@botiverse/raft-daemon 0.68.0 → 0.69.0

package/dist/{chunk-RXK2KSYW.js → chunk-URPIDKXK.js}

@@ -3288,6 +3288,7 @@ var AGENT_ACTIVITY_DETAIL_KINDS = [
   "compaction_stale",
   "reviewing_changes",
   "review_finished",
+  "review_stale",
   "runtime_reconnecting",
   "runtime_error",
   "runtime_crashed",
@@ -13560,11 +13561,12 @@ var RUNTIME_ERROR_DELIVERY_BACKOFF_MAX_MS = 5 * 6e4;
 var RUNTIME_ERROR_FINGERPRINT_FENCE_THRESHOLD = 3;
 var SPAWN_FAIL_BACKOFF_BASE_MS = 1e3;
 var SPAWN_FAIL_BACKOFF_MAX_MS = 3e4;
-var SPAWN_FAIL_BACKOFF_THRESHOLD = 3;
+var SPAWN_FAIL_BACKOFF_THRESHOLD = 0;
 var RUNNER_CREDENTIAL_MINT_BACKOFF_BASE_MS = 6e4;
 var RUNNER_CREDENTIAL_MINT_BACKOFF_MAX_MS = 10 * 6e4;
 var RUNNER_CREDENTIAL_MINT_BACKOFF_THRESHOLD = 0;
 var COMPACTION_STALE_MS = 5 * 6e4;
+var REVIEW_STALE_MS = 10 * 6e4;
 var RUNTIME_PROGRESS_STALE_MS = 15 * 6e4;
 var DEFAULT_RUNTIME_START_TIMEOUT_MS = 2 * 6e4;
 var DEFAULT_STALLED_RECOVERY_SIGTERM_TIMEOUT_MS = 1e4;
@@ -14653,10 +14655,10 @@ var AgentProcessManager = class _AgentProcessManager {
   }
   // ----- SPAWN-FAIL BACKOFF (per-agent) ----------------------------------
   // Anchored at auto_restart_from_idle (apm:3596) + runtime_profile_auto_restart (apm:4137).
-  // Generic spawn errors keep a threshold > 1 so single-transient hiccups behave as today.
-  // Runner credential mint failures are different: one outer spawn failure already means the
-  // inner credential-mint loop exhausted RUNNER_CREDENTIAL_MINT_MAX_ATTEMPTS. Those enter a
-  // longer cooldown immediately so repeated wakes don't keep burning three network fetches.
+  // One outer spawn failure enters cooldown immediately so repeated wakes cannot burn one
+  // full spawn attempt per delivery. Runner credential mint failures keep a longer cooldown
+  // because one outer failure already means the inner credential-mint loop exhausted
+  // RUNNER_CREDENTIAL_MINT_MAX_ATTEMPTS.
   // Successful spawn resets state. State lives outside AgentProcess because spawn fails BEFORE
   // ap exists and the rate-window must persist across stop/respawn (else stop/start churn
   // bypasses the cap — CC1 lifecycle pattern). Never advances any consume cursor or model-seen
@@ -14809,14 +14811,22 @@ var AgentProcessManager = class _AgentProcessManager {
       this.sendStdinNotification(agentId);
     }, delayMs);
   }
-  flushAsyncRejectedIdleDelivery(agentId) {
+  scheduleIdleInboxDeliveryRetry(agentId, ap, notificationCount, source, traceName) {
+    if (notificationCount <= 0 || !ap.driver.supportsStdinNotification || !ap.sessionId) return false;
+    ap.notifications.add(notificationCount);
+    ap.notifications.clearTimer();
+    return ap.notifications.schedule(() => {
+      this.flushIdleInboxDeliveryRetry(agentId, source, traceName);
+    }, this.stdinNotificationRetryMs);
+  }
+  flushIdleInboxDeliveryRetry(agentId, source, traceName) {
     const ap = this.agents.get(agentId);
     if (!ap) return false;
     const count = ap.notifications.takePendingAndClearTimer();
     if (count === 0) return false;
     if (!ap.driver.supportsStdinNotification || !ap.sessionId || ap.inbox.length === 0) {
       ap.notifications.add(count);
-      this.recordDaemonTrace("daemon.agent.stdin_delivery.async_rejected.retry", {
+      this.recordDaemonTrace(traceName, {
         agentId,
         runtime: ap.config.runtime,
         model: ap.config.model,
@@ -14837,7 +14847,7 @@ var AgentProcessManager = class _AgentProcessManager {
     ap.notifications.pruneContributedToPending(ap.inbox, ap.sessionId);
     const messages = ap.notifications.filterUncontributedMessages(ap.inbox, ap.sessionId);
     if (messages.length === 0) {
-      this.recordDaemonTrace("daemon.agent.stdin_delivery.async_rejected.retry", {
+      this.recordDaemonTrace(traceName, {
         agentId,
         runtime: ap.config.runtime,
         model: ap.config.model,
@@ -14851,16 +14861,16 @@ var AgentProcessManager = class _AgentProcessManager {
       return false;
     }
     this.commitApmIdleState(agentId, ap, false);
-    this.startRuntimeTrace(agentId, ap, "async-rejected-idle-delivery-retry", messages);
+    this.startRuntimeTrace(agentId, ap, source.replaceAll("_", "-"), messages);
     this.broadcastMessageReceivedActivity(agentId);
     const accepted = this.deliverInboxUpdateViaStdin(
       agentId,
       ap,
       messages,
       "idle",
-      "async_rejected_idle_delivery_retry"
+      source
     );
-    this.recordDaemonTrace("daemon.agent.stdin_delivery.async_rejected.retry", {
+    this.recordDaemonTrace(traceName, {
       agentId,
       runtime: ap.config.runtime,
       model: ap.config.model,
@@ -14873,6 +14883,13 @@ var AgentProcessManager = class _AgentProcessManager {
     });
     return accepted;
   }
+  flushAsyncRejectedIdleDelivery(agentId) {
+    return this.flushIdleInboxDeliveryRetry(
+      agentId,
+      "async_rejected_idle_delivery_retry",
+      "daemon.agent.stdin_delivery.async_rejected.retry"
+    );
+  }
   isApmIdle(ap) {
     return ap.gatedSteering.isIdle;
   }
@@ -14936,6 +14953,7 @@ var AgentProcessManager = class _AgentProcessManager {
     if (options.includeCompactionWatchdog ?? true) {
       this.clearCompactionWatchdog(ap);
     }
+    this.clearReviewWatchdog(ap);
     this.clearStalledRecoverySigtermWatchdog(ap);
   }
   clearRuntimeErrorDeliveryBackoffWithTrace(agentId, ap, resetSource) {
@@ -16112,6 +16130,7 @@ Use \`raft message read\` to catch up on the channels listed above, then stop. R
         readinessTransition: null,
         activation: { kind: "idle" },
         compaction: { kind: "none" },
+        review: { kind: "none" },
         runtimeProgress: new RuntimeProgressState(Date.now()),
         runtimeTraceSpan: null,
         runtimeTraceCounters: createRuntimeTraceCounters(),
@@ -18489,6 +18508,29 @@ Use \`raft message read\` to catch up on the channels listed above, then stop. R
     ap.compaction = { ...ap.compaction, watchdog: null };
     this.broadcastActivity(agentId, "working", "Context compaction still running; no finish event observed", [], void 0, "compaction_stale");
   }
+  startReviewWatchdog(agentId, ap) {
+    this.clearReviewWatchdog(ap);
+    const startedAt = Date.now();
+    const watchdog = setTimeout(() => {
+      this.markReviewStale(agentId, startedAt);
+    }, REVIEW_STALE_MS);
+    ap.review = { kind: "active", startedAt, watchdog };
+  }
+  clearReviewWatchdog(ap) {
+    if (ap.review.kind === "active" && ap.review.watchdog) {
+      clearTimeout(ap.review.watchdog);
+    }
+    ap.review = { kind: "none" };
+  }
+  markReviewStale(agentId, startedAt) {
+    const ap = this.agents.get(agentId);
+    if (!ap || ap.review.kind !== "active" || ap.review.startedAt !== startedAt) return;
+    ap.review = { ...ap.review, watchdog: null };
+    this.broadcastActivity(agentId, "working", "Review mode still active; no finish event observed", [], void 0, "review_stale");
+    const reduction = reduceApmGatedReview(ap.gatedSteering, { kind: "review_finished" });
+    this.commitGatedSteeringDecisionState(agentId, ap, reduction.nextState, { event: "review_finished", inferred: true });
+    this.flushReviewBoundaryMessages(agentId, ap);
+  }
   completeCompactionIfActive(agentId, detail = "Context compaction finished", options = {}) {
     const ap = this.agents.get(agentId);
     if (!ap || ap.compaction.kind !== "active") return;
@@ -19428,11 +19470,15 @@ Use \`raft message read\` to catch up on the channels listed above, then stop. R
         if (ap) {
           const reduction = reduceApmGatedReview(ap.gatedSteering, { kind: "review_started" });
           this.commitGatedSteeringDecisionState(agentId, ap, reduction.nextState, { event: "review_started" });
+          this.startReviewWatchdog(agentId, ap);
         }
         break;
       case "review_finished":
         this.flushPendingTrajectory(agentId);
-        if (ap) this.recordRuntimeTraceEvent(agentId, ap, "runtime.review_mode.finished");
+        if (ap) {
+          this.clearReviewWatchdog(ap);
+          this.recordRuntimeTraceEvent(agentId, ap, "runtime.review_mode.finished");
+        }
         this.broadcastActivity(agentId, "working", "Review finished", [], void 0, "review_finished");
         if (ap) {
           const reduction = reduceApmGatedReview(ap.gatedSteering, { kind: "review_finished" });
@@ -19953,6 +19999,15 @@ ${RESPONSE_TARGET_HINT}`;
       `${mode} inbox update`
     );
     if (!sendResult.ok) {
+      const retryNotificationCount = mode === "idle" && ap.driver.supportsStdinNotification && ap.sessionId ? messages.length : 0;
+      const retrySource = source.endsWith("_retry") ? source : `${source}_retry`;
+      const retryScheduled = mode === "idle" ? this.scheduleIdleInboxDeliveryRetry(
+        agentId,
+        ap,
+        retryNotificationCount,
+        retrySource,
+        "daemon.agent.stdin_delivery.idle_retry"
+      ) : false;
       if (mode === "idle") {
         this.commitApmIdleState(agentId, ap, true);
       }
@@ -19977,7 +20032,9 @@ ${RESPONSE_TARGET_HINT}`;
         outcome: runtimeSendFailureOutcome(sendResult),
         failure_reason: sendResult.reason,
         failure_error: sendResult.error,
-        requeued_messages_count: 0,
+        requeued_messages_count: retryNotificationCount,
+        retry_scheduled: retryScheduled,
+        notification_timer_present: ap.notifications.hasTimer,
         cursors_advanced: "none"
       }, "error");
       return false;
@@ -20162,6 +20219,14 @@ var systemClock = {
   clearTimeout: (timer) => clearTimeout(timer)
 };
 var INBOUND_WATCHDOG_MS = 7e4;
+function normalizeHandshakeReason(value) {
+  const raw = Array.isArray(value) ? value[0] : value;
+  if (typeof raw !== "string") return null;
+  const trimmed = raw.trim();
+  if (!trimmed) return null;
+  if (!/^[a-z0-9_:-]{1,80}$/i.test(trimmed)) return "invalid_header";
+  return trimmed;
+}
 function durationMsBucket(ms) {
   if (ms == null || !Number.isFinite(ms) || ms < 0) return "unknown";
   if (ms === 0) return "0";
@@ -20254,6 +20319,7 @@ var DaemonConnection = class {
     });
     const ws = this.options.wsFactory ? this.options.wsFactory(wsUrl, wsOptions) : new WebSocket(wsUrl, wsOptions);
     this.ws = ws;
+    let handshakeRejected = false;
     ws.on("open", () => {
       if (this.ws !== ws) return;
       if (!this.shouldConnect) return;
@@ -20314,8 +20380,27 @@ var DaemonConnection = class {
       this.options.onDisconnect();
       this.scheduleReconnect();
     });
+    ws.on("unexpected-response", (_request, response) => {
+      if (this.ws !== ws) return;
+      handshakeRejected = true;
+      const reason = normalizeHandshakeReason(response.headers["slock-reason"]);
+      const statusCode = response.statusCode ?? 0;
+      const reasonText = reason ? `, slock_reason=${reason}` : "";
+      logger.error(`[Daemon] WebSocket handshake rejected (status=${statusCode}${reasonText})`);
+      this.trace("daemon.connection.handshake_rejected", {
+        status_code: statusCode,
+        slock_reason_present: Boolean(reason),
+        slock_reason: reason
+      }, "error");
+      response.resume();
+      try {
+        ws.terminate();
+      } catch {
+      }
+    });
     ws.on("error", (err) => {
       if (this.ws !== ws) return;
+      if (handshakeRejected) return;
       logger.error(`[Daemon] WebSocket error: ${err.message}`);
       this.trace("daemon.connection.error", {
         error_class: err.name || "Error"
@@ -21408,7 +21493,7 @@ function resolveSlockCliPathOrEmpty(moduleUrl = import.meta.url) {
 }
 async function runBundledSlockCli(argv) {
   process.argv = [process.execPath, "slock", ...argv];
-  await import("./dist-Y3VRAU5E.js");
+  await import("./dist-P3SAWND7.js");
 }
 function detectRuntimes(tracer = noopTracer) {
   const ids = [];

package/dist/cli/index.js

@@ -44853,6 +44853,7 @@ var AGENT_ACTIVITY_DETAIL_KINDS = [
   "compaction_stale",
   "reviewing_changes",
   "review_finished",
+  "review_stale",
   "runtime_reconnecting",
   "runtime_error",
   "runtime_crashed",
@@ -51041,6 +51042,9 @@ function formatClaimResults(channel, data) {
       const msgShort = r.messageId ? r.messageId.slice(0, 8) : "";
       return `${label} (msg:${msgShort}): claimed`;
     }
+    if (r.reason === "already claimed by you") {
+      return `${label}: already claimed by you.`;
+    }
     return `${label}: FAILED \u2014 ${r.reason || "already claimed"}. Do not work on this task unless an owner/admin explicitly redirects it to you.`;
   });
   const succeeded = data.results.filter((r) => r.success).length;

package/dist/core.js

@@ -12,7 +12,7 @@ import {
   runBundledSlockCli,
   scanWorkspaceDirectories,
   subscribeDaemonLogs
-} from "./chunk-RXK2KSYW.js";
+} from "./chunk-URPIDKXK.js";
 export {
   DAEMON_CLI_USAGE,
   DAEMON_CORE_TRACE_ATTR_CONTRACTS,

package/dist/{dist-Y3VRAU5E.js → dist-P3SAWND7.js}

@@ -44382,6 +44382,7 @@ var AGENT_ACTIVITY_DETAIL_KINDS = [
   "compaction_stale",
   "reviewing_changes",
   "review_finished",
+  "review_stale",
   "runtime_reconnecting",
   "runtime_error",
   "runtime_crashed",
@@ -50421,6 +50422,9 @@ function formatClaimResults(channel, data) {
       const msgShort = r.messageId ? r.messageId.slice(0, 8) : "";
       return `${label} (msg:${msgShort}): claimed`;
     }
+    if (r.reason === "already claimed by you") {
+      return `${label}: already claimed by you.`;
+    }
     return `${label}: FAILED \u2014 ${r.reason || "already claimed"}. Do not work on this task unless an owner/admin explicitly redirects it to you.`;
   });
   const succeeded = data.results.filter((r) => r.success).length;

package/dist/index.js

@@ -3,7 +3,7 @@ import {
   DAEMON_CLI_USAGE,
   DaemonCore,
   parseDaemonCliArgs
-} from "./chunk-RXK2KSYW.js";
+} from "./chunk-URPIDKXK.js";
 
 // src/index.ts
 var parsedArgs = parseDaemonCliArgs(process.argv.slice(2));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@botiverse/raft-daemon",
-  "version": "0.68.0",
+  "version": "0.69.0",
   "type": "module",
   "bin": {
     "raft-daemon": "dist/raft-daemon.js",