res.json();
3223
- if (Array.isArray(body)) {
3224
- questions = body;
3225
- } else {
3226
- questionsPolledOk = false;
3227
- }
3228
- } else {
3229
- questionsPolledOk = false;
2777
+ questions = Array.isArray(body) ? body : [];
3230
2778
  }
3231
2779
  } catch {
3232
- questionsPolledOk = false;
3233
2780
  }
3234
2781
  for (const q of questions) {
3235
- if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
3236
- const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
3237
- if (paused) openQuestions.add(paused.evidentMessageId);
2782
+ if (q.sessionID !== sessionId) continue;
3238
2783
  if (watcher.reportedQuestions.has(q.id)) continue;
2784
+ const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
3239
2785
  const reported = await this.reportInteraction(
3240
2786
  watcher.conv.id,
3241
2787
  "question",
@@ -3249,22 +2795,14 @@ var ChannelDriver = class {
3249
2795
  const res = await this.fetchImpl(`${this.opencodeBase}/permission`);
3250
2796
  if (res.ok) {
3251
2797
  const body = await res.json();
3252
- if (Array.isArray(body)) {
3253
- permissions = body;
3254
- } else {
3255
- permissionsPolledOk = false;
3256
- }
3257
- } else {
3258
- permissionsPolledOk = false;
2798
+ permissions = Array.isArray(body) ? body : [];
3259
2799
  }
3260
2800
  } catch {
3261
- permissionsPolledOk = false;
3262
2801
  }
3263
2802
  for (const p of permissions) {
3264
- if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
3265
- const paused = this.attributeInteraction(watcher, p.messageID, messages);
3266
- if (paused) openPermissions.add(paused.evidentMessageId);
2803
+ if (p.sessionID !== sessionId) continue;
3267
2804
  if (watcher.reportedPermissions.has(p.id)) continue;
2805
+ const paused = this.attributeInteraction(watcher, p.messageID, messages);
3268
2806
  const reported = await this.reportInteraction(
3269
2807
  watcher.conv.id,
3270
2808
  "permission",
@@ -3273,128 +2811,6 @@ var ChannelDriver = class {
3273
2811
  );
3274
2812
  if (reported) watcher.reportedPermissions.add(p.id);
3275
2813
  }
3276
- return { openQuestions, openPermissions, questionsPolledOk, permissionsPolledOk };
3277
- }
3278
- /**
3279
- * True when `sessionId` is the `rootSessionId` itself OR a descendant of it —
3280
- * i.e. its `parentID` chain (resolved via `GET /session/:id`) reaches the
3281
- * watched root. Sub-agents spawned via the `task` tool run in child sessions,
3282
- * so their questions/permissions live under a different `sessionID` that must
3283
- * still be attributed to the root conversation the watcher owns.
3284
- *
3285
- * Parents are cached in `sessionParents` so we walk each session at most once;
3286
- * a bounded depth cap guards against a cycle or a pathological chain, and any
3287
- * fetch failure is treated as "not a descendant" (best-effort — the interaction
3288
- * simply isn't surfaced this tick and is retried next tick once resolvable).
3289
- */
3290
- async sessionBelongsTo(sessionId, rootSessionId) {
3291
- let current = sessionId;
3292
- for (let depth = 0; current && depth < 32; depth++) {
3293
- if (current === rootSessionId) return true;
3294
- const parent = await this.resolveSessionParent(current);
3295
- if (parent === null || parent === void 0) return false;
3296
- current = parent;
3297
- }
3298
- return false;
3299
- }
3300
- /**
3301
- * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
3302
- * `null` for a root session (no parent) and `undefined` when opencode is
3303
- * unreachable / the session can't be read (so the caller stops walking without
3304
- * caching a wrong answer — the next tick retries).
3305
- */
3306
- async resolveSessionParent(sessionId) {
3307
- const cached = this.sessionParents.get(sessionId);
3308
- if (cached !== void 0) return cached;
3309
- let parent = void 0;
3310
- try {
3311
- const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
3312
- if (res.ok) {
3313
- const body = await res.json();
3314
- parent = body && typeof body.parentID === "string" ? body.parentID : null;
3315
- }
3316
- } catch {
3317
- parent = void 0;
3318
- }
3319
- if (parent !== void 0) this.sessionParents.set(sessionId, parent);
3320
- return parent;
3321
- }
3322
- /**
3323
- * DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
3324
- * (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
3325
- *
3326
- * The PRIMARY recovery trigger is "preamble-pinned on recovery ⇒ idle" — a
3327
- * runner restart wipes OpenCode's in-memory `SessionStatus`/`Runner`, so a
3328
- * completed `finish: "tool-calls"` root reply encountered during re-adoption is
3329
- * idle by OpenCode's own definition and is re-dispatched. This method exists only
3330
- * so the WI-3 caller can VETO that re-dispatch in the rare case a descendant is
3331
- * provably in flight at the exact moment of recovery.
3332
- *
3333
- * "Alive" criterion (TIGHTENED): a descendant is alive only when it is PROVABLY,
3334
- * ACTIVELY generating — its LAST message is an assistant still mid-generation
3335
- * (`completed == null`, via `isSessionActivelyGenerating`). An
3336
- * INCOMPLETE-BUT-NOT-GENERATING child — last message a user message, or a
3337
- * completed `finish: "tool-calls"` step — is NOT alive after a restart (nothing
3338
- * is generating once the runner is gone), so it does NOT veto. (This is
3339
- * deliberately NOT `!isTurnComplete`, which also matches those dead-but-non-terminal
3340
- * shapes and would falsely veto — re-hanging the very turn this path recovers.)
3341
- *
3342
- * Return contract (encoded so WI-3 need not re-derive it):
3343
- * - `true` → a descendant is provably, actively generating (veto re-dispatch).
3344
- * - `false` → descendants exist but none is actively generating (the restart
3345
- * case), OR no descendant is found at all.
3346
- * - `null` → liveness is INDETERMINATE (enumeration via `listSessions` failed).
3347
- *
3348
- * ⚠️ `null` (UNKNOWN) MUST NOT be treated as "alive": WI-3 treats `null` the same
3349
- * as `false` and does NOT veto — a restart guarantees no live runner, so an
3350
- * indeterminate cross-check almost always means "couldn't reach a child that no
3351
- * longer exists". The inversion lives in the caller; this method just reports
3352
- * true/false/null faithfully.
3353
- *
3354
- * VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
3355
- * (already proven by the existing child-session interaction tests, via
3356
- * `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
3357
- * terminal state. We do NOT depend on any session-level `busy`/`idle` field —
3358
- * there is none on `GET /session/:id`; OpenCode's busy state is in-memory
3359
- * `SessionStatus` only.
3360
- */
3361
- async isAnyDescendantSessionAlive(rootSessionId) {
3362
- const sessions = await listSessions(this.port);
3363
- if (!sessions) {
3364
- this.log({
3365
- level: "error",
3366
- message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
3367
- });
3368
- return null;
3369
- }
3370
- for (const candidate of sessions) {
3371
- if (!candidate?.id || candidate.id === rootSessionId) continue;
3372
- if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
3373
- const childMsgs = await getSessionMessages(this.port, candidate.id);
3374
- if (isSessionActivelyGenerating(childMsgs)) {
3375
- return true;
3376
- }
3377
- }
3378
- return false;
3379
- }
3380
- /**
3381
- * Cheap decision-telemetry label for a running row's LAST correlated reply
3382
- * (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
3383
- * - `b1` — the reply itself is still in flight (`time.completed == null`) —
3384
- * the aborted-in-flight production bug after a restart.
3385
- * - `b2` — a COMPLETED reply pinned running only by `finish === "tool-calls"`
3386
- * (the sub-agent preamble — #253's shape).
3387
- * - `other` — any other shape (defensive; a running row is normally b1 or b2).
3388
- * Reads `info.time.completed` / `info.finish` (tolerating the legacy top-level
3389
- * shape) directly rather than re-importing the module-private `completedOf`/
3390
- * `finishOf` — this is a display label only, not a correctness predicate.
3391
- */
3392
- replyCompletionShape(reply) {
3393
- if (!reply) return "other";
3394
- const completed = reply.info?.time?.completed ?? reply.time?.completed;
3395
- if (completed == null) return "b1";
3396
- const finish = reply.info?.finish ?? reply.finish;
3397
- return finish === "tool-calls" ? "b2" : "other";
3398
2814
  }
3399
2815
  /**
3400
2816
  * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
@@ -3627,12 +3043,6 @@ var ChannelDriver = class {
3627
3043
  * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
3628
3044
  * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
3629
3045
  * context (no silent catch, per development-workflow).
3630
- *
3631
- * Returns whether the POST SUCCEEDED (2xx). Most callers ignore this (pure
3632
- * telemetry), but the `paused` liveness-clear uses it to know whether to
3633
- * RE-ASSERT on a later tick — a single dropped `paused` POST must not leave a
3634
- * stale `last_seen_alive_at` on a still-paused row (Bugbot "Failed paused signal
3635
- * leaves liveness").
3636
3046
  */
3637
3047
  async postSignal(conversationId, messageId, signal, extra) {
3638
3048
  try {
@@ -3651,9 +3061,7 @@ var ChannelDriver = class {
3651
3061
  conversation_id: conversationId,
3652
3062
  message_id: messageId
3653
3063
  });
3654
- return false;
3655
3064
  }
3656
- return true;
3657
3065
  } catch (err) {
3658
3066
  this.log({
3659
3067
  level: "error",
@@ -3661,7 +3069,6 @@ var ChannelDriver = class {
3661
3069
  conversation_id: conversationId,
3662
3070
  message_id: messageId
3663
3071
  });
3664
- return false;
3665
3072
  }
3666
3073
  }
3667
3074
  async persistSession(conversationId, sessionId) {
@@ -3934,25 +3341,6 @@ async function resolveAgentIdFromKey(authHeader) {
3934
3341
  return { error: `Failed to resolve agent from key: ${message}` };
3935
3342
  }
3936
3343
  }
3937
- async function notifyAgentDisconnected(agentId, authHeader) {
3938
- const apiUrl = getApiUrlConfig();
3939
- try {
3940
- const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
3941
- method: "POST",
3942
- headers: { Authorization: authHeader }
3943
- });
3944
- if (!response.ok) {
3945
- const serverMessage = await readErrorMessage(response);
3946
- return {
3947
- ok: false,
3948
- error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
3949
- };
3950
- }
3951
- return { ok: true };
3952
- } catch (error2) {
3953
- return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
3954
- }
3955
- }
3956
3344
  async function getAgentInfo(agentId, authHeader) {
3957
3345
  const apiUrl = getApiUrlConfig();
3958
3346
  try {
@@ -3999,7 +3387,7 @@ async function getAgentInfo(agentId, authHeader) {
3999
3387
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
4000
3388
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
4001
3389
  var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
4002
- var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
3390
+ var CHANNEL_SETTLE_MS = Number(process.env.EVIDENT_SETTLE_MS) || void 0;
4003
3391
  function log2(state, message, isError = false) {
4004
3392
  if (state.json) {
4005
3393
  console.log(
@@ -4154,7 +3542,7 @@ async function driveChannels(state, driver) {
4154
3542
  logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
4155
3543
  if (state.interactive) displayStatus(state);
4156
3544
  }
4157
- await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
3545
+ await new Promise((resolve) => setTimeout(resolve, CHANNEL_POLL_INTERVAL_MS));
4158
3546
  if (state.idleTimeout !== null && idlePolls >= 2) {
4159
3547
  const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
4160
3548
  if (idleMs > state.idleTimeout * 1e3) {
@@ -4165,122 +3553,8 @@ async function driveChannels(state, driver) {
4165
3553
  }
4166
3554
  }
4167
3555
  }
4168
- var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
4169
- async function runSweep(state, driver, config2) {
4170
- const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
4171
- try {
4172
- const sessions = await listSessions(state.port);
4173
- if (sessions === null) {
4174
- logActivity(state, {
4175
- type: "info",
4176
- message: `Session cleanup: could not list sessions (opencode unreachable); skipping this sweep (${mode})`
4177
- });
4178
- return;
4179
- }
4180
- const toDelete = selectSessionsToDelete(
4181
- sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
4182
- {
4183
- maxAgeMs: config2.maxAgeMs,
4184
- maxCount: config2.maxCount,
4185
- nowMs: Date.now(),
4186
- protectedIds: driver.protectedSessionIds()
4187
- }
4188
- );
4189
- const protectedNow = driver.protectedSessionIds();
4190
- let deleted = 0;
4191
- let failed = 0;
4192
- let skippedNewlyActive = 0;
4193
- for (const id of toDelete) {
4194
- if (protectedNow.has(id)) {
4195
- skippedNewlyActive++;
4196
- logActivity(state, {
4197
- type: "info",
4198
- message: `Session cleanup: skipping ${id} \u2014 became active/bound after selection (${mode})`
4199
- });
4200
- continue;
4201
- }
4202
- if (await deleteSession(state.port, id)) deleted++;
4203
- else failed++;
4204
- }
4205
- const failedNote = failed > 0 ? `, failed ${failed}` : "";
4206
- const skippedNote = skippedNewlyActive > 0 ? `, skipped ${skippedNewlyActive} newly-active` : "";
4207
- logActivity(state, {
4208
- type: "info",
4209
- message: `Session cleanup: inspected ${sessions.length}, deleted ${deleted}${failedNote}${skippedNote} (${mode})`
4210
- });
4211
- } catch (error2) {
4212
- const message = error2 instanceof Error ? error2.message : String(error2);
4213
- logActivity(state, {
4214
- type: "error",
4215
- error: `Session cleanup sweep failed (non-fatal, ${mode}): ${message}`
4216
- });
4217
- }
4218
- }
4219
- function scheduleSessionCleanup(state, driver, options) {
4220
- const config2 = resolveSessionCleanupConfig(
4221
- {
4222
- maxAge: options.sessionCleanupMaxAge,
4223
- maxCount: options.sessionCleanupMaxCount,
4224
- interval: options.sessionCleanupInterval
4225
- },
4226
- process.env
4227
- );
4228
- for (const warning2 of config2.warnings) {
4229
- logActivity(state, { type: "info", message: `Session cleanup: ${warning2}` });
4230
- }
4231
- if (!config2.enabled) return;
4232
- logActivity(state, {
4233
- type: "info",
4234
- message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
4235
- });
4236
- const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
4237
- const firstSweep = setTimeout(
4238
- () => void runSweep(state, driver, config2),
4239
- SESSION_CLEANUP_FIRST_SWEEP_MS
4240
- );
4241
- state.sessionCleanupTimers.push(interval, firstSweep);
4242
- }
4243
- async function notifyOffline(state) {
4244
- if (!state.agentId || !state.authHeader) return;
4245
- if (!state.connected) {
4246
- log2(state, "Skipping offline signal \u2014 this runner does not hold the live tunnel");
4247
- return;
4248
- }
4249
- const result = await notifyAgentDisconnected(state.agentId, state.authHeader);
4250
- if (result.ok) {
4251
- log2(state, "Notified Evident the agent is going offline");
4252
- } else {
4253
- logActivity(state, {
4254
- type: "error",
4255
- error: `Could not notify Evident of offline status (relay will still report it): ${result.error}`
4256
- });
4257
- if (state.interactive) displayStatus(state);
4258
- }
4259
- }
4260
- async function cleanup(state, opts = {}) {
3556
+ async function cleanup(state) {
4261
3557
  state.running = false;
4262
- for (const timer of state.sessionCleanupTimers) {
4263
- clearInterval(timer);
4264
- clearTimeout(timer);
4265
- }
4266
- state.sessionCleanupTimers = [];
4267
- if (opts.graceful && state.channelDriver) {
4268
- state.channelDriver.stop();
4269
- log2(state, "Draining in-flight channel work before shutdown...");
4270
- if (state.interactive) {
4271
- logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
4272
- displayStatus(state);
4273
- }
4274
- const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
4275
- if (!settled) {
4276
- logActivity(state, {
4277
- type: "info",
4278
- message: "Shutdown drain timed out with work still in flight \u2014 leaving it for restart recovery"
4279
- });
4280
- if (state.interactive) displayStatus(state);
4281
- }
4282
- }
4283
- await notifyOffline(state);
4284
3558
  if (state.connection) {
4285
3559
  state.connection.close();
4286
3560
  state.connection = null;
@@ -4311,13 +3585,10 @@ async function run(options) {
4311
3585
  opencodeVersion: null,
4312
3586
  opencodeProcess: null,
4313
3587
  connection: null,
4314
- channelDriver: null,
4315
3588
  running: true,
4316
- shuttingDown: false,
4317
3589
  activityLog: [],
4318
3590
  messageCount: 0,
4319
3591
  lastProxiedActivityAt: null,
4320
- sessionCleanupTimers: [],
4321
3592
  authHeader: ""
4322
3593
  };
4323
3594
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
@@ -4328,15 +3599,13 @@ async function run(options) {
4328
3599
  );
4329
3600
  }
4330
3601
  const handleSignal = async () => {
4331
- if (state.shuttingDown) return;
4332
- state.shuttingDown = true;
4333
3602
  if (state.interactive) {
4334
3603
  logActivity(state, { type: "info", message: "Shutting down..." });
4335
3604
  displayStatus(state);
4336
3605
  } else {
4337
3606
  log2(state, "Shutting down...");
4338
3607
  }
4339
- await cleanup(state, { graceful: true });
3608
+ await cleanup(state);
4340
3609
  await shutdownTelemetry();
4341
3610
  process.exit(0);
4342
3611
  };
@@ -4457,13 +3726,13 @@ async function run(options) {
4457
3726
  getAuthHeader: () => state.authHeader,
4458
3727
  conversationFilter: state.conversationFilter,
4459
3728
  stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
3729
+ settleMs: CHANNEL_SETTLE_MS,
4460
3730
  log: (entry) => logActivity(state, {
4461
3731
  type: entry.level === "error" ? "error" : "info",
4462
3732
  message: entry.message,
4463
3733
  error: entry.level === "error" ? entry.message : void 0
4464
3734
  })
4465
3735
  });
4466
- state.channelDriver = channelDriver;
4467
3736
  const connection = new RunnerConnection({
4468
3737
  agentId: state.agentId,
4469
3738
  getAuthHeader: () => state.authHeader,
@@ -4477,11 +3746,7 @@ async function run(options) {
4477
3746
  type: "info",
4478
3747
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
4479
3748
  });
4480
- emitAgentConnected(state.agentId, {
4481
- port: state.port,
4482
- cli_version: getCliVersion(),
4483
- opencode_version: state.opencodeVersion
4484
- });
3749
+ emitAgentConnected(state.agentId, { port: state.port });
4485
3750
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
4486
3751
  if (state.interactive) displayStatus(state);
4487
3752
  channelDriver.drainPending().then((processed) => {
@@ -4560,12 +3825,10 @@ async function run(options) {
4560
3825
  if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
4561
3826
  throw error2;
4562
3827
  }
4563
- scheduleSessionCleanup(state, channelDriver, options);
4564
3828
  if (!interactive || state.json) {
4565
3829
  log2(state, "Driving channel messages...");
4566
3830
  }
4567
3831
  await driveChannels(state, channelDriver);
4568
- if (state.shuttingDown) return;
4569
3832
  await cleanup(state);
4570
3833
  if (state.json) {
4571
3834
  console.log(
@@ -4580,7 +3843,6 @@ async function run(options) {
4580
3843
  await shutdownTelemetry();
4581
3844
  process.exit(0);
4582
3845
  } catch (error2) {
4583
- if (state.shuttingDown) return;
4584
3846
  await cleanup(state);
4585
3847
  const message = error2 instanceof Error ? error2.message : String(error2);
4586
3848
  if (state.json) {
@@ -4615,16 +3877,7 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
4615
3877
  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);
4616
3878
  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 }));
4617
3879
  program.command("whoami").description("Show the currently logged in user").action(whoami);
4618
- 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(
4619
- "--session-cleanup-max-age <duration>",
4620
- "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
4621
- ).option(
4622
- "--session-cleanup-max-count <n>",
4623
- "Keep only the newest N OpenCode sessions. Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_COUNT"
4624
- ).option(
4625
- "--session-cleanup-interval <duration>",
4626
- "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
4627
- ).action(
3880
+ 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(
4628
3881
  (options) => {
4629
3882
  run({
4630
3883
  agent: options.agent,
@@ -4632,11 +3885,7 @@ program.command("run").description("Connect to Evident and process messages").op
4632
3885
  verbose: options.verbose,
4633
3886
  conversation: options.conversation,
4634
3887
  idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
4635
- json: options.json,
4636
- // Raw strings — the resolver in run.ts single-sources parsing (M1).
4637
- sessionCleanupMaxAge: options.sessionCleanupMaxAge,
4638
- sessionCleanupMaxCount: options.sessionCleanupMaxCount,
4639
- sessionCleanupInterval: options.sessionCleanupInterval
3888
+ json: options.json
4640
3889
  });
4641
3890
  }
4642
3891
  );