@evident-ai/cli 3.0.1-dev.ec55c96 → 3.0.1-dev.edba747

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.js CHANGED
@@ -1,6 +1,7 @@
1
1
  #!/usr/bin/env node
2
2
 
3
3
  // src/index.ts
4
+ import { createRequire } from "module";
4
5
  import { Command } from "commander";
5
6
 
6
7
  // src/commands/login.ts
@@ -470,6 +471,12 @@ import chalk6 from "chalk";
470
471
  import ora3 from "ora";
471
472
  import { select as select3 } from "@inquirer/prompts";
472
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
+
473
480
  // ../../packages/types/src/telemetry/index.ts
474
481
  var TelemetryEventTypes = {
475
482
  // Agent activity events (shown in web UI activity log)
@@ -484,8 +491,34 @@ var TelemetryEventTypes = {
484
491
  var MAX_FRAME_BYTES = 256 * 1024;
485
492
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
486
493
 
494
+ // ../../packages/types/src/logging/index.ts
495
+ var CORRELATION_ID_HEADER = "x-evident-correlation-id";
496
+ function log(level, event, fields) {
497
+ const method = level === "debug" ? "log" : level;
498
+ try {
499
+ console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
500
+ } catch (err) {
501
+ console.error(
502
+ "[evident] log_serialize_failed",
503
+ event,
504
+ err instanceof Error ? err.message : String(err)
505
+ );
506
+ }
507
+ }
508
+ function stripQuery(url) {
509
+ try {
510
+ return new URL(url).pathname;
511
+ } catch {
512
+ const q = url.indexOf("?");
513
+ return q === -1 ? url : url.slice(0, q);
514
+ }
515
+ }
516
+
487
517
  // src/lib/telemetry.ts
488
- var CLI_VERSION = process.env.npm_package_version || "unknown";
518
+ var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
519
+ function getCliVersion() {
520
+ return CLI_VERSION;
521
+ }
489
522
  var eventBuffer = [];
490
523
  var flushTimeout = null;
491
524
  var isShuttingDown = false;
@@ -684,6 +717,19 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
684
717
  return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
685
718
  }
686
719
 
720
+ // src/lib/opencode/opencode-version-gate.ts
721
+ var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
722
+ function isQueueValidatedVersion(version2) {
723
+ if (!version2) return false;
724
+ return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
725
+ }
726
+ function buildOpenCodeVersionWarning(version2) {
727
+ if (isQueueValidatedVersion(version2)) return null;
728
+ const detected = version2 ? `v${version2}` : "unknown";
729
+ const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
730
+ return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
731
+ }
732
+
687
733
  // src/lib/opencode/process.ts
688
734
  import { execSync, spawn } from "child_process";
689
735
  var OPENCODE_PORT_RANGE = [4096, 4097, 4098, 4099, 4100];
@@ -996,23 +1042,37 @@ function roleOf(m) {
996
1042
  }
997
1043
  function completedOf(m) {
998
1044
  if (!m || typeof m !== "object") return void 0;
999
- return m.info?.time?.completed;
1045
+ return m.info?.time?.completed ?? m.time?.completed;
1000
1046
  }
1001
- async function getSessionMessages(port, sessionId) {
1002
- try {
1003
- const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
1004
- if (!res.ok) return null;
1005
- const body = await res.json();
1006
- return Array.isArray(body) ? body : null;
1007
- } catch {
1008
- return null;
1009
- }
1047
+ function createdOf(m) {
1048
+ if (!m || typeof m !== "object") return void 0;
1049
+ return m.info?.time?.created ?? m.time?.created;
1010
1050
  }
1011
- function isTurnComplete(messages) {
1012
- if (!messages || messages.length === 0) return false;
1013
- const last = messages[messages.length - 1];
1014
- if (roleOf(last) !== "assistant") return false;
1015
- return completedOf(last) != null;
1051
+ function idOf(m) {
1052
+ if (!m || typeof m !== "object") return void 0;
1053
+ if (typeof m.id === "string") return m.id;
1054
+ const infoId = m.info?.id;
1055
+ return typeof infoId === "string" ? infoId : void 0;
1056
+ }
1057
+ function parentIdOf(m) {
1058
+ if (!m || typeof m !== "object") return void 0;
1059
+ if (typeof m.parentID === "string") return m.parentID;
1060
+ const infoParent = m.info?.parentID;
1061
+ return typeof infoParent === "string" ? infoParent : void 0;
1062
+ }
1063
+ function finishOf(m) {
1064
+ if (!m || typeof m !== "object") return void 0;
1065
+ if (typeof m.finish === "string") return m.finish;
1066
+ const infoFinish = m.info?.finish;
1067
+ return typeof infoFinish === "string" ? infoFinish : void 0;
1068
+ }
1069
+ function errorOf(m) {
1070
+ if (!m || typeof m !== "object") return void 0;
1071
+ return m.info?.error ?? m.error;
1072
+ }
1073
+ function isAssistantInFlight(m) {
1074
+ if (completedOf(m) == null) return true;
1075
+ return finishOf(m) === "tool-calls";
1016
1076
  }
1017
1077
  async function createOpenCodeSession(port, directory) {
1018
1078
  const url = new URL(`${opencodeBase(port)}/session`);
@@ -1031,8 +1091,9 @@ async function createOpenCodeSession(port, directory) {
1031
1091
  const data = await response.json();
1032
1092
  return data.id;
1033
1093
  }
1034
- async function sendMessageToOpenCode(port, sessionId, content, options, hooks, maxWaitMs = 10 * 60 * 1e3) {
1094
+ async function sendPromptAsync(port, sessionId, content, options, messageId) {
1035
1095
  const body = {
1096
+ messageID: messageId,
1036
1097
  parts: [{ type: "text", text: content }]
1037
1098
  };
1038
1099
  if (options?.agent) {
@@ -1047,79 +1108,111 @@ async function sendMessageToOpenCode(port, sessionId, content, options, hooks, m
1047
1108
  };
1048
1109
  }
1049
1110
  }
1050
- let pollDone = false;
1051
- const reportedQuestions = /* @__PURE__ */ new Set();
1052
- const reportedPermissions = /* @__PURE__ */ new Set();
1053
- const pollInteractive = async () => {
1054
- while (!pollDone) {
1055
- await new Promise((resolve) => setTimeout(resolve, 1e3));
1056
- if (pollDone) break;
1057
- if (hooks?.onQuestion) {
1058
- try {
1059
- const res = await fetch(`${opencodeBase(port)}/question`);
1060
- if (res.ok) {
1061
- const questions = await res.json();
1062
- for (const q of questions) {
1063
- if (q.sessionID === sessionId && !reportedQuestions.has(q.id)) {
1064
- reportedQuestions.add(q.id);
1065
- await hooks.onQuestion(q);
1066
- }
1067
- }
1068
- }
1069
- } catch {
1070
- }
1071
- }
1072
- if (hooks?.onPermission) {
1073
- try {
1074
- const res = await fetch(`${opencodeBase(port)}/permission`);
1075
- if (res.ok) {
1076
- const permissions = await res.json();
1077
- for (const p of permissions) {
1078
- if (p.sessionID === sessionId && !reportedPermissions.has(p.id)) {
1079
- reportedPermissions.add(p.id);
1080
- await hooks.onPermission(p);
1081
- }
1082
- }
1083
- }
1084
- } catch {
1085
- }
1086
- }
1087
- }
1088
- };
1089
- const sendMessage = async () => {
1090
- const controller = new AbortController();
1091
- const timer = setTimeout(() => controller.abort(), maxWaitMs);
1092
- try {
1093
- const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`, {
1094
- method: "POST",
1095
- headers: { "Content-Type": "application/json" },
1096
- body: JSON.stringify(body),
1097
- signal: controller.signal
1098
- });
1099
- if (!res.ok) {
1100
- const text = await res.text().catch(() => "");
1101
- throw new Error(`OpenCode message failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1102
- }
1103
- const sessionRes = await fetch(`${opencodeBase(port)}/session/${sessionId}`).catch(
1104
- () => null
1105
- );
1106
- const session = sessionRes?.ok ? await sessionRes.json() : null;
1107
- const reportedInteraction = reportedQuestions.size > 0 || reportedPermissions.size > 0;
1108
- const turnComplete = isTurnComplete(await getSessionMessages(port, sessionId));
1109
- const awaitingInteraction = reportedInteraction && !turnComplete;
1110
- return { title: session?.title, awaitingInteraction };
1111
- } catch (err) {
1112
- if (err instanceof Error && err.name === "AbortError") {
1113
- throw new Error("Message processing timed out");
1114
- }
1115
- throw err;
1116
- } finally {
1117
- clearTimeout(timer);
1118
- pollDone = true;
1111
+ const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/prompt_async`, {
1112
+ method: "POST",
1113
+ headers: { "Content-Type": "application/json" },
1114
+ body: JSON.stringify(body)
1115
+ });
1116
+ if (res.status < 200 || res.status >= 300) {
1117
+ const text = await res.text().catch(() => "");
1118
+ throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1119
+ }
1120
+ }
1121
+ function findAssistantReplyAfter(messages, userMessageId) {
1122
+ if (!messages || messages.length === 0) return null;
1123
+ const byParent = messages.find(
1124
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) === userMessageId
1125
+ );
1126
+ if (byParent) return byParent;
1127
+ const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1128
+ if (userIndex === -1) return null;
1129
+ for (let i = userIndex + 1; i < messages.length; i++) {
1130
+ if (roleOf(messages[i]) === "assistant") return messages[i];
1131
+ }
1132
+ return null;
1133
+ }
1134
+ function findLastAssistantReplyFor(messages, userMessageId) {
1135
+ if (!messages || messages.length === 0) return null;
1136
+ let lastCorrelated = null;
1137
+ let lastNonErrored = null;
1138
+ for (let i = messages.length - 1; i >= 0; i--) {
1139
+ const m = messages[i];
1140
+ if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
1141
+ if (lastCorrelated === null) lastCorrelated = m;
1142
+ if (errorOf(m) == null) {
1143
+ lastNonErrored = m;
1144
+ break;
1145
+ }
1146
+ }
1147
+ if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
1148
+ const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1149
+ if (userIndex === -1) return null;
1150
+ let last = null;
1151
+ let lastOk = null;
1152
+ for (let i = userIndex + 1; i < messages.length; i++) {
1153
+ const role = roleOf(messages[i]);
1154
+ if (role === "user") break;
1155
+ if (role === "assistant") {
1156
+ last = messages[i];
1157
+ if (errorOf(messages[i]) == null) lastOk = messages[i];
1158
+ }
1159
+ }
1160
+ return lastOk ?? last;
1161
+ }
1162
+ function messageRunState(messages, userMessageId) {
1163
+ if (!messages || messages.length === 0) return "unknown";
1164
+ const hasUser = messages.some((m) => idOf(m) === userMessageId);
1165
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1166
+ if (!hasUser) {
1167
+ if (!reply) return "unknown";
1168
+ }
1169
+ if (!reply) return "queued";
1170
+ if (isAssistantInFlight(reply)) return "running";
1171
+ return errorOf(reply) != null ? "failed" : "done";
1172
+ }
1173
+ function messageError(messages, userMessageId) {
1174
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1175
+ const error2 = errorOf(reply);
1176
+ if (error2 == null) return null;
1177
+ if (typeof error2 === "string") return error2;
1178
+ if (typeof error2 === "object") {
1179
+ const e = error2;
1180
+ const dataMessage = e.data?.message;
1181
+ if (typeof dataMessage === "string") return dataMessage;
1182
+ if (typeof e.message === "string") return e.message;
1183
+ }
1184
+ return "The agent run failed.";
1185
+ }
1186
+ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1187
+ if (!messages || messages.length === 0) return false;
1188
+ return messages.some(
1189
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1190
+ );
1191
+ }
1192
+ function isSessionSettled(messages, now, settleMs) {
1193
+ if (!messages || messages.length === 0) return true;
1194
+ let newestAssistantCreated = null;
1195
+ for (const m of messages) {
1196
+ if (roleOf(m) !== "assistant") continue;
1197
+ if (isAssistantInFlight(m)) return false;
1198
+ const created = createdOf(m);
1199
+ if (typeof created === "number" && (newestAssistantCreated === null || created > newestAssistantCreated)) {
1200
+ newestAssistantCreated = created;
1119
1201
  }
1120
- };
1121
- const [result] = await Promise.all([sendMessage(), pollInteractive()]);
1122
- return result;
1202
+ }
1203
+ if (newestAssistantCreated === null) return true;
1204
+ return now - newestAssistantCreated >= settleMs;
1205
+ }
1206
+ function opencodeMessageIdFor2(queuedMessageId) {
1207
+ return opencodeMessageIdFor(queuedMessageId);
1208
+ }
1209
+ var NATIVE_ID_ALPHABET = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
1210
+ function nativeOpencodeMessageId() {
1211
+ let id = "msg_";
1212
+ for (let i = 0; i < 24; i++) {
1213
+ id += NATIVE_ID_ALPHABET[Math.floor(Math.random() * NATIVE_ID_ALPHABET.length)];
1214
+ }
1215
+ return id;
1123
1216
  }
1124
1217
 
1125
1218
  // src/lib/tunnel/connection.ts
@@ -1190,12 +1283,20 @@ var StreamForwarder = class {
1190
1283
  }
1191
1284
  async handleOpen(frame) {
1192
1285
  const { sid, method, path, headers, has_body } = frame;
1286
+ const correlationId = headers?.[CORRELATION_ID_HEADER];
1287
+ const startedAt = Date.now();
1193
1288
  if (path === TUNNEL_DRAIN_PING_PATH) {
1194
1289
  this.callbacks.onDrainPing?.();
1195
1290
  this.send({ type: "head", sid, status: 204, headers: {} });
1196
1291
  this.send({ type: "res_end", sid });
1197
1292
  return;
1198
1293
  }
1294
+ log("info", "agent_request", {
1295
+ correlation_id: correlationId,
1296
+ sid,
1297
+ method,
1298
+ path: stripQuery(path)
1299
+ });
1199
1300
  const ac = new AbortController();
1200
1301
  let bodyPromise;
1201
1302
  let pushBody;
@@ -1242,6 +1343,12 @@ var StreamForwarder = class {
1242
1343
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1243
1344
  });
1244
1345
  this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
1346
+ log("info", "agent_response", {
1347
+ correlation_id: correlationId,
1348
+ sid,
1349
+ status: upstream.status,
1350
+ duration_ms: Date.now() - startedAt
1351
+ });
1245
1352
  this.callbacks.onHead?.(sid, upstream.status);
1246
1353
  try {
1247
1354
  if (upstream.body) {
@@ -1498,6 +1605,12 @@ var RunnerConnection = class {
1498
1605
  };
1499
1606
 
1500
1607
  // src/lib/channels/driver.ts
1608
+ function messageIdOf(m) {
1609
+ if (!m || typeof m !== "object") return void 0;
1610
+ if (typeof m.id === "string") return m.id;
1611
+ const infoId = m.info?.id;
1612
+ return typeof infoId === "string" ? infoId : void 0;
1613
+ }
1501
1614
  var DEFAULT_RETRY_POLICY = {
1502
1615
  maxAttempts: 6,
1503
1616
  baseDelayMs: 500,
@@ -1505,12 +1618,24 @@ var DEFAULT_RETRY_POLICY = {
1505
1618
  };
1506
1619
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1507
1620
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1621
+ var DEFAULT_DISPATCH_CONFIRM_MS = 6e3;
1622
+ var DEFAULT_STUCK_QUEUED_MS = 6e4;
1623
+ var DEFAULT_STUCK_QUEUED_REDRIVE_MAX = 3;
1624
+ var DEFAULT_SETTLE_MS = 3500;
1508
1625
  var ChannelAuthError = class extends Error {
1509
1626
  constructor(message) {
1510
1627
  super(message);
1511
1628
  this.name = "ChannelAuthError";
1512
1629
  }
1513
1630
  };
1631
+ var ChannelTerminalError = class extends Error {
1632
+ status;
1633
+ constructor(message, status) {
1634
+ super(message);
1635
+ this.name = "ChannelTerminalError";
1636
+ this.status = status;
1637
+ }
1638
+ };
1514
1639
  function backoffDelay(attempt, policy) {
1515
1640
  const exp = policy.baseDelayMs * Math.pow(2, attempt);
1516
1641
  const capped = Math.min(policy.maxDelayMs, exp);
@@ -1531,15 +1656,63 @@ var ChannelDriver = class {
1531
1656
  sleep;
1532
1657
  pausedPollIntervalMs;
1533
1658
  pausedMaxWaitMs;
1659
+ dispatchConfirmMs;
1660
+ stuckQueuedMs;
1661
+ stuckQueuedRedriveMax;
1662
+ settleMs;
1663
+ now;
1534
1664
  /** Cache of conversationId → opencode sessionId. */
1535
1665
  sessions = /* @__PURE__ */ new Map();
1536
1666
  /**
1537
- * Outstanding paused-session watchers, keyed by `message.id` (WI-2-CLI).
1538
- * Single-flight per message: while a watcher is live for a message we never
1539
- * start a second one. The message stays `processing` for the watcher's lifetime
1540
- * so the `?status=pending` drain cannot double-claim it.
1667
+ * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1668
+ * session: one polling loop services all of that session's in-flight messages.
1669
+ * A session entry exists while it has any in-flight (dispatched-but-not-done)
1670
+ * message; it is removed once its in-flight set empties.
1541
1671
  */
1542
1672
  watchers = /* @__PURE__ */ new Map();
1673
+ /**
1674
+ * AUTHORITATIVE local dedup (WI-3): Evident message ids that have been
1675
+ * dispatched and are still in-flight. A message in this set is never
1676
+ * re-`prompt_async`-ed by a subsequent poll tick while it is queued/running.
1677
+ * Backed by a stable minted opencode `messageID` whose duplicate re-enqueue is
1678
+ * idempotent on opencode (PoC fact 9) — so even if this set is lost on restart,
1679
+ * a steady-state-poll re-dispatch will not double-run the message.
1680
+ */
1681
+ dispatched = /* @__PURE__ */ new Set();
1682
+ /**
1683
+ * Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
1684
+ * Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
1685
+ * so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
1686
+ * it is re-adopted and removed when its watcher settles or it is observed off
1687
+ * the processing list.
1688
+ */
1689
+ readopted = /* @__PURE__ */ new Set();
1690
+ /**
1691
+ * "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
1692
+ * re-adopted running/orphan row's watcher hit its `processed_at`-anchored
1693
+ * deadline (or an orphan whose window already elapsed): the still-`processing`
1694
+ * server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
1695
+ * drain until the 15-min cron resets it — spamming new turns.
1696
+ *
1697
+ * CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
1698
+ * does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
1699
+ * in opencode must still be delivered via `markDone` on the next drain — so
1700
+ * `readoptOne` computes `state` FIRST and this set is checked only on the
1701
+ * non-done path. It is cleared once the row leaves the processing list (cron
1702
+ * reset → it drains normally as `pending`), so it can never leak.
1703
+ */
1704
+ dontRedispatch = /* @__PURE__ */ new Set();
1705
+ /**
1706
+ * "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
1707
+ * re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
1708
+ * never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
1709
+ * that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
1710
+ * markDone failure must NOT land here (it must still retry next drain). Separate
1711
+ * from `dontRedispatch` because the two concerns are independent: a row can need
1712
+ * "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
1713
+ * the row leaves the processing list, exactly like `dontRedispatch`.
1714
+ */
1715
+ doneUndeliverable = /* @__PURE__ */ new Set();
1543
1716
  /**
1544
1717
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1545
1718
  * first session creation so drain-created sessions are rooted at the project
@@ -1562,6 +1735,11 @@ var ChannelDriver = class {
1562
1735
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1563
1736
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1564
1737
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
1738
+ this.dispatchConfirmMs = config2.dispatchConfirmMs ?? DEFAULT_DISPATCH_CONFIRM_MS;
1739
+ this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
1740
+ this.stuckQueuedRedriveMax = config2.stuckQueuedRedriveMax ?? DEFAULT_STUCK_QUEUED_REDRIVE_MAX;
1741
+ this.settleMs = config2.settleMs ?? DEFAULT_SETTLE_MS;
1742
+ this.now = config2.now ?? (() => Date.now());
1565
1743
  }
1566
1744
  /** The IPv4-loopback base URL for the local `opencode serve`. */
1567
1745
  get opencodeBase() {
@@ -1571,16 +1749,16 @@ var ChannelDriver = class {
1571
1749
  // Public API
1572
1750
  // -------------------------------------------------------------------------
1573
1751
  /**
1574
- * Drain all pending channel conversations once: poll → processcallback.
1752
+ * Drain all pending channel conversations once: poll → dispatchregister.
1575
1753
  * Called on tunnel `connected` (WI-CHAN-4) and on each poll tick by `run.ts`.
1576
1754
  * Re-entrant calls while a drain is in flight are skipped (return 0).
1577
1755
  *
1578
- * @returns the number of messages processed.
1756
+ * @returns the number of messages NEWLY dispatched to opencode's native queue.
1579
1757
  */
1580
1758
  async drainPending() {
1581
1759
  if (this.draining) return 0;
1582
1760
  this.draining = true;
1583
- let processed = 0;
1761
+ let dispatched = 0;
1584
1762
  try {
1585
1763
  const conversations = await this.getPendingConversations();
1586
1764
  if (conversations.length > 0) {
@@ -1591,95 +1769,106 @@ var ChannelDriver = class {
1591
1769
  });
1592
1770
  }
1593
1771
  for (const conv of conversations) {
1594
- processed += await this.processConversation(conv);
1772
+ dispatched += await this.processConversation(conv);
1595
1773
  }
1774
+ await this.readoptProcessing();
1596
1775
  } finally {
1597
1776
  this.draining = false;
1598
1777
  }
1599
- return processed;
1778
+ return dispatched;
1600
1779
  }
1601
1780
  /**
1602
- * Await all outstanding paused-session watchers (WI-2-CLI).
1781
+ * True while any per-session watcher has a non-empty in-flight dispatched set
1782
+ * (Task 3.7). `run.ts` treats this as NON-idle so `--idle-timeout` cannot exit
1783
+ * the process while a dispatched message is still queued/running — which would
1784
+ * kill the turn and orphan its reply.
1785
+ */
1786
+ hasInFlightWatchers() {
1787
+ for (const watcher of this.watchers.values()) {
1788
+ if (watcher.inFlight.size > 0) return true;
1789
+ }
1790
+ return false;
1791
+ }
1792
+ /**
1793
+ * Await all outstanding per-session watchers (WI-3).
1603
1794
  *
1604
- * In production the watchers are deliberately started-not-awaited so the drain
1605
- * loop never blocks on them and process exit is not held up (the cron recovers
1606
- * any abandoned ones). This helper exists primarily for deterministic tests
1607
- * that need to observe the watcher's effect (the `done` PATCH or its giving up)
1608
- * after a non-blocking `drainPending`. Watchers never reject, so this resolves.
1795
+ * In production the watcher loops are deliberately started-not-awaited so the
1796
+ * drain loop never blocks on them and process exit is not held up (the cron
1797
+ * recovers any abandoned ones). This helper exists primarily for deterministic
1798
+ * tests that need to observe a watcher's effect (the `processing`/`done` PATCH
1799
+ * or its giving up) after a non-blocking `drainPending`. Watcher loops never
1800
+ * reject, so this resolves.
1609
1801
  */
1610
1802
  async flushPausedWatchers() {
1611
- await Promise.all([...this.watchers.values()]);
1803
+ while (true) {
1804
+ const loops = [...this.watchers.values()].map((w) => w.loop).filter((l) => l != null);
1805
+ if (loops.length === 0) return;
1806
+ await Promise.all(loops);
1807
+ const stillLive = [...this.watchers.values()].some((w) => w.loop != null);
1808
+ if (!stillLive) return;
1809
+ }
1612
1810
  }
1613
1811
  // -------------------------------------------------------------------------
1614
- // Conversation processing
1812
+ // Conversation processing (WI-3 — async dispatch)
1615
1813
  // -------------------------------------------------------------------------
1814
+ /**
1815
+ * Dispatch each pending message for a conversation to opencode's native queue
1816
+ * via `prompt_async` (Task 3.2) and register it with the conversation's
1817
+ * per-session watcher. Does NOT block on the turn and does NOT call
1818
+ * `markProcessing` here — that fires from the watcher on running-start.
1819
+ *
1820
+ * @returns the count of messages NEWLY dispatched (not already in-flight).
1821
+ */
1616
1822
  async processConversation(conv) {
1617
1823
  const sessionId = await this.ensureSession(conv);
1618
1824
  const messages = await this.getPendingMessages(conv.id);
1619
- let processed = 0;
1825
+ let dispatched = 0;
1826
+ let skippedAlreadyDispatched = 0;
1620
1827
  for (const message of messages) {
1621
- const claimed = await this.markProcessing(conv.id, message.id, sessionId);
1622
- if (!claimed) {
1623
- this.log({
1624
- level: "info",
1625
- message: `Message ${message.id.slice(0, 8)} already claimed \u2014 skipping`,
1626
- conversation_id: conv.id,
1627
- message_id: message.id
1628
- });
1828
+ if (this.dispatched.has(message.id)) {
1829
+ skippedAlreadyDispatched += 1;
1629
1830
  continue;
1630
1831
  }
1832
+ const opencodeMessageId = opencodeMessageIdFor2(message.id);
1833
+ const options = {
1834
+ agent: message.opencode_agent ?? void 0,
1835
+ model: message.opencode_model ?? void 0
1836
+ };
1631
1837
  try {
1632
1838
  this.log({
1633
1839
  level: "info",
1634
- message: `Sending queued message ${message.id.slice(0, 8)} to OpenCode (session ${sessionId.slice(0, 8)})`,
1635
- conversation_id: conv.id,
1636
- message_id: message.id
1637
- });
1638
- const result = await sendMessageToOpenCode(
1639
- this.port,
1640
- sessionId,
1641
- message.content,
1642
- {
1643
- agent: message.opencode_agent ?? void 0,
1644
- model: message.opencode_model ?? void 0
1645
- },
1646
- {
1647
- onQuestion: (question) => this.reportInteraction(conv.id, "question", question),
1648
- onPermission: (permission) => this.reportInteraction(conv.id, "permission", permission)
1649
- }
1650
- );
1651
- if (result.awaitingInteraction) {
1652
- this.log({
1653
- level: "info",
1654
- message: `Message ${message.id.slice(0, 8)} paused awaiting interaction \u2014 watching session for completion`,
1655
- conversation_id: conv.id,
1656
- message_id: message.id
1657
- });
1658
- this.startPausedWatcher(conv, message, sessionId);
1659
- continue;
1660
- }
1661
- await this.confirmCompletion(sessionId);
1662
- await this.markDone(conv.id, message.id, sessionId);
1663
- processed += 1;
1664
- this.log({
1665
- level: "info",
1666
- message: `Message ${message.id.slice(0, 8)} processed`,
1840
+ message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
1667
1841
  conversation_id: conv.id,
1668
1842
  message_id: message.id
1669
1843
  });
1844
+ await sendPromptAsync(this.port, sessionId, message.content, options, opencodeMessageId);
1670
1845
  } catch (err) {
1671
1846
  if (err instanceof ChannelAuthError) throw err;
1847
+ this.dispatched.delete(message.id);
1672
1848
  await this.markFailed(conv.id, message.id).catch(() => {
1673
1849
  });
1674
1850
  this.log({
1675
1851
  level: "error",
1676
- message: `Message ${message.id.slice(0, 8)} failed: ${err instanceof Error ? err.message : String(err)}`,
1852
+ message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
1677
1853
  conversation_id: conv.id,
1678
1854
  message_id: message.id
1679
1855
  });
1856
+ continue;
1680
1857
  }
1858
+ this.dispatched.add(message.id);
1859
+ this.registerInFlight(conv, sessionId, message, opencodeMessageId);
1860
+ dispatched += 1;
1861
+ void this.postSignal(conv.id, message.id, "dispatched");
1681
1862
  }
1682
- return processed;
1863
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
1864
+ this.log({
1865
+ level: "error",
1866
+ message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
1867
+ conversation_id: conv.id
1868
+ });
1869
+ }
1870
+ this.ensureWatcherRunning(sessionId);
1871
+ return dispatched;
1683
1872
  }
1684
1873
  async ensureSession(conv) {
1685
1874
  const cached = this.sessions.get(conv.id);
@@ -1711,111 +1900,974 @@ var ChannelDriver = class {
1711
1900
  }
1712
1901
  return this.opencodeDirectory;
1713
1902
  }
1903
+ // -------------------------------------------------------------------------
1904
+ // Per-session watcher (WI-3)
1905
+ // -------------------------------------------------------------------------
1906
+ /** Register a freshly-dispatched message with its session's watcher state. */
1907
+ registerInFlight(conv, sessionId, message, opencodeMessageId) {
1908
+ let watcher = this.watchers.get(sessionId);
1909
+ if (!watcher) {
1910
+ watcher = {
1911
+ conv,
1912
+ inFlight: /* @__PURE__ */ new Map(),
1913
+ loop: null,
1914
+ reportedQuestions: /* @__PURE__ */ new Set(),
1915
+ reportedPermissions: /* @__PURE__ */ new Set()
1916
+ };
1917
+ this.watchers.set(sessionId, watcher);
1918
+ }
1919
+ const now = this.now();
1920
+ watcher.inFlight.set(message.id, {
1921
+ evidentMessageId: message.id,
1922
+ opencodeMessageId,
1923
+ message,
1924
+ dispatchedAt: now,
1925
+ deadline: now + this.pausedMaxWaitMs,
1926
+ started: false,
1927
+ done: false,
1928
+ stuckReported: false,
1929
+ redriveAttempts: 0,
1930
+ lastRedriveAt: null,
1931
+ redriveOpencodeMessageId: null,
1932
+ attemptedOpencodeMessageIds: [opencodeMessageId]
1933
+ });
1934
+ }
1714
1935
  /**
1715
- * Local reconcile: re-query `GET /session/:id/message` and check whether the
1716
- * last assistant message is message-level complete (`isTurnComplete`).
1936
+ * Register a RE-ADOPTED `processing` message with its session watcher
1937
+ * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
1938
+ * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
1939
+ * `now`: a row already `processing` for e.g. 5 min must give up ~5 min from now
1940
+ * (10 min after `processed_at`), not 10 min from now — otherwise its deadline
1941
+ * lands ~15 min after `processed_at`, coinciding with the cron reset →
1942
+ * double-drive race. `dispatchedAt` stays `now` (only the appear-guard uses it).
1943
+ *
1944
+ * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
1945
+ * `opencodeMessageId` is the id the watcher polls for a reply — for the orphan
1946
+ * fresh-run path these differ (a fresh opencode id under the same server row).
1717
1947
  *
1718
- * This is a PURE OBSERVABILITY probe on the normal (non-paused) path: the
1719
- * blocking POST already returned, and `markDone` causes the server to re-fetch
1720
- * the messages itself (via `extractTextFromMessages`) when delivering the
1721
- * reply — so this round-trip never gates delivery. We keep it only to surface a
1722
- * truthful diagnostic when opencode hasn't yet recorded a completed assistant
1723
- * turn at reconcile time, then proceed to `markDone` regardless. Uses the
1724
- * injected `fetchImpl` and reuses only the pure `isTurnComplete` predicate.
1948
+ * `started` is set true so the watcher does NOT re-`markProcessing` a row the
1949
+ * server already flipped to `processing`; the running/done transitions still
1950
+ * fire from the watcher's normal branches.
1951
+ */
1952
+ registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs, redriveOpencodeMessageId = null) {
1953
+ let watcher = this.watchers.get(sessionId);
1954
+ if (!watcher) {
1955
+ watcher = {
1956
+ conv,
1957
+ inFlight: /* @__PURE__ */ new Map(),
1958
+ loop: null,
1959
+ reportedQuestions: /* @__PURE__ */ new Set(),
1960
+ reportedPermissions: /* @__PURE__ */ new Set()
1961
+ };
1962
+ this.watchers.set(sessionId, watcher);
1963
+ }
1964
+ watcher.inFlight.set(message.id, {
1965
+ evidentMessageId: message.id,
1966
+ opencodeMessageId,
1967
+ message,
1968
+ dispatchedAt: this.now(),
1969
+ deadline: processedAtMs + this.pausedMaxWaitMs,
1970
+ // The server row is ALREADY `processing`; do not re-fire markProcessing.
1971
+ started: true,
1972
+ done: false,
1973
+ // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
1974
+ // AND the stuck-queued observer now INCLUDES re-adopted queued wedges: it
1975
+ // gates on `state === 'queued'` (turn produced no reply), not on `started`,
1976
+ // so a re-adopted row left wedged in `queued` still emits the signal once
1977
+ // (queued-followup-redrive, #210). A re-adopted `queued` wedge is ALSO
1978
+ // re-driven by the same path (ADR-0046 §c deferred the live-session redrive
1979
+ // to here) — `redriveAttempts` starts fresh so it gets the full budget.
1980
+ stuckReported: false,
1981
+ redriveAttempts: 0,
1982
+ lastRedriveAt: null,
1983
+ redriveOpencodeMessageId,
1984
+ attemptedOpencodeMessageIds: [opencodeMessageId]
1985
+ });
1986
+ }
1987
+ /**
1988
+ * Start (but do NOT await) the per-session watcher loop if it has in-flight
1989
+ * work and is not already running. Single-flight per session. The loop is
1990
+ * tracked on the watcher and cleared when it settles; it never rejects (fully
1991
+ * guarded), so a failed poll/callback can never crash the run loop — the cron
1992
+ * stays as the safety net.
1993
+ */
1994
+ ensureWatcherRunning(sessionId) {
1995
+ const watcher = this.watchers.get(sessionId);
1996
+ if (!watcher) return;
1997
+ if (watcher.loop) return;
1998
+ if (watcher.inFlight.size === 0) {
1999
+ this.watchers.delete(sessionId);
2000
+ return;
2001
+ }
2002
+ const loop = this.runWatcherLoop(sessionId, watcher).finally(() => {
2003
+ watcher.loop = null;
2004
+ if (watcher.inFlight.size === 0) {
2005
+ this.watchers.delete(sessionId);
2006
+ }
2007
+ });
2008
+ watcher.loop = loop;
2009
+ }
2010
+ /**
2011
+ * The per-session polling loop (WI-3). Once per tick it:
2012
+ * 1. polls `GET /session/:id/message` once and, per in-flight message,
2013
+ * computes `messageRunState` and fires markProcessing (queued→running) /
2014
+ * markDone (done) exactly once per transition;
2015
+ * 2. applies the idle-path re-dispatch guard (a dispatched message that never
2016
+ * APPEARS → re-dispatch — D1 obligation 2);
2017
+ * 3. polls `/question` + `/permission` (scoped to the session) and surfaces
2018
+ * NEW ones via `reportInteraction`, carrying the PAUSED message's own
2019
+ * `source_message_id`;
2020
+ * 4. drops messages that completed or timed out from the in-flight set.
2021
+ * Exits when the in-flight set empties. Never throws.
1725
2022
  */
1726
- async confirmCompletion(sessionId) {
2023
+ async runWatcherLoop(sessionId, watcher) {
1727
2024
  try {
1728
- const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
1729
- if (!res.ok) return;
1730
- const body = await res.json();
1731
- const messages = Array.isArray(body) ? body : null;
1732
- if (!isTurnComplete(messages)) {
2025
+ while (watcher.inFlight.size > 0) {
2026
+ await this.sleep(this.pausedPollIntervalMs);
2027
+ let messages = null;
2028
+ try {
2029
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
2030
+ if (res.ok) {
2031
+ const body = await res.json();
2032
+ messages = Array.isArray(body) ? body : null;
2033
+ }
2034
+ } catch {
2035
+ continue;
2036
+ }
2037
+ for (const inFlight of [...watcher.inFlight.values()]) {
2038
+ await this.serviceInFlightMessage(sessionId, watcher, inFlight, messages);
2039
+ }
2040
+ await this.pollInteractions(sessionId, watcher, messages);
2041
+ }
2042
+ } catch (err) {
2043
+ if (err instanceof ChannelAuthError) {
2044
+ this.log({
2045
+ level: "error",
2046
+ message: `Session watcher aborted on auth failure for session ${sessionId.slice(0, 8)} \u2014 clearing in-flight state for re-drive after re-auth: ${err.message}`,
2047
+ conversation_id: watcher.conv.id
2048
+ });
2049
+ for (const evidentMessageId of [...watcher.inFlight.keys()]) {
2050
+ this.readopted.delete(evidentMessageId);
2051
+ this.removeInFlight(watcher, evidentMessageId);
2052
+ }
2053
+ return;
2054
+ }
2055
+ this.log({
2056
+ level: "error",
2057
+ message: `Session watcher failed for session ${sessionId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
2058
+ conversation_id: watcher.conv.id
2059
+ });
2060
+ }
2061
+ }
2062
+ /**
2063
+ * Drive ONE in-flight message's lifecycle from the tick's message snapshot.
2064
+ * Fires markProcessing on queued→running and markDone on done (each once),
2065
+ * applies the idle-path re-dispatch guard, and removes the message from the
2066
+ * in-flight set on completion or timeout.
2067
+ */
2068
+ async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
2069
+ const conv = watcher.conv;
2070
+ const state = messageRunState(messages, inFlight.opencodeMessageId);
2071
+ if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
2072
+ let claimed;
2073
+ try {
2074
+ claimed = await this.markProcessing(
2075
+ conv.id,
2076
+ inFlight.evidentMessageId,
2077
+ sessionId,
2078
+ inFlight.redriveOpencodeMessageId
2079
+ );
2080
+ } catch (err) {
2081
+ if (err instanceof ChannelAuthError) throw err;
2082
+ this.log({
2083
+ level: "error",
2084
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} processing (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2085
+ conversation_id: conv.id,
2086
+ message_id: inFlight.evidentMessageId
2087
+ });
2088
+ return;
2089
+ }
2090
+ inFlight.started = true;
2091
+ if (!claimed) {
1733
2092
  this.log({
1734
2093
  level: "info",
1735
- message: `Session ${sessionId.slice(0, 8)} messages do not yet show a completed assistant turn on reconcile \u2014 delivering anyway`
2094
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} already marked processing \u2014 continuing`,
2095
+ conversation_id: conv.id,
2096
+ message_id: inFlight.evidentMessageId
1736
2097
  });
1737
2098
  }
1738
- } catch {
2099
+ }
2100
+ if (state === "done") {
2101
+ if (!inFlight.done) {
2102
+ this.log({
2103
+ level: "info",
2104
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
2105
+ conversation_id: conv.id,
2106
+ message_id: inFlight.evidentMessageId
2107
+ });
2108
+ try {
2109
+ await this.markDone(
2110
+ conv.id,
2111
+ inFlight.evidentMessageId,
2112
+ sessionId,
2113
+ inFlight.redriveOpencodeMessageId
2114
+ );
2115
+ } catch (err) {
2116
+ if (err instanceof ChannelAuthError) throw err;
2117
+ if (err instanceof ChannelTerminalError) {
2118
+ this.log({
2119
+ level: "error",
2120
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2121
+ conversation_id: conv.id,
2122
+ message_id: inFlight.evidentMessageId
2123
+ });
2124
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2125
+ return;
2126
+ }
2127
+ if (this.now() >= inFlight.deadline) {
2128
+ this.log({
2129
+ level: "error",
2130
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
2131
+ conversation_id: conv.id,
2132
+ message_id: inFlight.evidentMessageId
2133
+ });
2134
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2135
+ return;
2136
+ }
2137
+ this.log({
2138
+ level: "error",
2139
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2140
+ conversation_id: conv.id,
2141
+ message_id: inFlight.evidentMessageId
2142
+ });
2143
+ return;
2144
+ }
2145
+ inFlight.done = true;
2146
+ }
2147
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2148
+ return;
2149
+ }
2150
+ if (state === "failed") {
2151
+ if (!inFlight.done) {
2152
+ const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
2153
+ this.log({
2154
+ level: "error",
2155
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2156
+ conversation_id: conv.id,
2157
+ message_id: inFlight.evidentMessageId
2158
+ });
2159
+ try {
2160
+ await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
2161
+ } catch (err) {
2162
+ if (err instanceof ChannelAuthError) throw err;
2163
+ if (err instanceof ChannelTerminalError) {
2164
+ this.log({
2165
+ level: "error",
2166
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2167
+ conversation_id: conv.id,
2168
+ message_id: inFlight.evidentMessageId
2169
+ });
2170
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2171
+ return;
2172
+ }
2173
+ if (this.now() >= inFlight.deadline) {
2174
+ this.log({
2175
+ level: "error",
2176
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
2177
+ conversation_id: conv.id,
2178
+ message_id: inFlight.evidentMessageId
2179
+ });
2180
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2181
+ return;
2182
+ }
2183
+ this.log({
2184
+ level: "error",
2185
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2186
+ conversation_id: conv.id,
2187
+ message_id: inFlight.evidentMessageId
2188
+ });
2189
+ return;
2190
+ }
2191
+ inFlight.done = true;
2192
+ }
2193
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2194
+ return;
2195
+ }
2196
+ if (state === "unknown") {
2197
+ if (this.now() - inFlight.dispatchedAt >= this.dispatchConfirmMs) {
2198
+ await this.redispatchInFlight(conv.id, sessionId, inFlight);
2199
+ }
2200
+ }
2201
+ const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
2202
+ const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
2203
+ if (state === "queued" && pastStuckBound && sessionIdle) {
2204
+ if (!inFlight.stuckReported) {
2205
+ inFlight.stuckReported = true;
2206
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2207
+ stuck_for_ms: this.now() - inFlight.dispatchedAt
2208
+ });
2209
+ }
2210
+ const dueForRedrive = inFlight.lastRedriveAt == null || this.now() - inFlight.lastRedriveAt >= this.stuckQueuedMs;
2211
+ if (dueForRedrive && isSessionSettled(messages, this.now(), this.settleMs)) {
2212
+ const removed = await this.redriveStuckQueued(sessionId, watcher, inFlight, messages);
2213
+ if (removed) return;
2214
+ }
2215
+ }
2216
+ if (this.now() >= inFlight.deadline) {
2217
+ this.log({
2218
+ level: "info",
2219
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} did not complete within the watch window \u2014 leaving for the cron safety net`,
2220
+ conversation_id: conv.id,
2221
+ message_id: inFlight.evidentMessageId
2222
+ });
2223
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2224
+ watched_for_ms: this.now() - inFlight.dispatchedAt
2225
+ });
2226
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
1739
2227
  }
1740
2228
  }
1741
- // -------------------------------------------------------------------------
1742
- // Paused-session watcher (WI-2-CLI)
1743
- // -------------------------------------------------------------------------
1744
2229
  /**
1745
- * Start (but do NOT await) a watcher that resumes a paused turn to completion.
1746
- *
1747
- * Single-flight per message: if a watcher is already live for this message we
1748
- * skip. The returned watcher promise is tracked in `this.watchers` and removed
1749
- * when it settles; it never rejects (the body is fully guarded), so a failed
1750
- * poll/markDone can never crash the run loop — the cron stays as the safety net.
2230
+ * Re-dispatch a message whose user row never appeared (idle-path guard). Safe:
2231
+ * opencode treats a duplicate caller-supplied `messageID` as idempotent (PoC
2232
+ * fact 9) — one user message + one reply even if the original DID land. Resets
2233
+ * the dispatch timestamp so the guard doesn't immediately fire again.
1751
2234
  */
1752
- startPausedWatcher(conv, message, sessionId) {
1753
- if (this.watchers.has(message.id)) return;
1754
- const watcher = this.watchPausedSession(conv, message, sessionId).finally(() => {
1755
- this.watchers.delete(message.id);
2235
+ async redispatchInFlight(conversationId, sessionId, inFlight) {
2236
+ const options = {
2237
+ agent: inFlight.message.opencode_agent ?? void 0,
2238
+ model: inFlight.message.opencode_model ?? void 0
2239
+ };
2240
+ this.log({
2241
+ level: "info",
2242
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} not observed after dispatch \u2014 re-dispatching (idle-path guard)`,
2243
+ message_id: inFlight.evidentMessageId
1756
2244
  });
1757
- this.watchers.set(message.id, watcher);
2245
+ void this.postSignal(conversationId, inFlight.evidentMessageId, "redispatched");
2246
+ try {
2247
+ await sendPromptAsync(
2248
+ this.port,
2249
+ sessionId,
2250
+ inFlight.message.content,
2251
+ options,
2252
+ inFlight.opencodeMessageId
2253
+ );
2254
+ } catch (err) {
2255
+ this.log({
2256
+ level: "error",
2257
+ message: `Re-dispatch failed for message ${inFlight.evidentMessageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
2258
+ message_id: inFlight.evidentMessageId
2259
+ });
2260
+ }
2261
+ inFlight.dispatchedAt = this.now();
1758
2262
  }
1759
2263
  /**
1760
- * Poll `GET /session/:id/message` until the SAME paused turn is message-level
1761
- * complete the last message is an ASSISTANT message whose
1762
- * `info.time.completed` is set (the user answered the question/permission in
1763
- * opencode web and opencode finished the turn) then complete it via the
1764
- * EXISTING `markDone` PATCHNO re-send of the original message. The server
1765
- * re-fetches the assistant messages on completion, so the watcher does not
1766
- * pass any reply text.
2264
+ * RE-DRIVE a stuck-`queued` follow-up so opencode actually runs it
2265
+ * (queued-followup-redrive; the fix for the 2026-07-18 dev incident).
2266
+ *
2267
+ * The wedge: a follow-up `prompt_async`'d mid-turn / in the post-turn settling
2268
+ * window is orphanedits user message persists but the turn never runs
2269
+ * (`messageRunState === 'queued'`). Proven against real opencode 1.18.3 by an
2270
+ * isolated, interleaved experiment (`.harness/followup-send-mechanism-finding.md`):
2271
+ * the un-sticking variable is the `messageID` FORMAT. A follow-up carrying the
2272
+ * runner's CUSTOM underscore id (`msg_<sanitized-uuid>`) is NEVER picked up mid-
2273
+ * turn (0/24 trials); the SAME follow-up carrying a NATIVE-format id
2274
+ * (`msg_`+24 base62) runs once the prior turn settles (~50% locally, reliably on
2275
+ * dev). `parts[].id` and `agent`/`model` were proven IRRELEVANT.
2276
+ *
2277
+ * So we re-`prompt_async` into the SAME opencode session (opencode-web does
2278
+ * exactly this and it works — UPDATE 2 in the investigation doc), preserving the
2279
+ * conversation's history/continuity: we do NOT create a fresh session and we do
2280
+ * NOT overwrite the conversation's `opencode_session_id`. Only the opencode
2281
+ * user-message id changes — to a fresh native id per attempt (opencode's
2282
+ * caller-supplied id dedup is global+permanent, so each attempt needs a never-
2283
+ * seen id).
1767
2284
  *
1768
- * Fetch seam: the poll uses the injected `this.fetchImpl` (preserving the
1769
- * tests' injection) and reuses only the pure `isTurnComplete` predicate from
1770
- * `session.ts` we do NOT call `getSessionMessages` (which uses the global
1771
- * `fetch`) here.
2285
+ * Reply correlation is PRESERVED: the native id is random and NOT re-derivable
2286
+ * from the row id, so the runner carries it to the server (on markProcessing/
2287
+ * markDone via `redriveOpencodeMessageId`), which persists it on the row and
2288
+ * correlates the reply by THAT id (`conversation-notification.ts`). A normal,
2289
+ * never-re-driven message still correlates by the derived stable id — unchanged.
2290
+ *
2291
+ * Bounded to `stuckQueuedRedriveMax` attempts. On exhaustion the row is marked
2292
+ * FAILED (existing channel failure affordance) so the user is TOLD it could not
2293
+ * be answered rather than left silent, and it is removed from the in-flight set.
2294
+ *
2295
+ * @returns true if the message LEFT this watcher's in-flight set (gave up +
2296
+ * marked failed) — the caller then stops servicing it this tick. Returns false
2297
+ * when the follow-up stays in THIS watcher (re-driven in place, or the re-drive
2298
+ * send failed and the next window retries).
2299
+ */
2300
+ async redriveStuckQueued(sessionId, watcher, inFlight, messages) {
2301
+ if (await this.deliverIfAnyAttemptCompleted(sessionId, watcher, inFlight, messages)) {
2302
+ return true;
2303
+ }
2304
+ if (inFlight.redriveAttempts >= this.stuckQueuedRedriveMax) {
2305
+ this.log({
2306
+ level: "error",
2307
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} stuck queued after ${inFlight.redriveAttempts} re-drive attempt(s) \u2014 marking failed`,
2308
+ conversation_id: watcher.conv.id,
2309
+ message_id: inFlight.evidentMessageId
2310
+ });
2311
+ try {
2312
+ await this.markFailed(watcher.conv.id, inFlight.evidentMessageId);
2313
+ } catch (err) {
2314
+ if (err instanceof ChannelAuthError) throw err;
2315
+ this.log({
2316
+ level: "error",
2317
+ 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)}`,
2318
+ conversation_id: watcher.conv.id,
2319
+ message_id: inFlight.evidentMessageId
2320
+ });
2321
+ }
2322
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2323
+ return true;
2324
+ }
2325
+ inFlight.redriveAttempts += 1;
2326
+ inFlight.lastRedriveAt = this.now();
2327
+ this.log({
2328
+ level: "info",
2329
+ 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})`,
2330
+ conversation_id: watcher.conv.id,
2331
+ message_id: inFlight.evidentMessageId
2332
+ });
2333
+ void this.postSignal(watcher.conv.id, inFlight.evidentMessageId, "redriven", {
2334
+ redrive_attempt: inFlight.redriveAttempts
2335
+ });
2336
+ const options = {
2337
+ agent: inFlight.message.opencode_agent ?? void 0,
2338
+ model: inFlight.message.opencode_model ?? void 0
2339
+ };
2340
+ const nativeId = nativeOpencodeMessageId();
2341
+ try {
2342
+ await sendPromptAsync(this.port, sessionId, inFlight.message.content, options, nativeId);
2343
+ } catch (err) {
2344
+ if (err instanceof ChannelAuthError) throw err;
2345
+ inFlight.redriveAttempts -= 1;
2346
+ this.log({
2347
+ level: "error",
2348
+ message: `Re-drive failed for stuck message ${inFlight.evidentMessageId.slice(0, 8)} (will retry next window): ${err instanceof Error ? err.message : String(err)}`,
2349
+ conversation_id: watcher.conv.id,
2350
+ message_id: inFlight.evidentMessageId
2351
+ });
2352
+ return false;
2353
+ }
2354
+ inFlight.opencodeMessageId = nativeId;
2355
+ inFlight.redriveOpencodeMessageId = nativeId;
2356
+ inFlight.attemptedOpencodeMessageIds.push(nativeId);
2357
+ inFlight.started = false;
2358
+ return false;
2359
+ }
2360
+ /**
2361
+ * Finding 2 (Bugbot #217): scan EVERY opencode id this stuck message has been
2362
+ * driven under (`attemptedOpencodeMessageIds`) for a COMPLETED correlated reply
2363
+ * in the tick's snapshot; if one is found, markDone off THAT id (so the reply an
2364
+ * earlier re-drive attempt eventually produced is delivered) instead of
2365
+ * re-driving again or marking failed. Idempotent + guarded like the watcher's
2366
+ * done branch. `@returns` true when the message left the in-flight set (delivered
2367
+ * or terminally-undeliverable) — the caller then stops servicing it this tick.
1772
2368
  *
1773
- * Bounded by `pausedMaxWaitMs` (default 10 min, strictly < the 15-min cron
1774
- * reset): on timeout we STOP and leave the message `processing` so the cron
1775
- * remains the last-resort safety net. The whole body is wrapped so any
1776
- * poll/markDone failure is logged and swallowed — a watcher MUST NEVER throw
1777
- * out of the run loop.
2369
+ * The LATEST id is normally handled by `serviceInFlightMessage`'s own `done`
2370
+ * branch; this covers the ids a re-drive OVERWROTE, which that branch no longer
2371
+ * polls. `messageRunState === 'done'` means a completed, non-errored correlated
2372
+ * reply exists for that id.
1778
2373
  */
1779
- async watchPausedSession(conv, message, sessionId) {
1780
- const deadline = Date.now() + this.pausedMaxWaitMs;
2374
+ async deliverIfAnyAttemptCompleted(sessionId, watcher, inFlight, messages) {
2375
+ const completedId = inFlight.attemptedOpencodeMessageIds.find(
2376
+ (id) => messageRunState(messages, id) === "done"
2377
+ );
2378
+ if (!completedId) return false;
2379
+ if (inFlight.done) {
2380
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2381
+ return true;
2382
+ }
2383
+ this.log({
2384
+ level: "info",
2385
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed under an earlier re-drive attempt's id \u2014 marking done (not re-driving/failing)`,
2386
+ conversation_id: watcher.conv.id,
2387
+ message_id: inFlight.evidentMessageId
2388
+ });
1781
2389
  try {
1782
- while (Date.now() < deadline) {
1783
- await this.sleep(this.pausedPollIntervalMs);
1784
- let completed = false;
1785
- try {
1786
- const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
1787
- if (res.ok) {
1788
- const body = await res.json();
1789
- const messages = Array.isArray(body) ? body : null;
1790
- completed = isTurnComplete(messages);
2390
+ await this.markDone(watcher.conv.id, inFlight.evidentMessageId, sessionId, completedId);
2391
+ } catch (err) {
2392
+ if (err instanceof ChannelAuthError) throw err;
2393
+ this.log({
2394
+ level: "error",
2395
+ 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)}`,
2396
+ conversation_id: watcher.conv.id,
2397
+ message_id: inFlight.evidentMessageId
2398
+ });
2399
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2400
+ return true;
2401
+ }
2402
+ inFlight.done = true;
2403
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2404
+ return true;
2405
+ }
2406
+ // -------------------------------------------------------------------------
2407
+ // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
2408
+ // -------------------------------------------------------------------------
2409
+ /**
2410
+ * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
2411
+ *
2412
+ * The pending drain only re-drives `pending` rows; a message already flipped to
2413
+ * `processing` before the runner died is watched by nobody until the 15-min
2414
+ * cron resets it. Here we fetch those rows, and per row resolve its correlated
2415
+ * reply against opencode's OWN session store — completing, re-attaching, or
2416
+ * (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
2417
+ * is idempotent per message (Invariant 2): a row a watcher already tracks is
2418
+ * skipped in `readoptOne` — one driver, no double-drive.
2419
+ *
2420
+ * Only `ChannelAuthError` propagates (to `drainPending`, like the pending
2421
+ * path); every other early return LOGS a reason with context — no silent drop.
2422
+ */
2423
+ async readoptProcessing() {
2424
+ const rows = await this.getProcessingMessages();
2425
+ if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0) {
2426
+ const stillProcessing = new Set(rows.map((r) => r.id));
2427
+ for (const id of [...this.dontRedispatch, ...this.doneUndeliverable]) {
2428
+ if (!stillProcessing.has(id)) {
2429
+ const cleared = this.dontRedispatch.delete(id);
2430
+ const clearedUndeliverable = this.doneUndeliverable.delete(id);
2431
+ if (cleared || clearedUndeliverable) {
2432
+ this.log({
2433
+ level: "info",
2434
+ message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
2435
+ message_id: id
2436
+ });
1791
2437
  }
1792
- } catch {
2438
+ }
2439
+ }
2440
+ }
2441
+ if (rows.length === 0) return;
2442
+ const bySession = /* @__PURE__ */ new Map();
2443
+ for (const row of rows) {
2444
+ if (!row.opencode_session_id) {
2445
+ this.log({
2446
+ level: "error",
2447
+ message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
2448
+ conversation_id: row.conversation_id,
2449
+ message_id: row.id
2450
+ });
2451
+ continue;
2452
+ }
2453
+ const list = bySession.get(row.opencode_session_id) ?? [];
2454
+ list.push(row);
2455
+ bySession.set(row.opencode_session_id, list);
2456
+ }
2457
+ for (const [sessionId, sessionRows] of bySession) {
2458
+ let messages;
2459
+ try {
2460
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
2461
+ if (!res.ok) {
2462
+ this.log({
2463
+ level: "error",
2464
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
2465
+ });
2466
+ continue;
2467
+ }
2468
+ const body = await res.json();
2469
+ if (!Array.isArray(body)) {
2470
+ this.log({
2471
+ level: "error",
2472
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
2473
+ });
1793
2474
  continue;
1794
2475
  }
1795
- if (!completed) continue;
2476
+ messages = body;
2477
+ } catch (err) {
2478
+ this.log({
2479
+ level: "error",
2480
+ message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
2481
+ });
2482
+ continue;
2483
+ }
2484
+ for (const row of sessionRows) {
2485
+ await this.readoptOne(sessionId, row, messages);
2486
+ }
2487
+ }
2488
+ }
2489
+ /**
2490
+ * Re-adopt ONE `processing` row against the tick's session message snapshot
2491
+ * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
2492
+ *
2493
+ * Branches on `messageRunState(messages, effectiveId)`, where `effectiveId` is
2494
+ * the NATIVE id a prior lifetime's re-drive ran this row under
2495
+ * (`row.opencode_message_id`) if present, else the STABLE derived id
2496
+ * (`opencodeMessageIdFor(row.id)`). Consulting the native id is what makes a
2497
+ * re-driven row that ALREADY ran/completed resolve correctly on restart instead
2498
+ * of looking `queued` under the (never-run) stable id and being re-driven AGAIN
2499
+ * (Bugbot #217 Finding 3 — the duplicate-turn bug):
2500
+ * - `done` → `markDone` now (guarded like the watcher's done branch);
2501
+ * - `failed` → `markFailed` with the surfaced error (issue #182), so an
2502
+ * errored turn is reported failed on restart, NOT re-dispatched;
2503
+ * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
2504
+ * tracking the effective id so the reply correlates by it;
2505
+ * - `unknown` → re-dispatch the STABLE id + attach a watcher (orphan).
2506
+ *
2507
+ * Only `ChannelAuthError` propagates.
2508
+ */
2509
+ async readoptOne(sessionId, row, messages) {
2510
+ if (this.isTracked(sessionId, row.id)) {
2511
+ this.log({
2512
+ level: "info",
2513
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
2514
+ conversation_id: row.conversation_id,
2515
+ message_id: row.id
2516
+ });
2517
+ return;
2518
+ }
2519
+ const stableId = opencodeMessageIdFor2(row.id);
2520
+ const nativeRedriveId = row.opencode_message_id;
2521
+ const ocId = nativeRedriveId ?? stableId;
2522
+ const state = messageRunState(messages, ocId);
2523
+ if (state === "done") {
2524
+ if (this.doneUndeliverable.has(row.id)) {
1796
2525
  this.log({
1797
2526
  level: "info",
1798
- message: `Paused session ${sessionId.slice(0, 8)} completed \u2014 marking message ${message.id.slice(0, 8)} done`,
1799
- conversation_id: conv.id,
1800
- message_id: message.id
2527
+ message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
2528
+ conversation_id: row.conversation_id,
2529
+ message_id: row.id
1801
2530
  });
1802
- await this.markDone(conv.id, message.id, sessionId);
1803
2531
  return;
1804
2532
  }
1805
2533
  this.log({
1806
2534
  level: "info",
1807
- message: `Paused session ${sessionId.slice(0, 8)} did not complete within the watch window \u2014 leaving message ${message.id.slice(0, 8)} for the cron safety net`,
1808
- conversation_id: conv.id,
1809
- message_id: message.id
2535
+ message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
2536
+ conversation_id: row.conversation_id,
2537
+ message_id: row.id
1810
2538
  });
2539
+ try {
2540
+ await this.markDone(row.conversation_id, row.id, sessionId, nativeRedriveId);
2541
+ } catch (err) {
2542
+ if (err instanceof ChannelAuthError) throw err;
2543
+ if (err instanceof ChannelTerminalError) {
2544
+ this.doneUndeliverable.add(row.id);
2545
+ this.log({
2546
+ level: "error",
2547
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 parking until it leaves processing; leaving for the cron safety net: ${err.message}`,
2548
+ conversation_id: row.conversation_id,
2549
+ message_id: row.id
2550
+ });
2551
+ return;
2552
+ }
2553
+ this.log({
2554
+ level: "error",
2555
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2556
+ conversation_id: row.conversation_id,
2557
+ message_id: row.id
2558
+ });
2559
+ return;
2560
+ }
2561
+ this.dontRedispatch.delete(row.id);
2562
+ return;
2563
+ }
2564
+ if (state === "failed") {
2565
+ const error2 = messageError(messages, ocId) ?? void 0;
2566
+ this.log({
2567
+ level: "error",
2568
+ message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2569
+ conversation_id: row.conversation_id,
2570
+ message_id: row.id
2571
+ });
2572
+ try {
2573
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2);
2574
+ } catch (err) {
2575
+ if (err instanceof ChannelAuthError) throw err;
2576
+ if (err instanceof ChannelTerminalError) {
2577
+ this.doneUndeliverable.add(row.id);
2578
+ this.log({
2579
+ level: "error",
2580
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 parking until it leaves processing; leaving for the cron safety net: ${err.message}`,
2581
+ conversation_id: row.conversation_id,
2582
+ message_id: row.id
2583
+ });
2584
+ return;
2585
+ }
2586
+ this.log({
2587
+ level: "error",
2588
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2589
+ conversation_id: row.conversation_id,
2590
+ message_id: row.id
2591
+ });
2592
+ return;
2593
+ }
2594
+ this.dontRedispatch.delete(row.id);
2595
+ return;
2596
+ }
2597
+ if (this.dontRedispatch.has(row.id)) {
2598
+ this.log({
2599
+ level: "info",
2600
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
2601
+ conversation_id: row.conversation_id,
2602
+ message_id: row.id
2603
+ });
2604
+ return;
2605
+ }
2606
+ if (state === "running" || state === "queued") {
2607
+ const conv = this.convForRow(sessionId, row);
2608
+ const message = this.queuedMessageForRow(row);
2609
+ this.registerReadopted(
2610
+ conv,
2611
+ sessionId,
2612
+ message,
2613
+ ocId,
2614
+ this.processedAtMs(row),
2615
+ nativeRedriveId
2616
+ );
2617
+ this.dispatched.add(row.id);
2618
+ this.readopted.add(row.id);
2619
+ this.ensureWatcherRunning(sessionId);
2620
+ this.log({
2621
+ level: "info",
2622
+ message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (${nativeRedriveId ? "native re-drive id" : "stable id"}, no re-dispatch)`,
2623
+ conversation_id: row.conversation_id,
2624
+ message_id: row.id
2625
+ });
2626
+ return;
2627
+ }
2628
+ await this.forceReadoptRun(sessionId, row);
2629
+ }
2630
+ /**
2631
+ * Re-dispatch an orphaned (`unknown`) `processing` row (ADR-0046 Decision §2).
2632
+ *
2633
+ * The stable-id user message is absent from the session, so we (re-)dispatch with
2634
+ * the STABLE id (`opencodeMessageIdFor(row.id)`) — NOT a divergent per-attempt id.
2635
+ * This is what keeps the reply correlatable: the server's completion
2636
+ * notification looks for the reply under the stable id, so the fresh turn's reply
2637
+ * (which hangs off the stable id) is found and delivered. The residual
2638
+ * duplicate-incomplete-turn semantics (ADR §2, `.harness/restart-recovery-orphan-finding.md`)
2639
+ * are unchanged and, for an ABSENT id, cannot bite — there is no existing turn
2640
+ * to swallow the duplicate.
2641
+ *
2642
+ * `evidentMessageId = row.id` addresses the SERVER row; the stable
2643
+ * `opencodeMessageId` is what the watcher polls. Deadline anchored to
2644
+ * `processed_at` (Invariant 1).
2645
+ */
2646
+ async forceReadoptRun(sessionId, row) {
2647
+ const ocId = opencodeMessageIdFor2(row.id);
2648
+ if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
2649
+ this.dontRedispatch.add(row.id);
2650
+ this.log({
2651
+ level: "info",
2652
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned but its re-adopt window has already elapsed \u2014 not dispatching an unwatchable turn; parking until it leaves processing (cron will reset it)`,
2653
+ conversation_id: row.conversation_id,
2654
+ message_id: row.id
2655
+ });
2656
+ return;
2657
+ }
2658
+ const options = {
2659
+ agent: row.opencode_agent ?? void 0,
2660
+ model: row.opencode_model ?? void 0
2661
+ };
2662
+ this.log({
2663
+ level: "info",
2664
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching with the stable id`,
2665
+ conversation_id: row.conversation_id,
2666
+ message_id: row.id
2667
+ });
2668
+ try {
2669
+ await sendPromptAsync(this.port, sessionId, row.content, options, ocId);
1811
2670
  } catch (err) {
2671
+ if (err instanceof ChannelAuthError) throw err;
1812
2672
  this.log({
1813
2673
  level: "error",
1814
- message: `Paused-session watcher failed for message ${message.id.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
1815
- conversation_id: conv.id,
1816
- message_id: message.id
2674
+ message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2675
+ conversation_id: row.conversation_id,
2676
+ message_id: row.id
1817
2677
  });
2678
+ return;
1818
2679
  }
2680
+ const conv = this.convForRow(sessionId, row);
2681
+ const message = this.queuedMessageForRow(row);
2682
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
2683
+ this.dispatched.add(row.id);
2684
+ this.readopted.add(row.id);
2685
+ this.ensureWatcherRunning(sessionId);
2686
+ }
2687
+ /**
2688
+ * True if `evidentMessageId` is already being driven — either in the
2689
+ * authoritative `dispatched` set or a live watcher's in-flight set for this
2690
+ * session (Invariant 2, WI-5). Either signal means a watcher owns the row.
2691
+ */
2692
+ isTracked(sessionId, evidentMessageId) {
2693
+ if (this.dispatched.has(evidentMessageId)) return true;
2694
+ const watcher = this.watchers.get(sessionId);
2695
+ return watcher?.inFlight.has(evidentMessageId) ?? false;
2696
+ }
2697
+ /**
2698
+ * Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
2699
+ * deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
2700
+ * for `processing` rows, but if it is somehow null/unparseable fall back to
2701
+ * `now` (defensive) AND log — a fallback means the anchor is weaker than
2702
+ * intended, which is worth surfacing.
2703
+ */
2704
+ processedAtMs(row) {
2705
+ const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
2706
+ if (!Number.isNaN(parsed)) return parsed;
2707
+ this.log({
2708
+ level: "error",
2709
+ message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
2710
+ conversation_id: row.conversation_id,
2711
+ message_id: row.id
2712
+ });
2713
+ return this.now();
2714
+ }
2715
+ /** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
2716
+ convForRow(sessionId, row) {
2717
+ return {
2718
+ id: row.conversation_id,
2719
+ agent_id: this.agentId,
2720
+ opencode_session_id: sessionId,
2721
+ pending_message_count: 0,
2722
+ oldest_pending_at: row.processed_at
2723
+ };
2724
+ }
2725
+ /** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
2726
+ queuedMessageForRow(row) {
2727
+ return {
2728
+ id: row.id,
2729
+ content: row.content,
2730
+ status: "processing",
2731
+ opencode_agent: row.opencode_agent,
2732
+ opencode_model: row.opencode_model,
2733
+ source_message_id: row.source_message_id,
2734
+ slack_user_id: row.slack_user_id
2735
+ };
2736
+ }
2737
+ /**
2738
+ * Remove a message from the in-flight set AND the authoritative dispatched
2739
+ * set. Once the in-flight set empties, the watcher loop's `while` guard exits
2740
+ * and its `.finally` removes the session entry from `this.watchers`.
2741
+ *
2742
+ * Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
2743
+ * (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
2744
+ * cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
2745
+ * re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
2746
+ * re-adopted message that completed (`done`) needs no marker — it's leaving
2747
+ * `processing`. This suppresses only re-dispatch: if its reply later completes,
2748
+ * the done branch still delivers it (Bugbot #202).
2749
+ */
2750
+ removeInFlight(watcher, evidentMessageId) {
2751
+ const inFlight = watcher.inFlight.get(evidentMessageId);
2752
+ if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
2753
+ this.dontRedispatch.add(evidentMessageId);
2754
+ this.log({
2755
+ level: "info",
2756
+ message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
2757
+ conversation_id: watcher.conv.id,
2758
+ message_id: evidentMessageId
2759
+ });
2760
+ }
2761
+ watcher.inFlight.delete(evidentMessageId);
2762
+ this.dispatched.delete(evidentMessageId);
2763
+ }
2764
+ /**
2765
+ * Poll `/question` + `/permission` (scoped to the session) and surface NEW ones
2766
+ * via `reportInteraction` (Task 3.5), carrying the PAUSED message's own
2767
+ * `source_message_id` so the server @mentions the correct person under
2768
+ * concurrency. Dedups by interaction id across ticks (reused per-session sets).
2769
+ *
2770
+ * The interaction is attributed to the in-flight message it paused on. opencode
2771
+ * stamps a `messageID` on a permission (and `tool.messageID` on a question) =
2772
+ * the assistant message id, whose `parentID` is the user message id — but the
2773
+ * simplest robust attribution here is: the single in-flight message that is
2774
+ * RUNNING (not done) is the one that paused. With one running message that is
2775
+ * unambiguous; with several we prefer an explicit messageID match, else the
2776
+ * oldest running message.
2777
+ */
2778
+ async pollInteractions(sessionId, watcher, messages) {
2779
+ let questions = [];
2780
+ try {
2781
+ const res = await this.fetchImpl(`${this.opencodeBase}/question`);
2782
+ if (res.ok) {
2783
+ const body = await res.json();
2784
+ questions = Array.isArray(body) ? body : [];
2785
+ }
2786
+ } catch {
2787
+ }
2788
+ for (const q of questions) {
2789
+ if (q.sessionID !== sessionId) continue;
2790
+ if (watcher.reportedQuestions.has(q.id)) continue;
2791
+ const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
2792
+ const reported = await this.reportInteraction(
2793
+ watcher.conv.id,
2794
+ "question",
2795
+ q,
2796
+ paused?.message.source_message_id ?? void 0
2797
+ );
2798
+ if (reported) watcher.reportedQuestions.add(q.id);
2799
+ }
2800
+ let permissions = [];
2801
+ try {
2802
+ const res = await this.fetchImpl(`${this.opencodeBase}/permission`);
2803
+ if (res.ok) {
2804
+ const body = await res.json();
2805
+ permissions = Array.isArray(body) ? body : [];
2806
+ }
2807
+ } catch {
2808
+ }
2809
+ for (const p of permissions) {
2810
+ if (p.sessionID !== sessionId) continue;
2811
+ if (watcher.reportedPermissions.has(p.id)) continue;
2812
+ const paused = this.attributeInteraction(watcher, p.messageID, messages);
2813
+ const reported = await this.reportInteraction(
2814
+ watcher.conv.id,
2815
+ "permission",
2816
+ p,
2817
+ paused?.message.source_message_id ?? void 0
2818
+ );
2819
+ if (reported) watcher.reportedPermissions.add(p.id);
2820
+ }
2821
+ }
2822
+ /**
2823
+ * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
2824
+ *
2825
+ * The interaction carries `interactionMessageId` — the ASSISTANT message id
2826
+ * that raised it (a question's `tool.messageID` / a permission's `messageID`).
2827
+ * That assistant message is the reply to ONE of our minted user messages
2828
+ * (correlated by `parentID`, GATE-B). So when we have the tick's message
2829
+ * snapshot, we resolve each running in-flight message's correlated assistant
2830
+ * reply (`findAssistantReplyAfter`) and match its id against
2831
+ * `interactionMessageId` — giving an EXACT attribution even with several
2832
+ * messages in flight concurrently in one session.
2833
+ *
2834
+ * We fall back to the oldest running message ONLY when no exact match is
2835
+ * possible (the id is absent, the snapshot is missing, or the reply has not yet
2836
+ * been correlated). With a single running message either path is exact. Never
2837
+ * throws.
2838
+ *
2839
+ * Attribution must NOT depend on our own `started` PATCH flag: opencode can
2840
+ * START a turn AND raise a question/permission BEFORE our next tick fires
2841
+ * `markProcessing` (which sets `started`). Relying on `started` would leave the
2842
+ * running set empty in that window and let the server fall back to "newest
2843
+ * processing/pending" — possibly @mentioning a FOLLOW-UP author rather than the
2844
+ * person whose active turn actually paused. So we derive "running" from the
2845
+ * tick's `messages` snapshot via `messageRunState` instead.
2846
+ */
2847
+ attributeInteraction(watcher, interactionMessageId, messages) {
2848
+ const inFlight = [...watcher.inFlight.values()].filter((m) => !m.done);
2849
+ if (inFlight.length === 0) return void 0;
2850
+ if (interactionMessageId && messages) {
2851
+ const exact = inFlight.find((m) => {
2852
+ const reply = findAssistantReplyAfter(messages, m.opencodeMessageId);
2853
+ return reply != null && messageIdOf(reply) === interactionMessageId;
2854
+ });
2855
+ if (exact) return exact;
2856
+ }
2857
+ const byOldest = (a, b) => a.dispatchedAt - b.dispatchedAt;
2858
+ if (messages) {
2859
+ const runningPerSnapshot = inFlight.filter(
2860
+ (m) => messageRunState(messages, m.opencodeMessageId) === "running"
2861
+ );
2862
+ if (runningPerSnapshot.length > 0) {
2863
+ return runningPerSnapshot.sort(byOldest)[0];
2864
+ }
2865
+ }
2866
+ const startedRunning = inFlight.filter((m) => m.started);
2867
+ if (startedRunning.length > 0) {
2868
+ return startedRunning.sort(byOldest)[0];
2869
+ }
2870
+ return inFlight.sort(byOldest)[0];
1819
2871
  }
1820
2872
  // -------------------------------------------------------------------------
1821
2873
  // Evident API calls (combinedAuth thread routes)
@@ -1849,52 +2901,182 @@ var ChannelDriver = class {
1849
2901
  }
1850
2902
  return await res.json();
1851
2903
  }
1852
- async markProcessing(conversationId, messageId, sessionId) {
2904
+ /**
2905
+ * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
2906
+ * The pending path (`getPendingConversations`/`getPendingMessages`) only
2907
+ * surfaces `pending` rows, so a message already `processing` when the runner
2908
+ * died is invisible to it — this dedicated endpoint returns exactly those rows
2909
+ * with the fields the re-adopt path needs (`processed_at`,
2910
+ * `opencode_session_id`, routing).
2911
+ *
2912
+ * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
2913
+ * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
2914
+ * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
2915
+ * next tick retries.
2916
+ */
2917
+ async getProcessingMessages() {
2918
+ const res = await this.fetchImpl(
2919
+ `${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
2920
+ { headers: { Authorization: this.getAuthHeader() } }
2921
+ );
2922
+ this.assertAuth(res, "fetching processing messages");
2923
+ if (!res.ok) {
2924
+ throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
2925
+ }
2926
+ const data = await res.json();
2927
+ let messages = data.messages ?? [];
2928
+ if (this.conversationFilter) {
2929
+ messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
2930
+ }
2931
+ return messages;
2932
+ }
2933
+ /**
2934
+ * EXISTING combinedAuth route — now fired by the watcher on queued→running
2935
+ * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
2936
+ * opencode_session_id}` → `notifyMessageStarted` (hourglass→runner swap +
2937
+ * deep-linked "View in Evident" notice).
2938
+ *
2939
+ * Return/throw contract (consumed by the watcher's swap-to-running guard):
2940
+ * - returns `true` → the server transitioned the row to processing;
2941
+ * - returns `false` → the server gave a DEFINITIVE "already-processing"
2942
+ * answer (a non-retryable, non-auth status — e.g. a
2943
+ * conflict because a duplicate already transitioned it),
2944
+ * so the caller treats it as already-started and does NOT
2945
+ * retry;
2946
+ * - throws `ChannelAuthError` on 401/403 (terminal auth failure);
2947
+ * - throws on a TRANSIENT failure (retryable 5xx/429 status, or a
2948
+ * network-level error from `fetch`) — i.e. NO definitive server response —
2949
+ * so the caller leaves the message un-started and retries the swap on the
2950
+ * next tick.
2951
+ * A single attempt (no internal retry): the watcher's per-tick loop is the
2952
+ * retry vehicle for the swap-to-running.
2953
+ */
2954
+ async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
1853
2955
  const res = await this.fetchImpl(
1854
2956
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
1855
2957
  {
1856
2958
  method: "PATCH",
1857
2959
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1858
- // Include the opencode session id so the server can deep-link the
1859
- // "View in Evident" notice straight to the conversation (the session
1860
- // already exists by now — ensureSession runs before claiming).
1861
- body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
2960
+ body: JSON.stringify({
2961
+ status: "processing",
2962
+ opencode_session_id: sessionId,
2963
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
2964
+ })
1862
2965
  }
1863
2966
  );
1864
2967
  this.assertAuth(res, "marking message as processing");
1865
- return res.ok;
2968
+ if (res.ok) return true;
2969
+ if (isRetryableStatus(res.status)) {
2970
+ throw new Error(`marking message as processing: HTTP ${res.status}`);
2971
+ }
2972
+ return false;
1866
2973
  }
1867
2974
  /**
1868
- * EXISTING combinedAuth completion route — idempotent + retried (WI-CHAN-2).
1869
- * `PATCH .../messages/:id {status:'done', opencode_session_id}`. The server's
2975
+ * EXISTING combinedAuth completion route — idempotent (WI-CHAN-2). `PATCH
2976
+ * .../messages/:id {status:'done', opencode_session_id}`. The server's
1870
2977
  * `queued_conversation_messages.status`/`processed_at` gate makes a re-call
1871
- * for an already-`done` message a no-op (no double Slack post).
2978
+ * for an already-`done` message a no-op (no double Slack post). Fired by the
2979
+ * watcher on per-message completion (Task 3.4) — no `confirmCompletion`
2980
+ * round-trip (we already observed completion via the message list).
2981
+ *
2982
+ * SINGLE ATTEMPT (no in-call `callWithRetry` backoff). The per-session watcher
2983
+ * services its in-flight messages SEQUENTIALLY within a tick
2984
+ * (`runWatcherLoop` → `serviceInFlightMessage`), so a long multi-attempt
2985
+ * backoff here would BLOCK sibling messages in the SAME session/tick: while
2986
+ * message A's done PATCH burned its internal retries, message B could not be
2987
+ * swapped to running even though opencode had already started it. Instead this
2988
+ * does ONE PATCH and surfaces the SAME outcome contract the watcher's markDone
2989
+ * handler already relies on, leaning on the per-tick retry across ticks
2990
+ * (bounded by `inFlight.deadline`) rather than an in-call retry:
2991
+ * - resolves (`void`) → the server transitioned the row to done
2992
+ * (or idempotently confirmed already-done);
2993
+ * - throws `ChannelAuthError` → 401/403 (terminal auth failure → loop
2994
+ * cleanup, Finding 1);
2995
+ * - throws `ChannelTerminalError`→ non-retryable, non-auth 4xx (will never
2996
+ * succeed → straight to the cron, Finding 4);
2997
+ * - throws a plain `Error` → TRANSIENT 5xx/429 or a network-level error
2998
+ * (no definitive server response → the
2999
+ * watcher retries next tick within the
3000
+ * deadline, Finding 4).
3001
+ */
3002
+ async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
3003
+ const res = await this.fetchImpl(
3004
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
3005
+ {
3006
+ method: "PATCH",
3007
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
3008
+ body: JSON.stringify({
3009
+ status: "done",
3010
+ opencode_session_id: sessionId,
3011
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
3012
+ })
3013
+ }
3014
+ );
3015
+ this.assertAuth(res, "marking message as done");
3016
+ if (res.ok) return;
3017
+ if (isRetryableStatus(res.status)) {
3018
+ throw new Error(`marking message as done: HTTP ${res.status}`);
3019
+ }
3020
+ throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
3021
+ }
3022
+ /**
3023
+ * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
3024
+ * when provided (issue #182): a bare `markFailed(conv, msg)` sends
3025
+ * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
3026
+ * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
3027
+ * failure reason reaches the channel.
1872
3028
  */
1873
- async markDone(conversationId, messageId, sessionId) {
3029
+ async markFailed(conversationId, messageId, sessionId, error2) {
3030
+ const body = { status: "failed" };
3031
+ if (sessionId !== void 0) body.opencode_session_id = sessionId;
3032
+ if (error2 !== void 0) body.error = error2;
1874
3033
  await this.callWithRetry(
1875
- "marking message as done",
3034
+ "marking message as failed",
1876
3035
  () => this.fetchImpl(
1877
3036
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
1878
3037
  {
1879
3038
  method: "PATCH",
1880
3039
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1881
- body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
3040
+ body: JSON.stringify(body)
1882
3041
  }
1883
3042
  )
1884
3043
  );
1885
3044
  }
1886
- async markFailed(conversationId, messageId) {
1887
- await this.callWithRetry(
1888
- "marking message as failed",
1889
- () => this.fetchImpl(
1890
- `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
3045
+ /**
3046
+ * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
3047
+ * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
3048
+ * — the server records it via `log()` (no DB write, no notification). This is
3049
+ * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
3050
+ * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
3051
+ * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
3052
+ * context (no silent catch, per development-workflow).
3053
+ */
3054
+ async postSignal(conversationId, messageId, signal, extra) {
3055
+ try {
3056
+ const res = await this.fetchImpl(
3057
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
1891
3058
  {
1892
- method: "PATCH",
3059
+ method: "POST",
1893
3060
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1894
- body: JSON.stringify({ status: "failed" })
3061
+ body: JSON.stringify({ signal, ...extra })
1895
3062
  }
1896
- )
1897
- );
3063
+ );
3064
+ if (!res.ok) {
3065
+ this.log({
3066
+ level: "error",
3067
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
3068
+ conversation_id: conversationId,
3069
+ message_id: messageId
3070
+ });
3071
+ }
3072
+ } catch (err) {
3073
+ this.log({
3074
+ level: "error",
3075
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
3076
+ conversation_id: conversationId,
3077
+ message_id: messageId
3078
+ });
3079
+ }
1898
3080
  }
1899
3081
  async persistSession(conversationId, sessionId) {
1900
3082
  const res = await this.fetchImpl(
@@ -1909,10 +3091,17 @@ var ChannelDriver = class {
1909
3091
  }
1910
3092
  /**
1911
3093
  * EXISTING combinedAuth interaction route (WI-CHAN-3) — idempotent + retried.
1912
- * `POST .../interactive-event {type, data}`. The server persists the
1913
- * interaction and posts a link to the proxied opencode-web conversation.
3094
+ * `POST .../interactive-event {type, data, source_message_id?}`. The server
3095
+ * persists the interaction and posts a link to the proxied opencode-web
3096
+ * conversation, @mentioning the user who triggered THIS message's turn.
3097
+ *
3098
+ * WI-3 / WI-4 contract: `source_message_id` is the PAUSED message's own Slack
3099
+ * ts (`message.source_message_id`). The server resolves the @mention from that
3100
+ * message's user FIRST (falling back to the old "newest processing" precedence
3101
+ * only when absent), so the correct person is mentioned under concurrency. It
3102
+ * is OPTIONAL for back-compat with older clients / legacy rows.
1914
3103
  */
1915
- async reportInteraction(conversationId, type, data) {
3104
+ async reportInteraction(conversationId, type, data, sourceMessageId) {
1916
3105
  try {
1917
3106
  await this.callWithRetry(
1918
3107
  "reporting interactive event",
@@ -1921,7 +3110,9 @@ var ChannelDriver = class {
1921
3110
  {
1922
3111
  method: "POST",
1923
3112
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
1924
- body: JSON.stringify({ type, data })
3113
+ body: JSON.stringify(
3114
+ sourceMessageId ? { type, data, source_message_id: sourceMessageId } : { type, data }
3115
+ )
1925
3116
  }
1926
3117
  )
1927
3118
  );
@@ -1930,6 +3121,7 @@ var ChannelDriver = class {
1930
3121
  message: `${type} surfaced to channel (id: ${data.id.slice(0, 8)})`,
1931
3122
  conversation_id: conversationId
1932
3123
  });
3124
+ return true;
1933
3125
  } catch (err) {
1934
3126
  if (err instanceof ChannelAuthError) throw err;
1935
3127
  this.log({
@@ -1937,6 +3129,7 @@ var ChannelDriver = class {
1937
3129
  message: `Failed to surface ${type}: ${err instanceof Error ? err.message : String(err)}`,
1938
3130
  conversation_id: conversationId
1939
3131
  });
3132
+ return false;
1940
3133
  }
1941
3134
  }
1942
3135
  // -------------------------------------------------------------------------
@@ -1975,8 +3168,9 @@ var ChannelDriver = class {
1975
3168
  await this.sleep(backoffDelay(attempt, this.retry));
1976
3169
  continue;
1977
3170
  }
3171
+ break;
1978
3172
  }
1979
- throw new Error(`${context}: HTTP ${res.status}`);
3173
+ throw new ChannelTerminalError(`${context}: HTTP ${res.status}`, res.status);
1980
3174
  }
1981
3175
  throw lastError instanceof Error ? lastError : new Error(`${context}: exhausted retries`);
1982
3176
  }
@@ -2199,7 +3393,9 @@ async function getAgentInfo(agentId, authHeader) {
2199
3393
  // src/commands/run.ts
2200
3394
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
2201
3395
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
2202
- function log(state, message, isError = false) {
3396
+ var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
3397
+ var CHANNEL_SETTLE_MS = Number(process.env.EVIDENT_SETTLE_MS) || void 0;
3398
+ function log2(state, message, isError = false) {
2203
3399
  if (state.json) {
2204
3400
  console.log(
2205
3401
  JSON.stringify({
@@ -2224,9 +3420,9 @@ function logActivity(state, entry) {
2224
3420
  }
2225
3421
  if (!state.interactive) {
2226
3422
  if (entry.type === "error") {
2227
- log(state, entry.error ?? "Unknown error", true);
3423
+ log2(state, entry.error ?? "Unknown error", true);
2228
3424
  } else if (entry.type === "info" && entry.message) {
2229
- log(state, entry.message);
3425
+ log2(state, entry.message);
2230
3426
  }
2231
3427
  }
2232
3428
  }
@@ -2312,6 +3508,7 @@ async function handleAuthError(state, error2) {
2312
3508
  }
2313
3509
  async function driveChannels(state, driver) {
2314
3510
  let idlePolls = 0;
3511
+ let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
2315
3512
  while (state.running) {
2316
3513
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
2317
3514
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2321,9 +3518,11 @@ async function driveChannels(state, driver) {
2321
3518
  try {
2322
3519
  const processed = await driver.drainPending();
2323
3520
  state.messageCount += processed;
2324
- if (processed > 0) {
3521
+ const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
3522
+ lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
3523
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
2325
3524
  idlePolls = 0;
2326
- if (state.interactive) displayStatus(state);
3525
+ if (processed > 0 && state.interactive) displayStatus(state);
2327
3526
  } else if (state.idleTimeout !== null) {
2328
3527
  idlePolls++;
2329
3528
  if (idlePolls === 1) {
@@ -2373,7 +3572,7 @@ async function cleanup(state) {
2373
3572
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
2374
3573
  displayStatus(state);
2375
3574
  } else {
2376
- log(state, "Stopped OpenCode process");
3575
+ log2(state, "Stopped OpenCode process");
2377
3576
  }
2378
3577
  state.opencodeProcess = null;
2379
3578
  }
@@ -2396,10 +3595,11 @@ async function run(options) {
2396
3595
  running: true,
2397
3596
  activityLog: [],
2398
3597
  messageCount: 0,
3598
+ lastProxiedActivityAt: null,
2399
3599
  authHeader: ""
2400
3600
  };
2401
3601
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
2402
- log(
3602
+ log2(
2403
3603
  state,
2404
3604
  "Warning: No --idle-timeout set in CI environment. The runner will poll indefinitely until the job times out. Consider adding --idle-timeout 30 to avoid wasting runner minutes.",
2405
3605
  false
@@ -2410,7 +3610,7 @@ async function run(options) {
2410
3610
  logActivity(state, { type: "info", message: "Shutting down..." });
2411
3611
  displayStatus(state);
2412
3612
  } else {
2413
- log(state, "Shutting down...");
3613
+ log2(state, "Shutting down...");
2414
3614
  }
2415
3615
  await cleanup(state);
2416
3616
  await shutdownTelemetry();
@@ -2443,7 +3643,7 @@ async function run(options) {
2443
3643
  const resolved = await resolveAgentIdFromKey(state.authHeader);
2444
3644
  if (resolved.agent_id) {
2445
3645
  state.agentId = resolved.agent_id;
2446
- log(state, `Resolved agent ID from key: ${state.agentId}`);
3646
+ log2(state, `Resolved agent ID from key: ${state.agentId}`);
2447
3647
  if (state.interactive && !state.json) {
2448
3648
  logActivity(state, {
2449
3649
  type: "info",
@@ -2506,14 +3706,21 @@ async function run(options) {
2506
3706
  port: state.port,
2507
3707
  interactive: state.interactive,
2508
3708
  agentId: state.agentId,
2509
- log: (message) => log(state, message)
3709
+ log: (message) => log2(state, message)
2510
3710
  });
2511
3711
  state.port = oc.port;
2512
3712
  state.opencodeProcess = oc.process;
2513
3713
  state.opencodeVersion = oc.version;
2514
3714
  state.opencodeConnected = oc.process !== null || oc.version !== null;
2515
- const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
2516
- ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
3715
+ const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
3716
+ ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
3717
+ const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
3718
+ if (versionWarning) {
3719
+ log2(state, versionWarning, false);
3720
+ if (state.interactive && !state.json) {
3721
+ logActivity(state, { type: "info", message: versionWarning });
3722
+ }
3723
+ }
2517
3724
  } catch (error2) {
2518
3725
  ocSpinner?.fail(error2.message);
2519
3726
  throw error2;
@@ -2525,6 +3732,8 @@ async function run(options) {
2525
3732
  apiUrl: getApiUrlConfig(),
2526
3733
  getAuthHeader: () => state.authHeader,
2527
3734
  conversationFilter: state.conversationFilter,
3735
+ stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
3736
+ settleMs: CHANNEL_SETTLE_MS,
2528
3737
  log: (entry) => logActivity(state, {
2529
3738
  type: entry.level === "error" ? "error" : "info",
2530
3739
  message: entry.message,
@@ -2544,7 +3753,11 @@ async function run(options) {
2544
3753
  type: "info",
2545
3754
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
2546
3755
  });
2547
- emitAgentConnected(state.agentId, { port: state.port });
3756
+ emitAgentConnected(state.agentId, {
3757
+ port: state.port,
3758
+ cli_version: getCliVersion(),
3759
+ opencode_version: state.opencodeVersion
3760
+ });
2548
3761
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
2549
3762
  if (state.interactive) displayStatus(state);
2550
3763
  channelDriver.drainPending().then((processed) => {
@@ -2578,9 +3791,14 @@ async function run(options) {
2578
3791
  logActivity(state, { type: "error", error: error2 });
2579
3792
  if (state.interactive) displayStatus(state);
2580
3793
  },
2581
- // Web traffic is proxied transparently; only note opencode is live.
3794
+ // Web traffic is proxied transparently; note opencode is live and stamp
3795
+ // proxied activity so the idle loop treats interactive proxy use as work.
3796
+ // Fires per forwarded response head (incl. every SSE open) and excludes
3797
+ // the internal drain-ping, so an actively-used proxy keeps the timer
3798
+ // fresh while a lone idle SSE with no follow-up requests still ages out.
2582
3799
  onResponse: () => {
2583
3800
  state.opencodeConnected = true;
3801
+ state.lastProxiedActivityAt = Date.now();
2584
3802
  },
2585
3803
  // A channel message was queued and the api-worker pinged us over the
2586
3804
  // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2619,7 +3837,7 @@ async function run(options) {
2619
3837
  throw error2;
2620
3838
  }
2621
3839
  if (!interactive || state.json) {
2622
- log(state, "Driving channel messages...");
3840
+ log2(state, "Driving channel messages...");
2623
3841
  }
2624
3842
  await driveChannels(state, channelDriver);
2625
3843
  await cleanup(state);
@@ -2631,7 +3849,7 @@ async function run(options) {
2631
3849
  })
2632
3850
  );
2633
3851
  } else if (!interactive) {
2634
- log(state, `Completed. Processed ${state.messageCount} message(s).`);
3852
+ log2(state, `Completed. Processed ${state.messageCount} message(s).`);
2635
3853
  }
2636
3854
  await shutdownTelemetry();
2637
3855
  process.exit(0);
@@ -2653,8 +3871,9 @@ async function run(options) {
2653
3871
  }
2654
3872
 
2655
3873
  // src/index.ts
3874
+ var { version } = createRequire(import.meta.url)("../package.json");
2656
3875
  var program = new Command();
2657
- program.name("evident").description("Run OpenCode locally and connect it to Evident").version("0.1.0").option(
3876
+ program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
2658
3877
  "--endpoint <url>",
2659
3878
  "Evident API base URL (default: production; e.g. http://localhost:3001)"
2660
3879
  ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {