@evident-ai/cli 3.0.1-dev.0590aa5 → 3.0.1-dev.07ee447

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
@@ -484,8 +485,34 @@ var TelemetryEventTypes = {
484
485
  var MAX_FRAME_BYTES = 256 * 1024;
485
486
  var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
486
487
 
488
+ // ../../packages/types/src/logging/index.ts
489
+ var CORRELATION_ID_HEADER = "x-evident-correlation-id";
490
+ function log(level, event, fields) {
491
+ const method = level === "debug" ? "log" : level;
492
+ try {
493
+ console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
494
+ } catch (err) {
495
+ console.error(
496
+ "[evident] log_serialize_failed",
497
+ event,
498
+ err instanceof Error ? err.message : String(err)
499
+ );
500
+ }
501
+ }
502
+ function stripQuery(url) {
503
+ try {
504
+ return new URL(url).pathname;
505
+ } catch {
506
+ const q = url.indexOf("?");
507
+ return q === -1 ? url : url.slice(0, q);
508
+ }
509
+ }
510
+
487
511
  // src/lib/telemetry.ts
488
- var CLI_VERSION = process.env.npm_package_version || "unknown";
512
+ var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
513
+ function getCliVersion() {
514
+ return CLI_VERSION;
515
+ }
489
516
  var eventBuffer = [];
490
517
  var flushTimeout = null;
491
518
  var isShuttingDown = false;
@@ -685,14 +712,14 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
685
712
  }
686
713
 
687
714
  // src/lib/opencode/opencode-version-gate.ts
688
- var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
689
- function isQueueValidatedVersion(version) {
690
- if (!version) return false;
691
- return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version);
692
- }
693
- function buildOpenCodeVersionWarning(version) {
694
- if (isQueueValidatedVersion(version)) return null;
695
- const detected = version ? `v${version}` : "unknown";
715
+ var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
716
+ function isQueueValidatedVersion(version2) {
717
+ if (!version2) return false;
718
+ return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
719
+ }
720
+ function buildOpenCodeVersionWarning(version2) {
721
+ if (isQueueValidatedVersion(version2)) return null;
722
+ const detected = version2 ? `v${version2}` : "unknown";
696
723
  const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
697
724
  return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
698
725
  }
@@ -1009,7 +1036,11 @@ function roleOf(m) {
1009
1036
  }
1010
1037
  function completedOf(m) {
1011
1038
  if (!m || typeof m !== "object") return void 0;
1012
- return m.info?.time?.completed;
1039
+ return m.info?.time?.completed ?? m.time?.completed;
1040
+ }
1041
+ function createdOf(m) {
1042
+ if (!m || typeof m !== "object") return void 0;
1043
+ return m.info?.time?.created ?? m.time?.created;
1013
1044
  }
1014
1045
  function idOf(m) {
1015
1046
  if (!m || typeof m !== "object") return void 0;
@@ -1029,6 +1060,24 @@ function finishOf(m) {
1029
1060
  const infoFinish = m.info?.finish;
1030
1061
  return typeof infoFinish === "string" ? infoFinish : void 0;
1031
1062
  }
1063
+ function errorOf(m) {
1064
+ if (!m || typeof m !== "object") return void 0;
1065
+ return m.info?.error ?? m.error;
1066
+ }
1067
+ function isAssistantInFlight(m) {
1068
+ if (completedOf(m) == null) return true;
1069
+ return finishOf(m) === "tool-calls";
1070
+ }
1071
+ async function getSessionMessages(port, sessionId) {
1072
+ try {
1073
+ const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
1074
+ if (!res.ok) return null;
1075
+ const body = await res.json();
1076
+ return Array.isArray(body) ? body : null;
1077
+ } catch {
1078
+ return null;
1079
+ }
1080
+ }
1032
1081
  async function createOpenCodeSession(port, directory) {
1033
1082
  const url = new URL(`${opencodeBase(port)}/session`);
1034
1083
  if (directory && directory.trim()) {
@@ -1046,9 +1095,16 @@ async function createOpenCodeSession(port, directory) {
1046
1095
  const data = await response.json();
1047
1096
  return data.id;
1048
1097
  }
1049
- async function sendPromptAsync(port, sessionId, content, options, messageId) {
1098
+ function messageText(m) {
1099
+ if (!m || !Array.isArray(m.parts)) return "";
1100
+ return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
1101
+ }
1102
+ async function sendPromptAsync(port, sessionId, content, options) {
1103
+ const before = await getSessionMessages(port, sessionId);
1104
+ const knownUserIds = new Set(
1105
+ (before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
1106
+ );
1050
1107
  const body = {
1051
- messageID: messageId,
1052
1108
  parts: [{ type: "text", text: content }]
1053
1109
  };
1054
1110
  if (options?.agent) {
@@ -1072,6 +1128,29 @@ async function sendPromptAsync(port, sessionId, content, options, messageId) {
1072
1128
  const text = await res.text().catch(() => "");
1073
1129
  throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
1074
1130
  }
1131
+ const READ_BACK_ATTEMPTS = 5;
1132
+ const READ_BACK_DELAY_MS = 150;
1133
+ for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
1134
+ const after = await getSessionMessages(port, sessionId);
1135
+ if (after) {
1136
+ let best = null;
1137
+ for (const m of after) {
1138
+ if (roleOf(m) !== "user") continue;
1139
+ const id = idOf(m);
1140
+ if (typeof id !== "string" || knownUserIds.has(id)) continue;
1141
+ if (messageText(m) !== content) continue;
1142
+ const created = createdOf(m) ?? 0;
1143
+ if (best === null || created > best.created) {
1144
+ best = { id, created };
1145
+ }
1146
+ }
1147
+ if (best) return best.id;
1148
+ }
1149
+ if (attempt < READ_BACK_ATTEMPTS - 1) {
1150
+ await new Promise((resolve) => setTimeout(resolve, READ_BACK_DELAY_MS));
1151
+ }
1152
+ }
1153
+ return null;
1075
1154
  }
1076
1155
  function findAssistantReplyAfter(messages, userMessageId) {
1077
1156
  if (!messages || messages.length === 0) return null;
@@ -1088,19 +1167,31 @@ function findAssistantReplyAfter(messages, userMessageId) {
1088
1167
  }
1089
1168
  function findLastAssistantReplyFor(messages, userMessageId) {
1090
1169
  if (!messages || messages.length === 0) return null;
1170
+ let lastCorrelated = null;
1171
+ let lastNonErrored = null;
1091
1172
  for (let i = messages.length - 1; i >= 0; i--) {
1092
1173
  const m = messages[i];
1093
- if (roleOf(m) === "assistant" && parentIdOf(m) === userMessageId) return m;
1174
+ if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
1175
+ if (lastCorrelated === null) lastCorrelated = m;
1176
+ if (errorOf(m) == null) {
1177
+ lastNonErrored = m;
1178
+ break;
1179
+ }
1094
1180
  }
1181
+ if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
1095
1182
  const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
1096
1183
  if (userIndex === -1) return null;
1097
1184
  let last = null;
1185
+ let lastOk = null;
1098
1186
  for (let i = userIndex + 1; i < messages.length; i++) {
1099
1187
  const role = roleOf(messages[i]);
1100
1188
  if (role === "user") break;
1101
- if (role === "assistant") last = messages[i];
1189
+ if (role === "assistant") {
1190
+ last = messages[i];
1191
+ if (errorOf(messages[i]) == null) lastOk = messages[i];
1192
+ }
1102
1193
  }
1103
- return last;
1194
+ return lastOk ?? last;
1104
1195
  }
1105
1196
  function messageRunState(messages, userMessageId) {
1106
1197
  if (!messages || messages.length === 0) return "unknown";
@@ -1110,13 +1201,27 @@ function messageRunState(messages, userMessageId) {
1110
1201
  if (!reply) return "unknown";
1111
1202
  }
1112
1203
  if (!reply) return "queued";
1113
- if (completedOf(reply) == null) return "running";
1114
- if (finishOf(reply) === "tool-calls") return "running";
1115
- return "done";
1204
+ if (isAssistantInFlight(reply)) return "running";
1205
+ return errorOf(reply) != null ? "failed" : "done";
1116
1206
  }
1117
- function opencodeMessageIdFor(queuedMessageId) {
1118
- const sanitized = queuedMessageId.replace(/[^a-zA-Z0-9]/g, "_");
1119
- return `msg_${sanitized}`;
1207
+ function messageError(messages, userMessageId) {
1208
+ const reply = findLastAssistantReplyFor(messages, userMessageId);
1209
+ const error2 = errorOf(reply);
1210
+ if (error2 == null) return null;
1211
+ if (typeof error2 === "string") return error2;
1212
+ if (typeof error2 === "object") {
1213
+ const e = error2;
1214
+ const dataMessage = e.data?.message;
1215
+ if (typeof dataMessage === "string") return dataMessage;
1216
+ if (typeof e.message === "string") return e.message;
1217
+ }
1218
+ return "The agent run failed.";
1219
+ }
1220
+ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
1221
+ if (!messages || messages.length === 0) return false;
1222
+ return messages.some(
1223
+ (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
1224
+ );
1120
1225
  }
1121
1226
 
1122
1227
  // src/lib/tunnel/connection.ts
@@ -1187,12 +1292,20 @@ var StreamForwarder = class {
1187
1292
  }
1188
1293
  async handleOpen(frame) {
1189
1294
  const { sid, method, path, headers, has_body } = frame;
1295
+ const correlationId = headers?.[CORRELATION_ID_HEADER];
1296
+ const startedAt = Date.now();
1190
1297
  if (path === TUNNEL_DRAIN_PING_PATH) {
1191
1298
  this.callbacks.onDrainPing?.();
1192
1299
  this.send({ type: "head", sid, status: 204, headers: {} });
1193
1300
  this.send({ type: "res_end", sid });
1194
1301
  return;
1195
1302
  }
1303
+ log("info", "agent_request", {
1304
+ correlation_id: correlationId,
1305
+ sid,
1306
+ method,
1307
+ path: stripQuery(path)
1308
+ });
1196
1309
  const ac = new AbortController();
1197
1310
  let bodyPromise;
1198
1311
  let pushBody;
@@ -1239,6 +1352,12 @@ var StreamForwarder = class {
1239
1352
  if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
1240
1353
  });
1241
1354
  this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
1355
+ log("info", "agent_response", {
1356
+ correlation_id: correlationId,
1357
+ sid,
1358
+ status: upstream.status,
1359
+ duration_ms: Date.now() - startedAt
1360
+ });
1242
1361
  this.callbacks.onHead?.(sid, upstream.status);
1243
1362
  try {
1244
1363
  if (upstream.body) {
@@ -1508,7 +1627,7 @@ var DEFAULT_RETRY_POLICY = {
1508
1627
  };
1509
1628
  var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
1510
1629
  var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
1511
- var DEFAULT_DISPATCH_CONFIRM_MS = 6e3;
1630
+ var DEFAULT_STUCK_QUEUED_MS = 6e4;
1512
1631
  var ChannelAuthError = class extends Error {
1513
1632
  constructor(message) {
1514
1633
  super(message);
@@ -1543,10 +1662,18 @@ var ChannelDriver = class {
1543
1662
  sleep;
1544
1663
  pausedPollIntervalMs;
1545
1664
  pausedMaxWaitMs;
1546
- dispatchConfirmMs;
1665
+ stuckQueuedMs;
1547
1666
  now;
1548
1667
  /** Cache of conversationId → opencode sessionId. */
1549
1668
  sessions = /* @__PURE__ */ new Map();
1669
+ /**
1670
+ * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
1671
+ * longer idempotent (no caller-supplied `messageID`), and its read-back picks
1672
+ * "the one new user row" — which is only unambiguous if no OTHER dispatch into
1673
+ * the SAME session interleaves its snapshot→POST→read-back. This map chains each
1674
+ * session's dispatches so they run serially; distinct sessions stay concurrent.
1675
+ */
1676
+ sessionDispatchLocks = /* @__PURE__ */ new Map();
1550
1677
  /**
1551
1678
  * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
1552
1679
  * session: one polling loop services all of that session's in-flight messages.
@@ -1563,6 +1690,54 @@ var ChannelDriver = class {
1563
1690
  * a steady-state-poll re-dispatch will not double-run the message.
1564
1691
  */
1565
1692
  dispatched = /* @__PURE__ */ new Set();
1693
+ /**
1694
+ * Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
1695
+ * Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
1696
+ * so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
1697
+ * it is re-adopted and removed when its watcher settles or it is observed off
1698
+ * the processing list.
1699
+ */
1700
+ readopted = /* @__PURE__ */ new Set();
1701
+ /**
1702
+ * "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
1703
+ * re-adopted running/orphan row's watcher hit its `processed_at`-anchored
1704
+ * deadline (or an orphan whose window already elapsed): the still-`processing`
1705
+ * server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
1706
+ * drain until the 15-min cron resets it — spamming new turns.
1707
+ *
1708
+ * CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
1709
+ * does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
1710
+ * in opencode must still be delivered via `markDone` on the next drain — so
1711
+ * `readoptOne` computes `state` FIRST and this set is checked only on the
1712
+ * non-done path. It is cleared once the row leaves the processing list (cron
1713
+ * reset → it drains normally as `pending`), so it can never leak.
1714
+ */
1715
+ dontRedispatch = /* @__PURE__ */ new Set();
1716
+ /**
1717
+ * "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
1718
+ * re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
1719
+ * never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
1720
+ * that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
1721
+ * markDone failure must NOT land here (it must still retry next drain). Separate
1722
+ * from `dontRedispatch` because the two concerns are independent: a row can need
1723
+ * "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
1724
+ * the row leaves the processing list, exactly like `dontRedispatch`.
1725
+ */
1726
+ doneUndeliverable = /* @__PURE__ */ new Set();
1727
+ /**
1728
+ * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
1729
+ * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
1730
+ * is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
1731
+ * persist hasn't landed before tick N+1 re-reads the still-null
1732
+ * `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
1733
+ * A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
1734
+ * short-circuits while it is present, so a null-id row is re-dispatched AT MOST
1735
+ * ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
1736
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
1737
+ * re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
1738
+ * so the NEXT tick may retry exactly once more).
1739
+ */
1740
+ awaitingReadopt = /* @__PURE__ */ new Set();
1566
1741
  /**
1567
1742
  * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
1568
1743
  * first session creation so drain-created sessions are rooted at the project
@@ -1585,7 +1760,7 @@ var ChannelDriver = class {
1585
1760
  this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
1586
1761
  this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
1587
1762
  this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
1588
- this.dispatchConfirmMs = config2.dispatchConfirmMs ?? DEFAULT_DISPATCH_CONFIRM_MS;
1763
+ this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
1589
1764
  this.now = config2.now ?? (() => Date.now());
1590
1765
  }
1591
1766
  /** The IPv4-loopback base URL for the local `opencode serve`. */
@@ -1618,6 +1793,7 @@ var ChannelDriver = class {
1618
1793
  for (const conv of conversations) {
1619
1794
  dispatched += await this.processConversation(conv);
1620
1795
  }
1796
+ await this.readoptProcessing();
1621
1797
  } finally {
1622
1798
  this.draining = false;
1623
1799
  }
@@ -1669,15 +1845,17 @@ var ChannelDriver = class {
1669
1845
  const sessionId = await this.ensureSession(conv);
1670
1846
  const messages = await this.getPendingMessages(conv.id);
1671
1847
  let dispatched = 0;
1848
+ let skippedAlreadyDispatched = 0;
1672
1849
  for (const message of messages) {
1673
1850
  if (this.dispatched.has(message.id)) {
1851
+ skippedAlreadyDispatched += 1;
1674
1852
  continue;
1675
1853
  }
1676
- const opencodeMessageId = opencodeMessageIdFor(message.id);
1677
1854
  const options = {
1678
1855
  agent: message.opencode_agent ?? void 0,
1679
1856
  model: message.opencode_model ?? void 0
1680
1857
  };
1858
+ let opencodeMessageId;
1681
1859
  try {
1682
1860
  this.log({
1683
1861
  level: "info",
@@ -1685,7 +1863,10 @@ var ChannelDriver = class {
1685
1863
  conversation_id: conv.id,
1686
1864
  message_id: message.id
1687
1865
  });
1688
- await sendPromptAsync(this.port, sessionId, message.content, options, opencodeMessageId);
1866
+ opencodeMessageId = await this.dispatchLocked(
1867
+ sessionId,
1868
+ () => sendPromptAsync(this.port, sessionId, message.content, options)
1869
+ );
1689
1870
  } catch (err) {
1690
1871
  if (err instanceof ChannelAuthError) throw err;
1691
1872
  this.dispatched.delete(message.id);
@@ -1699,9 +1880,26 @@ var ChannelDriver = class {
1699
1880
  });
1700
1881
  continue;
1701
1882
  }
1883
+ if (opencodeMessageId === null) {
1884
+ this.log({
1885
+ level: "error",
1886
+ message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
1887
+ conversation_id: conv.id,
1888
+ message_id: message.id
1889
+ });
1890
+ continue;
1891
+ }
1702
1892
  this.dispatched.add(message.id);
1703
1893
  this.registerInFlight(conv, sessionId, message, opencodeMessageId);
1704
1894
  dispatched += 1;
1895
+ void this.postSignal(conv.id, message.id, "dispatched");
1896
+ }
1897
+ if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
1898
+ this.log({
1899
+ level: "error",
1900
+ 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).`,
1901
+ conversation_id: conv.id
1902
+ });
1705
1903
  }
1706
1904
  this.ensureWatcherRunning(sessionId);
1707
1905
  return dispatched;
@@ -1739,6 +1937,25 @@ var ChannelDriver = class {
1739
1937
  // -------------------------------------------------------------------------
1740
1938
  // Per-session watcher (WI-3)
1741
1939
  // -------------------------------------------------------------------------
1940
+ /**
1941
+ * Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
1942
+ * opencode session (Task 2.1a), so two dispatches into the SAME session can
1943
+ * never interleave and mis-correlate their read-backs. Distinct sessions run
1944
+ * concurrently. The chained tail intentionally ignores the prior result/error
1945
+ * (each dispatch reports its own outcome to its caller).
1946
+ */
1947
+ dispatchLocked(sessionId, fn) {
1948
+ const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
1949
+ const run2 = prior.then(fn, fn);
1950
+ this.sessionDispatchLocks.set(
1951
+ sessionId,
1952
+ run2.then(
1953
+ () => void 0,
1954
+ () => void 0
1955
+ )
1956
+ );
1957
+ return run2;
1958
+ }
1742
1959
  /** Register a freshly-dispatched message with its session's watcher state. */
1743
1960
  registerInFlight(conv, sessionId, message, opencodeMessageId) {
1744
1961
  let watcher = this.watchers.get(sessionId);
@@ -1760,7 +1977,54 @@ var ChannelDriver = class {
1760
1977
  dispatchedAt: now,
1761
1978
  deadline: now + this.pausedMaxWaitMs,
1762
1979
  started: false,
1763
- done: false
1980
+ done: false,
1981
+ stuckReported: false
1982
+ });
1983
+ }
1984
+ /**
1985
+ * Register a RE-ADOPTED `processing` message with its session watcher
1986
+ * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
1987
+ * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
1988
+ * `now`: a row already `processing` for e.g. 5 min must give up ~5 min from now
1989
+ * (10 min after `processed_at`), not 10 min from now — otherwise its deadline
1990
+ * lands ~15 min after `processed_at`, coinciding with the cron reset →
1991
+ * double-drive race. `dispatchedAt` stays `now` (only the appear-guard uses it).
1992
+ *
1993
+ * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
1994
+ * `opencodeMessageId` is the id the watcher polls for a reply — for the orphan
1995
+ * fresh-run path these differ (a fresh opencode id under the same server row).
1996
+ *
1997
+ * `started` is set true so the watcher does NOT re-`markProcessing` a row the
1998
+ * server already flipped to `processing`; the running/done transitions still
1999
+ * fire from the watcher's normal branches.
2000
+ */
2001
+ registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
2002
+ let watcher = this.watchers.get(sessionId);
2003
+ if (!watcher) {
2004
+ watcher = {
2005
+ conv,
2006
+ inFlight: /* @__PURE__ */ new Map(),
2007
+ loop: null,
2008
+ reportedQuestions: /* @__PURE__ */ new Set(),
2009
+ reportedPermissions: /* @__PURE__ */ new Set()
2010
+ };
2011
+ this.watchers.set(sessionId, watcher);
2012
+ }
2013
+ watcher.inFlight.set(message.id, {
2014
+ evidentMessageId: message.id,
2015
+ opencodeMessageId,
2016
+ message,
2017
+ dispatchedAt: this.now(),
2018
+ deadline: processedAtMs + this.pausedMaxWaitMs,
2019
+ // The server row is ALREADY `processing`; do not re-fire markProcessing.
2020
+ started: true,
2021
+ done: false,
2022
+ // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
2023
+ // AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
2024
+ // on `state === 'queued'` (turn produced no reply), not on `started`, so a
2025
+ // re-adopted row left wedged in `queued` still emits the signal once
2026
+ // (#210/#220 observability).
2027
+ stuckReported: false
1764
2028
  });
1765
2029
  }
1766
2030
  /**
@@ -1826,6 +2090,7 @@ var ChannelDriver = class {
1826
2090
  conversation_id: watcher.conv.id
1827
2091
  });
1828
2092
  for (const evidentMessageId of [...watcher.inFlight.keys()]) {
2093
+ this.readopted.delete(evidentMessageId);
1829
2094
  this.removeInFlight(watcher, evidentMessageId);
1830
2095
  }
1831
2096
  return;
@@ -1846,10 +2111,15 @@ var ChannelDriver = class {
1846
2111
  async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
1847
2112
  const conv = watcher.conv;
1848
2113
  const state = messageRunState(messages, inFlight.opencodeMessageId);
1849
- if ((state === "running" || state === "done") && !inFlight.started) {
2114
+ if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
1850
2115
  let claimed;
1851
2116
  try {
1852
- claimed = await this.markProcessing(conv.id, inFlight.evidentMessageId, sessionId);
2117
+ claimed = await this.markProcessing(
2118
+ conv.id,
2119
+ inFlight.evidentMessageId,
2120
+ sessionId,
2121
+ inFlight.opencodeMessageId
2122
+ );
1853
2123
  } catch (err) {
1854
2124
  if (err instanceof ChannelAuthError) throw err;
1855
2125
  this.log({
@@ -1879,7 +2149,12 @@ var ChannelDriver = class {
1879
2149
  message_id: inFlight.evidentMessageId
1880
2150
  });
1881
2151
  try {
1882
- await this.markDone(conv.id, inFlight.evidentMessageId, sessionId);
2152
+ await this.markDone(
2153
+ conv.id,
2154
+ inFlight.evidentMessageId,
2155
+ sessionId,
2156
+ inFlight.opencodeMessageId
2157
+ );
1883
2158
  } catch (err) {
1884
2159
  if (err instanceof ChannelAuthError) throw err;
1885
2160
  if (err instanceof ChannelTerminalError) {
@@ -1915,10 +2190,59 @@ var ChannelDriver = class {
1915
2190
  this.removeInFlight(watcher, inFlight.evidentMessageId);
1916
2191
  return;
1917
2192
  }
1918
- if (state === "unknown") {
1919
- if (this.now() - inFlight.dispatchedAt >= this.dispatchConfirmMs) {
1920
- await this.redispatchInFlight(sessionId, inFlight);
2193
+ if (state === "failed") {
2194
+ if (!inFlight.done) {
2195
+ const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
2196
+ this.log({
2197
+ level: "error",
2198
+ message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2199
+ conversation_id: conv.id,
2200
+ message_id: inFlight.evidentMessageId
2201
+ });
2202
+ try {
2203
+ await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
2204
+ } catch (err) {
2205
+ if (err instanceof ChannelAuthError) throw err;
2206
+ if (err instanceof ChannelTerminalError) {
2207
+ this.log({
2208
+ level: "error",
2209
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
2210
+ conversation_id: conv.id,
2211
+ message_id: inFlight.evidentMessageId
2212
+ });
2213
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2214
+ return;
2215
+ }
2216
+ if (this.now() >= inFlight.deadline) {
2217
+ this.log({
2218
+ level: "error",
2219
+ 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)}`,
2220
+ conversation_id: conv.id,
2221
+ message_id: inFlight.evidentMessageId
2222
+ });
2223
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2224
+ return;
2225
+ }
2226
+ this.log({
2227
+ level: "error",
2228
+ message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
2229
+ conversation_id: conv.id,
2230
+ message_id: inFlight.evidentMessageId
2231
+ });
2232
+ return;
2233
+ }
2234
+ inFlight.done = true;
1921
2235
  }
2236
+ this.removeInFlight(watcher, inFlight.evidentMessageId);
2237
+ return;
2238
+ }
2239
+ const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
2240
+ const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
2241
+ if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
2242
+ inFlight.stuckReported = true;
2243
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
2244
+ stuck_for_ms: this.now() - inFlight.dispatchedAt
2245
+ });
1922
2246
  }
1923
2247
  if (this.now() >= inFlight.deadline) {
1924
2248
  this.log({
@@ -1927,48 +2251,386 @@ var ChannelDriver = class {
1927
2251
  conversation_id: conv.id,
1928
2252
  message_id: inFlight.evidentMessageId
1929
2253
  });
2254
+ void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
2255
+ watched_for_ms: this.now() - inFlight.dispatchedAt
2256
+ });
1930
2257
  this.removeInFlight(watcher, inFlight.evidentMessageId);
1931
2258
  }
1932
2259
  }
2260
+ // -------------------------------------------------------------------------
2261
+ // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
2262
+ // -------------------------------------------------------------------------
2263
+ /**
2264
+ * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
2265
+ *
2266
+ * The pending drain only re-drives `pending` rows; a message already flipped to
2267
+ * `processing` before the runner died is watched by nobody until the 15-min
2268
+ * cron resets it. Here we fetch those rows, and per row resolve its correlated
2269
+ * reply against opencode's OWN session store — completing, re-attaching, or
2270
+ * (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
2271
+ * is idempotent per message (Invariant 2): a row a watcher already tracks is
2272
+ * skipped in `readoptOne` — one driver, no double-drive.
2273
+ *
2274
+ * Only `ChannelAuthError` propagates (to `drainPending`, like the pending
2275
+ * path); every other early return LOGS a reason with context — no silent drop.
2276
+ */
2277
+ async readoptProcessing() {
2278
+ const rows = await this.getProcessingMessages();
2279
+ if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0) {
2280
+ const stillProcessing = new Set(rows.map((r) => r.id));
2281
+ for (const id of [...this.dontRedispatch, ...this.doneUndeliverable]) {
2282
+ if (!stillProcessing.has(id)) {
2283
+ const cleared = this.dontRedispatch.delete(id);
2284
+ const clearedUndeliverable = this.doneUndeliverable.delete(id);
2285
+ if (cleared || clearedUndeliverable) {
2286
+ this.log({
2287
+ level: "info",
2288
+ message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
2289
+ message_id: id
2290
+ });
2291
+ }
2292
+ }
2293
+ }
2294
+ }
2295
+ if (rows.length === 0) return;
2296
+ const bySession = /* @__PURE__ */ new Map();
2297
+ for (const row of rows) {
2298
+ if (!row.opencode_session_id) {
2299
+ this.log({
2300
+ level: "error",
2301
+ message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
2302
+ conversation_id: row.conversation_id,
2303
+ message_id: row.id
2304
+ });
2305
+ continue;
2306
+ }
2307
+ const list = bySession.get(row.opencode_session_id) ?? [];
2308
+ list.push(row);
2309
+ bySession.set(row.opencode_session_id, list);
2310
+ }
2311
+ for (const [sessionId, sessionRows] of bySession) {
2312
+ let messages;
2313
+ try {
2314
+ const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
2315
+ if (!res.ok) {
2316
+ this.log({
2317
+ level: "error",
2318
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
2319
+ });
2320
+ continue;
2321
+ }
2322
+ const body = await res.json();
2323
+ if (!Array.isArray(body)) {
2324
+ this.log({
2325
+ level: "error",
2326
+ message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
2327
+ });
2328
+ continue;
2329
+ }
2330
+ messages = body;
2331
+ } catch (err) {
2332
+ this.log({
2333
+ level: "error",
2334
+ message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
2335
+ });
2336
+ continue;
2337
+ }
2338
+ for (const row of sessionRows) {
2339
+ await this.readoptOne(sessionId, row, messages);
2340
+ }
2341
+ }
2342
+ }
2343
+ /**
2344
+ * Re-adopt ONE `processing` row against the tick's session message snapshot
2345
+ * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
2346
+ *
2347
+ * Branches on `messageRunState(messages, row.opencode_message_id)` — the
2348
+ * opencode-assigned user-message id persisted on the first `processing` PATCH
2349
+ * (#218). A row with a NULL stored id (dispatched but the read-back never landed
2350
+ * before the restart) has no id to correlate → treated as an orphan and
2351
+ * re-dispatched (at most once, see `forceReadoptRun`):
2352
+ * - `done` → `markDone` now (guarded like the watcher's done branch);
2353
+ * - `failed` → `markFailed` with the surfaced error (issue #182), so an
2354
+ * errored turn is reported failed on restart, NOT re-dispatched;
2355
+ * - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
2356
+ * tracking the stored id so the reply correlates by it;
2357
+ * - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
2358
+ *
2359
+ * Only `ChannelAuthError` propagates.
2360
+ */
2361
+ async readoptOne(sessionId, row, messages) {
2362
+ if (this.isTracked(sessionId, row.id)) {
2363
+ this.log({
2364
+ level: "info",
2365
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
2366
+ conversation_id: row.conversation_id,
2367
+ message_id: row.id
2368
+ });
2369
+ return;
2370
+ }
2371
+ const ocId = row.opencode_message_id;
2372
+ const state = messageRunState(messages, ocId ?? "");
2373
+ if (state === "done") {
2374
+ if (this.doneUndeliverable.has(row.id)) {
2375
+ this.log({
2376
+ level: "info",
2377
+ message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
2378
+ conversation_id: row.conversation_id,
2379
+ message_id: row.id
2380
+ });
2381
+ return;
2382
+ }
2383
+ this.log({
2384
+ level: "info",
2385
+ message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
2386
+ conversation_id: row.conversation_id,
2387
+ message_id: row.id
2388
+ });
2389
+ try {
2390
+ await this.markDone(row.conversation_id, row.id, sessionId, ocId);
2391
+ } catch (err) {
2392
+ if (err instanceof ChannelAuthError) throw err;
2393
+ if (err instanceof ChannelTerminalError) {
2394
+ this.doneUndeliverable.add(row.id);
2395
+ this.log({
2396
+ level: "error",
2397
+ 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}`,
2398
+ conversation_id: row.conversation_id,
2399
+ message_id: row.id
2400
+ });
2401
+ return;
2402
+ }
2403
+ this.log({
2404
+ level: "error",
2405
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2406
+ conversation_id: row.conversation_id,
2407
+ message_id: row.id
2408
+ });
2409
+ return;
2410
+ }
2411
+ this.dontRedispatch.delete(row.id);
2412
+ return;
2413
+ }
2414
+ if (state === "failed") {
2415
+ const error2 = messageError(messages, ocId ?? "") ?? void 0;
2416
+ this.log({
2417
+ level: "error",
2418
+ message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
2419
+ conversation_id: row.conversation_id,
2420
+ message_id: row.id
2421
+ });
2422
+ try {
2423
+ await this.markFailed(row.conversation_id, row.id, sessionId, error2);
2424
+ } catch (err) {
2425
+ if (err instanceof ChannelAuthError) throw err;
2426
+ if (err instanceof ChannelTerminalError) {
2427
+ this.doneUndeliverable.add(row.id);
2428
+ this.log({
2429
+ level: "error",
2430
+ 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}`,
2431
+ conversation_id: row.conversation_id,
2432
+ message_id: row.id
2433
+ });
2434
+ return;
2435
+ }
2436
+ this.log({
2437
+ level: "error",
2438
+ message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2439
+ conversation_id: row.conversation_id,
2440
+ message_id: row.id
2441
+ });
2442
+ return;
2443
+ }
2444
+ this.dontRedispatch.delete(row.id);
2445
+ return;
2446
+ }
2447
+ if (this.dontRedispatch.has(row.id)) {
2448
+ this.log({
2449
+ level: "info",
2450
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
2451
+ conversation_id: row.conversation_id,
2452
+ message_id: row.id
2453
+ });
2454
+ return;
2455
+ }
2456
+ if ((state === "running" || state === "queued") && ocId) {
2457
+ const conv = this.convForRow(sessionId, row);
2458
+ const message = this.queuedMessageForRow(row);
2459
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
2460
+ this.dispatched.add(row.id);
2461
+ this.readopted.add(row.id);
2462
+ this.ensureWatcherRunning(sessionId);
2463
+ this.log({
2464
+ level: "info",
2465
+ message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
2466
+ conversation_id: row.conversation_id,
2467
+ message_id: row.id
2468
+ });
2469
+ return;
2470
+ }
2471
+ await this.forceReadoptRun(sessionId, row);
2472
+ }
1933
2473
  /**
1934
- * Re-dispatch a message whose user row never appeared (idle-path guard). Safe:
1935
- * opencode treats a duplicate caller-supplied `messageID` as idempotent (PoC
1936
- * fact 9) one user message + one reply even if the original DID land. Resets
1937
- * the dispatch timestamp so the guard doesn't immediately fire again.
2474
+ * Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
2475
+ *
2476
+ * #218/WI-5: the row's user message is absent (never kept, or a null stored id),
2477
+ * so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
2478
+ * read it back, and register the watcher under the assigned id so the reply
2479
+ * correlates server-side.
2480
+ *
2481
+ * ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
2482
+ * id). Without a guard, if this dispatches on tick N but the read-back+persist
2483
+ * hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
2484
+ * tick N+1 would dispatch AGAIN → duplicate user turns. The `awaitingReadopt`
2485
+ * latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
2486
+ * short-circuit while the row is latched; clear it on a successful dispatch (the
2487
+ * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
2488
+ * re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
2489
+ * may retry exactly once more).
2490
+ *
2491
+ * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
2492
+ * `processed_at` (Invariant 1).
1938
2493
  */
1939
- async redispatchInFlight(sessionId, inFlight) {
2494
+ async forceReadoptRun(sessionId, row) {
2495
+ if (this.awaitingReadopt.has(row.id)) {
2496
+ this.log({
2497
+ level: "info",
2498
+ message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
2499
+ conversation_id: row.conversation_id,
2500
+ message_id: row.id
2501
+ });
2502
+ return;
2503
+ }
2504
+ if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
2505
+ this.dontRedispatch.add(row.id);
2506
+ this.log({
2507
+ level: "info",
2508
+ 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)`,
2509
+ conversation_id: row.conversation_id,
2510
+ message_id: row.id
2511
+ });
2512
+ return;
2513
+ }
1940
2514
  const options = {
1941
- agent: inFlight.message.opencode_agent ?? void 0,
1942
- model: inFlight.message.opencode_model ?? void 0
2515
+ agent: row.opencode_agent ?? void 0,
2516
+ model: row.opencode_model ?? void 0
1943
2517
  };
1944
2518
  this.log({
1945
2519
  level: "info",
1946
- message: `Message ${inFlight.evidentMessageId.slice(0, 8)} not observed after dispatch \u2014 re-dispatching (idle-path guard)`,
1947
- message_id: inFlight.evidentMessageId
2520
+ message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
2521
+ conversation_id: row.conversation_id,
2522
+ message_id: row.id
1948
2523
  });
2524
+ this.awaitingReadopt.add(row.id);
2525
+ let ocId;
1949
2526
  try {
1950
- await sendPromptAsync(
1951
- this.port,
2527
+ ocId = await this.dispatchLocked(
1952
2528
  sessionId,
1953
- inFlight.message.content,
1954
- options,
1955
- inFlight.opencodeMessageId
2529
+ () => sendPromptAsync(this.port, sessionId, row.content, options)
1956
2530
  );
1957
2531
  } catch (err) {
2532
+ this.awaitingReadopt.delete(row.id);
2533
+ if (err instanceof ChannelAuthError) throw err;
1958
2534
  this.log({
1959
2535
  level: "error",
1960
- message: `Re-dispatch failed for message ${inFlight.evidentMessageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
1961
- message_id: inFlight.evidentMessageId
2536
+ message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
2537
+ conversation_id: row.conversation_id,
2538
+ message_id: row.id
2539
+ });
2540
+ return;
2541
+ }
2542
+ if (ocId === null) {
2543
+ this.awaitingReadopt.delete(row.id);
2544
+ this.log({
2545
+ level: "error",
2546
+ message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next drain`,
2547
+ conversation_id: row.conversation_id,
2548
+ message_id: row.id
1962
2549
  });
2550
+ return;
1963
2551
  }
1964
- inFlight.dispatchedAt = this.now();
2552
+ const conv = this.convForRow(sessionId, row);
2553
+ const message = this.queuedMessageForRow(row);
2554
+ this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
2555
+ this.dispatched.add(row.id);
2556
+ this.readopted.add(row.id);
2557
+ this.awaitingReadopt.delete(row.id);
2558
+ this.ensureWatcherRunning(sessionId);
2559
+ }
2560
+ /**
2561
+ * True if `evidentMessageId` is already being driven — either in the
2562
+ * authoritative `dispatched` set or a live watcher's in-flight set for this
2563
+ * session (Invariant 2, WI-5). Either signal means a watcher owns the row.
2564
+ */
2565
+ isTracked(sessionId, evidentMessageId) {
2566
+ if (this.dispatched.has(evidentMessageId)) return true;
2567
+ const watcher = this.watchers.get(sessionId);
2568
+ return watcher?.inFlight.has(evidentMessageId) ?? false;
2569
+ }
2570
+ /**
2571
+ * Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
2572
+ * deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
2573
+ * for `processing` rows, but if it is somehow null/unparseable fall back to
2574
+ * `now` (defensive) AND log — a fallback means the anchor is weaker than
2575
+ * intended, which is worth surfacing.
2576
+ */
2577
+ processedAtMs(row) {
2578
+ const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
2579
+ if (!Number.isNaN(parsed)) return parsed;
2580
+ this.log({
2581
+ level: "error",
2582
+ message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
2583
+ conversation_id: row.conversation_id,
2584
+ message_id: row.id
2585
+ });
2586
+ return this.now();
2587
+ }
2588
+ /** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
2589
+ convForRow(sessionId, row) {
2590
+ return {
2591
+ id: row.conversation_id,
2592
+ agent_id: this.agentId,
2593
+ opencode_session_id: sessionId,
2594
+ pending_message_count: 0,
2595
+ oldest_pending_at: row.processed_at
2596
+ };
2597
+ }
2598
+ /** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
2599
+ queuedMessageForRow(row) {
2600
+ return {
2601
+ id: row.id,
2602
+ content: row.content,
2603
+ status: "processing",
2604
+ opencode_agent: row.opencode_agent,
2605
+ opencode_model: row.opencode_model,
2606
+ source_message_id: row.source_message_id,
2607
+ slack_user_id: row.slack_user_id
2608
+ };
1965
2609
  }
1966
2610
  /**
1967
2611
  * Remove a message from the in-flight set AND the authoritative dispatched
1968
2612
  * set. Once the in-flight set empties, the watcher loop's `while` guard exits
1969
2613
  * and its `.finally` removes the session entry from `this.watchers`.
2614
+ *
2615
+ * Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
2616
+ * (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
2617
+ * cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
2618
+ * re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
2619
+ * re-adopted message that completed (`done`) needs no marker — it's leaving
2620
+ * `processing`. This suppresses only re-dispatch: if its reply later completes,
2621
+ * the done branch still delivers it (Bugbot #202).
1970
2622
  */
1971
2623
  removeInFlight(watcher, evidentMessageId) {
2624
+ const inFlight = watcher.inFlight.get(evidentMessageId);
2625
+ if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
2626
+ this.dontRedispatch.add(evidentMessageId);
2627
+ this.log({
2628
+ level: "info",
2629
+ message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
2630
+ conversation_id: watcher.conv.id,
2631
+ message_id: evidentMessageId
2632
+ });
2633
+ }
1972
2634
  watcher.inFlight.delete(evidentMessageId);
1973
2635
  this.dispatched.delete(evidentMessageId);
1974
2636
  }
@@ -2112,6 +2774,35 @@ var ChannelDriver = class {
2112
2774
  }
2113
2775
  return await res.json();
2114
2776
  }
2777
+ /**
2778
+ * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
2779
+ * The pending path (`getPendingConversations`/`getPendingMessages`) only
2780
+ * surfaces `pending` rows, so a message already `processing` when the runner
2781
+ * died is invisible to it — this dedicated endpoint returns exactly those rows
2782
+ * with the fields the re-adopt path needs (`processed_at`,
2783
+ * `opencode_session_id`, routing).
2784
+ *
2785
+ * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
2786
+ * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
2787
+ * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
2788
+ * next tick retries.
2789
+ */
2790
+ async getProcessingMessages() {
2791
+ const res = await this.fetchImpl(
2792
+ `${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
2793
+ { headers: { Authorization: this.getAuthHeader() } }
2794
+ );
2795
+ this.assertAuth(res, "fetching processing messages");
2796
+ if (!res.ok) {
2797
+ throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
2798
+ }
2799
+ const data = await res.json();
2800
+ let messages = data.messages ?? [];
2801
+ if (this.conversationFilter) {
2802
+ messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
2803
+ }
2804
+ return messages;
2805
+ }
2115
2806
  /**
2116
2807
  * EXISTING combinedAuth route — now fired by the watcher on queued→running
2117
2808
  * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -2133,13 +2824,17 @@ var ChannelDriver = class {
2133
2824
  * A single attempt (no internal retry): the watcher's per-tick loop is the
2134
2825
  * retry vehicle for the swap-to-running.
2135
2826
  */
2136
- async markProcessing(conversationId, messageId, sessionId) {
2827
+ async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
2137
2828
  const res = await this.fetchImpl(
2138
2829
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
2139
2830
  {
2140
2831
  method: "PATCH",
2141
2832
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2142
- body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
2833
+ body: JSON.stringify({
2834
+ status: "processing",
2835
+ opencode_session_id: sessionId,
2836
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
2837
+ })
2143
2838
  }
2144
2839
  );
2145
2840
  this.assertAuth(res, "marking message as processing");
@@ -2177,13 +2872,17 @@ var ChannelDriver = class {
2177
2872
  * watcher retries next tick within the
2178
2873
  * deadline, Finding 4).
2179
2874
  */
2180
- async markDone(conversationId, messageId, sessionId) {
2875
+ async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
2181
2876
  const res = await this.fetchImpl(
2182
2877
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
2183
2878
  {
2184
2879
  method: "PATCH",
2185
2880
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2186
- body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
2881
+ body: JSON.stringify({
2882
+ status: "done",
2883
+ opencode_session_id: sessionId,
2884
+ ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
2885
+ })
2187
2886
  }
2188
2887
  );
2189
2888
  this.assertAuth(res, "marking message as done");
@@ -2193,7 +2892,17 @@ var ChannelDriver = class {
2193
2892
  }
2194
2893
  throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
2195
2894
  }
2196
- async markFailed(conversationId, messageId) {
2895
+ /**
2896
+ * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
2897
+ * when provided (issue #182): a bare `markFailed(conv, msg)` sends
2898
+ * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
2899
+ * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
2900
+ * failure reason reaches the channel.
2901
+ */
2902
+ async markFailed(conversationId, messageId, sessionId, error2) {
2903
+ const body = { status: "failed" };
2904
+ if (sessionId !== void 0) body.opencode_session_id = sessionId;
2905
+ if (error2 !== void 0) body.error = error2;
2197
2906
  await this.callWithRetry(
2198
2907
  "marking message as failed",
2199
2908
  () => this.fetchImpl(
@@ -2201,11 +2910,47 @@ var ChannelDriver = class {
2201
2910
  {
2202
2911
  method: "PATCH",
2203
2912
  headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2204
- body: JSON.stringify({ status: "failed" })
2913
+ body: JSON.stringify(body)
2205
2914
  }
2206
2915
  )
2207
2916
  );
2208
2917
  }
2918
+ /**
2919
+ * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
2920
+ * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
2921
+ * — the server records it via `log()` (no DB write, no notification). This is
2922
+ * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
2923
+ * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
2924
+ * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
2925
+ * context (no silent catch, per development-workflow).
2926
+ */
2927
+ async postSignal(conversationId, messageId, signal, extra) {
2928
+ try {
2929
+ const res = await this.fetchImpl(
2930
+ `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
2931
+ {
2932
+ method: "POST",
2933
+ headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
2934
+ body: JSON.stringify({ signal, ...extra })
2935
+ }
2936
+ );
2937
+ if (!res.ok) {
2938
+ this.log({
2939
+ level: "error",
2940
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
2941
+ conversation_id: conversationId,
2942
+ message_id: messageId
2943
+ });
2944
+ }
2945
+ } catch (err) {
2946
+ this.log({
2947
+ level: "error",
2948
+ message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
2949
+ conversation_id: conversationId,
2950
+ message_id: messageId
2951
+ });
2952
+ }
2953
+ }
2209
2954
  async persistSession(conversationId, sessionId) {
2210
2955
  const res = await this.fetchImpl(
2211
2956
  `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}`,
@@ -2521,7 +3266,8 @@ async function getAgentInfo(agentId, authHeader) {
2521
3266
  // src/commands/run.ts
2522
3267
  var MAX_ACTIVITY_LOG_ENTRIES = 10;
2523
3268
  var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
2524
- function log(state, message, isError = false) {
3269
+ var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
3270
+ function log2(state, message, isError = false) {
2525
3271
  if (state.json) {
2526
3272
  console.log(
2527
3273
  JSON.stringify({
@@ -2546,9 +3292,9 @@ function logActivity(state, entry) {
2546
3292
  }
2547
3293
  if (!state.interactive) {
2548
3294
  if (entry.type === "error") {
2549
- log(state, entry.error ?? "Unknown error", true);
3295
+ log2(state, entry.error ?? "Unknown error", true);
2550
3296
  } else if (entry.type === "info" && entry.message) {
2551
- log(state, entry.message);
3297
+ log2(state, entry.message);
2552
3298
  }
2553
3299
  }
2554
3300
  }
@@ -2634,6 +3380,7 @@ async function handleAuthError(state, error2) {
2634
3380
  }
2635
3381
  async function driveChannels(state, driver) {
2636
3382
  let idlePolls = 0;
3383
+ let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
2637
3384
  while (state.running) {
2638
3385
  if (state.connection?.reconnecting && state.connection.reconnectPromise) {
2639
3386
  logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2643,7 +3390,9 @@ async function driveChannels(state, driver) {
2643
3390
  try {
2644
3391
  const processed = await driver.drainPending();
2645
3392
  state.messageCount += processed;
2646
- if (processed > 0 || driver.hasInFlightWatchers()) {
3393
+ const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
3394
+ lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
3395
+ if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
2647
3396
  idlePolls = 0;
2648
3397
  if (processed > 0 && state.interactive) displayStatus(state);
2649
3398
  } else if (state.idleTimeout !== null) {
@@ -2695,7 +3444,7 @@ async function cleanup(state) {
2695
3444
  logActivity(state, { type: "info", message: "Stopped OpenCode process" });
2696
3445
  displayStatus(state);
2697
3446
  } else {
2698
- log(state, "Stopped OpenCode process");
3447
+ log2(state, "Stopped OpenCode process");
2699
3448
  }
2700
3449
  state.opencodeProcess = null;
2701
3450
  }
@@ -2718,10 +3467,11 @@ async function run(options) {
2718
3467
  running: true,
2719
3468
  activityLog: [],
2720
3469
  messageCount: 0,
3470
+ lastProxiedActivityAt: null,
2721
3471
  authHeader: ""
2722
3472
  };
2723
3473
  if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
2724
- log(
3474
+ log2(
2725
3475
  state,
2726
3476
  "Warning: No --idle-timeout set in CI environment. The runner will poll indefinitely until the job times out. Consider adding --idle-timeout 30 to avoid wasting runner minutes.",
2727
3477
  false
@@ -2732,7 +3482,7 @@ async function run(options) {
2732
3482
  logActivity(state, { type: "info", message: "Shutting down..." });
2733
3483
  displayStatus(state);
2734
3484
  } else {
2735
- log(state, "Shutting down...");
3485
+ log2(state, "Shutting down...");
2736
3486
  }
2737
3487
  await cleanup(state);
2738
3488
  await shutdownTelemetry();
@@ -2765,7 +3515,7 @@ async function run(options) {
2765
3515
  const resolved = await resolveAgentIdFromKey(state.authHeader);
2766
3516
  if (resolved.agent_id) {
2767
3517
  state.agentId = resolved.agent_id;
2768
- log(state, `Resolved agent ID from key: ${state.agentId}`);
3518
+ log2(state, `Resolved agent ID from key: ${state.agentId}`);
2769
3519
  if (state.interactive && !state.json) {
2770
3520
  logActivity(state, {
2771
3521
  type: "info",
@@ -2828,17 +3578,17 @@ async function run(options) {
2828
3578
  port: state.port,
2829
3579
  interactive: state.interactive,
2830
3580
  agentId: state.agentId,
2831
- log: (message) => log(state, message)
3581
+ log: (message) => log2(state, message)
2832
3582
  });
2833
3583
  state.port = oc.port;
2834
3584
  state.opencodeProcess = oc.process;
2835
3585
  state.opencodeVersion = oc.version;
2836
3586
  state.opencodeConnected = oc.process !== null || oc.version !== null;
2837
- const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
2838
- ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
3587
+ const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
3588
+ ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
2839
3589
  const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
2840
3590
  if (versionWarning) {
2841
- log(state, versionWarning, false);
3591
+ log2(state, versionWarning, false);
2842
3592
  if (state.interactive && !state.json) {
2843
3593
  logActivity(state, { type: "info", message: versionWarning });
2844
3594
  }
@@ -2854,6 +3604,7 @@ async function run(options) {
2854
3604
  apiUrl: getApiUrlConfig(),
2855
3605
  getAuthHeader: () => state.authHeader,
2856
3606
  conversationFilter: state.conversationFilter,
3607
+ stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
2857
3608
  log: (entry) => logActivity(state, {
2858
3609
  type: entry.level === "error" ? "error" : "info",
2859
3610
  message: entry.message,
@@ -2873,7 +3624,11 @@ async function run(options) {
2873
3624
  type: "info",
2874
3625
  message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
2875
3626
  });
2876
- emitAgentConnected(state.agentId, { port: state.port });
3627
+ emitAgentConnected(state.agentId, {
3628
+ port: state.port,
3629
+ cli_version: getCliVersion(),
3630
+ opencode_version: state.opencodeVersion
3631
+ });
2877
3632
  if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
2878
3633
  if (state.interactive) displayStatus(state);
2879
3634
  channelDriver.drainPending().then((processed) => {
@@ -2907,9 +3662,14 @@ async function run(options) {
2907
3662
  logActivity(state, { type: "error", error: error2 });
2908
3663
  if (state.interactive) displayStatus(state);
2909
3664
  },
2910
- // Web traffic is proxied transparently; only note opencode is live.
3665
+ // Web traffic is proxied transparently; note opencode is live and stamp
3666
+ // proxied activity so the idle loop treats interactive proxy use as work.
3667
+ // Fires per forwarded response head (incl. every SSE open) and excludes
3668
+ // the internal drain-ping, so an actively-used proxy keeps the timer
3669
+ // fresh while a lone idle SSE with no follow-up requests still ages out.
2911
3670
  onResponse: () => {
2912
3671
  state.opencodeConnected = true;
3672
+ state.lastProxiedActivityAt = Date.now();
2913
3673
  },
2914
3674
  // A channel message was queued and the api-worker pinged us over the
2915
3675
  // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2948,7 +3708,7 @@ async function run(options) {
2948
3708
  throw error2;
2949
3709
  }
2950
3710
  if (!interactive || state.json) {
2951
- log(state, "Driving channel messages...");
3711
+ log2(state, "Driving channel messages...");
2952
3712
  }
2953
3713
  await driveChannels(state, channelDriver);
2954
3714
  await cleanup(state);
@@ -2960,7 +3720,7 @@ async function run(options) {
2960
3720
  })
2961
3721
  );
2962
3722
  } else if (!interactive) {
2963
- log(state, `Completed. Processed ${state.messageCount} message(s).`);
3723
+ log2(state, `Completed. Processed ${state.messageCount} message(s).`);
2964
3724
  }
2965
3725
  await shutdownTelemetry();
2966
3726
  process.exit(0);
@@ -2982,8 +3742,9 @@ async function run(options) {
2982
3742
  }
2983
3743
 
2984
3744
  // src/index.ts
3745
+ var { version } = createRequire(import.meta.url)("../package.json");
2985
3746
  var program = new Command();
2986
- program.name("evident").description("Run OpenCode locally and connect it to Evident").version("0.1.0").option(
3747
+ program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
2987
3748
  "--endpoint <url>",
2988
3749
  "Evident API base URL (default: production; e.g. http://localhost:3001)"
2989
3750
  ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {