@evident-ai/cli 3.0.1-dev.0c07f2c → 3.0.1-dev.0d0721c

package/dist/index.js

@@ -285,14 +285,14 @@ function blank() {
   console.log();
 }
 function waitForEnter(prompt = "Press Enter to continue...") {
-  return new Promise((resolve) => {
+  return new Promise((resolve2) => {
     process.stdout.write(chalk.dim(prompt));
     const handler = () => {
       process.stdin.removeListener("data", handler);
       process.stdin.setRawMode?.(false);
       process.stdin.pause();
       console.log();
-      resolve();
+      resolve2();
     };
     if (process.stdin.isTTY) {
       process.stdin.setRawMode?.(true);
@@ -302,7 +302,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
   });
 }
 function sleep(ms) {
-  return new Promise((resolve) => setTimeout(resolve, ms));
+  return new Promise((resolve2) => setTimeout(resolve2, ms));
 }
 
 // src/commands/login.ts
@@ -376,19 +376,19 @@ async function tokenLogin() {
   console.log("Visit your Evident dashboard to generate a CLI token.");
   blank();
   process.stdout.write("Paste token: ");
-  const token = await new Promise((resolve) => {
+  const token = await new Promise((resolve2) => {
     let data = "";
     process.stdin.setEncoding("utf8");
     process.stdin.on("data", (chunk) => {
       data += chunk;
     });
     process.stdin.on("end", () => {
-      resolve(data.trim());
+      resolve2(data.trim());
     });
     if (process.stdin.isTTY) {
       process.stdin.once("data", (chunk) => {
         process.stdin.pause();
-        resolve(chunk.toString().trim());
+        resolve2(chunk.toString().trim());
       });
       process.stdin.resume();
     }
@@ -471,12 +471,6 @@ import chalk6 from "chalk";
 import ora3 from "ora";
 import { select as select3 } from "@inquirer/prompts";
 
-// ../../packages/types/src/opencode/index.ts
-function opencodeMessageIdFor(queuedMessageId) {
-  const sanitized = queuedMessageId.replace(/[^a-zA-Z0-9]/g, "_");
-  return `msg_${sanitized}`;
-}
-
 // ../../packages/types/src/telemetry/index.ts
 var TelemetryEventTypes = {
   // Agent activity events (shown in web UI activity log)
@@ -515,7 +509,10 @@ function stripQuery(url) {
 }
 
 // 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;
@@ -709,13 +706,13 @@ async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
     if (health.healthy) {
       return health;
     }
-    await new Promise((resolve) => setTimeout(resolve, 1e3));
+    await new Promise((resolve2) => setTimeout(resolve2, 1e3));
   }
   return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
 }
 
 // src/lib/opencode/opencode-version-gate.ts
-var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11"];
+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);
@@ -1039,7 +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;
@@ -1059,10 +1060,66 @@ 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;
+  }
+}
+function sessionLastActivityMs(session) {
+  const candidates = [
+    session.time?.updated,
+    session.time?.created,
+    session.time_updated,
+    session.time_created,
+    session.updated,
+    session.created
+  ];
+  for (const c of candidates) {
+    if (typeof c === "number" && Number.isFinite(c)) return c;
+  }
+  return null;
+}
+async function listSessions(port) {
+  try {
+    const res = await fetch(`${opencodeBase(port)}/session`);
+    if (!res.ok) return null;
+    const body = await res.json();
+    return Array.isArray(body) ? body : null;
+  } catch {
+    return null;
+  }
+}
+async function deleteSession(port, id) {
+  try {
+    const res = await fetch(`${opencodeBase(port)}/session/${id}`, { method: "DELETE" });
+    return res.status >= 200 && res.status < 300;
+  } catch {
+    return false;
+  }
+}
+async function sessionExists(port, id) {
+  try {
+    const res = await fetch(`${opencodeBase(port)}/session/${id}`);
+    if (res.status >= 200 && res.status < 300) return true;
+    if (res.status === 404) return false;
+    return null;
+  } catch {
+    return null;
+  }
+}
 async function createOpenCodeSession(port, directory) {
   const url = new URL(`${opencodeBase(port)}/session`);
   if (directory && directory.trim()) {
@@ -1080,9 +1137,16 @@ 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) {
@@ -1106,6 +1170,29 @@ 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((resolve2) => setTimeout(resolve2, READ_BACK_DELAY_MS));
+    }
+  }
+  return null;
 }
 function findAssistantReplyAfter(messages, userMessageId) {
   if (!messages || messages.length === 0) return null;
@@ -1122,19 +1209,31 @@ 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";
@@ -1144,7 +1243,21 @@ function messageRunState(messages, userMessageId) {
     if (!reply) return "unknown";
   }
   if (!reply) return "queued";
-  return isAssistantInFlight(reply) ? "running" : "done";
+  if (isAssistantInFlight(reply)) return "running";
+  return errorOf(reply) != null ? "failed" : "done";
+}
+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;
@@ -1152,8 +1265,109 @@ function hasRunningAssistantExcept(messages, exceptUserMessageId) {
     (m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
   );
 }
-function opencodeMessageIdFor2(queuedMessageId) {
-  return opencodeMessageIdFor(queuedMessageId);
+
+// src/lib/opencode/session-cleanup.ts
+var DURATION_UNIT_MS = {
+  s: 1e3,
+  m: 60 * 1e3,
+  h: 60 * 60 * 1e3,
+  d: 24 * 60 * 60 * 1e3
+};
+function parseDurationMs(input) {
+  const trimmed = input.trim();
+  const match = /^(\d+)([smhd])$/.exec(trimmed);
+  if (!match) {
+    throw new Error(
+      `Invalid duration "${input}": expected <number><unit> where unit is one of s, m, h, d (e.g. "7d", "24h", "30m", "90s").`
+    );
+  }
+  const value = Number(match[1]);
+  if (value <= 0) {
+    throw new Error(`Invalid duration "${input}": must be a positive value.`);
+  }
+  return value * DURATION_UNIT_MS[match[2]];
+}
+function selectSessionsToDelete(sessions, opts) {
+  const { maxAgeMs, maxCount, nowMs, protectedIds } = opts;
+  if (maxAgeMs === void 0 && maxCount === void 0) return [];
+  const ageEligible = (s) => {
+    if (maxAgeMs === void 0) return false;
+    if (s.lastActivityMs === null) return true;
+    return nowMs - s.lastActivityMs > maxAgeMs;
+  };
+  const countEligibleIds = /* @__PURE__ */ new Set();
+  if (maxCount !== void 0) {
+    const byActivityDesc = [...sessions].sort(
+      (a, b) => (b.lastActivityMs ?? -Infinity) - (a.lastActivityMs ?? -Infinity)
+    );
+    for (const s of byActivityDesc.slice(maxCount)) {
+      countEligibleIds.add(s.id);
+    }
+  }
+  const toDelete = [];
+  for (const s of sessions) {
+    if (protectedIds.has(s.id)) continue;
+    if (ageEligible(s) || countEligibleIds.has(s.id)) {
+      toDelete.push(s.id);
+    }
+  }
+  return toDelete;
+}
+var DEFAULT_INTERVAL = "1h";
+function resolve(flag, envValue, fallback) {
+  return flag ?? envValue ?? fallback;
+}
+function parseMaxCount(input) {
+  const trimmed = input.trim();
+  if (!/^\d+$/.test(trimmed)) {
+    throw new Error(`Invalid max-count "${input}": expected a positive integer.`);
+  }
+  const value = Number(trimmed);
+  if (value <= 0) {
+    throw new Error(`Invalid max-count "${input}": must be greater than 0.`);
+  }
+  return value;
+}
+function resolveSessionCleanupConfig(flags, env = process.env) {
+  const warnings = [];
+  const maxAgeRaw = resolve(flags.maxAge, env.EVIDENT_SESSION_CLEANUP_MAX_AGE);
+  const maxCountRaw = resolve(flags.maxCount, env.EVIDENT_SESSION_CLEANUP_MAX_COUNT);
+  const intervalRaw = resolve(
+    flags.interval,
+    env.EVIDENT_SESSION_CLEANUP_INTERVAL,
+    DEFAULT_INTERVAL
+  );
+  let maxAgeMs;
+  if (maxAgeRaw !== void 0) {
+    try {
+      maxAgeMs = parseDurationMs(maxAgeRaw);
+    } catch (err) {
+      warnings.push(
+        `Ignoring invalid --session-cleanup-max-age: ${err instanceof Error ? err.message : String(err)}`
+      );
+    }
+  }
+  let maxCount;
+  if (maxCountRaw !== void 0) {
+    try {
+      maxCount = parseMaxCount(maxCountRaw);
+    } catch (err) {
+      warnings.push(
+        `Ignoring invalid --session-cleanup-max-count: ${err instanceof Error ? err.message : String(err)}`
+      );
+    }
+  }
+  let intervalMs;
+  try {
+    intervalMs = parseDurationMs(intervalRaw ?? DEFAULT_INTERVAL);
+  } catch (err) {
+    warnings.push(
+      `Ignoring invalid --session-cleanup-interval, using default ${DEFAULT_INTERVAL}: ${err instanceof Error ? err.message : String(err)}`
+    );
+    intervalMs = parseDurationMs(DEFAULT_INTERVAL);
+  }
+  const enabled = maxAgeMs !== void 0 || maxCount !== void 0;
+  return { enabled, maxAgeMs, maxCount, intervalMs, warnings };
 }
 
 // src/lib/tunnel/connection.ts
@@ -1232,24 +1446,26 @@ var StreamForwarder = class {
       this.send({ type: "res_end", sid });
       return;
     }
-    log("info", "agent_request", {
-      correlation_id: correlationId,
-      sid,
-      method,
-      path: stripQuery(path)
-    });
+    if (process.env.DEBUG) {
+      log("debug", "agent_request", {
+        correlation_id: correlationId,
+        sid,
+        method,
+        path: stripQuery(path)
+      });
+    }
     const ac = new AbortController();
     let bodyPromise;
     let pushBody;
     let endBody;
     if (has_body) {
       const chunks = [];
-      bodyPromise = new Promise((resolve) => {
+      bodyPromise = new Promise((resolve2) => {
         pushBody = (buf) => {
           chunks.push(buf);
         };
         endBody = () => {
-          resolve(Buffer.concat(chunks));
+          resolve2(Buffer.concat(chunks));
         };
       });
     }
@@ -1284,12 +1500,14 @@ var StreamForwarder = class {
       if (!STRIP_RES.has(key.toLowerCase())) resHeaders[key] = value;
     });
     this.send({ type: "head", sid, status: upstream.status, headers: resHeaders });
-    log("info", "agent_response", {
-      correlation_id: correlationId,
-      sid,
-      status: upstream.status,
-      duration_ms: Date.now() - startedAt
-    });
+    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) {
@@ -1365,7 +1583,7 @@ function connectTunnel(options) {
   } = options;
   const tunnelUrl = getTunnelUrlConfig();
   const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
-  return new Promise((resolve, reject) => {
+  return new Promise((resolve2, reject) => {
     const ws = new WebSocket2(url, {
       headers: {
         Authorization: authHeader
@@ -1430,7 +1648,7 @@ function connectTunnel(options) {
           clearTimeout(connectionTimeout);
           const connectedAgentId = message.agent_id ?? agentId;
           onConnected?.(connectedAgentId);
-          resolve({
+          resolve2({
             ws,
             close: () => ws.close(1e3, "CLI shutdown")
           });
@@ -1559,7 +1777,6 @@ 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) {
@@ -1595,11 +1812,18 @@ 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.
@@ -1650,6 +1874,20 @@ var ChannelDriver = class {
    * 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
@@ -1657,8 +1895,33 @@ 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;
@@ -1672,7 +1935,6 @@ 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());
   }
@@ -1691,8 +1953,21 @@ 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();
@@ -1704,6 +1979,7 @@ var ChannelDriver = class {
         });
       }
       for (const conv of conversations) {
+        if (this.stopped) break;
         dispatched += await this.processConversation(conv);
       }
       await this.readoptProcessing();
@@ -1724,6 +2000,73 @@ var ChannelDriver = class {
     }
     return false;
   }
+  /**
+   * OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
+   * exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
+   * `watchers` entry whose `inFlight` set is non-empty — the same predicate
+   * `hasInFlightWatchers()` uses, lifted to return the ids.
+   *
+   * Deliberately does NOT include `this.sessions` (the permanent, never-pruned
+   * conversation→session cache). Protecting every bound-but-idle session there
+   * would shield nearly every session and defeat cleanup — AND it is unnecessary:
+   * `ensureSession` is self-healing (it recreates a session whose id no longer
+   * exists), so deleting an idle bound session is harmless — the conversation's
+   * next turn transparently rebinds a fresh one. The only thing worth protecting
+   * is a session with a turn ACTIVELY in flight right now: tearing that down
+   * mid-turn would strand the running `prompt_async`. Idle sessions are fair game.
+   */
+  protectedSessionIds() {
+    const ids = /* @__PURE__ */ new Set();
+    for (const [sessionId, watcher] of this.watchers) {
+      if (watcher.inFlight.size > 0) ids.add(sessionId);
+    }
+    return ids;
+  }
+  /**
+   * 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).
    *
@@ -1760,15 +2103,16 @@ var ChannelDriver = class {
     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 = opencodeMessageIdFor2(message.id);
       const options = {
         agent: message.opencode_agent ?? void 0,
         model: message.opencode_model ?? void 0
       };
+      let opencodeMessageId;
       try {
         this.log({
           level: "info",
@@ -1776,10 +2120,23 @@ 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);
+        if (await sessionExists(this.port, sessionId) === false) {
+          this.sessions.delete(conv.id);
+          this.log({
+            level: "info",
+            message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
+            conversation_id: conv.id,
+            message_id: message.id
+          });
+          break;
+        }
         await this.markFailed(conv.id, message.id).catch(() => {
         });
         this.log({
@@ -1790,6 +2147,15 @@ 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;
@@ -1806,16 +2172,33 @@ var ChannelDriver = class {
     return dispatched;
   }
   async ensureSession(conv) {
-    const cached = this.sessions.get(conv.id);
-    if (cached) return cached;
-    if (conv.opencode_session_id) {
-      this.sessions.set(conv.id, conv.opencode_session_id);
-      return conv.opencode_session_id;
+    const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
+    if (bound) {
+      const exists = await sessionExists(this.port, bound);
+      if (exists === false) {
+        this.log({
+          level: "info",
+          message: `OpenCode session ${bound} for conversation ${conv.id.slice(0, 8)} no longer exists (deleted or DB reset) \u2014 creating a fresh session and rebinding.`,
+          conversation_id: conv.id
+        });
+        this.sessions.delete(conv.id);
+        return this.createAndBindSession(conv.id);
+      }
+      this.sessions.set(conv.id, bound);
+      return bound;
     }
+    return this.createAndBindSession(conv.id);
+  }
+  /**
+   * Create a fresh OpenCode session for a conversation, cache the binding, and
+   * best-effort persist it server-side. Shared by the first-ever bind and the
+   * self-heal recreate path in `ensureSession`.
+   */
+  async createAndBindSession(conversationId) {
     const directory = await this.resolveOpenCodeDirectory();
     const sessionId = await createOpenCodeSession(this.port, directory);
-    this.sessions.set(conv.id, sessionId);
-    await this.persistSession(conv.id, sessionId).catch(() => {
+    this.sessions.set(conversationId, sessionId);
+    await this.persistSession(conversationId, sessionId).catch(() => {
     });
     return sessionId;
   }
@@ -1838,6 +2221,25 @@ 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);
@@ -1902,10 +2304,10 @@ var ChannelDriver = class {
       started: true,
       done: false,
       // Not yet reported stuck-queued. The once-guard (`stuckReported`) applies,
-      // AND the stuck-queued observer now 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
-      // (queued-followup-redrive, #210).
+      // 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
     });
   }
@@ -1993,10 +2395,15 @@ 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({
@@ -2026,7 +2433,12 @@ 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) {
@@ -2062,12 +2474,55 @@ 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;
     }
-    if (state === "queued" && !inFlight.stuckReported && this.now() - inFlight.dispatchedAt >= this.stuckQueuedMs && !hasRunningAssistantExcept(messages, inFlight.opencodeMessageId)) {
+    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
@@ -2080,41 +2535,11 @@ var ChannelDriver = class {
         conversation_id: conv.id,
         message_id: inFlight.evidentMessageId
       });
-      this.removeInFlight(watcher, inFlight.evidentMessageId);
-    }
-  }
-  /**
-   * 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.
-   */
-  async redispatchInFlight(sessionId, inFlight) {
-    const options = {
-      agent: inFlight.message.opencode_agent ?? void 0,
-      model: inFlight.message.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
-    });
-    try {
-      await sendPromptAsync(
-        this.port,
-        sessionId,
-        inFlight.message.content,
-        options,
-        inFlight.opencodeMessageId
-      );
-    } catch (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
+      void this.postSignal(conv.id, inFlight.evidentMessageId, "gave_up", {
+        watched_for_ms: this.now() - inFlight.dispatchedAt
       });
+      this.removeInFlight(watcher, inFlight.evidentMessageId);
     }
-    inFlight.dispatchedAt = this.now();
   }
   // -------------------------------------------------------------------------
   // Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
@@ -2203,10 +2628,17 @@ var ChannelDriver = class {
    * 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, opencodeMessageIdFor(row.id))`:
+   * 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);
-   *   - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch);
-   *   - `unknown`          → re-dispatch the STABLE id + attach a watcher (orphan).
+   *   - `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.
    */
@@ -2220,8 +2652,8 @@ var ChannelDriver = class {
       });
       return;
     }
-    const ocId = opencodeMessageIdFor2(row.id);
-    const state = messageRunState(messages, ocId);
+    const ocId = row.opencode_message_id;
+    const state = messageRunState(messages, ocId ?? "");
     if (state === "done") {
       if (this.doneUndeliverable.has(row.id)) {
         this.log({
@@ -2239,7 +2671,7 @@ var ChannelDriver = class {
         message_id: row.id
       });
       try {
-        await this.markDone(row.conversation_id, row.id, sessionId);
+        await this.markDone(row.conversation_id, row.id, sessionId, ocId);
       } catch (err) {
         if (err instanceof ChannelAuthError) throw err;
         if (err instanceof ChannelTerminalError) {
@@ -2263,6 +2695,39 @@ var ChannelDriver = class {
       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",
@@ -2272,7 +2737,7 @@ var ChannelDriver = class {
       });
       return;
     }
-    if (state === "running" || state === "queued") {
+    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));
@@ -2281,7 +2746,7 @@ var ChannelDriver = class {
       this.ensureWatcherRunning(sessionId);
       this.log({
         level: "info",
-        message: `Re-adopt: message ${row.id.slice(0, 8)} ${state} \u2014 re-attached watcher (stable id, no re-dispatch)`,
+        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
       });
@@ -2290,23 +2755,45 @@ var ChannelDriver = class {
     await this.forceReadoptRun(sessionId, row);
   }
   /**
-   * Re-dispatch an orphaned (`unknown`) `processing` row (ADR-0046 Decision §2).
+   * 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.
    *
-   * The stable-id user message is absent from the session, so we (re-)dispatch with
-   * the STABLE id (`opencodeMessageIdFor(row.id)`) — NOT a divergent per-attempt id.
-   * This is what keeps the reply correlatable: the server's completion
-   * notification looks for the reply under the stable id, so the fresh turn's reply
-   * (which hangs off the stable id) is found and delivered. The residual
-   * duplicate-incomplete-turn semantics (ADR §2, `.harness/restart-recovery-orphan-finding.md`)
-   * are unchanged and, for an ABSENT id, cannot bite — there is no existing turn
-   * to swallow the duplicate.
+   * ⚠️ 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; the stable
-   * `opencodeMessageId` is what the watcher polls. Deadline anchored to
+   * `evidentMessageId = row.id` addresses the SERVER row. Deadline anchored to
    * `processed_at` (Invariant 1).
    */
   async forceReadoptRun(sessionId, row) {
-    const ocId = opencodeMessageIdFor2(row.id);
+    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({
@@ -2323,13 +2810,19 @@ var ChannelDriver = class {
     };
     this.log({
       level: "info",
-      message: `Re-adopt: message ${row.id.slice(0, 8)} orphaned (user message absent) \u2014 re-dispatching with the stable id`,
+      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, sessionId, row.content, options, ocId);
+      ocId = await this.dispatchLocked(
+        sessionId,
+        () => sendPromptAsync(this.port, sessionId, row.content, options)
+      );
     } catch (err) {
+      this.awaitingReadopt.delete(row.id);
       if (err instanceof ChannelAuthError) throw err;
       this.log({
         level: "error",
@@ -2339,11 +2832,22 @@ var ChannelDriver = class {
       });
       return;
     }
+    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);
   }
   /**
@@ -2448,8 +2952,8 @@ 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,
@@ -2469,8 +2973,8 @@ 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,
@@ -2481,6 +2985,50 @@ 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/${sessionId}`);
+      if (res.ok) {
+        const body = await res.json();
+        parent = body && typeof body.parentID === "string" ? body.parentID : null;
+      }
+    } catch {
+      parent = void 0;
+    }
+    if (parent !== void 0) this.sessionParents.set(sessionId, parent);
+    return parent;
+  }
   /**
    * Attribute a surfaced interaction to the in-flight message it paused on (M-1).
    *
@@ -2613,13 +3161,17 @@ var ChannelDriver = class {
    * A single attempt (no internal retry): the watcher's per-tick loop is the
    * retry vehicle for the swap-to-running.
    */
-  async markProcessing(conversationId, messageId, sessionId) {
+  async markProcessing(conversationId, messageId, sessionId, opencodeMessageId) {
     const res = await this.fetchImpl(
       `${this.apiUrl}/agents/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
       {
         method: "PATCH",
         headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
-        body: JSON.stringify({ status: "processing", opencode_session_id: sessionId })
+        body: JSON.stringify({
+          status: "processing",
+          opencode_session_id: sessionId,
+          ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {}
+        })
       }
     );
     this.assertAuth(res, "marking message as processing");
@@ -2657,13 +3209,17 @@ var ChannelDriver = class {
    *                                    watcher retries next tick within the
    *                                    deadline, Finding 4).
    */
-  async markDone(conversationId, messageId, sessionId) {
+  async markDone(conversationId, messageId, sessionId, opencodeMessageId) {
     const res = await this.fetchImpl(
       `${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");
@@ -2673,7 +3229,17 @@ 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(
@@ -2681,7 +3247,7 @@ var ChannelDriver = class {
         {
           method: "PATCH",
           headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
-          body: JSON.stringify({ status: "failed" })
+          body: JSON.stringify(body)
         }
       )
     );
@@ -2992,6 +3558,25 @@ 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 {
@@ -3037,6 +3622,8 @@ 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;
+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(
@@ -3191,7 +3778,7 @@ async function driveChannels(state, driver) {
       logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
       if (state.interactive) displayStatus(state);
     }
-    await new Promise((resolve) => setTimeout(resolve, CHANNEL_POLL_INTERVAL_MS));
+    await new Promise((resolve2) => setTimeout(resolve2, CHANNEL_POLL_INTERVAL_MS));
     if (state.idleTimeout !== null && idlePolls >= 2) {
       const idleMs = idlePolls * CHANNEL_POLL_INTERVAL_MS;
       if (idleMs > state.idleTimeout * 1e3) {
@@ -3202,8 +3789,122 @@ async function driveChannels(state, driver) {
     }
   }
 }
-async function cleanup(state) {
+var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
+async function runSweep(state, driver, config2) {
+  const mode = `age=${config2.maxAgeMs ?? "\u2014"} count=${config2.maxCount ?? "\u2014"}`;
+  try {
+    const sessions = await listSessions(state.port);
+    if (sessions === null) {
+      logActivity(state, {
+        type: "info",
+        message: `Session cleanup: could not list sessions (opencode unreachable); skipping this sweep (${mode})`
+      });
+      return;
+    }
+    const toDelete = selectSessionsToDelete(
+      sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
+      {
+        maxAgeMs: config2.maxAgeMs,
+        maxCount: config2.maxCount,
+        nowMs: Date.now(),
+        protectedIds: driver.protectedSessionIds()
+      }
+    );
+    const protectedNow = driver.protectedSessionIds();
+    let deleted = 0;
+    let failed = 0;
+    let skippedNewlyActive = 0;
+    for (const id of toDelete) {
+      if (protectedNow.has(id)) {
+        skippedNewlyActive++;
+        logActivity(state, {
+          type: "info",
+          message: `Session cleanup: skipping ${id} \u2014 became active/bound after selection (${mode})`
+        });
+        continue;
+      }
+      if (await deleteSession(state.port, id)) deleted++;
+      else failed++;
+    }
+    const failedNote = failed > 0 ? `, failed ${failed}` : "";
+    const skippedNote = skippedNewlyActive > 0 ? `, skipped ${skippedNewlyActive} newly-active` : "";
+    logActivity(state, {
+      type: "info",
+      message: `Session cleanup: inspected ${sessions.length}, deleted ${deleted}${failedNote}${skippedNote} (${mode})`
+    });
+  } catch (error2) {
+    const message = error2 instanceof Error ? error2.message : String(error2);
+    logActivity(state, {
+      type: "error",
+      error: `Session cleanup sweep failed (non-fatal, ${mode}): ${message}`
+    });
+  }
+}
+function scheduleSessionCleanup(state, driver, options) {
+  const config2 = resolveSessionCleanupConfig(
+    {
+      maxAge: options.sessionCleanupMaxAge,
+      maxCount: options.sessionCleanupMaxCount,
+      interval: options.sessionCleanupInterval
+    },
+    process.env
+  );
+  for (const warning2 of config2.warnings) {
+    logActivity(state, { type: "info", message: `Session cleanup: ${warning2}` });
+  }
+  if (!config2.enabled) return;
+  logActivity(state, {
+    type: "info",
+    message: `Session cleanup enabled (age=${config2.maxAgeMs ?? "\u2014"}, count=${config2.maxCount ?? "\u2014"}, interval=${config2.intervalMs}ms)`
+  });
+  const interval = setInterval(() => void runSweep(state, driver, config2), config2.intervalMs);
+  const firstSweep = setTimeout(
+    () => void runSweep(state, driver, config2),
+    SESSION_CLEANUP_FIRST_SWEEP_MS
+  );
+  state.sessionCleanupTimers.push(interval, firstSweep);
+}
+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;
+  for (const timer of state.sessionCleanupTimers) {
+    clearInterval(timer);
+    clearTimeout(timer);
+  }
+  state.sessionCleanupTimers = [];
+  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;
@@ -3234,10 +3935,13 @@ async function run(options) {
     opencodeVersion: null,
     opencodeProcess: null,
     connection: null,
+    channelDriver: null,
     running: true,
+    shuttingDown: false,
     activityLog: [],
     messageCount: 0,
     lastProxiedActivityAt: null,
+    sessionCleanupTimers: [],
     authHeader: ""
   };
   if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
@@ -3248,13 +3952,15 @@ async function run(options) {
     );
   }
   const handleSignal = async () => {
+    if (state.shuttingDown) return;
+    state.shuttingDown = true;
     if (state.interactive) {
       logActivity(state, { type: "info", message: "Shutting down..." });
       displayStatus(state);
     } else {
       log2(state, "Shutting down...");
     }
-    await cleanup(state);
+    await cleanup(state, { graceful: true });
     await shutdownTelemetry();
     process.exit(0);
   };
@@ -3374,12 +4080,14 @@ 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,
@@ -3393,7 +4101,11 @@ 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) => {
@@ -3472,10 +4184,12 @@ async function run(options) {
       if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
       throw error2;
     }
+    scheduleSessionCleanup(state, channelDriver, options);
     if (!interactive || state.json) {
       log2(state, "Driving channel messages...");
     }
     await driveChannels(state, channelDriver);
+    if (state.shuttingDown) return;
     await cleanup(state);
     if (state.json) {
       console.log(
@@ -3490,6 +4204,7 @@ async function run(options) {
     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) {
@@ -3524,7 +4239,16 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
 program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
 program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
 program.command("whoami").description("Show the currently logged in user").action(whoami);
-program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Agent ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option("-v, --verbose", "Show detailed request/response information").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").action(
+program.command("run").description("Connect to Evident and process messages").option("-a, --agent [id]", "Agent ID to connect to (optional when EVIDENT_AGENT_KEY is set)").option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option("-v, --verbose", "Show detailed request/response information").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option("--json", "Output in JSON format").option(
+  "--session-cleanup-max-age <duration>",
+  "Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
+).option(
+  "--session-cleanup-max-count <n>",
+  "Keep only the newest N OpenCode sessions. Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_COUNT"
+).option(
+  "--session-cleanup-interval <duration>",
+  "How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
+).action(
   (options) => {
     run({
       agent: options.agent,
@@ -3532,7 +4256,11 @@ program.command("run").description("Connect to Evident and process messages").op
       verbose: options.verbose,
       conversation: options.conversation,
       idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
-      json: options.json
+      json: options.json,
+      // Raw strings — the resolver in run.ts single-sources parsing (M1).
+      sessionCleanupMaxAge: options.sessionCleanupMaxAge,
+      sessionCleanupMaxCount: options.sessionCleanupMaxCount,
+      sessionCleanupInterval: options.sessionCleanupInterval
     });
   }
 );