@adhdev/daemon-core 0.9.82-rc.23 → 0.9.82-rc.24

package/dist/mesh/mesh-events.d.ts

@@ -9,12 +9,12 @@ export interface PendingMeshCoordinatorEvent {
     queuedAt: number;
 }
 export declare function queuePendingMeshCoordinatorEvent(event: PendingMeshCoordinatorEvent): boolean;
-/** Drain and return all pending coordinator events, clearing the queue. */
-export declare function drainPendingMeshCoordinatorEvents(): PendingMeshCoordinatorEvent[];
+/** Drain and return all pending coordinator events for meshId, removing them from disk. */
+export declare function drainPendingMeshCoordinatorEvents(meshId?: string): PendingMeshCoordinatorEvent[];
 /** Peek at pending coordinator events without draining (non-destructive). */
-export declare function getPendingMeshCoordinatorEvents(): readonly PendingMeshCoordinatorEvent[];
-/** Explicitly clear all pending coordinator events. */
-export declare function clearPendingMeshCoordinatorEvents(): void;
+export declare function getPendingMeshCoordinatorEvents(meshId?: string): readonly PendingMeshCoordinatorEvent[];
+/** Explicitly clear all pending coordinator events for a mesh. */
+export declare function clearPendingMeshCoordinatorEvents(meshId?: string): void;
 export declare function tryAssignQueueTask(components: DaemonComponents, meshId: string, nodeId: string, sessionId: string, providerType: string): boolean;
 /**
  * Triggers a queue check for all nodes in the mesh.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@adhdev/daemon-core",
-  "version": "0.9.82-rc.23",
+  "version": "0.9.82-rc.24",
   "description": "ADHDev daemon core — CDP, IDE detection, providers, command execution",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

package/src/commands/router.ts

@@ -964,6 +964,8 @@ export interface CommandRouterDeps {
     sessionHostControl?: SessionHostControlPlane | null;
     /** Selected-coordinator mesh peer telemetry surface for target daemons, when supported by the runtime. */
     getMeshPeerConnectionStatus?: (daemonId: string) => Record<string, unknown> | null;
+    /** Dispatch a command to a remote mesh node via P2P/relay. Injected by cloud runtime; absent in standalone. */
+    dispatchMeshCommand?: (daemonId: string, cmd: string, args: Record<string, unknown>) => Promise<unknown>;
 }
 
 export interface CommandRouterResult {
@@ -1688,7 +1690,8 @@ export class DaemonCommandRouter {
             }
 
             case 'get_pending_mesh_events': {
-                const events = drainPendingMeshCoordinatorEvents();
+                const meshId = typeof args?.meshId === 'string' ? args.meshId.trim() : '';
+                const events = drainPendingMeshCoordinatorEvents(meshId || undefined);
                 return { success: true, events };
             }
 
@@ -3344,29 +3347,52 @@ export class DaemonCommandRouter {
                         }
                         if (workspace) {
                             if (!fs.existsSync(workspace)) {
-                                if (applyCachedInlineMeshNodeStatus(status, node)) {
-                                    status.launchReady = !!daemonId && (readStringValue(status.machineStatus) === 'online' || isSelfNode);
-                                    nodeStatuses.push(status);
-                                    continue;
-                                }
-                                if (meshRecord?.source === 'inline_cache' && !isSelfNode) {
-                                    status.launchReady = !!daemonId && (readStringValue(status.machineStatus) === 'online' || isSelfNode);
-                                    nodeStatuses.push(status);
-                                    continue;
+                                // Workspace not local — attempt a P2P git probe for remote nodes.
+                                let remoteProbeApplied = false;
+                                if (!isSelfNode && daemonId && this.deps.dispatchMeshCommand) {
+                                    try {
+                                        const remoteResult = await Promise.race([
+                                            this.deps.dispatchMeshCommand(daemonId, 'git_status', { workspace }),
+                                            new Promise<never>((_, reject) => setTimeout(() => reject(new Error('timeout')), 8000)),
+                                        ]) as any;
+                                        const remoteGit = remoteResult?.status ?? remoteResult?.git ?? remoteResult;
+                                        if (remoteGit && typeof remoteGit === 'object' && typeof remoteGit.isGitRepo === 'boolean') {
+                                            status.git = remoteGit;
+                                            status.health = remoteGit.isGitRepo
+                                                ? deriveMeshNodeHealthFromGit(remoteGit as unknown as Record<string, unknown>)
+                                                : 'degraded';
+                                            remoteProbeApplied = true;
+                                        }
+                                    } catch {
+                                        // Probe timed out or P2P unavailable — fall back to cached status
+                                    }
                                 }
-                            }
-                            try {
-                                const gitStatus = await getGitRepoStatus(workspace, { timeoutMs: 10_000, refreshUpstream: true });
-                                status.git = gitStatus;
-                                if (gitStatus.isGitRepo) {
-                                    status.health = deriveMeshNodeHealthFromGit(gitStatus as unknown as Record<string, unknown>);
-                                } else {
-                                    status.health = 'degraded';
-                                    if (gitStatus.error && !status.error) status.error = gitStatus.error;
+                                if (!remoteProbeApplied) {
+                                    if (applyCachedInlineMeshNodeStatus(status, node)) {
+                                        status.launchReady = !!daemonId && (readStringValue(status.machineStatus) === 'online' || isSelfNode);
+                                        nodeStatuses.push(status);
+                                        continue;
+                                    }
+                                    if (meshRecord?.source === 'inline_cache' && !isSelfNode) {
+                                        status.launchReady = !!daemonId && (readStringValue(status.machineStatus) === 'online' || isSelfNode);
+                                        nodeStatuses.push(status);
+                                        continue;
+                                    }
                                 }
-                            } catch {
-                                if (!applyCachedInlineMeshNodeStatus(status, node)) {
-                                    status.health = 'degraded';
+                            } else {
+                                try {
+                                    const gitStatus = await getGitRepoStatus(workspace, { timeoutMs: 10_000, refreshUpstream: true });
+                                    status.git = gitStatus;
+                                    if (gitStatus.isGitRepo) {
+                                        status.health = deriveMeshNodeHealthFromGit(gitStatus as unknown as Record<string, unknown>);
+                                    } else {
+                                        status.health = 'degraded';
+                                        if (gitStatus.error && !status.error) status.error = gitStatus.error;
+                                    }
+                                } catch {
+                                    if (!applyCachedInlineMeshNodeStatus(status, node)) {
+                                        status.health = 'degraded';
+                                    }
                                 }
                             }
                         } else {

package/src/mesh/mesh-events.ts

@@ -1,31 +1,47 @@
+import { appendFileSync, existsSync, readFileSync, unlinkSync } 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';
 
 // ---------------------------------------------------------------------------
 // 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}`
 
+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
+// 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 {
@@ -38,30 +54,53 @@ export interface PendingMeshCoordinatorEvent {
     queuedAt: number;
 }
 
-const MAX_PENDING_EVENTS = 50;
-const pendingMeshCoordinatorEvents: PendingMeshCoordinatorEvent[] = [];
+function getPendingEventsPath(meshId: string): string {
+    const safe = meshId.replace(/[^a-zA-Z0-9_-]/g, '_');
+    return join(getLedgerDir(), `${safe}.pending-events.jsonl`);
+}
 
 export function queuePendingMeshCoordinatorEvent(event: PendingMeshCoordinatorEvent): boolean {
-    if (pendingMeshCoordinatorEvents.length >= MAX_PENDING_EVENTS) {
+    try {
+        appendFileSync(getPendingEventsPath(event.meshId), 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);
+/** Drain and return all pending coordinator events for meshId, removing them from disk. */
+export function drainPendingMeshCoordinatorEvents(meshId?: string): PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    const path = getPendingEventsPath(meshId);
+    if (!existsSync(path)) return [];
+    try {
+        const raw = readFileSync(path, 'utf-8');
+        try { unlinkSync(path); } catch { /* concurrent drain already removed it */ }
+        return raw.split('\n').filter(Boolean).flatMap(line => {
+            try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
+        });
+    } catch { return []; }
 }
 
 /** Peek at pending coordinator events without draining (non-destructive). */
-export function getPendingMeshCoordinatorEvents(): readonly PendingMeshCoordinatorEvent[] {
-    return pendingMeshCoordinatorEvents.slice();
+export function getPendingMeshCoordinatorEvents(meshId?: string): readonly PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    const path = getPendingEventsPath(meshId);
+    if (!existsSync(path)) return [];
+    try {
+        const raw = readFileSync(path, 'utf-8');
+        return raw.split('\n').filter(Boolean).flatMap(line => {
+            try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
+        });
+    } catch { return []; }
 }
 
-/** Explicitly clear all pending coordinator events. */
-export function clearPendingMeshCoordinatorEvents(): void {
-    pendingMeshCoordinatorEvents.splice(0);
+/** Explicitly clear all pending coordinator events for a mesh. */
+export function clearPendingMeshCoordinatorEvents(meshId?: string): void {
+    if (!meshId) return;
+    const path = getPendingEventsPath(meshId);
+    if (existsSync(path)) try { unlinkSync(path); } catch { /* already removed */ }
 }
 
 function readNonEmptyString(value: unknown): string {
@@ -180,7 +219,16 @@ 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;
         }
@@ -614,10 +662,11 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
             const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed');
             completedTaskForLedger = completedTask ? { id: completedTask.id } : null;
             if (nodeId && providerType) {
-                // Short delay to allow completion event to propagate before pulling next
-                setTimeout(() => {
+                // Queue state is already updated above; setImmediate avoids the
+                // 500 ms artificial delay while still deferring past this call frame.
+                setImmediate(() => {
                     tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
-                }, 500);
+                });
             }
         }
     } else if (args.event === 'agent:ready') {
@@ -658,13 +707,15 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         }
         
         if (sessionId && nodeId && providerType) {
-            remoteIdleSessions.set(`${nodeId}:${sessionId}`, { nodeId, sessionId, providerType });
-            setTimeout(() => {
+            sweepExpiredRemoteIdleSessions();
+            remoteIdleSessions.set(`${nodeId}:${sessionId}`, {
+                nodeId, sessionId, providerType,
+                expiresAt: Date.now() + REMOTE_IDLE_SESSION_TTL_MS,
+            });
+            setImmediate(() => {
                 const assigned = tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
-                if (assigned) {
-                    remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
-                }
-            }, 500);
+                if (assigned) remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
+            });
         }
     } else if (args.event === 'agent:generating_started') {
         const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);

package/src/mesh/mesh-work-queue.ts

@@ -1,4 +1,4 @@
-import { existsSync, writeFileSync, readFileSync } from 'fs';
+import { existsSync, writeFileSync, readFileSync, openSync, closeSync, unlinkSync } from 'fs';
 import { join } from 'path';
 import { randomUUID } from 'crypto';
 import { getLedgerDir } from './mesh-ledger.js';
@@ -50,6 +50,31 @@ function getQueuePath(meshId: string): string {
     return join(getLedgerDir(), `${safe}.queue.json`);
 }
 
+function getLockPath(meshId: string): string {
+    const safe = meshId.replace(/[^a-zA-Z0-9_-]/g, '_');
+    return join(getLedgerDir(), `${safe}.queue.lock`);
+}
+
+/**
+ * Simple advisory file lock using O_EXCL (atomic create) for queue mutations.
+ * Retries up to 10 times at 30 ms intervals; proceeds without lock on timeout
+ * to prevent deadlock (best-effort — far better than no locking at all).
+ */
+function withQueueLock<T>(meshId: string, fn: () => T): T {
+    const lockPath = getLockPath(meshId);
+    let fd = -1;
+    for (let i = 0; i < 10; i++) {
+        try { fd = openSync(lockPath, 'wx'); break; } catch {
+            const deadline = Date.now() + 30;
+            while (Date.now() < deadline) { /* spin */ }
+        }
+    }
+    try { return fn(); } finally {
+        if (fd !== -1) try { closeSync(fd); } catch { /* noop */ }
+        try { unlinkSync(lockPath); } catch { /* already removed */ }
+    }
+}
+
 function readQueue(meshId: string): MeshWorkQueueEntry[] {
     const path = getQueuePath(meshId);
     if (!existsSync(path)) return [];
@@ -74,20 +99,22 @@ export function enqueueTask(
     message: string,
     opts?: { targetNodeId?: string; targetSessionId?: string }
 ): MeshWorkQueueEntry {
-    const queue = readQueue(meshId);
-    const entry: MeshWorkQueueEntry = {
-        id: randomUUID(),
-        meshId,
-        message,
-        status: 'pending',
-        targetNodeId: opts?.targetNodeId,
-        targetSessionId: opts?.targetSessionId,
-        createdAt: new Date().toISOString(),
-        updatedAt: new Date().toISOString(),
-    };
-    queue.push(entry);
-    writeQueue(meshId, queue);
-    return entry;
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        const entry: MeshWorkQueueEntry = {
+            id: randomUUID(),
+            meshId,
+            message,
+            status: 'pending',
+            targetNodeId: opts?.targetNodeId,
+            targetSessionId: opts?.targetSessionId,
+            createdAt: new Date().toISOString(),
+            updatedAt: new Date().toISOString(),
+        };
+        queue.push(entry);
+        writeQueue(meshId, queue);
+        return entry;
+    });
 }
 
 /**
@@ -106,39 +133,29 @@ export function getQueue(meshId: string, opts?: { status?: MeshTaskStatus[] }):
  * Find the next pending task that this node is allowed to claim, and mark it as assigned.
  */
 export function claimNextTask(meshId: string, nodeId: string, sessionId: string): MeshWorkQueueEntry | null {
-    const queue = readQueue(meshId);
-
-    // A worker must finish or fail its current queued assignment before it can
-    // claim another one. maxParallelTasks limits total mesh concurrency; it is
-    // not permission for one node/session to accumulate multiple assigned items.
-    const hasActiveAssignment = queue.some(q => q.status === 'assigned' && (
-        q.assignedSessionId === sessionId || q.assignedNodeId === nodeId
-    ));
-    if (hasActiveAssignment) return null;
-    
-    // Find highest priority task:
-    // 1. Pending tasks explicitly targeted at this runtime session
-    // 2. Pending tasks explicitly targeted at this node (but not another session)
-    // 3. Pending tasks with no target node/session
-    let targetIdx = queue.findIndex(q => q.status === 'pending' && q.targetSessionId === sessionId);
-    if (targetIdx === -1) {
-        targetIdx = queue.findIndex(q => q.status === 'pending' && q.targetNodeId === nodeId && !q.targetSessionId);
-    }
-    if (targetIdx === -1) {
-        targetIdx = queue.findIndex(q => q.status === 'pending' && !q.targetNodeId && !q.targetSessionId);
-    }
-
-    if (targetIdx === -1) return null;
-
-    const entry = queue[targetIdx];
-    entry.status = 'assigned';
-    entry.assignedNodeId = nodeId;
-    entry.assignedSessionId = sessionId;
-    entry.dispatchTimestamp = new Date().toISOString();
-    entry.updatedAt = new Date().toISOString();
-
-    writeQueue(meshId, queue);
-    return entry;
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        const hasActiveAssignment = queue.some(q => q.status === 'assigned' && (
+            q.assignedSessionId === sessionId || q.assignedNodeId === nodeId
+        ));
+        if (hasActiveAssignment) return null;
+        let targetIdx = queue.findIndex(q => q.status === 'pending' && q.targetSessionId === sessionId);
+        if (targetIdx === -1) {
+            targetIdx = queue.findIndex(q => q.status === 'pending' && q.targetNodeId === nodeId && !q.targetSessionId);
+        }
+        if (targetIdx === -1) {
+            targetIdx = queue.findIndex(q => q.status === 'pending' && !q.targetNodeId && !q.targetSessionId);
+        }
+        if (targetIdx === -1) return null;
+        const entry = queue[targetIdx];
+        entry.status = 'assigned';
+        entry.assignedNodeId = nodeId;
+        entry.assignedSessionId = sessionId;
+        entry.dispatchTimestamp = new Date().toISOString();
+        entry.updatedAt = new Date().toISOString();
+        writeQueue(meshId, queue);
+        return entry;
+    });
 }
 
 /**
@@ -150,14 +167,15 @@ export function updateTaskStatus(
     taskId: string,
     status: MeshTaskStatus,
 ): MeshWorkQueueEntry | null {
-    const queue = readQueue(meshId);
-    const idx = queue.findIndex(q => q.id === taskId);
-    if (idx === -1) return null;
-
-    queue[idx].status = status;
-    queue[idx].updatedAt = new Date().toISOString();
-    writeQueue(meshId, queue);
-    return queue[idx];
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        const idx = queue.findIndex(q => q.id === taskId);
+        if (idx === -1) return null;
+        queue[idx].status = status;
+        queue[idx].updatedAt = new Date().toISOString();
+        writeQueue(meshId, queue);
+        return queue[idx];
+    });
 }
 
 export function recordTaskAutoLaunch(
@@ -165,17 +183,16 @@ export function recordTaskAutoLaunch(
     taskId: string,
     autoLaunch: Omit<NonNullable<MeshWorkQueueEntry['autoLaunch']>, 'updatedAt'>,
 ): MeshWorkQueueEntry | null {
-    const queue = readQueue(meshId);
-    const idx = queue.findIndex(q => q.id === taskId);
-    if (idx === -1) return null;
-    const now = new Date().toISOString();
-    queue[idx].autoLaunch = {
-        ...autoLaunch,
-        updatedAt: now,
-    };
-    queue[idx].updatedAt = now;
-    writeQueue(meshId, queue);
-    return queue[idx];
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        const idx = queue.findIndex(q => q.id === taskId);
+        if (idx === -1) return null;
+        const now = new Date().toISOString();
+        queue[idx].autoLaunch = { ...autoLaunch, updatedAt: now };
+        queue[idx].updatedAt = now;
+        writeQueue(meshId, queue);
+        return queue[idx];
+    });
 }
 
 /**
@@ -186,17 +203,18 @@ export function cancelTask(
     taskId: string,
     opts?: { reason?: string },
 ): MeshWorkQueueEntry | null {
-    const queue = readQueue(meshId);
-    const idx = queue.findIndex(q => q.id === taskId);
-    if (idx === -1) return null;
-
-    const now = new Date().toISOString();
-    queue[idx].status = 'cancelled';
-    queue[idx].updatedAt = now;
-    queue[idx].cancelledAt = now;
-    if (opts?.reason) queue[idx].cancelReason = opts.reason;
-    writeQueue(meshId, queue);
-    return queue[idx];
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        const idx = queue.findIndex(q => q.id === taskId);
+        if (idx === -1) return null;
+        const now = new Date().toISOString();
+        queue[idx].status = 'cancelled';
+        queue[idx].updatedAt = now;
+        queue[idx].cancelledAt = now;
+        if (opts?.reason) queue[idx].cancelReason = opts.reason;
+        writeQueue(meshId, queue);
+        return queue[idx];
+    });
 }
 
 /**
@@ -214,27 +232,28 @@ export function requeueTask(
         clearTargetSession?: boolean;
     },
 ): MeshWorkQueueEntry | null {
-    const queue = readQueue(meshId);
-    const idx = queue.findIndex(q => q.id === taskId);
-    if (idx === -1) return null;
-
-    const entry = queue[idx];
-    const now = new Date().toISOString();
-    entry.status = 'pending';
-    delete entry.assignedNodeId;
-    delete entry.assignedSessionId;
-    delete entry.cancelledAt;
-    delete entry.cancelReason;
-    if (opts?.clearTargetNode) delete entry.targetNodeId;
-    if (typeof opts?.targetNodeId === 'string') entry.targetNodeId = opts.targetNodeId;
-    if (opts?.clearTargetSession !== false) delete entry.targetSessionId;
-    if (typeof opts?.targetSessionId === 'string') entry.targetSessionId = opts.targetSessionId;
-    entry.updatedAt = now;
-    entry.requeuedAt = now;
-    entry.requeueCount = (entry.requeueCount || 0) + 1;
-    if (opts?.reason) entry.requeueReason = opts.reason;
-    writeQueue(meshId, queue);
-    return entry;
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        const idx = queue.findIndex(q => q.id === taskId);
+        if (idx === -1) return null;
+        const entry = queue[idx];
+        const now = new Date().toISOString();
+        entry.status = 'pending';
+        delete entry.assignedNodeId;
+        delete entry.assignedSessionId;
+        delete entry.cancelledAt;
+        delete entry.cancelReason;
+        if (opts?.clearTargetNode) delete entry.targetNodeId;
+        if (typeof opts?.targetNodeId === 'string') entry.targetNodeId = opts.targetNodeId;
+        if (opts?.clearTargetSession !== false) delete entry.targetSessionId;
+        if (typeof opts?.targetSessionId === 'string') entry.targetSessionId = opts.targetSessionId;
+        entry.updatedAt = now;
+        entry.requeuedAt = now;
+        entry.requeueCount = (entry.requeueCount || 0) + 1;
+        if (opts?.reason) entry.requeueReason = opts.reason;
+        writeQueue(meshId, queue);
+        return entry;
+    });
 }
 
 /**
@@ -245,28 +264,22 @@ export function updateSessionTaskStatus(
     sessionId: string,
     status: MeshTaskStatus,
 ): MeshWorkQueueEntry | null {
-    const queue = readQueue(meshId);
-    // Collect all assigned tasks for this session, then pick the one with the
-    // most recent dispatchTimestamp (or updatedAt fallback for legacy entries).
-    // This prevents completing the wrong task when multiple tasks were assigned
-    // to the same session in rapid succession.
-    let bestIdx = -1;
-    let bestTime = 0;
-    for (let i = queue.length - 1; i >= 0; i--) {
-        if (queue[i].assignedSessionId === sessionId && queue[i].status === 'assigned') {
-            const time = new Date(queue[i].dispatchTimestamp || queue[i].updatedAt).getTime();
-            if (time > bestTime) {
-                bestTime = time;
-                bestIdx = i;
+    return withQueueLock(meshId, () => {
+        const queue = readQueue(meshId);
+        let bestIdx = -1;
+        let bestTime = 0;
+        for (let i = queue.length - 1; i >= 0; i--) {
+            if (queue[i].assignedSessionId === sessionId && queue[i].status === 'assigned') {
+                const time = new Date(queue[i].dispatchTimestamp || queue[i].updatedAt).getTime();
+                if (time > bestTime) { bestTime = time; bestIdx = i; }
             }
         }
-    }
-    if (bestIdx === -1) return null;
-
-    queue[bestIdx].status = status;
-    queue[bestIdx].updatedAt = new Date().toISOString();
-    writeQueue(meshId, queue);
-    return queue[bestIdx];
+        if (bestIdx === -1) return null;
+        queue[bestIdx].status = status;
+        queue[bestIdx].updatedAt = new Date().toISOString();
+        writeQueue(meshId, queue);
+        return queue[bestIdx];
+    });
 }
 
 export interface MeshWorkQueueStats {