@evident-ai/cli 3.0.1-dev.7a6d5aa → 3.0.1-dev.7bf63a1

package/dist/index.js

@@ -1,6 +1,7 @@
 #!/usr/bin/env node
 
 // src/index.ts
+import { createRequire } from "module";
 import { Command } from "commander";
 
 // src/commands/login.ts
@@ -484,8 +485,34 @@ var TelemetryEventTypes = {
 var MAX_FRAME_BYTES = 256 * 1024;
 var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
 
+// ../../packages/types/src/logging/index.ts
+var CORRELATION_ID_HEADER = "x-evident-correlation-id";
+function log(level, event, fields) {
+  const method = level === "debug" ? "log" : level;
+  try {
+    console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
+  } catch (err) {
+    console.error(
+      "[evident] log_serialize_failed",
+      event,
+      err instanceof Error ? err.message : String(err)
+    );
+  }
+}
+function stripQuery(url) {
+  try {
+    return new URL(url).pathname;
+  } catch {
+    const q = url.indexOf("?");
+    return q === -1 ? url : url.slice(0, q);
+  }
+}
+
 // src/lib/telemetry.ts
-var CLI_VERSION = process.env.npm_package_version || "unknown";
+var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
+function getCliVersion() {
+  return CLI_VERSION;
+}
 var eventBuffer = [];
 var flushTimeout = null;
 var isShuttingDown = false;
@@ -685,14 +712,14 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
 }
 
 // src/lib/opencode/opencode-version-gate.ts
-var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
-function isQueueValidatedVersion(version) {
-  if (!version) return false;
-  return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version);
-}
-function buildOpenCodeVersionWarning(version) {
-  if (isQueueValidatedVersion(version)) return null;
-  const detected = version ? `v${version}` : "unknown";
+var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
+function isQueueValidatedVersion(version2) {
+  if (!version2) return false;
+  return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
+}
+function buildOpenCodeVersionWarning(version2) {
+  if (isQueueValidatedVersion(version2)) return null;
+  const detected = version2 ? `v${version2}` : "unknown";
   const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
   return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack/WhatsApp) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
 }
@@ -1009,7 +1036,11 @@ function roleOf(m) {
 }
 function completedOf(m) {
   if (!m || typeof m !== "object") return void 0;
-  return m.info?.time?.completed;
+  return m.info?.time?.completed ?? m.time?.completed;
+}
+function createdOf(m) {
+  if (!m || typeof m !== "object") return void 0;
+  return m.info?.time?.created ?? m.time?.created;
 }
 function idOf(m) {
   if (!m || typeof m !== "object") return void 0;
@@ -1029,6 +1060,24 @@ function finishOf(m) {
   const infoFinish = m.info?.finish;
   return typeof infoFinish === "string" ? infoFinish : void 0;
 }
+function errorOf(m) {
+  if (!m || typeof m !== "object") return void 0;
+  return m.info?.error ?? m.error;
+}
+function isAssistantInFlight(m) {
+  if (completedOf(m) == null) return true;
+  return finishOf(m) === "tool-calls";
+}
+async function getSessionMessages(port, sessionId) {
+  try {
+    const res = await fetch(`${opencodeBase(port)}/session/${sessionId}/message`);
+    if (!res.ok) return null;
+    const body = await res.json();
+    return Array.isArray(body) ? body : null;
+  } catch {
+    return null;
+  }
+}
 async function createOpenCodeSession(port, directory) {
   const url = new URL(`${opencodeBase(port)}/session`);
   if (directory && directory.trim()) {
@@ -1046,9 +1095,16 @@ async function createOpenCodeSession(port, directory) {
   const data = await response.json();
   return data.id;
 }
-async function sendPromptAsync(port, sessionId, content, options, messageId) {
+function messageText(m) {
+  if (!m || !Array.isArray(m.parts)) return "";
+  return m.parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("");
+}
+async function sendPromptAsync(port, sessionId, content, options) {
+  const before = await getSessionMessages(port, sessionId);
+  const knownUserIds = new Set(
+    (before ?? []).filter((m) => roleOf(m) === "user").map((m) => idOf(m)).filter((id) => typeof id === "string")
+  );
   const body = {
-    messageID: messageId,
     parts: [{ type: "text", text: content }]
   };
   if (options?.agent) {
@@ -1072,6 +1128,29 @@ async function sendPromptAsync(port, sessionId, content, options, messageId) {
     const text = await res.text().catch(() => "");
     throw new Error(`OpenCode prompt_async failed: HTTP ${res.status}${text ? `: ${text}` : ""}`);
   }
+  const READ_BACK_ATTEMPTS = 5;
+  const READ_BACK_DELAY_MS = 150;
+  for (let attempt = 0; attempt < READ_BACK_ATTEMPTS; attempt++) {
+    const after = await getSessionMessages(port, sessionId);
+    if (after) {
+      let best = null;
+      for (const m of after) {
+        if (roleOf(m) !== "user") continue;
+        const id = idOf(m);
+        if (typeof id !== "string" || knownUserIds.has(id)) continue;
+        if (messageText(m) !== content) continue;
+        const created = createdOf(m) ?? 0;
+        if (best === null || created > best.created) {
+          best = { id, created };
+        }
+      }
+      if (best) return best.id;
+    }
+    if (attempt < READ_BACK_ATTEMPTS - 1) {
+      await new Promise((resolve) => setTimeout(resolve, READ_BACK_DELAY_MS));
+    }
+  }
+  return null;
 }
 function findAssistantReplyAfter(messages, userMessageId) {
   if (!messages || messages.length === 0) return null;
@@ -1088,19 +1167,31 @@ function findAssistantReplyAfter(messages, userMessageId) {
 }
 function findLastAssistantReplyFor(messages, userMessageId) {
   if (!messages || messages.length === 0) return null;
+  let lastCorrelated = null;
+  let lastNonErrored = null;
   for (let i = messages.length - 1; i >= 0; i--) {
     const m = messages[i];
-    if (roleOf(m) === "assistant" && parentIdOf(m) === userMessageId) return m;
+    if (roleOf(m) !== "assistant" || parentIdOf(m) !== userMessageId) continue;
+    if (lastCorrelated === null) lastCorrelated = m;
+    if (errorOf(m) == null) {
+      lastNonErrored = m;
+      break;
+    }
   }
+  if (lastCorrelated) return lastNonErrored ?? lastCorrelated;
   const userIndex = messages.findIndex((m) => idOf(m) === userMessageId);
   if (userIndex === -1) return null;
   let last = null;
+  let lastOk = null;
   for (let i = userIndex + 1; i < messages.length; i++) {
     const role = roleOf(messages[i]);
     if (role === "user") break;
-    if (role === "assistant") last = messages[i];
+    if (role === "assistant") {
+      last = messages[i];
+      if (errorOf(messages[i]) == null) lastOk = messages[i];
+    }
   }
-  return last;
+  return lastOk ?? last;
 }
 function messageRunState(messages, userMessageId) {
   if (!messages || messages.length === 0) return "unknown";
@@ -1110,13 +1201,27 @@ function messageRunState(messages, userMessageId) {
     if (!reply) return "unknown";
   }
   if (!reply) return "queued";
-  if (completedOf(reply) == null) return "running";
-  if (finishOf(reply) === "tool-calls") return "running";
-  return "done";
+  if (isAssistantInFlight(reply)) return "running";
+  return errorOf(reply) != null ? "failed" : "done";
 }
-function opencodeMessageIdFor(queuedMessageId) {
-  const sanitized = queuedMessageId.replace(/[^a-zA-Z0-9]/g, "_");
-  return `msg_${sanitized}`;
+function messageError(messages, userMessageId) {
+  const reply = findLastAssistantReplyFor(messages, userMessageId);
+  const error2 = errorOf(reply);
+  if (error2 == null) return null;
+  if (typeof error2 === "string") return error2;
+  if (typeof error2 === "object") {
+    const e = error2;
+    const dataMessage = e.data?.message;
+    if (typeof dataMessage === "string") return dataMessage;
+    if (typeof e.message === "string") return e.message;
+  }
+  return "The agent run failed.";
+}
+function hasRunningAssistantExcept(messages, exceptUserMessageId) {
+  if (!messages || messages.length === 0) return false;
+  return messages.some(
+    (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
+  );
 }
 
 // src/lib/tunnel/connection.ts
@@ -1187,12 +1292,22 @@ var StreamForwarder = class {
   }
   async handleOpen(frame) {
     const { sid, method, path, headers, has_body } = frame;
+    const correlationId = headers?.[CORRELATION_ID_HEADER];
+    const startedAt = Date.now();
     if (path === TUNNEL_DRAIN_PING_PATH) {
       this.callbacks.onDrainPing?.();
       this.send({ type: "head", sid, status: 204, headers: {} });
       this.send({ type: "res_end", sid });
       return;
     }
+    if (process.env.DEBUG) {
+      log("debug", "agent_request", {
+        correlation_id: correlationId,
+        sid,
+        method,
+        path: stripQuery(path)
+      });
+    }
     const ac = new AbortController();
     let bodyPromise;
     let pushBody;
@@ -1239,6 +1354,14 @@ var StreamForwarder = class {
       if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
     });
     this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
+    if (process.env.DEBUG) {
+      log("debug", "agent_response", {
+        correlation_id: correlationId,
+        sid,
+        status: upstream.status,
+        duration_ms: Date.now() - startedAt
+      });
+    }
     this.callbacks.onHead?.(sid, upstream.status);
     try {
       if (upstream.body) {
@@ -1508,7 +1631,7 @@ var DEFAULT_RETRY_POLICY = {
 };
 var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
 var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
-var DEFAULT_DISPATCH_CONFIRM_MS = 6e3;
+var DEFAULT_STUCK_QUEUED_MS = 6e4;
 var ChannelAuthError = class extends Error {
   constructor(message) {
     super(message);
@@ -1543,10 +1666,18 @@ var ChannelDriver = class {
   sleep;
   pausedPollIntervalMs;
   pausedMaxWaitMs;
-  dispatchConfirmMs;
+  stuckQueuedMs;
   now;
   /** Cache of conversationId → opencode sessionId. */
   sessions = /* @__PURE__ */ new Map();
+  /**
+   * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
+   * longer idempotent (no caller-supplied `messageID`), and its read-back picks
+   * "the one new user row" — which is only unambiguous if no OTHER dispatch into
+   * the SAME session interleaves its snapshot→POST→read-back. This map chains each
+   * session's dispatches so they run serially; distinct sessions stay concurrent.
+   */
+  sessionDispatchLocks = /* @__PURE__ */ new Map();
   /**
    * Per-SESSION watchers (WI-3), keyed by opencode sessionId. Single-flight per
    * session: one polling loop services all of that session's in-flight messages.
@@ -1563,6 +1694,54 @@ var ChannelDriver = class {
    * a steady-state-poll re-dispatch will not double-run the message.
    */
   dispatched = /* @__PURE__ */ new Set();
+  /**
+   * Re-adopted (ADR-0046) Evident message ids currently tracked by a watcher.
+   * Used only to distinguish a RE-ADOPTED give-up from a normal-dispatch give-up
+   * so the former can be parked in `dontRedispatch` (Bug 2). A row is added when
+   * it is re-adopted and removed when its watcher settles or it is observed off
+   * the processing list.
+   */
+  readopted = /* @__PURE__ */ new Set();
+  /**
+   * "Don't re-DISPATCH / re-attach this orphan again" (Bug 2/5). Set when a
+   * re-adopted running/orphan row's watcher hit its `processed_at`-anchored
+   * deadline (or an orphan whose window already elapsed): the still-`processing`
+   * server row would otherwise be re-adopted (and re-dispatched) on EVERY ~2s
+   * drain until the 15-min cron resets it — spamming new turns.
+   *
+   * CRITICAL (Bugbot #202): this suppresses ONLY the dispatch/re-attach paths, it
+   * does NOT suppress DONE delivery. A row parked here whose reply later COMPLETES
+   * in opencode must still be delivered via `markDone` on the next drain — so
+   * `readoptOne` computes `state` FIRST and this set is checked only on the
+   * non-done path. It is cleared once the row leaves the processing list (cron
+   * reset → it drains normally as `pending`), so it can never leak.
+   */
+  dontRedispatch = /* @__PURE__ */ new Set();
+  /**
+   * "markDone for this row is TERMINALLY undeliverable" (Bug 4). Set ONLY when a
+   * re-adopted DONE row's `markDone` returned a terminal 4xx (a status that will
+   * never succeed). Checked at the TOP of the `done` branch so we do NOT re-attempt
+   * that markDone every ~2s drain while the row stays `processing`. A TRANSIENT
+   * markDone failure must NOT land here (it must still retry next drain). Separate
+   * from `dontRedispatch` because the two concerns are independent: a row can need
+   * "stop re-dispatching" without "stop delivering", and vice versa. Cleared once
+   * the row leaves the processing list, exactly like `dontRedispatch`.
+   */
+  doneUndeliverable = /* @__PURE__ */ new Set();
+  /**
+   * "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
+   * Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
+   * is NOT idempotent: if `forceReadoptRun` dispatches on tick N but the read-back +
+   * persist hasn't landed before tick N+1 re-reads the still-null
+   * `row.opencode_message_id`, tick N+1 would dispatch AGAIN → duplicate user turns.
+   * A row is added here right before its `sendPromptAsync` and `forceReadoptRun`
+   * short-circuits while it is present, so a null-id row is re-dispatched AT MOST
+   * ONCE per outstanding read-back. Cleared on a SUCCESSFUL dispatch+read-back (the
+   * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
+   * re-entry) OR on a failed/unresolved dispatch (the message is genuinely un-sent,
+   * so the NEXT tick may retry exactly once more).
+   */
+  awaitingReadopt = /* @__PURE__ */ new Set();
   /**
    * Cache of the opencode root directory (from `GET /path`). Resolved lazily on
    * first session creation so drain-created sessions are rooted at the project
@@ -1570,8 +1749,33 @@ var ChannelDriver = class {
    * not yet resolved; `null` = resolved-but-unavailable (don't keep retrying).
    */
   opencodeDirectory = void 0;
+  /**
+   * Cache of opencode `sessionId → parentID` (its parent session, or `null` when
+   * the session is a root with no parent). Sub-agents spawned via the `task` tool
+   * run in CHILD sessions whose `parentID` chains up to the Evident-created
+   * (watched) session; we resolve this once per session so a child-session
+   * question/permission can be attributed to the watched session's subtree
+   * (`sessionBelongsTo`) instead of being dropped by an exact-id filter. A missing
+   * entry = not yet resolved; `null` = resolved root (stop walking).
+   */
+  sessionParents = /* @__PURE__ */ new Map();
   /** Serialises drains so a reconnect during a drain doesn't double-process. */
   draining = false;
+  /**
+   * The currently-executing `drainPending()` promise, or null when idle. Lets a
+   * graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
+   * is about to dispatch is not missed by the `hasInFlightWatchers()` check (a
+   * drain that entered before `stop()` still registers its watcher).
+   */
+  activeDrain = null;
+  /**
+   * Set by `stop()` on graceful shutdown. Once stopped, `drainPending` no longer
+   * dispatches NEW work (it returns 0 immediately) — but the per-session watcher
+   * loops already running keep going so in-flight turns can finish and deliver
+   * their reply. `run.ts` awaits `waitForInFlight()` before it closes the tunnel
+   * and stops opencode.
+   */
+  stopped = false;
   constructor(config2) {
     this.agentId = config2.agentId;
     this.port = config2.port;
@@ -1585,7 +1789,7 @@ var ChannelDriver = class {
     this.sleep = config2.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
     this.pausedPollIntervalMs = config2.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
     this.pausedMaxWaitMs = config2.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
-    this.dispatchConfirmMs = config2.dispatchConfirmMs ?? DEFAULT_DISPATCH_CONFIRM_MS;
+    this.stuckQueuedMs = config2.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
     this.now = config2.now ?? (() => Date.now());
   }
   /** The IPv4-loopback base URL for the local `opencode serve`. */
@@ -1603,8 +1807,21 @@ var ChannelDriver = class {
    * @returns the number of messages NEWLY dispatched to opencode's native queue.
    */
   async drainPending() {
+    if (this.stopped) return 0;
     if (this.draining) return 0;
     this.draining = true;
+    const run2 = this.runDrain();
+    this.activeDrain = run2.then(
+      () => {
+        this.activeDrain = null;
+      },
+      () => {
+        this.activeDrain = null;
+      }
+    );
+    return run2;
+  }
+  async runDrain() {
     let dispatched = 0;
     try {
       const conversations = await this.getPendingConversations();
@@ -1616,8 +1833,10 @@ var ChannelDriver = class {
         });
       }
       for (const conv of conversations) {
+        if (this.stopped) break;
         dispatched += await this.processConversation(conv);
       }
+      await this.readoptProcessing();
     } finally {
       this.draining = false;
     }
@@ -1635,6 +1854,51 @@ var ChannelDriver = class {
     }
     return false;
   }
+  /**
+   * Begin a graceful stop: stop accepting NEW channel work. Idempotent. After
+   * this, `drainPending()` is a no-op (returns 0), so no new message is dispatched
+   * — but the watcher loops already tracking in-flight turns keep running, so a
+   * turn that has finished (or is about to) still fires `markDone` and delivers
+   * its reply. Pair with `waitForInFlight()` to bound how long shutdown waits.
+   */
+  stop() {
+    this.stopped = true;
+  }
+  /**
+   * Wait (up to `timeoutMs`) for in-flight watcher work to settle during a
+   * graceful shutdown, so a turn whose reply is ready — or completes within the
+   * window — is delivered before the process exits, instead of being cut off and
+   * left for the ADR-0046 restart-recovery path.
+   *
+   * Bounded on purpose: the watcher's own give-up deadline is up to 10 minutes,
+   * far longer than a shutdown grace period (e.g. Fargate's SIGTERM→SIGKILL
+   * window). We poll `hasInFlightWatchers()` and return as soon as the in-flight
+   * set empties OR the timeout elapses. Anything still in flight at the timeout is
+   * safe to abandon — it stays `processing` server-side and is re-adopted on the
+   * next runner start (ADR-0046).
+   *
+   * @returns true if all in-flight work settled within the window; false if the
+   *          timeout elapsed with work still in flight.
+   */
+  async waitForInFlight(timeoutMs) {
+    const deadline = this.now() + timeoutMs;
+    const step = Math.min(this.pausedPollIntervalMs, 250);
+    if (this.activeDrain) {
+      let drainSettled = false;
+      void this.activeDrain.then(() => {
+        drainSettled = true;
+      });
+      while (!drainSettled) {
+        if (this.now() >= deadline) return false;
+        await this.sleep(step);
+      }
+    }
+    while (this.hasInFlightWatchers()) {
+      if (this.now() >= deadline) return false;
+      await this.sleep(step);
+    }
+    return true;
+  }
   /**
    * Await all outstanding per-session watchers (WI-3).
    *
@@ -1669,15 +1933,18 @@ var ChannelDriver = class {
     const sessionId = await this.ensureSession(conv);
     const messages = await this.getPendingMessages(conv.id);
     let dispatched = 0;
+    let skippedAlreadyDispatched = 0;
     for (const message of messages) {
+      if (this.stopped) break;
       if (this.dispatched.has(message.id)) {
+        skippedAlreadyDispatched += 1;
         continue;
       }
-      const opencodeMessageId = opencodeMessageIdFor(message.id);
       const options = {
         agent: message.opencode_agent ?? void 0,
         model: message.opencode_model ?? void 0
       };
+      let opencodeMessageId;
       try {
         this.log({
           level: "info",
@@ -1685,7 +1952,10 @@ var ChannelDriver = class {
           conversation_id: conv.id,
           message_id: message.id
         });
-        await sendPromptAsync(this.port, sessionId, message.content, options, opencodeMessageId);
+        opencodeMessageId = await this.dispatchLocked(
+          sessionId,
+          () => sendPromptAsync(this.port, sessionId, message.content, options)
+        );
       } catch (err) {
         if (err instanceof ChannelAuthError) throw err;
         this.dispatched.delete(message.id);
@@ -1699,9 +1969,26 @@ var ChannelDriver = class {
         });
         continue;
       }
+      if (opencodeMessageId === null) {
+        this.log({
+          level: "error",
+          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`,
+          conversation_id: conv.id,
+          message_id: message.id
+        });
+        continue;
+      }
       this.dispatched.add(message.id);
       this.registerInFlight(conv, sessionId, message, opencodeMessageId);
       dispatched += 1;
+      void this.postSignal(conv.id, message.id, "dispatched");
+    }
+    if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
+      this.log({
+        level: "error",
+        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).`,
+        conversation_id: conv.id
+      });
     }
     this.ensureWatcherRunning(sessionId);
     return dispatched;
@@ -1739,6 +2026,25 @@ var ChannelDriver = class {
   // -------------------------------------------------------------------------
   // Per-session watcher (WI-3)
   // -------------------------------------------------------------------------
+  /**
+   * Run one dispatch (`sendPromptAsync` snapshot→POST→read-back) serialized per
+   * opencode session (Task 2.1a), so two dispatches into the SAME session can
+   * never interleave and mis-correlate their read-backs. Distinct sessions run
+   * concurrently. The chained tail intentionally ignores the prior result/error
+   * (each dispatch reports its own outcome to its caller).
+   */
+  dispatchLocked(sessionId, fn) {
+    const prior = this.sessionDispatchLocks.get(sessionId) ?? Promise.resolve();
+    const run2 = prior.then(fn, fn);
+    this.sessionDispatchLocks.set(
+      sessionId,
+      run2.then(
+        () => void 0,
+        () => void 0
+      )
+    );
+    return run2;
+  }
   /** Register a freshly-dispatched message with its session's watcher state. */
   registerInFlight(conv, sessionId, message, opencodeMessageId) {
     let watcher = this.watchers.get(sessionId);
@@ -1760,7 +2066,54 @@ var ChannelDriver = class {
       dispatchedAt: now,
       deadline: now + this.pausedMaxWaitMs,
       started: false,
-      done: false
+      done: false,
+      stuckReported: false
+    });
+  }
+  /**
+   * Register a RE-ADOPTED `processing` message with its session watcher
+   * (ADR-0046, WI-4). Mirrors `registerInFlight` but anchors the give-up
+   * `deadline` to the row's SERVER-SIDE `processed_at` (Invariant 1), NEVER to
+   * `now`: a row already `processing` for e.g. 5 min must give up ~5 min from now
+   * (10 min after `processed_at`), not 10 min from now — otherwise its deadline
+   * lands ~15 min after `processed_at`, coinciding with the cron reset →
+   * double-drive race. `dispatchedAt` stays `now` (only the appear-guard uses it).
+   *
+   * `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
+   * `opencodeMessageId` is the id the watcher polls for a reply — for the orphan
+   * fresh-run path these differ (a fresh opencode id under the same server row).
+   *
+   * `started` is set true so the watcher does NOT re-`markProcessing` a row the
+   * server already flipped to `processing`; the running/done transitions still
+   * fire from the watcher's normal branches.
+   */
+  registerReadopted(conv, sessionId, message, opencodeMessageId, processedAtMs) {
+    let watcher = this.watchers.get(sessionId);
+    if (!watcher) {
+      watcher = {
+        conv,
+        inFlight: /* @__PURE__ */ new Map(),
+        loop: null,
+        reportedQuestions: /* @__PURE__ */ new Set(),
+        reportedPermissions: /* @__PURE__ */ new Set()
+      };
+      this.watchers.set(sessionId, watcher);
+    }
+    watcher.inFlight.set(message.id, {
+      evidentMessageId: message.id,
+      opencodeMessageId,
+      message,
+      dispatchedAt: this.now(),
+      deadline: processedAtMs + this.pausedMaxWaitMs,
+      // The server row is ALREADY `processing`; do not re-fire markProcessing.
+      started: true,
+      done: false,
+      // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
+      // AND the stuck-queued observer INCLUDES re-adopted queued wedges: it gates
+      // on `state === 'queued'` (turn produced no reply), not on `started`, so a
+      // re-adopted row left wedged in `queued` still emits the signal once
+      // (#210/#220 observability).
+      stuckReported: false
     });
   }
   /**
@@ -1826,6 +2179,7 @@ var ChannelDriver = class {
           conversation_id: watcher.conv.id
         });
         for (const evidentMessageId of [...watcher.inFlight.keys()]) {
+          this.readopted.delete(evidentMessageId);
           this.removeInFlight(watcher, evidentMessageId);
         }
         return;
@@ -1846,10 +2200,15 @@ var ChannelDriver = class {
   async serviceInFlightMessage(sessionId, watcher, inFlight, messages) {
     const conv = watcher.conv;
     const state = messageRunState(messages, inFlight.opencodeMessageId);
-    if ((state === "running" || state === "done") && !inFlight.started) {
+    if ((state === "running" || state === "done" || state === "failed") && !inFlight.started) {
       let claimed;
       try {
-        claimed = await this.markProcessing(conv.id, inFlight.evidentMessageId, sessionId);
+        claimed = await this.markProcessing(
+          conv.id,
+          inFlight.evidentMessageId,
+          sessionId,
+          inFlight.opencodeMessageId
+        );
       } catch (err) {
         if (err instanceof ChannelAuthError) throw err;
         this.log({
@@ -1879,7 +2238,12 @@ var ChannelDriver = class {
           message_id: inFlight.evidentMessageId
         });
         try {
-          await this.markDone(conv.id, inFlight.evidentMessageId, sessionId);
+          await this.markDone(
+            conv.id,
+            inFlight.evidentMessageId,
+            sessionId,
+            inFlight.opencodeMessageId
+          );
         } catch (err) {
           if (err instanceof ChannelAuthError) throw err;
           if (err instanceof ChannelTerminalError) {
@@ -1915,10 +2279,59 @@ var ChannelDriver = class {
       this.removeInFlight(watcher, inFlight.evidentMessageId);
       return;
     }
-    if (state === "unknown") {
-      if (this.now() - inFlight.dispatchedAt >= this.dispatchConfirmMs) {
-        await this.redispatchInFlight(sessionId, inFlight);
+    if (state === "failed") {
+      if (!inFlight.done) {
+        const error2 = messageError(messages, inFlight.opencodeMessageId) ?? void 0;
+        this.log({
+          level: "error",
+          message: `Message ${inFlight.evidentMessageId.slice(0, 8)} errored \u2014 marking failed: ${error2 ?? "(no error text)"}`,
+          conversation_id: conv.id,
+          message_id: inFlight.evidentMessageId
+        });
+        try {
+          await this.markFailed(conv.id, inFlight.evidentMessageId, sessionId, error2);
+        } catch (err) {
+          if (err instanceof ChannelAuthError) throw err;
+          if (err instanceof ChannelTerminalError) {
+            this.log({
+              level: "error",
+              message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
+              conversation_id: conv.id,
+              message_id: inFlight.evidentMessageId
+            });
+            this.removeInFlight(watcher, inFlight.evidentMessageId);
+            return;
+          }
+          if (this.now() >= inFlight.deadline) {
+            this.log({
+              level: "error",
+              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)}`,
+              conversation_id: conv.id,
+              message_id: inFlight.evidentMessageId
+            });
+            this.removeInFlight(watcher, inFlight.evidentMessageId);
+            return;
+          }
+          this.log({
+            level: "error",
+            message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} failed (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
+            conversation_id: conv.id,
+            message_id: inFlight.evidentMessageId
+          });
+          return;
+        }
+        inFlight.done = true;
       }
+      this.removeInFlight(watcher, inFlight.evidentMessageId);
+      return;
+    }
+    const pastStuckBound = this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs;
+    const sessionIdle = state === "queued" && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId);
+    if (state === "queued" && pastStuckBound && sessionIdle && !inFlight.stuckReported) {
+      inFlight.stuckReported = true;
+      void this.postSignal(conv.id, inFlight.evidentMessageId, "stuck_queued", {
+        stuck_for_ms: this.now() - inFlight.dispatchedAt
+      });
     }
     if (this.now() >= inFlight.deadline) {
       this.log({
@@ -1927,48 +2340,395 @@ var ChannelDriver = class {
         conversation_id: conv.id,
         message_id: inFlight.evidentMessageId
       });
+      void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
+        watched_for_ms: this.now() - inFlight.dispatchedAt
+      });
       this.removeInFlight(watcher, inFlight.evidentMessageId);
     }
   }
+  // -------------------------------------------------------------------------
+  // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
+  // -------------------------------------------------------------------------
   /**
-   * Re-dispatch a message whose user row never appeared (idle-path guard). Safe:
-   * opencode treats a duplicate caller-supplied `messageID` as idempotent (PoC
-   * fact 9) — one user message + one reply even if the original DID land. Resets
-   * the dispatch timestamp so the guard doesn't immediately fire again.
+   * Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
+   *
+   * The pending drain only re-drives `pending` rows; a message already flipped to
+   * `processing` before the runner died is watched by nobody until the 15-min
+   * cron resets it. Here we fetch those rows, and per row resolve its correlated
+   * reply against opencode's OWN session store — completing, re-attaching, or
+   * (for an orphan) forcing a genuine fresh run. Runs on EVERY drain tick, so it
+   * is idempotent per message (Invariant 2): a row a watcher already tracks is
+   * skipped in `readoptOne` — one driver, no double-drive.
+   *
+   * Only `ChannelAuthError` propagates (to `drainPending`, like the pending
+   * path); every other early return LOGS a reason with context — no silent drop.
    */
-  async redispatchInFlight(sessionId, inFlight) {
+  async readoptProcessing() {
+    const rows = await this.getProcessingMessages();
+    if (this.dontRedispatch.size > 0 || this.doneUndeliverable.size > 0) {
+      const stillProcessing = new Set(rows.map((r) => r.id));
+      for (const id of [...this.dontRedispatch, ...this.doneUndeliverable]) {
+        if (!stillProcessing.has(id)) {
+          const cleared = this.dontRedispatch.delete(id);
+          const clearedUndeliverable = this.doneUndeliverable.delete(id);
+          if (cleared || clearedUndeliverable) {
+            this.log({
+              level: "info",
+              message: `Re-adopt: message ${id.slice(0, 8)} left the processing list (cron reset) \u2014 cleared gave-up marker`,
+              message_id: id
+            });
+          }
+        }
+      }
+    }
+    if (rows.length === 0) return;
+    const bySession = /* @__PURE__ */ new Map();
+    for (const row of rows) {
+      if (!row.opencode_session_id) {
+        this.log({
+          level: "error",
+          message: `Cannot re-adopt processing message ${row.id.slice(0, 8)} \u2014 no opencode session id; leaving for the cron safety net`,
+          conversation_id: row.conversation_id,
+          message_id: row.id
+        });
+        continue;
+      }
+      const list = bySession.get(row.opencode_session_id) ?? [];
+      list.push(row);
+      bySession.set(row.opencode_session_id, list);
+    }
+    for (const [sessionId, sessionRows] of bySession) {
+      let messages;
+      try {
+        const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
+        if (!res.ok) {
+          this.log({
+            level: "error",
+            message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned HTTP ${res.status} \u2014 skipping this session this tick`
+          });
+          continue;
+        }
+        const body = await res.json();
+        if (!Array.isArray(body)) {
+          this.log({
+            level: "error",
+            message: `Re-adopt: polling session ${sessionId.slice(0, 8)} returned a non-array message body \u2014 skipping this session this tick`
+          });
+          continue;
+        }
+        messages = body;
+      } catch (err) {
+        this.log({
+          level: "error",
+          message: `Re-adopt: failed to poll session ${sessionId.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`
+        });
+        continue;
+      }
+      for (const row of sessionRows) {
+        await this.readoptOne(sessionId, row, messages);
+      }
+    }
+  }
+  /**
+   * Re-adopt ONE `processing` row against the tick's session message snapshot
+   * (ADR-0046 Decision §1/§2). Idempotent: skips a row already being driven.
+   *
+   * Branches on `messageRunState(messages, row.opencode_message_id)` — the
+   * opencode-assigned user-message id persisted on the first `processing` PATCH
+   * (#218). A row with a NULL stored id (dispatched but the read-back never landed
+   * before the restart) has no id to correlate → treated as an orphan and
+   * re-dispatched (at most once, see `forceReadoptRun`):
+   *   - `done`             → `markDone` now (guarded like the watcher's done branch);
+   *   - `failed`           → `markFailed` with the surfaced error (issue #182), so an
+   *                          errored turn is reported failed on restart, NOT re-dispatched;
+   *   - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
+   *                          tracking the stored id so the reply correlates by it;
+   *   - `unknown`/null id  → re-dispatch (opencode assigns a fresh id) + attach a watcher.
+   *
+   * Only `ChannelAuthError` propagates.
+   */
+  async readoptOne(sessionId, row, messages) {
+    if (this.isTracked(sessionId, row.id)) {
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} already tracked in-flight \u2014 skipping (owned by the watcher loop)`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
+    const ocId = row.opencode_message_id;
+    const state = messageRunState(messages, ocId ?? "");
+    if (state === "done") {
+      if (this.doneUndeliverable.has(row.id)) {
+        this.log({
+          level: "info",
+          message: `Re-adopt: message ${row.id.slice(0, 8)} markDone is terminally undeliverable \u2014 left to the cron; skipping until it leaves processing`,
+          conversation_id: row.conversation_id,
+          message_id: row.id
+        });
+        return;
+      }
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} completed while unwatched \u2014 marking done`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      try {
+        await this.markDone(row.conversation_id, row.id, sessionId, ocId);
+      } catch (err) {
+        if (err instanceof ChannelAuthError) throw err;
+        if (err instanceof ChannelTerminalError) {
+          this.doneUndeliverable.add(row.id);
+          this.log({
+            level: "error",
+            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}`,
+            conversation_id: row.conversation_id,
+            message_id: row.id
+          });
+          return;
+        }
+        this.log({
+          level: "error",
+          message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} done (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
+          conversation_id: row.conversation_id,
+          message_id: row.id
+        });
+        return;
+      }
+      this.dontRedispatch.delete(row.id);
+      return;
+    }
+    if (state === "failed") {
+      const error2 = messageError(messages, ocId ?? "") ?? void 0;
+      this.log({
+        level: "error",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      try {
+        await this.markFailed(row.conversation_id, row.id, sessionId, error2);
+      } catch (err) {
+        if (err instanceof ChannelAuthError) throw err;
+        if (err instanceof ChannelTerminalError) {
+          this.doneUndeliverable.add(row.id);
+          this.log({
+            level: "error",
+            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}`,
+            conversation_id: row.conversation_id,
+            message_id: row.id
+          });
+          return;
+        }
+        this.log({
+          level: "error",
+          message: `Re-adopt: failed to mark message ${row.id.slice(0, 8)} failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
+          conversation_id: row.conversation_id,
+          message_id: row.id
+        });
+        return;
+      }
+      this.dontRedispatch.delete(row.id);
+      return;
+    }
+    if (this.dontRedispatch.has(row.id)) {
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} already gave up \u2014 left to the cron; skipping until it leaves processing`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
+    if ((state === "running" || state === "queued") && ocId) {
+      const conv = this.convForRow(sessionId, row);
+      const message = this.queuedMessageForRow(row);
+      this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
+      this.dispatched.add(row.id);
+      this.readopted.add(row.id);
+      this.ensureWatcherRunning(sessionId);
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stored id, no re-dispatch)`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
+    await this.forceReadoptRun(sessionId, row);
+  }
+  /**
+   * Re-dispatch an orphaned (`unknown`/null-id) `processing` row (ADR-0046 §2).
+   *
+   * #218/WI-5: the row's user message is absent (never kept, or a null stored id),
+   * so we re-`prompt_async` WITHOUT a caller id (opencode assigns a monotonic one),
+   * read it back, and register the watcher under the assigned id so the reply
+   * correlates server-side.
+   *
+   * ⚠️ AT-MOST-ONCE (High-2): dispatch is no longer idempotent (no caller-supplied
+   * id). Without a guard, if this dispatches on tick N but the read-back+persist
+   * hasn't landed before tick N+1 re-reads the still-null `opencode_message_id`,
+   * tick N+1 would dispatch AGAIN → duplicate user turns. The `awaitingReadopt`
+   * latch makes a null-id row re-dispatched AT MOST ONCE per outstanding read-back:
+   * short-circuit while the row is latched; clear it on a successful dispatch (the
+   * row is then tracked in `dispatched`, so `readoptOne`'s early skip prevents
+   * re-entry) OR on a failed/unresolved dispatch (genuinely un-sent → the next tick
+   * may retry exactly once more).
+   *
+   * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
+   * `processed_at` (Invariant 1).
+   */
+  async forceReadoptRun(sessionId, row) {
+    if (this.stopped) {
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned but the runner is stopping \u2014 not starting a fresh turn; leaving for restart recovery`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
+    if (this.awaitingReadopt.has(row.id)) {
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${row.id.slice(0, 8)} already has a re-dispatch awaiting read-back \u2014 skipping (at most once)`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
+    if (this.processedAtMs(row) + this.pausedMaxWaitMs <= this.now()) {
+      this.dontRedispatch.add(row.id);
+      this.log({
+        level: "info",
+        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)`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
     const options = {
-      agent: inFlight.message.opencode_agent ?? void 0,
-      model: inFlight.message.opencode_model ?? void 0
+      agent: row.opencode_agent ?? void 0,
+      model: row.opencode_model ?? void 0
     };
     this.log({
       level: "info",
-      message: `Message ${inFlight.evidentMessageId.slice(0, 8)} not observed after dispatch \u2014 re-dispatching (idle-path guard)`,
-      message_id: inFlight.evidentMessageId
+      message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching (opencode assigns a fresh id)`,
+      conversation_id: row.conversation_id,
+      message_id: row.id
     });
+    this.awaitingReadopt.add(row.id);
+    let ocId;
     try {
-      await sendPromptAsync(
-        this.port,
+      ocId = await this.dispatchLocked(
         sessionId,
-        inFlight.message.content,
-        options,
-        inFlight.opencodeMessageId
+        () => sendPromptAsync(this.port, sessionId, row.content, options)
       );
     } catch (err) {
+      this.awaitingReadopt.delete(row.id);
+      if (err instanceof ChannelAuthError) throw err;
       this.log({
         level: "error",
-        message: `Re-dispatch failed for message ${inFlight.evidentMessageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
-        message_id: inFlight.evidentMessageId
+        message: `Re-adopt: re-dispatch failed for message ${row.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
       });
+      return;
     }
-    inFlight.dispatchedAt = this.now();
+    if (ocId === null) {
+      this.awaitingReadopt.delete(row.id);
+      this.log({
+        level: "error",
+        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`,
+        conversation_id: row.conversation_id,
+        message_id: row.id
+      });
+      return;
+    }
+    const conv = this.convForRow(sessionId, row);
+    const message = this.queuedMessageForRow(row);
+    this.registerReadopted(conv, sessionId, message, ocId, this.processedAtMs(row));
+    this.dispatched.add(row.id);
+    this.readopted.add(row.id);
+    this.awaitingReadopt.delete(row.id);
+    this.ensureWatcherRunning(sessionId);
+  }
+  /**
+   * True if `evidentMessageId` is already being driven — either in the
+   * authoritative `dispatched` set or a live watcher's in-flight set for this
+   * session (Invariant 2, WI-5). Either signal means a watcher owns the row.
+   */
+  isTracked(sessionId, evidentMessageId) {
+    if (this.dispatched.has(evidentMessageId)) return true;
+    const watcher = this.watchers.get(sessionId);
+    return watcher?.inFlight.has(evidentMessageId) ?? false;
+  }
+  /**
+   * Parse a re-adopt row's `processed_at` (ISO string) to epoch ms for the
+   * deadline anchor (Invariant 1). The endpoint guarantees `processed_at` is set
+   * for `processing` rows, but if it is somehow null/unparseable fall back to
+   * `now` (defensive) AND log — a fallback means the anchor is weaker than
+   * intended, which is worth surfacing.
+   */
+  processedAtMs(row) {
+    const parsed = row.processed_at ? Date.parse(row.processed_at) : NaN;
+    if (!Number.isNaN(parsed)) return parsed;
+    this.log({
+      level: "error",
+      message: `Re-adopt: message ${row.id.slice(0, 8)} has null/unparseable processed_at (${String(row.processed_at)}) \u2014 anchoring deadline to now (defensive)`,
+      conversation_id: row.conversation_id,
+      message_id: row.id
+    });
+    return this.now();
+  }
+  /** Build the `PendingConversation` shape the watcher needs from a re-adopt row. */
+  convForRow(sessionId, row) {
+    return {
+      id: row.conversation_id,
+      agent_id: this.agentId,
+      opencode_session_id: sessionId,
+      pending_message_count: 0,
+      oldest_pending_at: row.processed_at
+    };
+  }
+  /** Build the `QueuedMessage` shape the watcher/re-dispatch needs from a re-adopt row. */
+  queuedMessageForRow(row) {
+    return {
+      id: row.id,
+      content: row.content,
+      status: "processing",
+      opencode_agent: row.opencode_agent,
+      opencode_model: row.opencode_model,
+      source_message_id: row.source_message_id,
+      slack_user_id: row.slack_user_id
+    };
   }
   /**
    * Remove a message from the in-flight set AND the authoritative dispatched
    * set. Once the in-flight set empties, the watcher loop's `while` guard exits
    * and its `.finally` removes the session entry from `this.watchers`.
+   *
+   * Bug 2: if a RE-ADOPTED message is removed WITHOUT having completed
+   * (`!inFlight.done` — i.e. a give-up: deadline reached, or markDone left to the
+   * cron), park it in `dontRedispatch` so the next drain does NOT re-adopt (and
+   * re-dispatch) the still-`processing` row every ~2s until the 15-min cron. A
+   * re-adopted message that completed (`done`) needs no marker — it's leaving
+   * `processing`. This suppresses only re-dispatch: if its reply later completes,
+   * the done branch still delivers it (Bugbot #202).
    */
   removeInFlight(watcher, evidentMessageId) {
+    const inFlight = watcher.inFlight.get(evidentMessageId);
+    if (this.readopted.delete(evidentMessageId) && inFlight && !inFlight.done) {
+      this.dontRedispatch.add(evidentMessageId);
+      this.log({
+        level: "info",
+        message: `Re-adopt: message ${evidentMessageId.slice(0, 8)} gave up \u2014 parking until it leaves the processing list (cron reset)`,
+        conversation_id: watcher.conv.id,
+        message_id: evidentMessageId
+      });
+    }
     watcher.inFlight.delete(evidentMessageId);
     this.dispatched.delete(evidentMessageId);
   }
@@ -1997,8 +2757,8 @@ var ChannelDriver = class {
     } catch {
     }
     for (const q of questions) {
-      if (q.sessionID !== sessionId) continue;
       if (watcher.reportedQuestions.has(q.id)) continue;
+      if (!await this.sessionBelongsTo(q.sessionID, sessionId)) continue;
       const paused = this.attributeInteraction(watcher, q.tool?.messageID, messages);
       const reported = await this.reportInteraction(
         watcher.conv.id,
@@ -2018,8 +2778,8 @@ var ChannelDriver = class {
     } catch {
     }
     for (const p of permissions) {
-      if (p.sessionID !== sessionId) continue;
       if (watcher.reportedPermissions.has(p.id)) continue;
+      if (!await this.sessionBelongsTo(p.sessionID, sessionId)) continue;
       const paused = this.attributeInteraction(watcher, p.messageID, messages);
       const reported = await this.reportInteraction(
         watcher.conv.id,
@@ -2030,6 +2790,50 @@ var ChannelDriver = class {
       if (reported) watcher.reportedPermissions.add(p.id);
     }
   }
+  /**
+   * True when `sessionId` is the `rootSessionId` itself OR a descendant of it —
+   * i.e. its `parentID` chain (resolved via `GET /session/:id`) reaches the
+   * watched root. Sub-agents spawned via the `task` tool run in child sessions,
+   * so their questions/permissions live under a different `sessionID` that must
+   * still be attributed to the root conversation the watcher owns.
+   *
+   * Parents are cached in `sessionParents` so we walk each session at most once;
+   * a bounded depth cap guards against a cycle or a pathological chain, and any
+   * fetch failure is treated as "not a descendant" (best-effort — the interaction
+   * simply isn't surfaced this tick and is retried next tick once resolvable).
+   */
+  async sessionBelongsTo(sessionId, rootSessionId) {
+    let current = sessionId;
+    for (let depth = 0; current && depth < 32; depth++) {
+      if (current === rootSessionId) return true;
+      const parent = await this.resolveSessionParent(current);
+      if (parent === null || parent === void 0) return false;
+      current = parent;
+    }
+    return false;
+  }
+  /**
+   * Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
+   * `null` for a root session (no parent) and `undefined` when opencode is
+   * unreachable / the session can't be read (so the caller stops walking without
+   * caching a wrong answer — the next tick retries).
+   */
+  async resolveSessionParent(sessionId) {
+    const cached = this.sessionParents.get(sessionId);
+    if (cached !== void 0) return cached;
+    let parent = void 0;
+    try {
+      const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}`);
+      if (res.ok) {
+        const body = await res.json();
+        parent = body && typeof body.parentID === "string" ? body.parentID : null;
+      }
+    } catch {
+      parent = void 0;
+    }
+    if (parent !== void 0) this.sessionParents.set(sessionId, parent);
+    return parent;
+  }
   /**
    * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
    *
@@ -2112,6 +2916,35 @@ var ChannelDriver = class {
     }
     return await res.json();
   }
+  /**
+   * Fetch this agent's `processing` messages for re-adoption (ADR-0046, WI-1).
+   * The pending path (`getPendingConversations`/`getPendingMessages`) only
+   * surfaces `pending` rows, so a message already `processing` when the runner
+   * died is invisible to it — this dedicated endpoint returns exactly those rows
+   * with the fields the re-adopt path needs (`processed_at`,
+   * `opencode_session_id`, routing).
+   *
+   * Response is an OBJECT WRAPPER `{ messages: [...] }` (snake_case) — NOT a bare
+   * array. Propagates `ChannelAuthError` on 401/403; throws a plain `Error` on
+   * other non-ok so `drainPending`'s try/finally leaves `draining` false and the
+   * next tick retries.
+   */
+  async getProcessingMessages() {
+    const res = await this.fetchImpl(
+      `${this.apiUrl}/agents/${this.agentId}/conversations/processing`,
+      { headers: { Authorization: this.getAuthHeader() } }
+    );
+    this.assertAuth(res, "fetching processing messages");
+    if (!res.ok) {
+      throw new Error(`Failed to get processing messages: HTTP ${res.status}`);
+    }
+    const data = await res.json();
+    let messages = data.messages ?? [];
+    if (this.conversationFilter) {
+      messages = messages.filter((m) => m.conversation_id === this.conversationFilter);
+    }
+    return messages;
+  }
   /**
    * EXISTING combinedAuth route — now fired by the watcher on queued→running
    * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -2133,13 +2966,17 @@ var ChannelDriver = class {
    * A single attempt (no internal retry): the watcher's per-tick loop is the
    * retry vehicle for the swap-to-running.
    */
-  async markProcessing(conversationId, messageId, sessionId) {
+  async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
     const res = await this.fetchImpl(
       `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
       {
         method: "PATCH",
         headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
-        body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
+        body: JSON.stringify({
+          status: "processing",
+          opencode_session_id: sessionId,
+          ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
+        })
       }
     );
     this.assertAuth(res, "marking message as processing");
@@ -2177,13 +3014,17 @@ var ChannelDriver = class {
    *                                    watcher retries next tick within the
    *                                    deadline, Finding 4).
    */
-  async markDone(conversationId, messageId, sessionId) {
+  async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
     const res = await this.fetchImpl(
       `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
       {
         method: "PATCH",
         headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
-        body: JSON.stringify({ status: "done", opencode_session_id: sessionId })
+        body: JSON.stringify({
+          status: "done",
+          opencode_session_id: sessionId,
+          ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
+        })
       }
     );
     this.assertAuth(res, "marking message as done");
@@ -2193,7 +3034,17 @@ var ChannelDriver = class {
     }
     throw new ChannelTerminalError(`marking message as done: HTTP ${res.status}`, res.status);
   }
-  async markFailed(conversationId, messageId) {
+  /**
+   * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
+   * when provided (issue #182): a bare `markFailed(conv, msg)` sends
+   * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
+   * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
+   * failure reason reaches the channel.
+   */
+  async markFailed(conversationId, messageId, sessionId, error2) {
+    const body = { status: "failed" };
+    if (sessionId !== void 0) body.opencode_session_id = sessionId;
+    if (error2 !== void 0) body.error = error2;
     await this.callWithRetry(
       "marking message as failed",
       () => this.fetchImpl(
@@ -2201,11 +3052,47 @@ var ChannelDriver = class {
         {
           method: "PATCH",
           headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
-          body: JSON.stringify({ status: "failed" })
+          body: JSON.stringify(body)
         }
       )
     );
   }
+  /**
+   * Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
+   * (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
+   * — the server records it via `log()` (no DB write, no notification). This is
+   * fire-and-forget: it MUST NEVER throw into the drain or the watcher tick, and
+   * MUST NOT use `callWithRetry` (a telemetry ping must not block the sequential
+   * watcher tick — one attempt is enough). A failure is SWALLOWED but LOGGED with
+   * context (no silent catch, per development-workflow).
+   */
+  async postSignal(conversationId, messageId, signal, extra) {
+    try {
+      const res = await this.fetchImpl(
+        `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
+        {
+          method: "POST",
+          headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
+          body: JSON.stringify({ signal, ...extra })
+        }
+      );
+      if (!res.ok) {
+        this.log({
+          level: "error",
+          message: `Signal '${signal}' for message ${messageId.slice(0, 8)} returned HTTP ${res.status} (telemetry-only, ignored)`,
+          conversation_id: conversationId,
+          message_id: messageId
+        });
+      }
+    } catch (err) {
+      this.log({
+        level: "error",
+        message: `Signal '${signal}' for message ${messageId.slice(0, 8)} failed (telemetry-only, ignored): ${err instanceof Error ? err.message : String(err)}`,
+        conversation_id: conversationId,
+        message_id: messageId
+      });
+    }
+  }
   async persistSession(conversationId, sessionId) {
     const res = await this.fetchImpl(
       `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}`,
@@ -2476,6 +3363,25 @@ async function resolveAgentIdFromKey(authHeader) {
     return { error: `Failed to resolve agent from key: ${message}` };
   }
 }
+async function notifyAgentDisconnected(agentId, authHeader) {
+  const apiUrl = getApiUrlConfig();
+  try {
+    const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
+      method: "POST",
+      headers: { Authorization: authHeader }
+    });
+    if (!response.ok) {
+      const serverMessage = await readErrorMessage(response);
+      return {
+        ok: false,
+        error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
+      };
+    }
+    return { ok: true };
+  } catch (error2) {
+    return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
+  }
+}
 async function getAgentInfo(agentId, authHeader) {
   const apiUrl = getApiUrlConfig();
   try {
@@ -2521,7 +3427,9 @@ async function getAgentInfo(agentId, authHeader) {
 // src/commands/run.ts
 var MAX_ACTIVITY_LOG_ENTRIES = 10;
 var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
-function log(state, message, isError = false) {
+var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
+var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
+function log2(state, message, isError = false) {
   if (state.json) {
     console.log(
       JSON.stringify({
@@ -2546,9 +3454,9 @@ function logActivity(state, entry) {
   }
   if (!state.interactive) {
     if (entry.type === "error") {
-      log(state, entry.error ?? "Unknown error", true);
+      log2(state, entry.error ?? "Unknown error", true);
     } else if (entry.type === "info" && entry.message) {
-      log(state, entry.message);
+      log2(state, entry.message);
     }
   }
 }
@@ -2634,6 +3542,7 @@ async function handleAuthError(state, error2) {
 }
 async function driveChannels(state, driver) {
   let idlePolls = 0;
+  let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
   while (state.running) {
     if (state.connection?.reconnecting && state.connection.reconnectPromise) {
       logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
@@ -2643,7 +3552,9 @@ async function driveChannels(state, driver) {
     try {
       const processed = await driver.drainPending();
       state.messageCount += processed;
-      if (processed > 0 || driver.hasInFlightWatchers()) {
+      const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
+      lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
+      if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity) {
         idlePolls = 0;
         if (processed > 0 && state.interactive) displayStatus(state);
       } else if (state.idleTimeout !== null) {
@@ -2683,8 +3594,42 @@ async function driveChannels(state, driver) {
     }
   }
 }
-async function cleanup(state) {
+async function notifyOffline(state) {
+  if (!state.agentId || !state.authHeader) return;
+  if (!state.connected) {
+    log2(state, "Skipping offline signal \u2014 this runner does not hold the live tunnel");
+    return;
+  }
+  const result = await notifyAgentDisconnected(state.agentId, state.authHeader);
+  if (result.ok) {
+    log2(state, "Notified Evident the agent is going offline");
+  } else {
+    logActivity(state, {
+      type: "error",
+      error: `Could not notify Evident of offline status (relay will still report it): ${result.error}`
+    });
+    if (state.interactive) displayStatus(state);
+  }
+}
+async function cleanup(state, opts = {}) {
   state.running = false;
+  if (opts.graceful && state.channelDriver) {
+    state.channelDriver.stop();
+    log2(state, "Draining in-flight channel work before shutdown...");
+    if (state.interactive) {
+      logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
+      displayStatus(state);
+    }
+    const settled = await state.channelDriver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS);
+    if (!settled) {
+      logActivity(state, {
+        type: "info",
+        message: "Shutdown drain timed out with work still in flight \u2014 leaving it for restart recovery"
+      });
+      if (state.interactive) displayStatus(state);
+    }
+  }
+  await notifyOffline(state);
   if (state.connection) {
     state.connection.close();
     state.connection = null;
@@ -2695,7 +3640,7 @@ async function cleanup(state) {
       logActivity(state, { type: "info", message: "Stopped OpenCode process" });
       displayStatus(state);
     } else {
-      log(state, "Stopped OpenCode process");
+      log2(state, "Stopped OpenCode process");
     }
     state.opencodeProcess = null;
   }
@@ -2715,26 +3660,31 @@ async function run(options) {
     opencodeVersion: null,
     opencodeProcess: null,
     connection: null,
+    channelDriver: null,
     running: true,
+    shuttingDown: false,
     activityLog: [],
     messageCount: 0,
+    lastProxiedActivityAt: null,
     authHeader: ""
   };
   if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
-    log(
+    log2(
       state,
       "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.",
       false
     );
   }
   const handleSignal = async () => {
+    if (state.shuttingDown) return;
+    state.shuttingDown = true;
     if (state.interactive) {
       logActivity(state, { type: "info", message: "Shutting down..." });
       displayStatus(state);
     } else {
-      log(state, "Shutting down...");
+      log2(state, "Shutting down...");
     }
-    await cleanup(state);
+    await cleanup(state, { graceful: true });
     await shutdownTelemetry();
     process.exit(0);
   };
@@ -2765,7 +3715,7 @@ async function run(options) {
         const resolved = await resolveAgentIdFromKey(state.authHeader);
         if (resolved.agent_id) {
           state.agentId = resolved.agent_id;
-          log(state, `Resolved agent ID from key: ${state.agentId}`);
+          log2(state, `Resolved agent ID from key: ${state.agentId}`);
           if (state.interactive && !state.json) {
             logActivity(state, {
               type: "info",
@@ -2828,17 +3778,17 @@ async function run(options) {
         port: state.port,
         interactive: state.interactive,
         agentId: state.agentId,
-        log: (message) => log(state, message)
+        log: (message) => log2(state, message)
       });
       state.port = oc.port;
       state.opencodeProcess = oc.process;
       state.opencodeVersion = oc.version;
       state.opencodeConnected = oc.process !== null || oc.version !== null;
-      const version = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
-      ocSpinner?.succeed(`OpenCode running on port ${state.port}${version}`);
+      const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
+      ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
       const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
       if (versionWarning) {
-        log(state, versionWarning, false);
+        log2(state, versionWarning, false);
         if (state.interactive && !state.json) {
           logActivity(state, { type: "info", message: versionWarning });
         }
@@ -2854,12 +3804,14 @@ async function run(options) {
       apiUrl: getApiUrlConfig(),
       getAuthHeader: () => state.authHeader,
       conversationFilter: state.conversationFilter,
+      stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
       log: (entry) => logActivity(state, {
         type: entry.level === "error" ? "error" : "info",
         message: entry.message,
         error: entry.level === "error" ? entry.message : void 0
       })
     });
+    state.channelDriver = channelDriver;
     const connection = new RunnerConnection({
       agentId: state.agentId,
       getAuthHeader: () => state.authHeader,
@@ -2873,7 +3825,11 @@ async function run(options) {
             type: "info",
             message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (agent: ${agentId})`
           });
-          emitAgentConnected(state.agentId, { port: state.port });
+          emitAgentConnected(state.agentId, {
+            port: state.port,
+            cli_version: getCliVersion(),
+            opencode_version: state.opencodeVersion
+          });
           if (!isReconnect) tunnelSpinner?.succeed("Tunnel connected");
           if (state.interactive) displayStatus(state);
           channelDriver.drainPending().then((processed) => {
@@ -2907,9 +3863,14 @@ async function run(options) {
           logActivity(state, { type: "error", error: error2 });
           if (state.interactive) displayStatus(state);
         },
-        // Web traffic is proxied transparently; only note opencode is live.
+        // Web traffic is proxied transparently; note opencode is live and stamp
+        // proxied activity so the idle loop treats interactive proxy use as work.
+        // Fires per forwarded response head (incl. every SSE open) and excludes
+        // the internal drain-ping, so an actively-used proxy keeps the timer
+        // fresh while a lone idle SSE with no follow-up requests still ages out.
         onResponse: () => {
           state.opencodeConnected = true;
+          state.lastProxiedActivityAt = Date.now();
         },
         // A channel message was queued and the api-worker pinged us over the
         // tunnel to drain immediately instead of waiting for the next poll tick.
@@ -2948,9 +3909,10 @@ async function run(options) {
       throw error2;
     }
     if (!interactive || state.json) {
-      log(state, "Driving channel messages...");
+      log2(state, "Driving channel messages...");
     }
     await driveChannels(state, channelDriver);
+    if (state.shuttingDown) return;
     await cleanup(state);
     if (state.json) {
       console.log(
@@ -2960,11 +3922,12 @@ async function run(options) {
         })
       );
     } else if (!interactive) {
-      log(state, `Completed. Processed ${state.messageCount} message(s).`);
+      log2(state, `Completed. Processed ${state.messageCount} message(s).`);
     }
     await shutdownTelemetry();
     process.exit(0);
   } catch (error2) {
+    if (state.shuttingDown) return;
     await cleanup(state);
     const message = error2 instanceof Error ? error2.message : String(error2);
     if (state.json) {
@@ -2982,8 +3945,9 @@ async function run(options) {
 }
 
 // src/index.ts
+var { version } = createRequire(import.meta.url)("../package.json");
 var program = new Command();
-program.name("evident").description("Run OpenCode locally and connect it to Evident").version("0.1.0").option(
+program.name("evident").description("Run OpenCode locally and connect it to Evident").version(version).option(
   "--endpoint <url>",
   "Evident API base URL (default: production; e.g. http://localhost:3001)"
 ).option("--tunnel <url>", "Tunnel WebSocket URL (default: production; e.g. ws://localhost:8787)").hook("preAction", (thisCommand) => {