@adhdev/daemon-core 0.9.82-rc.18 → 0.9.82-rc.180

package/src/mesh/mesh-events.ts

@@ -1,31 +1,76 @@
+import { appendFileSync, existsSync, readFileSync, renameSync, statSync, unlinkSync, writeFileSync } from 'fs';
+import { join } from 'path';
 import type { DaemonComponents } from '../boot/daemon-lifecycle.js';
 import { loadConfig } from '../config/config.js';
 import { getMesh, getMeshByRepo } from '../config/mesh-config.js';
 import { detectCLI } from '../detection/cli-detector.js';
 import { LOG } from '../logging/logger.js';
-import { appendLedgerEntry, buildTaskCompletionEvidence, getSessionRecoveryContext, isIntentionalCleanupStopEntry, readLedgerEntries } from './mesh-ledger.js';
+import { appendLedgerEntry, buildTaskCompletionEvidence, getLedgerDir, getSessionRecoveryContext, isIntentionalCleanupStopEntry, readLedgerEntries } from './mesh-ledger.js';
 import type { MeshLedgerKind, SessionRecoveryContext } from './mesh-ledger.js';
-import { claimNextTask, updateSessionTaskStatus, enqueueTask, updateTaskStatus, getQueue, recordTaskAutoLaunch } from './mesh-work-queue.js';
+import { buildMeshNodeCapabilityTags, claimNextTask, updateSessionTaskStatus, enqueueTask, updateTaskStatus, getQueue, recordTaskAutoLaunch, updateDirectDispatchStatus, cleanupTerminalDirectDispatches, getActiveDirectDispatches } from './mesh-work-queue.js';
+import { BeadsDB } from './beads-db.js';
+import { fastForwardMeshNode } from './mesh-fast-forward.js';
 
 // ---------------------------------------------------------------------------
 // Remote Node Idle Session Tracking
 // ---------------------------------------------------------------------------
-// Tracks remote sessions that emitted 'agent:ready' so triggerMeshQueue 
-// can assign tasks to them.
+// Tracks remote sessions that emitted 'agent:ready' so triggerMeshQueue
+// can assign tasks to them. Each entry carries an expiresAt timestamp;
+// entries are swept on insertion to prevent unbounded growth.
 // ---------------------------------------------------------------------------
 interface RemoteIdleSession {
     nodeId: string;
     sessionId: string;
     providerType: string;
+    expiresAt: number;
 }
+const REMOTE_IDLE_SESSION_TTL_MS = 5 * 60 * 1000; // 5 minutes
 const remoteIdleSessions = new Map<string, RemoteIdleSession>(); // key: `${nodeId}:${sessionId}`
 
 // ---------------------------------------------------------------------------
-// MCP coordinator pending-event queue
+// Workspace-to-mesh lookup cache
+// ---------------------------------------------------------------------------
+// getMeshByRepo is called on every coordinator event for sessions without
+// meshNodeFor. Cache results for 5 seconds to avoid repeated config reads.
+const meshByWorkspaceCache = new Map<string, { mesh: any; cachedAt: number }>();
+const MESH_WORKSPACE_CACHE_TTL_MS = 5_000;
+const IDLE_AUTO_FAST_FORWARD_THROTTLE_MS = 30 * 60 * 1000;
+const idleAutoFastForwardLastAttempt = new Map<string, number>();
+
+function getCachedMeshByWorkspace(workspace: string): any {
+    const now = Date.now();
+    const cached = meshByWorkspaceCache.get(workspace);
+    if (cached && now - cached.cachedAt < MESH_WORKSPACE_CACHE_TTL_MS) return cached.mesh;
+    const mesh = getMeshByRepo(workspace);
+    meshByWorkspaceCache.set(workspace, { mesh, cachedAt: now });
+    return mesh;
+}
+
+function readWorkerResultMetadata(event: Record<string, unknown>): Record<string, unknown> | undefined {
+    return readRecord(event.workerResult) || readRecord(event.meshWorkerResult) || readRecord(event.structuredResult);
+}
+
+export function __resetIdleAutoFastForwardForTests(): void {
+    idleAutoFastForwardLastAttempt.clear();
+}
+
+function sweepExpiredRemoteIdleSessions(): void {
+    const now = Date.now();
+    for (const [key, session] of remoteIdleSessions) {
+        if (session.expiresAt <= now) remoteIdleSessions.delete(key);
+    }
+}
+
+// ---------------------------------------------------------------------------
+// MCP coordinator pending-event queue — FILE-BASED PERSISTENCE
 // ---------------------------------------------------------------------------
 // When a mesh event fires but no CLI coordinator session is registered (e.g.
-// the coordinator is Claude Code running via MCP), we buffer the event here.
-// The MCP server drains this queue on every mesh_status / mesh_send_task poll.
+// the coordinator is Claude Code running via MCP), we persist the event to a
+// per-mesh JSONL file so it survives daemon restarts. The 50-entry hard cap
+// is removed; the file is drained atomically on each get_pending_mesh_events
+// call and limited to 100 KB to prevent runaway growth.
+//
+// File: <ledgerDir>/<meshId>.pending-events.jsonl
 // ---------------------------------------------------------------------------
 
 export interface PendingMeshCoordinatorEvent {
@@ -35,39 +80,272 @@ export interface PendingMeshCoordinatorEvent {
     nodeId?: string;
     workspace?: string;
     metadataEvent: Record<string, unknown>;
+    coordinatorMessage?: string;
     queuedAt: number;
+    /**
+     * When set, this event is intended for a specific coordinator daemon.
+     * Coordinators on other daemons should ignore it during drain.
+     * Absent on legacy events — treated as broadcast to any coordinator.
+     */
+    targetCoordinatorDaemonId?: string;
+}
+
+const REFINE_TERMINAL_EVENTS = new Set(['refine:completed', 'refine:failed']);
+
+function readRefineJobId(event: { metadataEvent?: Record<string, unknown> } | Record<string, unknown>): string {
+    const metadata = readRecord((event as any).metadataEvent) || event as Record<string, unknown>;
+    const result = readRecord(metadata.result);
+    const refineJob = readRecord(result?.refineJob);
+    return readNonEmptyString(metadata.jobId) || readNonEmptyString(refineJob?.jobId);
 }
 
-const MAX_PENDING_EVENTS = 50;
-const pendingMeshCoordinatorEvents: PendingMeshCoordinatorEvent[] = [];
+function buildRefineTerminalEventFingerprint(meshId: string, eventName: string, metadataEvent: Record<string, unknown>): string {
+    const jobId = readRefineJobId({ metadataEvent });
+    return jobId && REFINE_TERMINAL_EVENTS.has(eventName) ? `${meshId}::${eventName}::${jobId}` : '';
+}
+
+function hasPendingRefineTerminalEventDuplicate(event: PendingMeshCoordinatorEvent): boolean {
+    if (!REFINE_TERMINAL_EVENTS.has(event.event)) return false;
+    const jobId = readRefineJobId(event);
+    if (!jobId) return false;
+    return readPendingMeshCoordinatorEventsFromDisk(event.meshId).some((pending) =>
+        pending.event === event.event && readRefineJobId(pending) === jobId,
+    );
+}
+
+function buildPendingEventFingerprint(event: PendingMeshCoordinatorEvent): string {
+    const metadata = readRecord(event.metadataEvent) || {};
+    const sessionId = resolveEventSessionId(metadata);
+    const providerSessionId = readNonEmptyString(metadata.providerSessionId);
+    const taskId = readNonEmptyString(metadata.taskId) || readNonEmptyString(readRecord(metadata.payload)?.taskId);
+    const jobId = readRefineJobId(event);
+    const timestamp = metadata.timestamp !== undefined && metadata.timestamp !== null ? String(metadata.timestamp) : '';
+    return [
+        event.meshId,
+        event.event,
+        event.nodeId || '',
+        sessionId || '',
+        providerSessionId || '',
+        taskId || '',
+        jobId || '',
+        timestamp || '',
+    ].join('::');
+}
+
+function hasPendingCoordinatorEventDuplicate(event: PendingMeshCoordinatorEvent): boolean {
+    const fingerprint = buildPendingEventFingerprint(event);
+    if (!fingerprint.trim()) return false;
+    return readPendingMeshCoordinatorEventsFromDisk(event.meshId).some((pending) => buildPendingEventFingerprint(pending) === fingerprint);
+}
+
+function getPendingEventsPath(meshId: string, coordinatorDaemonId?: string): string {
+    const safe = meshId.replace(/[^a-zA-Z0-9_-]/g, '_');
+    if (coordinatorDaemonId) {
+        const safeDaemon = coordinatorDaemonId.replace(/[^a-zA-Z0-9_-]/g, '_');
+        return join(getLedgerDir(), `${safe}-${safeDaemon}.pending-events.jsonl`);
+    }
+    return join(getLedgerDir(), `${safe}.pending-events.jsonl`);
+}
+
+function readPendingMeshCoordinatorEventsFromDisk(meshId?: string, coordinatorDaemonId?: string): PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    // Read coordinator-scoped file first; fall back to legacy shared file.
+    const paths = coordinatorDaemonId
+        ? [getPendingEventsPath(meshId, coordinatorDaemonId), getPendingEventsPath(meshId)]
+        : [getPendingEventsPath(meshId)];
+    const events: PendingMeshCoordinatorEvent[] = [];
+    for (const path of paths) {
+        if (!existsSync(path)) continue;
+        try {
+            const raw = readFileSync(path, 'utf-8');
+            const parsed = raw.split('\n').filter(Boolean).flatMap(line => {
+                try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
+            });
+            // If reading the shared file, filter to events that target this coordinator or are unscoped.
+            const filtered = (coordinatorDaemonId && path === getPendingEventsPath(meshId))
+                ? parsed.filter(e => !e.targetCoordinatorDaemonId || e.targetCoordinatorDaemonId === coordinatorDaemonId)
+                : parsed;
+            events.push(...filtered);
+        } catch { /* skip unreadable files */ }
+    }
+    return events;
+}
+
+function refineTerminalEventFromLedger(meshId: string, pending: readonly PendingMeshCoordinatorEvent[]): PendingMeshCoordinatorEvent[] {
+    const acceptedJobIds = new Set(
+        pending
+            .filter(event => event.event === 'refine:accepted')
+            .map(event => readRefineJobId(event))
+            .filter(Boolean),
+    );
+    if (acceptedJobIds.size === 0) return [];
+    const existingTerminalJobIds = new Set(
+        pending
+            .filter(event => REFINE_TERMINAL_EVENTS.has(event.event))
+            .map(event => `${event.event}:${readRefineJobId(event)}`)
+            .filter(value => !value.endsWith(':')),
+    );
+    const backfilled: PendingMeshCoordinatorEvent[] = [];
+    const entries = readLedgerEntries(meshId, { tail: 200 });
+    for (let i = entries.length - 1; i >= 0; i--) {
+        const entry = entries[i];
+        if (entry.kind !== 'task_completed' && entry.kind !== 'task_failed') continue;
+        const payload = readRecord(entry.payload);
+        if (payload?.source !== 'refine_mesh_node_async_job') continue;
+        const refineJob = readRecord(payload.refineJob);
+        const jobId = readNonEmptyString(refineJob?.jobId);
+        if (!jobId || !acceptedJobIds.has(jobId)) continue;
+        const eventName = entry.kind === 'task_completed' ? 'refine:completed' : 'refine:failed';
+        if (existingTerminalJobIds.has(`${eventName}:${jobId}`)) continue;
+        existingTerminalJobIds.add(`${eventName}:${jobId}`);
+        const result = readRecord(payload.result);
+        const metadataEvent = {
+            source: 'refine_mesh_node_async_job',
+            jobId,
+            interactionId: readNonEmptyString(refineJob?.interactionId),
+            meshId,
+            nodeId: readNonEmptyString(refineJob?.nodeId) || entry.nodeId,
+            targetDaemonId: readNonEmptyString(refineJob?.targetDaemonId),
+            workspace: readNonEmptyString(refineJob?.workspace),
+            status: eventName === 'refine:completed' ? 'completed' : 'failed',
+            startedAt: readNonEmptyString(refineJob?.startedAt),
+            completedAt: readNonEmptyString(refineJob?.completedAt) || entry.timestamp,
+            retryOfJobId: readNonEmptyString(refineJob?.retryOfJobId) || readNonEmptyString(payload.retryOfJobId),
+            ...(result ? { result } : {}),
+        };
+        backfilled.push({
+            event: eventName,
+            meshId,
+            nodeLabel: readNonEmptyString(refineJob?.nodeId) || entry.nodeId || 'refine job',
+            nodeId: readNonEmptyString(refineJob?.nodeId) || entry.nodeId,
+            workspace: readNonEmptyString(refineJob?.workspace),
+            metadataEvent,
+            coordinatorMessage: buildMeshSystemMessage({
+                event: eventName,
+                nodeLabel: readNonEmptyString(refineJob?.nodeId) || entry.nodeId || 'refine job',
+                metadataEvent,
+            }),
+            queuedAt: Date.now(),
+        });
+    }
+    return backfilled.reverse();
+}
+
+function reconcilePendingMeshCoordinatorEvents(meshId: string, events: PendingMeshCoordinatorEvent[]): PendingMeshCoordinatorEvent[] {
+    const backfilled = refineTerminalEventFromLedger(meshId, events);
+    if (backfilled.length === 0) return events;
+    const terminalJobIds = new Set(backfilled.map(event => readRefineJobId(event)).filter(Boolean));
+    return [
+        ...events.filter(event => !(event.event === 'refine:accepted' && terminalJobIds.has(readRefineJobId(event)))),
+        ...backfilled,
+    ];
+}
+
+const MAX_PENDING_EVENTS_BYTES = 100 * 1024; // 100 KB — keep the pending file small
+const MAX_PENDING_EVENTS_KEEP = 50;           // keep the last 50 events when trimming
+
+function trimPendingEventsIfNeeded(path: string): void {
+    try {
+        if (!existsSync(path)) return;
+        if (statSync(path).size <= MAX_PENDING_EVENTS_BYTES) return;
+        const lines = readFileSync(path, 'utf-8').split('\n').filter(Boolean);
+        if (lines.length <= MAX_PENDING_EVENTS_KEEP) return;
+        writeFileSync(path, lines.slice(-MAX_PENDING_EVENTS_KEEP).join('\n') + '\n', 'utf-8');
+    } catch { /* best-effort; if trim fails, append still proceeds */ }
+}
 
 export function queuePendingMeshCoordinatorEvent(event: PendingMeshCoordinatorEvent): boolean {
-    if (pendingMeshCoordinatorEvents.length >= MAX_PENDING_EVENTS) {
+    try {
+        if (hasPendingRefineTerminalEventDuplicate(event)) {
+            LOG.info('MeshEvents', `Suppressed duplicate pending ${event.event} for refine job ${readRefineJobId(event)}`);
+            return true;
+        }
+        if (hasPendingCoordinatorEventDuplicate(event)) {
+            LOG.info('MeshEvents', `Suppressed duplicate pending ${event.event} for mesh ${event.meshId}`);
+            return true;
+        }
+        // Write to the coordinator-scoped file when the target coordinator is known;
+        // fall back to the shared file for legacy/unscoped events.
+        const path = getPendingEventsPath(event.meshId, event.targetCoordinatorDaemonId);
+        trimPendingEventsIfNeeded(path);
+        appendFileSync(path, JSON.stringify(event) + '\n', 'utf-8');
+        return true;
+    } catch (e: any) {
+        LOG.warn('MeshEvents', `Failed to persist pending coordinator event: ${e?.message || e}`);
         return false;
     }
-    pendingMeshCoordinatorEvents.push(event);
-    return true;
 }
 
-/** Drain and return all pending coordinator events, clearing the queue. */
-export function drainPendingMeshCoordinatorEvents(): PendingMeshCoordinatorEvent[] {
-    return pendingMeshCoordinatorEvents.splice(0);
+// Atomically rename the file before reading so concurrent drains can't both consume
+// the same events. renameSync is atomic on POSIX (same filesystem); only one caller
+// wins the rename — the other gets ENOENT and returns null, preventing duplicate delivery.
+function atomicDrainFile(path: string): string | null {
+    const tmpPath = `${path}.draining`;
+    try {
+        renameSync(path, tmpPath);
+    } catch {
+        return null; // another drain already renamed it, or file doesn't exist
+    }
+    try {
+        const content = readFileSync(tmpPath, 'utf-8');
+        try { unlinkSync(tmpPath); } catch { /* already cleaned up */ }
+        return content;
+    } catch {
+        try { unlinkSync(tmpPath); } catch { /* best-effort cleanup */ }
+        return null;
+    }
+}
+
+/** Drain and return all pending coordinator events for meshId, removing them from disk. */
+export function drainPendingMeshCoordinatorEvents(meshId?: string, coordinatorDaemonId?: string): PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    const paths = coordinatorDaemonId
+        ? [getPendingEventsPath(meshId, coordinatorDaemonId), getPendingEventsPath(meshId)]
+        : [getPendingEventsPath(meshId)];
+    const all: PendingMeshCoordinatorEvent[] = [];
+    for (const path of paths) {
+        const content = atomicDrainFile(path);
+        if (!content) continue;
+        const parsed = content.split('\n').filter(Boolean).flatMap(line => {
+            try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
+        });
+        // If reading the shared file, filter to events that target this coordinator or are unscoped.
+        const filtered = (coordinatorDaemonId && path === getPendingEventsPath(meshId))
+            ? parsed.filter(e => !e.targetCoordinatorDaemonId || e.targetCoordinatorDaemonId === coordinatorDaemonId)
+            : parsed;
+        all.push(...filtered);
+    }
+    if (all.length === 0) return [];
+    return reconcilePendingMeshCoordinatorEvents(meshId, all);
 }
 
 /** Peek at pending coordinator events without draining (non-destructive). */
-export function getPendingMeshCoordinatorEvents(): readonly PendingMeshCoordinatorEvent[] {
-    return pendingMeshCoordinatorEvents.slice();
+export function getPendingMeshCoordinatorEvents(meshId?: string, coordinatorDaemonId?: string): readonly PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    return reconcilePendingMeshCoordinatorEvents(meshId, readPendingMeshCoordinatorEventsFromDisk(meshId, coordinatorDaemonId));
 }
 
-/** Explicitly clear all pending coordinator events. */
-export function clearPendingMeshCoordinatorEvents(): void {
-    pendingMeshCoordinatorEvents.splice(0);
+/** Explicitly clear all pending coordinator events for a mesh (and coordinator if scoped). */
+export function clearPendingMeshCoordinatorEvents(meshId?: string, coordinatorDaemonId?: string): void {
+    if (!meshId) return;
+    const paths = coordinatorDaemonId
+        ? [getPendingEventsPath(meshId, coordinatorDaemonId), getPendingEventsPath(meshId)]
+        : [getPendingEventsPath(meshId)];
+    for (const path of paths) {
+        if (existsSync(path)) try { unlinkSync(path); } catch { /* already removed */ }
+    }
 }
 
 function readNonEmptyString(value: unknown): string {
     return typeof value === 'string' && value.trim() ? value.trim() : '';
 }
 
+function readRecord(value: unknown): Record<string, unknown> | undefined {
+    return value && typeof value === 'object' && !Array.isArray(value)
+        ? value as Record<string, unknown>
+        : undefined;
+}
+
 function resolveEventSessionId(event: Record<string, unknown>, fallback?: unknown): string {
     return readNonEmptyString(event.targetSessionId)
         || readNonEmptyString(event.sessionId)
@@ -82,6 +360,9 @@ const MESH_COORDINATOR_EVENTS = new Set([
     'agent:stopped',
     'agent:ready',
     'monitor:long_generating',
+    'refine:accepted',
+    'refine:completed',
+    'refine:failed',
 ]);
 
 const EVENT_TO_LEDGER_KIND: Record<string, MeshLedgerKind> = {
@@ -96,10 +377,21 @@ function isMeshCoordinatorEvent(eventName: unknown): eventName is string {
 }
 
 function formatCompletionMetadata(event: Record<string, unknown>): string {
+    const completionDiagnostic = event.completionDiagnostic && typeof event.completionDiagnostic === 'object'
+        ? event.completionDiagnostic as Record<string, unknown>
+        : null;
+    const diagnosticReason = completionDiagnostic
+        ? readNonEmptyString(completionDiagnostic.blockReason) || 'present'
+        : '';
+    const finalAssistantPresent = typeof completionDiagnostic?.finalAssistantPresent === 'boolean'
+        ? String(completionDiagnostic.finalAssistantPresent)
+        : '';
     const parts = [
         readNonEmptyString(event.targetSessionId) ? `session_id=${readNonEmptyString(event.targetSessionId)}` : '',
         readNonEmptyString(event.providerType) ? `provider=${readNonEmptyString(event.providerType)}` : '',
         readNonEmptyString(event.providerSessionId) ? `provider_session_id=${readNonEmptyString(event.providerSessionId)}` : '',
+        diagnosticReason ? `completion_diagnostic=${diagnosticReason}` : '',
+        finalAssistantPresent ? `final_assistant=${finalAssistantPresent}` : '',
     ].filter(Boolean);
     return parts.length > 0 ? ` (${parts.join('; ')})` : '';
 }
@@ -126,7 +418,7 @@ function isIntentionalCleanupStopMetadata(event: Record<string, unknown>): boole
 function hasRecentIntentionalCleanupStop(meshId: string, sessionId?: string, nodeId?: string): boolean {
     if (!sessionId && !nodeId) return false;
     const cutoff = Date.now() - INTENTIONAL_CLEANUP_STOP_SUPPRESSION_MS;
-    const entries = readLedgerEntries(meshId);
+    const entries = readLedgerEntries(meshId, { tail: 200 });
     for (let i = entries.length - 1; i >= 0; i--) {
         const entry = entries[i];
         const timestamp = new Date(entry.timestamp).getTime();
@@ -150,6 +442,226 @@ function shouldSuppressIntentionalCleanupStop(args: {
     return hasRecentIntentionalCleanupStop(args.meshId, args.sessionId, args.nodeId);
 }
 
+const RECENT_COMPLETION_FINGERPRINT_TTL_MS = 10 * 60 * 1000;
+
+function hasFingerprintSeen(fingerprint: string): boolean {
+    try {
+        return BeadsDB.getInstance().hasCompletionFingerprint(fingerprint);
+    } catch {
+        return false;
+    }
+}
+
+function recordFingerprintSeen(fingerprint: string): void {
+    try {
+        const db = BeadsDB.getInstance();
+        db.recordCompletionFingerprint(fingerprint, RECENT_COMPLETION_FINGERPRINT_TTL_MS);
+        db.sweepExpiredFingerprints();
+    } catch { /* best-effort; duplicate events are preferable to a crash */ }
+}
+
+function readEventTimestamp(value: unknown): number | null {
+    if (typeof value === 'number' && Number.isFinite(value)) return value;
+    if (typeof value === 'string' && value.trim()) {
+        const numeric = Number(value);
+        if (Number.isFinite(numeric)) return numeric;
+        const parsed = Date.parse(value);
+        if (Number.isFinite(parsed)) return parsed;
+    }
+    return null;
+}
+
+function buildMeshCompletionFingerprint(args: {
+    meshId: string;
+    event: string;
+    sessionId: string;
+    providerType?: string;
+    providerSessionId?: string;
+    timestamp?: number | null;
+    finalSummary?: string;
+    /** When set, scopes the fingerprint to a specific coordinator daemon so
+     *  two coordinators processing events from their respective workers don't
+     *  suppress each other's completion events. */
+    coordinatorDaemonId?: string;
+}): string {
+    const timestampPart = Number.isFinite(args.timestamp)
+        ? String(args.timestamp)
+        : readNonEmptyString(args.finalSummary).slice(0, 200);
+    return [
+        args.meshId,
+        args.event,
+        args.sessionId,
+        args.providerType || '',
+        args.providerSessionId || '',
+        timestampPart,
+        args.coordinatorDaemonId || '',
+    ].join('::');
+}
+
+function isDuplicateMeshCompletionEvent(args: {
+    meshId: string;
+    event: string;
+    sessionId: string;
+    providerType?: string;
+    providerSessionId?: string;
+    timestamp?: number | null;
+    finalSummary?: string;
+    coordinatorDaemonId?: string;
+}): boolean {
+    const fingerprint = buildMeshCompletionFingerprint(args);
+    if (!fingerprint) return false;
+    if (hasFingerprintSeen(fingerprint)) return true;
+    recordFingerprintSeen(fingerprint);
+    return false;
+}
+
+function isDuplicateMeshApprovalEvent(args: {
+    meshId: string;
+    sessionId: string;
+    providerType?: string;
+    timestamp?: number | null;
+    modalMessage?: string;
+    modalButtons?: unknown;
+}): boolean {
+    const modalButtons = Array.isArray(args.modalButtons)
+        ? args.modalButtons.map(button => String(button).trim()).filter(Boolean)
+        : [];
+    const approvalIdentity = Number.isFinite(args.timestamp)
+        ? String(args.timestamp)
+        : JSON.stringify({ message: args.modalMessage || '', buttons: modalButtons });
+    if (!approvalIdentity || approvalIdentity === '{"message":"","buttons":[]}') return false;
+    const fingerprint = [
+        args.meshId,
+        'agent:waiting_approval',
+        args.sessionId,
+        args.providerType || '',
+        approvalIdentity,
+    ].join('::');
+    if (hasFingerprintSeen(fingerprint)) return true;
+    recordFingerprintSeen(fingerprint);
+    return false;
+}
+
+function isDuplicateRefineTerminalEvent(meshId: string, eventName: string, metadataEvent: Record<string, unknown>): boolean {
+    const fingerprint = buildRefineTerminalEventFingerprint(meshId, eventName, metadataEvent);
+    if (!fingerprint) return false;
+    if (hasFingerprintSeen(fingerprint)) return true;
+    recordFingerprintSeen(fingerprint);
+    return false;
+}
+
+function findRecentTerminalLedgerEvidence(args: {
+    meshId: string;
+    sessionId?: string;
+    nodeId?: string;
+}): { id: string; kind: MeshLedgerKind; payload: Record<string, unknown>; timestamp: string } | null {
+    if (!args.sessionId && !args.nodeId) return null;
+    // Tail-limit: 200 entries gives a wide enough window to catch terminal events for active
+    // sessions while avoiding a full O(n) scan. If a terminal is older than 200 entries,
+    // the BeadsDB fingerprint dedup will still block duplicate processing downstream.
+    const entries = readLedgerEntries(args.meshId, { tail: 200 });
+    for (let i = entries.length - 1; i >= 0; i--) {
+        const entry = entries[i];
+        if (entry.kind !== 'task_completed' && entry.kind !== 'task_failed' && entry.kind !== 'task_stalled') continue;
+        if (args.sessionId && entry.sessionId === args.sessionId) {
+            return { id: entry.id, kind: entry.kind, payload: entry.payload || {}, timestamp: entry.timestamp };
+        }
+        if (!args.sessionId && args.nodeId && entry.nodeId === args.nodeId) {
+            return { id: entry.id, kind: entry.kind, payload: entry.payload || {}, timestamp: entry.timestamp };
+        }
+    }
+    return null;
+}
+
+// Returns true when a task_dispatched entry for the given session appears AFTER the terminal
+// entry (identified by terminalId) in ledger order. Positional (append) order is used rather
+// than timestamp comparison because both entries may share the same millisecond.
+function hasDispatchAfterTerminal(meshId: string, sessionId: string, terminalId: string): boolean {
+    // 200-entry window matches findRecentTerminalLedgerEvidence for consistency.
+    const entries = readLedgerEntries(meshId, { tail: 200 });
+    let pastTerminal = false;
+    for (const entry of entries) {
+        if (!pastTerminal) {
+            if (entry.id === terminalId) pastTerminal = true;
+            continue;
+        }
+        if (entry.kind === 'task_dispatched' && entry.sessionId === sessionId) return true;
+    }
+    return false;
+}
+
+function hasUnterminalDirectDispatchLedgerEntry(meshId: string, sessionId: string): boolean {
+    // Some dispatch paths can persist task_dispatched before the direct-dispatch DB row is
+    // available. Recover routing from ledger order so coordinator self-targets still emit
+    // task_completed and pendingCoordinatorEvents.
+    const entries = readLedgerEntries(meshId, { tail: 200 });
+    for (let i = entries.length - 1; i >= 0; i--) {
+        const entry = entries[i];
+        if (entry.sessionId !== sessionId) continue;
+        if (entry.kind === 'task_completed' || entry.kind === 'task_failed' || entry.kind === 'task_stalled') {
+            return false;
+        }
+        if (entry.kind === 'task_dispatched' && entry.payload?.source === 'direct') {
+            return true;
+        }
+    }
+    return false;
+}
+
+function buildLongGeneratingCompletionReconciliation(args: {
+    meshId: string;
+    nodeId?: string;
+    nodeLabel: string;
+    metadataEvent: Record<string, unknown>;
+    sourceInstanceId?: string;
+}): Record<string, unknown> | null {
+    const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
+    const nodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId);
+    const providerType = readNonEmptyString(args.metadataEvent.providerType);
+    const providerSessionId = readNonEmptyString(args.metadataEvent.providerSessionId);
+    const workerResult = readWorkerResultMetadata(args.metadataEvent);
+    const completionDiagnostic = readRecord(args.metadataEvent.completionDiagnostic);
+    const finalSummary = readNonEmptyString(args.metadataEvent.finalSummary);
+    const status = readNonEmptyString(args.metadataEvent.status).toLowerCase();
+    const explicitCompletionEvidence = Boolean(
+        finalSummary
+        || workerResult
+        || completionDiagnostic?.finalAssistantPresent === true
+        || status === 'idle'
+        || status === 'ready'
+        || status === 'completed',
+    );
+    if (explicitCompletionEvidence) {
+        return {
+            ...args.metadataEvent,
+            targetSessionId: sessionId,
+            providerType,
+            providerSessionId,
+            finalSummary,
+            source: 'long_generating_reconciliation',
+            reconciledFromEvent: 'monitor:long_generating',
+            timestamp: args.metadataEvent.timestamp ?? Date.now(),
+            completionDiagnostic: {
+                ...(completionDiagnostic || {}),
+                reconciliationReason: 'provider_completion_evidence',
+            },
+        };
+    }
+
+    const terminal = findRecentTerminalLedgerEvidence({
+        meshId: args.meshId,
+        sessionId: sessionId || undefined,
+        nodeId: nodeId || undefined,
+    });
+    if (!terminal) return null;
+    return {
+        ...args.metadataEvent,
+        source: 'long_generating_terminal_ledger_suppression',
+        terminalLedgerKind: terminal.kind,
+        terminalLedgerAt: terminal.timestamp,
+    };
+}
+
 
 export function tryAssignQueueTask(
     components: DaemonComponents,
@@ -158,7 +670,10 @@ export function tryAssignQueueTask(
     sessionId: string,
     providerType: string
 ): boolean {
-    const task = claimNextTask(meshId, nodeId, sessionId);
+    const mesh = getMeshWithCache(components, meshId);
+    const node = mesh?.nodes.find((n: any) => n.id === nodeId);
+    const capabilityTags = buildMeshNodeCapabilityTags(node, providerType);
+    const task = claimNextTask(meshId, nodeId, sessionId, capabilityTags);
     if (!task) {
         return false;
     }
@@ -166,9 +681,6 @@ export function tryAssignQueueTask(
     LOG.info('MeshQueue', `Node ${nodeId} (${sessionId}) pulled task ${task.id}`);
 
     // Check if the node is remote
-    const mesh = getMeshWithCache(components, meshId);
-    const node = mesh?.nodes.find((n: any) => n.id === nodeId);
-    
     // If the node is explicitly remote and we have a dispatch mechanism, route via P2P
     if (node?.daemonId && components.dispatchMeshCommand) {
         const isLocalNode = components.cliManager.adapters.has(sessionId);
@@ -180,7 +692,16 @@ export function tryAssignQueueTask(
                 message: task.message,
             }).catch((e: any) => {
                 LOG.error('MeshQueue', `Failed to dispatch task via P2P to remote node ${nodeId}: ${e?.message}`);
-                updateTaskStatus(meshId, task.id, 'failed');
+                // Revert to pending so the task can be retried rather than permanently failing
+                updateTaskStatus(meshId, task.id, 'pending');
+                try {
+                    appendLedgerEntry(meshId, {
+                        kind: 'dispatch_failed' as any,
+                        nodeId,
+                        sessionId,
+                        payload: { taskId: task.id, error: e?.message, retryable: true },
+                    });
+                } catch { /* ledger write is best-effort */ }
             });
             return true;
         }
@@ -204,6 +725,13 @@ const autoLaunchInProgress = new Set<string>();
 const autoLaunchCooldownUntil = new Map<string, number>();
 const AUTO_LAUNCH_COOLDOWN_MS = 5_000;
 
+function sweepExpiredCooldowns(): void {
+    const now = Date.now();
+    for (const [key, until] of autoLaunchCooldownUntil) {
+        if (now >= until) autoLaunchCooldownUntil.delete(key);
+    }
+}
+
 function normalizeProviderPriority(policy: unknown): string[] {
     const raw = policy && typeof policy === 'object' && !Array.isArray(policy)
         ? (policy as Record<string, unknown>).providerPriority
@@ -277,6 +805,10 @@ function nodeHasActiveAssignment(meshId: string, nodeId: string): boolean {
     return getQueue(meshId, { status: ['assigned'] as any }).some(task => task.assignedNodeId === nodeId);
 }
 
+function sessionHasActiveAssignment(meshId: string, sessionId: string): boolean {
+    return getQueue(meshId, { status: ['assigned'] as any }).some(task => task.assignedSessionId === sessionId);
+}
+
 function liveSessionCountForNode(components: DaemonComponents, meshId: string, nodeId: string): number {
     return components.instanceManager.getByCategory('cli').filter((inst: any) => {
         const state = inst.getState();
@@ -406,12 +938,14 @@ async function maybeAutoLaunchOneQueueSession(components: DaemonComponents, mesh
             const nodeId = readNonEmptyString(node?.id);
             if (!nodeId) continue;
             const launchKey = `${meshId}:${nodeId}`;
+            const now = Date.now();
             const cooldownUntil = autoLaunchCooldownUntil.get(launchKey) || 0;
+            if (cooldownUntil > 0 && now >= cooldownUntil) autoLaunchCooldownUntil.delete(launchKey);
             if (autoLaunchInProgress.has(launchKey)) {
                 markAutoLaunch(meshId, task.id, { status: 'skipped', reason: 'auto_launch_in_progress', nodeId });
                 continue;
             }
-            if (Date.now() < cooldownUntil) {
+            if (now < cooldownUntil) {
                 markAutoLaunch(meshId, task.id, { status: 'skipped', reason: 'auto_launch_cooldown', nodeId });
                 continue;
             }
@@ -461,13 +995,13 @@ async function maybeAutoLaunchOneQueueSession(components: DaemonComponents, mesh
                 if (!launchResult?.success) {
                     const reason = launchResult?.error || 'launch_cli_failed';
                     markAutoLaunch(meshId, task.id, { status: 'failed', reason, nodeId, providerType: resolved.providerType });
-                    autoLaunchCooldownUntil.set(launchKey, Date.now() + AUTO_LAUNCH_COOLDOWN_MS);
+                    autoLaunchCooldownUntil.set(launchKey, Date.now() + AUTO_LAUNCH_COOLDOWN_MS); sweepExpiredCooldowns();
                     return false;
                 }
                 const sessionId = readNonEmptyString(launchResult.sessionId) || readNonEmptyString(launchResult.id) || readNonEmptyString(launchResult.runtimeSessionId);
                 if (!sessionId) {
                     markAutoLaunch(meshId, task.id, { status: 'failed', reason: 'launch_missing_session_id', nodeId, providerType: resolved.providerType });
-                    autoLaunchCooldownUntil.set(launchKey, Date.now() + AUTO_LAUNCH_COOLDOWN_MS);
+                    autoLaunchCooldownUntil.set(launchKey, Date.now() + AUTO_LAUNCH_COOLDOWN_MS); sweepExpiredCooldowns();
                     return false;
                 }
                 markAutoLaunch(meshId, task.id, { status: 'completed', nodeId, providerType: resolved.providerType, sessionId });
@@ -534,6 +1068,72 @@ export async function triggerMeshQueue(components: DaemonComponents, meshId: str
     await maybeAutoLaunchOneQueueSession(components, meshId, mesh);
 }
 
+async function maybeAutoFastForwardIdleNode(components: DaemonComponents, args: {
+    meshId: string;
+    nodeId: string;
+    sessionId?: string;
+    providerType?: string;
+}): Promise<void> {
+    const mesh = getMeshWithCache(components, args.meshId);
+    const node = mesh?.nodes?.find((candidate: any) => candidate?.id === args.nodeId || candidate?.nodeId === args.nodeId);
+    const workspace = readNonEmptyString(node?.workspace);
+    if (!workspace) return;
+    if (!existsSync(workspace)) return;
+
+    const throttleKey = `${args.meshId}:${args.nodeId}`;
+    const now = Date.now();
+    const lastAttempt = idleAutoFastForwardLastAttempt.get(throttleKey) || 0;
+    if (now - lastAttempt < IDLE_AUTO_FAST_FORWARD_THROTTLE_MS) return;
+    idleAutoFastForwardLastAttempt.set(throttleKey, now);
+
+    const submoduleIgnorePaths = Array.isArray(node?.policy?.submoduleIgnorePaths)
+        ? node.policy.submoduleIgnorePaths.filter((value: unknown): value is string => typeof value === 'string')
+        : undefined;
+    try {
+        const dryRun = await fastForwardMeshNode({
+            meshId: args.meshId,
+            nodeId: args.nodeId,
+            workspace,
+            execute: false,
+            dryRun: true,
+            updateSubmodules: false,
+            submoduleIgnorePaths,
+            trigger: 'idle_auto',
+        });
+        if (!dryRun || dryRun.code !== 'fast_forward_available' || dryRun.allowed !== true) return;
+        await fastForwardMeshNode({
+            meshId: args.meshId,
+            nodeId: args.nodeId,
+            workspace,
+            execute: true,
+            dryRun: false,
+            updateSubmodules: false,
+            submoduleIgnorePaths,
+            trigger: 'idle_auto',
+        });
+    } catch (e: any) {
+        LOG.warn('MeshFastForward', `Idle auto fast-forward check failed for ${args.nodeId}: ${e?.message || e}`);
+    }
+}
+
+function runIdleMaintenanceThenAssignQueue(components: DaemonComponents, args: {
+    meshId: string;
+    nodeId: string;
+    sessionId: string;
+    providerType: string;
+}): void {
+    setImmediate(() => {
+        maybeAutoFastForwardIdleNode(components, args)
+            .finally(() => {
+                try {
+                    tryAssignQueueTask(components, args.meshId, args.nodeId, args.sessionId, args.providerType);
+                } catch (e: any) {
+                    LOG.warn('MeshQueue', `Failed to assign idle queue task after maintenance for ${args.nodeId}: ${e?.message || e}`);
+                }
+            });
+    });
+}
+
 function buildMeshSystemMessage(args: {
     event: string;
     nodeLabel: string;
@@ -542,6 +1142,9 @@ function buildMeshSystemMessage(args: {
 }): string {
     const metadata = formatCompletionMetadata(args.metadataEvent);
     if (args.event === 'agent:generating_completed') {
+        if (args.metadataEvent.source === 'long_generating_reconciliation') {
+            return `[System] ${args.nodeLabel} already has completion evidence${metadata}. The long-generating monitor reconciled the terminal handoff and marked the session complete; wait for the queued completion event/status refresh before doing any manual transcript check.`;
+        }
         return `[System] ${args.nodeLabel} has completed its task and is now idle${metadata}. This completion came from the agent status event path; use mesh_read_chat once to review its final progress, but do not poll repeatedly.`;
     }
     if (args.event === 'agent:waiting_approval') {
@@ -573,7 +1176,72 @@ function buildMeshSystemMessage(args: {
         return `[System] ${args.nodeLabel} has stopped${metadata}. Use mesh_read_chat once if you need to inspect its last output.`;
     }
     if (args.event === 'monitor:long_generating') {
-        return `[System] ${args.nodeLabel} has been generating for a long time${metadata}. Use mesh_read_chat once for a status check, but do not poll repeatedly.`;
+        return `[System] ${args.nodeLabel} is still reported as generating after a long interval${metadata}. Wait for pendingCoordinatorEvents or a completion/status event; if the user explicitly asks for status, make one bounded status check and then wait again.`;
+    }
+    if (args.event === 'refine:accepted') {
+        const jobId = readRefineJobId({ metadataEvent: args.metadataEvent });
+        return `[System] Refinery accepted async job${jobId ? ` ${jobId}` : ''} for ${args.nodeLabel}. Completion/failure will be delivered as a terminal refine event; do not poll repeatedly.`;
+    }
+    if (args.event === 'refine:completed') {
+        const jobId = readRefineJobId({ metadataEvent: args.metadataEvent });
+        const result = readRecord(args.metadataEvent.result);
+        const validationSummary = readRecord(result?.validationSummary);
+        const patchEquivalence = readRecord(result?.patchEquivalence);
+        const finalConvergence = readRecord(result?.finalBranchConvergenceState);
+        const validationStatus = readNonEmptyString(validationSummary?.status);
+        const patchStatus = readNonEmptyString(patchEquivalence?.status)
+            || (patchEquivalence?.equivalent === true ? 'passed' : '');
+        const into = readNonEmptyString(result?.into);
+        const branch = readNonEmptyString(result?.branch);
+        const mergeStatus = result?.merged === true ? 'merged' : readNonEmptyString(finalConvergence?.status);
+        const convergenceStatus = readNonEmptyString(finalConvergence?.status);
+        const nextStep = readNonEmptyString(result?.nextStep)
+            || readNonEmptyString(finalConvergence?.nextStep)
+            || 'Continue from the updated mesh state.';
+        const details = [
+            jobId ? `job_id=${jobId}` : '',
+            branch && into ? `${branch}→${into}` : '',
+            validationStatus ? `validation=${validationStatus}` : '',
+            patchStatus ? `patch_equivalence=${patchStatus}` : '',
+            mergeStatus ? `merge=${mergeStatus}` : '',
+            convergenceStatus ? `final_convergence=${convergenceStatus}` : '',
+        ].filter(Boolean).join('; ');
+        return `[System] Refinery async job for ${args.nodeLabel} completed successfully${details ? ` (${details})` : ''}.\nNext step: ${nextStep}`;
+    }
+    if (args.event === 'refine:failed') {
+        const jobId = readRefineJobId({ metadataEvent: args.metadataEvent });
+        const result = readRecord(args.metadataEvent.result);
+        const validationSummary = readRecord(result?.validationSummary);
+        const patchEquivalence = readRecord(result?.patchEquivalence);
+        const finalConvergence = readRecord(result?.finalBranchConvergenceState);
+        const code = readNonEmptyString(result?.code);
+        const error = readNonEmptyString(result?.error);
+        const validationStatus = readNonEmptyString(validationSummary?.status);
+        const patchStatus = readNonEmptyString(patchEquivalence?.status)
+            || (patchEquivalence?.equivalent === true ? 'passed' : '');
+        const mergeStatus = result?.merged === true
+            ? 'merged'
+            : finalConvergence?.merged === false
+                ? 'not_merged'
+                : '';
+        const convergenceStatus = readNonEmptyString(result?.convergenceStatus)
+            || readNonEmptyString(finalConvergence?.status);
+        const blockedReason = readNonEmptyString(result?.blockedReason);
+        const nextStep = readNonEmptyString(result?.nextStep) || readNonEmptyString(finalConvergence?.nextStep);
+        const details = [
+            jobId ? `job_id=${jobId}` : '',
+            code ? `code=${code}` : '',
+            validationStatus ? `validation=${validationStatus}` : '',
+            patchStatus ? `patch_equivalence=${patchStatus}` : '',
+            mergeStatus ? `merge=${mergeStatus}` : '',
+            convergenceStatus ? `convergence=${convergenceStatus}` : '',
+            blockedReason ? `reason=${blockedReason}` : '',
+        ].filter(Boolean).join('; ');
+        const parts = [
+            `[System] Refinery async job for ${args.nodeLabel} failed${details ? ` (${details})` : ''}${error ? `: ${error}` : '.'}`,
+            nextStep ? `Next step: ${nextStep}` : 'Review the terminal refine event/ledger before retrying.',
+        ];
+        return parts.join('\n');
     }
     return '';
 }
@@ -588,6 +1256,15 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
 }) {
     const eventSessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
     const eventNodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId);
+
+    // Resolve coordinator ownership early — used in fingerprinting and coordinator routing.
+    const sourceSession = args.sourceInstanceId
+        ? components.instanceManager.getInstance(args.sourceInstanceId)
+        : undefined;
+    const workerCoordinatorDaemonId = readNonEmptyString(
+        (sourceSession?.getState()?.settings as Record<string, unknown>)?.meshCoordinatorDaemonId,
+    );
+    const localDaemonId = readNonEmptyString(loadConfig().machineId);
     const intentionalCleanupStop = shouldSuppressIntentionalCleanupStop({
         event: args.event,
         meshId: args.meshId,
@@ -603,6 +1280,97 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         return { success: true, forwarded: 0, suppressed: true, intentionalCleanupStop: true };
     }
 
+    if (args.event === 'monitor:long_generating') {
+        const reconciledCompletion = buildLongGeneratingCompletionReconciliation({
+            meshId: args.meshId,
+            nodeId: args.nodeId,
+            nodeLabel: args.nodeLabel,
+            metadataEvent: args.metadataEvent,
+            sourceInstanceId: args.sourceInstanceId,
+        });
+        if (reconciledCompletion?.source === 'long_generating_reconciliation') {
+            LOG.info('MeshEvents', `Reconciled long-generating monitor to completion for session ${eventSessionId || '(unknown session)'}`);
+            return injectMeshSystemMessage(components, {
+                ...args,
+                event: 'agent:generating_completed',
+                metadataEvent: reconciledCompletion,
+            });
+        }
+        if (reconciledCompletion?.source === 'long_generating_terminal_ledger_suppression') {
+            LOG.info('MeshEvents', `Suppressed long-generating monitor because terminal ledger evidence already exists for session ${eventSessionId || '(unknown session)'}`);
+            return {
+                success: true,
+                forwarded: 0,
+                suppressed: true,
+                terminalLedgerEvidence: true,
+                terminalLedgerKind: reconciledCompletion.terminalLedgerKind,
+            };
+        }
+    }
+
+    if (isDuplicateRefineTerminalEvent(args.meshId, args.event, args.metadataEvent)) {
+        LOG.info('MeshEvents', `Suppressed duplicate ${args.event} for refine job ${readRefineJobId({ metadataEvent: args.metadataEvent })}`);
+        return { success: true, forwarded: 0, suppressed: true, duplicateRefineTerminalEvent: true };
+    }
+
+    const eventTimestamp = readEventTimestamp(args.metadataEvent.timestamp);
+    if (args.event === 'agent:waiting_approval' && eventSessionId) {
+        const duplicateApproval = isDuplicateMeshApprovalEvent({
+            meshId: args.meshId,
+            sessionId: eventSessionId,
+            providerType: readNonEmptyString(args.metadataEvent.providerType) || undefined,
+            timestamp: eventTimestamp,
+            modalMessage: readNonEmptyString(args.metadataEvent.modalMessage) || undefined,
+            modalButtons: args.metadataEvent.modalButtons,
+        });
+        if (duplicateApproval) {
+            LOG.info('MeshEvents', `Suppressed duplicate approval event for mesh ${args.meshId} session ${eventSessionId}`);
+            return { success: true, forwarded: 0, suppressed: true, duplicateApproval: true };
+        }
+    }
+    if (args.event === 'agent:generating_completed' && eventSessionId) {
+        const terminal = findRecentTerminalLedgerEvidence({
+            meshId: args.meshId,
+            sessionId: eventSessionId,
+            nodeId: eventNodeId || undefined,
+        });
+        if (terminal?.kind === 'task_completed' && !sessionHasActiveAssignment(args.meshId, eventSessionId)) {
+            // If a new task_dispatched was recorded for this session after the prior terminal,
+            // this completion belongs to the new task — never suppress it as a duplicate.
+            const newDispatchAfterTerminal = hasDispatchAfterTerminal(args.meshId, eventSessionId, terminal.id);
+            if (!newDispatchAfterTerminal) {
+                const terminalProviderSessionId = readNonEmptyString(terminal.payload.providerSessionId);
+                const terminalFinalSummary = readNonEmptyString(terminal.payload.finalSummary);
+                const eventProviderSessionId = readNonEmptyString(args.metadataEvent.providerSessionId);
+                const eventFinalSummary = readNonEmptyString(args.metadataEvent.finalSummary);
+                if (
+                    (terminalProviderSessionId && terminalProviderSessionId === eventProviderSessionId)
+                    || (terminalFinalSummary && terminalFinalSummary === eventFinalSummary)
+                    || args.metadataEvent.source === 'long_generating_reconciliation'
+                ) {
+                    LOG.info('MeshEvents', `Suppressed duplicate completion with existing terminal ledger evidence for mesh ${args.meshId} session ${eventSessionId}`);
+                    return { success: true, forwarded: 0, suppressed: true, duplicateCompletion: true, terminalLedgerEvidence: true };
+                }
+            }
+        }
+        const duplicateCompletion = isDuplicateMeshCompletionEvent({
+            meshId: args.meshId,
+            event: args.event,
+            sessionId: eventSessionId,
+            providerType: readNonEmptyString(args.metadataEvent.providerType) || undefined,
+            providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
+            timestamp: eventTimestamp,
+            finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
+            // Scope dedup to the coordinator daemon so two coordinators for the same mesh
+            // don't suppress each other's completion events via shared fingerprint table.
+            coordinatorDaemonId: workerCoordinatorDaemonId || undefined,
+        });
+        if (duplicateCompletion) {
+            LOG.info('MeshEvents', `Suppressed duplicate completion for mesh ${args.meshId} session ${eventSessionId}`);
+            return { success: true, forwarded: 0, suppressed: true, duplicateCompletion: true };
+        }
+    }
+
     // ── Task Queue & Ledger ──
     let completedTaskForLedger: { id?: string } | null = null;
     if (args.event === 'agent:generating_completed') {
@@ -611,24 +1379,31 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         const providerType = readNonEmptyString(args.metadataEvent.providerType);
         
         if (sessionId) {
-            const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed');
+            const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed', {
+                occurredAt: eventTimestamp !== null ? new Date(eventTimestamp).toISOString() : undefined,
+            });
             completedTaskForLedger = completedTask ? { id: completedTask.id } : null;
+            updateDirectDispatchStatus(args.meshId, sessionId, 'completed');
+            setImmediate(() => cleanupTerminalDirectDispatches());
             if (nodeId && providerType) {
-                // Short delay to allow completion event to propagate before pulling next
-                setTimeout(() => {
-                    tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
-                }, 500);
+                runIdleMaintenanceThenAssignQueue(components, { meshId: args.meshId, nodeId, sessionId, providerType });
             }
         }
     } else if (args.event === 'agent:ready') {
         const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
         const nodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId);
         const providerType = readNonEmptyString(args.metadataEvent.providerType);
-        const completedTask = sessionId
+        const providerSessionId = readNonEmptyString(args.metadataEvent.providerSessionId) || undefined;
+        const finalSummary = readNonEmptyString(args.metadataEvent.finalSummary) || undefined;
+        const workerResult = readWorkerResultMetadata(args.metadataEvent);
+        const hasCompletionEvidence = !!finalSummary || !!workerResult;
+        const completedTask = sessionId && hasCompletionEvidence
             ? updateSessionTaskStatus(args.meshId, sessionId, 'completed')
             : null;
-        if (completedTask) {
+        if (completedTask && sessionId) {
             completedTaskForLedger = { id: completedTask.id };
+            updateDirectDispatchStatus(args.meshId, sessionId, 'completed');
+            setImmediate(() => cleanupTerminalDirectDispatches());
             try {
                 appendLedgerEntry(args.meshId, {
                     kind: 'task_completed',
@@ -640,15 +1415,17 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
                         nodeLabel: args.nodeLabel,
                         taskId: completedTask.id,
                         completedViaReady: true,
-                        providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
-                        finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
+                        providerSessionId,
+                        finalSummary,
+                        workerResult,
                         evidence: buildTaskCompletionEvidence({
                             event: 'agent:ready',
                             nodeId,
                             sessionId,
                             providerType: providerType || undefined,
-                            providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
-                            finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
+                            providerSessionId,
+                            finalSummary,
+                            workerResult,
                         }),
                     },
                 });
@@ -658,13 +1435,22 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         }
         
         if (sessionId && nodeId && providerType) {
-            remoteIdleSessions.set(`${nodeId}:${sessionId}`, { nodeId, sessionId, providerType });
-            setTimeout(() => {
-                const assigned = tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
-                if (assigned) {
-                    remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
-                }
-            }, 500);
+            sweepExpiredRemoteIdleSessions();
+            remoteIdleSessions.set(`${nodeId}:${sessionId}`, {
+                nodeId, sessionId, providerType,
+                expiresAt: Date.now() + REMOTE_IDLE_SESSION_TTL_MS,
+            });
+            setImmediate(() => {
+                maybeAutoFastForwardIdleNode(components, { meshId: args.meshId, nodeId, sessionId, providerType })
+                    .finally(() => {
+                        try {
+                            const assigned = tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
+                            if (assigned) remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
+                        } catch (e: any) {
+                            LOG.warn('MeshQueue', `Failed to assign idle queue task after maintenance for ${nodeId}: ${e?.message || e}`);
+                        }
+                    });
+            });
         }
     } else if (args.event === 'agent:generating_started') {
         const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
@@ -672,6 +1458,10 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         if (sessionId && nodeId) {
             remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
         }
+        if (sessionId) {
+            // Mark direct dispatch as acknowledged — the session started generating.
+            updateDirectDispatchStatus(args.meshId, sessionId, 'acked');
+        }
     } else if (args.event === 'agent:stopped') {
         const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
         const nodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId);
@@ -679,7 +1469,9 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
             remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
         }
         if (sessionId) {
-            updateSessionTaskStatus(args.meshId, sessionId, 'failed');
+            const failedTask = updateSessionTaskStatus(args.meshId, sessionId, 'failed');
+            completedTaskForLedger = failedTask ? { id: failedTask.id } : null;
+            updateDirectDispatchStatus(args.meshId, sessionId, 'failed');
         }
     }
 
@@ -689,14 +1481,18 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
             const ledgerNodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId) || undefined;
             const ledgerSessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId) || undefined;
             const ledgerProviderType = readNonEmptyString(args.metadataEvent.providerType) || undefined;
+            const providerSessionId = readNonEmptyString(args.metadataEvent.providerSessionId) || undefined;
+            const finalSummary = readNonEmptyString(args.metadataEvent.finalSummary) || undefined;
+            const workerResult = readWorkerResultMetadata(args.metadataEvent);
             const completionEvidence = ledgerKind === 'task_completed' && ledgerNodeId && ledgerSessionId
                 ? buildTaskCompletionEvidence({
                     event: 'agent:generating_completed',
                     nodeId: ledgerNodeId,
                     sessionId: ledgerSessionId,
                     providerType: ledgerProviderType,
-                    providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
-                    finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
+                    providerSessionId,
+                    finalSummary,
+                    workerResult,
                 })
                 : undefined;
             appendLedgerEntry(args.meshId, {
@@ -708,8 +1504,12 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
                     event: args.event,
                     nodeLabel: args.nodeLabel,
                     taskId: completedTaskForLedger?.id || undefined,
-                    providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
-                    finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
+                    providerSessionId,
+                    finalSummary,
+                    workerResult,
+                    completionDiagnostic: args.metadataEvent.completionDiagnostic && typeof args.metadataEvent.completionDiagnostic === 'object'
+                        ? args.metadataEvent.completionDiagnostic
+                        : undefined,
                     evidence: completionEvidence,
                 },
             });
@@ -782,15 +1582,33 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         }
     }
 
+    const messageText = buildMeshSystemMessage({
+        event: args.event,
+        nodeLabel: args.nodeLabel,
+        metadataEvent: args.metadataEvent,
+        recoveryContext,
+    });
+    if (!messageText) return { success: false, error: 'unsupported mesh event' };
+
     const coordinatorInstances = components.instanceManager.getByCategory('cli').filter((inst) => {
         const instState = inst.getState();
         if (instState.settings?.meshCoordinatorFor !== args.meshId) return false;
         if (args.sourceInstanceId && instState.instanceId === args.sourceInstanceId) return false;
+        // If the worker knows which coordinator daemon launched it, only route to coordinators
+        // on that specific daemon. This prevents cross-contamination when multiple coordinator
+        // sessions run simultaneously for the same mesh on different daemons.
+        if (workerCoordinatorDaemonId && localDaemonId && workerCoordinatorDaemonId !== localDaemonId) return false;
         return true;
     });
 
+    // Refine terminal events (refine:completed, refine:failed) are coordinator-delivered
+    // synchronously; only buffer them for MCP when no CLI coordinator is present.
+    // Agent runtime events (agent:*) use dual delivery so both CLI and MCP coordinators
+    // receive them regardless of whether a live CLI coordinator session is active.
+    const isRefineTerminalEvent = REFINE_TERMINAL_EVENTS.has(args.event);
+
     if (coordinatorInstances.length === 0) {
-        // No CLI coordinator session found — buffer for MCP-based coordinators.
+        // No local CLI coordinator — buffer for MCP-based coordinator on the target daemon.
         if (queuePendingMeshCoordinatorEvent({
                 event: args.event,
                 meshId: args.meshId,
@@ -801,20 +1619,69 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
                     ...args.metadataEvent,
                     ...(recoveryContext ? { recoveryContext } : {}),
                 },
+                coordinatorMessage: messageText,
                 queuedAt: Date.now(),
+                // Scope to the coordinator daemon that launched this worker so drain
+                // by other coordinators on the same daemon doesn't consume this event.
+                ...(workerCoordinatorDaemonId ? { targetCoordinatorDaemonId: workerCoordinatorDaemonId } : {}),
             })) {
-            LOG.info('MeshEvents', `Queued ${args.event} for MCP coordinator (mesh ${args.meshId})`);
+            LOG.info('MeshEvents', `Queued ${args.event} for MCP coordinator (mesh ${args.meshId}${workerCoordinatorDaemonId ? `, coordinator daemon ${workerCoordinatorDaemonId}` : ''})`);
         }
         return { success: true, forwarded: 0 };
     }
 
-    const messageText = buildMeshSystemMessage({
-        event: args.event,
-        nodeLabel: args.nodeLabel,
-        metadataEvent: args.metadataEvent,
-        recoveryContext,
+    // CLI coordinator is present. For non-refine events, also buffer for MCP coordinators
+    // that poll via get_pending_mesh_events (dual delivery). Refine terminal events are
+    // forwarded directly only — they must not accumulate in the pending queue when a live
+    // coordinator already received them.
+    //
+    // Exception: if ALL live CLI coordinator instances are in a generating/active state
+    // when a refine terminal event fires (e.g. a Codex CLI coordinator that triggered an
+    // async refine job and is still in the generating turn that sent it), the coordinator
+    // cannot immediately receive send_message input. Buffer the event to the pending queue
+    // so it is available via get_pending_mesh_events when the coordinator returns to idle.
+    // Critically: do NOT attempt send_message injection into a generating PTY coordinator for
+    // terminal refine events — injecting text into an active PTY can corrupt the input stream
+    // and leave the coordinator stuck in generating state, unable to process the refine result.
+    const allCoordinatorsGenerating = isRefineTerminalEvent && coordinatorInstances.every((inst) => {
+        const s = inst.getState();
+        const status = readNonEmptyString(s.status).toLowerCase();
+        const activeChatStatus = readNonEmptyString(s.activeChat?.status).toLowerCase();
+        return status === 'generating' || status === 'streaming' || status === 'long_generating'
+            || activeChatStatus === 'generating' || activeChatStatus === 'streaming';
     });
-    if (!messageText) return { success: false, error: 'unsupported mesh event' };
+
+    if (!isRefineTerminalEvent || allCoordinatorsGenerating) {
+        if (queuePendingMeshCoordinatorEvent({
+                event: args.event,
+                meshId: args.meshId,
+                nodeLabel: args.nodeLabel,
+                nodeId: args.nodeId || undefined,
+                workspace: readNonEmptyString(args.metadataEvent.workspace),
+                metadataEvent: {
+                    ...args.metadataEvent,
+                    ...(recoveryContext ? { recoveryContext } : {}),
+                },
+                coordinatorMessage: messageText,
+                queuedAt: Date.now(),
+                ...(workerCoordinatorDaemonId ? { targetCoordinatorDaemonId: workerCoordinatorDaemonId } : {}),
+            })) {
+            if (allCoordinatorsGenerating) {
+                LOG.info('MeshEvents', `Queued ${args.event} for generating CLI coordinator (mesh ${args.meshId}) — will be delivered via get_pending_mesh_events when coordinator returns to idle`);
+            } else {
+                LOG.info('MeshEvents', `Queued ${args.event} for MCP coordinator (mesh ${args.meshId})`);
+            }
+        }
+    }
+
+    // When all CLI coordinators are actively generating and a terminal refine event fires,
+    // skip send_message injection entirely. The event is already buffered to the pending queue.
+    // Injecting into a generating PTY coordinator can corrupt its input stream and cause it to
+    // remain stuck in generating state, never processing the refine result.
+    // The coordinator will drain pending events via get_pending_mesh_events on its next idle cycle.
+    if (allCoordinatorsGenerating) {
+        return { success: true, forwarded: 0, bufferedForGeneratingCoordinator: true };
+    }
 
     for (const coord of coordinatorInstances) {
         const coordState = coord.getState();
@@ -845,6 +1712,19 @@ export function handleMeshForwardEvent(components: DaemonComponents, payload: Re
             providerType: readNonEmptyString(payload.providerType),
             providerSessionId: readNonEmptyString(payload.providerSessionId),
             finalSummary: readNonEmptyString(payload.finalSummary) || readNonEmptyString(payload.summary),
+            jobId: readNonEmptyString(payload.jobId),
+            interactionId: readNonEmptyString(payload.interactionId),
+            status: readNonEmptyString(payload.status),
+            targetDaemonId: readNonEmptyString(payload.targetDaemonId),
+            startedAt: readNonEmptyString(payload.startedAt),
+            completedAt: readNonEmptyString(payload.completedAt),
+            retryOfJobId: readNonEmptyString(payload.retryOfJobId),
+            ...(payload.result && typeof payload.result === 'object' && !Array.isArray(payload.result) ? { result: payload.result } : {}),
+            ...(payload.completionDiagnostic && typeof payload.completionDiagnostic === 'object' && !Array.isArray(payload.completionDiagnostic) ? { completionDiagnostic: payload.completionDiagnostic } : {}),
+            ...(payload.workerResult && typeof payload.workerResult === 'object' && !Array.isArray(payload.workerResult) ? { workerResult: payload.workerResult } : {}),
+            ...(payload.meshWorkerResult && typeof payload.meshWorkerResult === 'object' && !Array.isArray(payload.meshWorkerResult) ? { meshWorkerResult: payload.meshWorkerResult } : {}),
+            ...(payload.structuredResult && typeof payload.structuredResult === 'object' && !Array.isArray(payload.structuredResult) ? { structuredResult: payload.structuredResult } : {}),
+            ...(payload.timestamp !== undefined ? { timestamp: payload.timestamp } : {}),
             intentional: payload.intentional === true,
             intentionalStop: payload.intentionalStop === true,
             operatorCleanup: payload.operatorCleanup === true,
@@ -872,21 +1752,56 @@ export function setupMeshEventForwarding(components: DaemonComponents) {
         if (!workspace) return;
         const settings = state.settings && typeof state.settings === 'object' ? state.settings as Record<string, unknown> : {};
 
-        // Coordinator sessions must never inject events into themselves.
-        // A coordinator instance carries meshCoordinatorFor but not meshNodeFor/launchedByCoordinator.
-        if (readNonEmptyString(settings.meshCoordinatorFor)) return;
+        // Coordinator sessions normally must not inject events into themselves, but a
+        // coordinator session can also be the direct-dispatch target of mesh_send_task
+        // issued by another coordinator (e.g. a remote orchestrator delegated a task to
+        // this local coordinator). In that case the completion event must still flow into
+        // injectMeshSystemMessage so the ledger records task_completed and the *other*
+        // coordinator's pendingCoordinatorEvents queue is populated. Skip only when this
+        // session is purely a coordinator with no in-flight direct dispatch waiting on it.
+        const coordinatorMeshId = readNonEmptyString(settings.meshCoordinatorFor);
+        let meshIdFromDirectDispatch = '';
+        if (coordinatorMeshId) {
+            try {
+                const activeDispatches = getActiveDirectDispatches(coordinatorMeshId);
+                if (activeDispatches.some(d => d.sessionId === instanceId) || hasUnterminalDirectDispatchLedgerEntry(coordinatorMeshId, instanceId)) {
+                    meshIdFromDirectDispatch = coordinatorMeshId;
+                }
+            } catch { /* best-effort */ }
+            if (!meshIdFromDirectDispatch) return;
+        } else {
+            // Plain session (no meshNodeFor / meshCoordinatorFor / launchedByCoordinator)
+            // that received a direct dispatch via mesh_send_task: the dispatcher logs
+            // the dispatch against the workspace-resolved mesh but never stamps the
+            // target session's settings. Without this fallback, the worker's
+            // generating_completed event silently drops here and task_completed
+            // never lands in the ledger — coordinator sees the task as still in
+            // flight forever. Resolve the mesh via workspace and accept the event
+            // when an active direct dispatch points at this session.
+            const workspaceMesh = getCachedMeshByWorkspace(workspace);
+            const workspaceMeshId = readNonEmptyString(workspaceMesh?.id);
+            if (workspaceMeshId) {
+                try {
+                    const activeDispatches = getActiveDirectDispatches(workspaceMeshId);
+                    if (activeDispatches.some(d => d.sessionId === instanceId) || hasUnterminalDirectDispatchLedgerEntry(workspaceMeshId, instanceId)) {
+                        meshIdFromDirectDispatch = workspaceMeshId;
+                    }
+                } catch { /* best-effort */ }
+            }
+        }
 
-        const meshIdFromRuntime = readNonEmptyString(settings.meshNodeFor);
+        const meshIdFromRuntime = readNonEmptyString(settings.meshNodeFor) || meshIdFromDirectDispatch;
 
         // Only forward events for sessions that were explicitly launched as mesh-node delegates
-        // (meshNodeFor set by mesh_launch_session) or that carry the launchedByCoordinator flag.
+        // (meshNodeFor set by mesh_launch_session), carry the launchedByCoordinator flag, or
+        // have an active direct-dispatch entry (mesh_send_task to a pre-existing session).
         // Do NOT fall back to workspace-based mesh lookup: that would pick up coordinator sessions
         // and any other CLI session that happens to share the same workspace, causing spurious
         // system-message injection into the coordinator's own conversation.
         const isMeshDelegate = Boolean(meshIdFromRuntime || settings.launchedByCoordinator);
         if (!isMeshDelegate) return;
 
-        const mesh = meshIdFromRuntime ? getMeshWithCache(components, meshIdFromRuntime) : getMeshByRepo(workspace);
+        const mesh = meshIdFromRuntime ? getMeshWithCache(components, meshIdFromRuntime) : getCachedMeshByWorkspace(workspace);
         const meshId = meshIdFromRuntime || readNonEmptyString(mesh?.id);
         if (!meshId) return;