@adhdev/daemon-core 0.9.82-rc.21 → 0.9.82-rc.211

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

@@ -1,24 +1,91 @@
-import { existsSync, writeFileSync, readFileSync } from 'fs';
-import { join } from 'path';
 import { randomUUID } from 'crypto';
-import { getLedgerDir } from './mesh-ledger.js';
+import { requireMeshHostQueueOwner } from './mesh-host-ownership.js';
+import type { RepoMeshDaemonRole } from '../repo-mesh-types.js';
+import { MeshRuntimeStore } from './mesh-runtime-store.js';
+import { getMesh } from '../config/mesh-config.js';
 
 export type MeshTaskStatus = 'pending' | 'assigned' | 'completed' | 'failed' | 'cancelled';
 export type MeshActiveTaskStatus = Extract<MeshTaskStatus, 'pending' | 'assigned'>;
 export type MeshHistoricalTaskStatus = Extract<MeshTaskStatus, 'completed' | 'failed' | 'cancelled'>;
+export type MeshTaskMode = 'code_change' | 'validation' | 'live_debug_readonly' | 'launch_app' | 'convergence';
 
 export const ACTIVE_MESH_QUEUE_STATUSES: MeshActiveTaskStatus[] = ['pending', 'assigned'];
 export const HISTORICAL_MESH_QUEUE_STATUSES: MeshHistoricalTaskStatus[] = ['completed', 'failed', 'cancelled'];
+export const MESH_TASK_MODES: MeshTaskMode[] = ['code_change', 'validation', 'live_debug_readonly', 'launch_app', 'convergence'];
+
+export interface MeshTaskModeValidationResult {
+    valid: boolean;
+    taskMode?: MeshTaskMode;
+    violations: string[];
+    allowedOperations?: string[];
+}
+
+const LIVE_DEBUG_READONLY_FORBIDDEN: Array<{ label: string; pattern: RegExp }> = [
+    { label: 'source_edit', pattern: /\b(edit|modify|patch|apply\s+patch|write\s+(?:to\s+)?(?:file|source)|overwrite|delete\s+file|remove\s+file|create\s+file|touch\s+file)\b/i },
+    { label: 'git_mutation', pattern: /\b(?:git\s+(?:add|commit|push|reset|rebase|clean|checkout|switch|merge|tag|restore|rm|mv|stash|worktree\s+(?:add|remove|move))|push\b)/i },
+    { label: 'checkpoint', pattern: /\b(checkpoint|mesh_checkpoint)\b/i },
+    { label: 'deploy_or_version_bump', pattern: /\b(deploy|wrangler\s+deploy|version[-\s]?bump|npm\s+version|release|npm\s+publish|yarn\s+publish|pnpm\s+publish)\b/i },
+    { label: 'destructive_shell', pattern: /\b(rm\s+-rf|mv\s+\S+\s+\S+|truncate\s|tee\s+\S+|sed\s+-i|shred\b)\b/i },
+    { label: 'package_install', pattern: /\b(npm\s+(?:install|i|add|link|uninstall|remove)|yarn\s+(?:add|remove|link)|pnpm\s+(?:add|remove|link)|pip\s+install|brew\s+install|apt\s+install|cargo\s+install)\b/i },
+    { label: 'container_mutation', pattern: /\b(docker\s+(?:build|run|exec|push|tag|rmi|rm|create|start|stop|kill)|kubectl\s+(?:apply|delete|patch|replace|create|scale))\b/i },
+];
+
+export function normalizeMeshTaskMode(value: unknown): MeshTaskMode | undefined {
+    if (typeof value !== 'string') return undefined;
+    const normalized = value.trim() as MeshTaskMode;
+    return (MESH_TASK_MODES as string[]).includes(normalized) ? normalized : undefined;
+}
+
+export function validateMeshTaskModeRequest(mode: unknown, message: string): MeshTaskModeValidationResult {
+    const taskMode = normalizeMeshTaskMode(mode);
+    if (!taskMode) {
+        return { valid: true, violations: [] };
+    }
+    if (taskMode !== 'live_debug_readonly') {
+        return { valid: true, taskMode, violations: [] };
+    }
+    const violations = LIVE_DEBUG_READONLY_FORBIDDEN
+        .filter(rule => rule.pattern.test(message || ''))
+        .map(rule => rule.label);
+    return {
+        valid: violations.length === 0,
+        taskMode,
+        violations,
+        allowedOperations: [
+            'process/log/window/port/session inspection',
+            'read-only filesystem listing/reading',
+            'status probes and keep-running handle reporting',
+            'diagnostic summaries without source edits, commits, checkpoints, pushes, deploys, resets, rebases, or destructive cleanups',
+        ],
+    };
+}
 
 export interface MeshWorkQueueEntry {
     id: string;
     meshId: string;
     message: string;
     status: MeshTaskStatus;
+    taskMode?: MeshTaskMode;
     /** If specified, only this node can claim the task (used by legacy mesh_send_task) */
     targetNodeId?: string;
     /** If specified, only this runtime session can claim the task */
     targetSessionId?: string;
+    /** If specified, a node must expose all tags before it can claim the task. */
+    requiredTags?: string[];
+    /**
+     * M1: ids of tasks that must reach 'completed' before this task is claimable.
+     * Forward references (ids not yet enqueued) are allowed for batch flows and
+     * simply keep the task waiting until the referenced task exists and completes.
+     */
+    dependsOn?: string[];
+    /** M1/M3: mission this task belongs to (joins mesh_missions). */
+    missionId?: string;
+    /**
+     * M1: why this task is held back (e.g. "dependency_failed:<taskId>").
+     * Only set by the system on dependency failure under the 'block' policy;
+     * waiting-on-dependency state is computed at view time, not stored.
+     */
+    blockedReason?: string;
     /** The node that actually claimed and is executing the task */
     assignedNodeId?: string;
     /** The session currently executing the task */
@@ -30,6 +97,8 @@ export interface MeshWorkQueueEntry {
     requeueReason?: string;
     requeuedAt?: string;
     requeueCount?: number;
+    /** Max automatic requeue attempts. When requeueCount reaches this, task is auto-failed. */
+    maxRetries?: number;
     /** Last automatic queue session spin-up attempt, for mesh_view_queue/debug visibility. */
     autoLaunch?: {
         status: 'skipped' | 'started' | 'failed' | 'completed';
@@ -45,25 +114,112 @@ export interface MeshWorkQueueEntry {
     updatedAt: string;
 }
 
-function getQueuePath(meshId: string): string {
-    const safe = meshId.replace(/[^a-zA-Z0-9_-]/g, '_');
-    return join(getLedgerDir(), `${safe}.queue.json`);
+export interface MeshQueueMutationOptions {
+    ownerRole?: RepoMeshDaemonRole;
+}
+
+export function normalizeMeshCapabilityTags(value: unknown): string[] {
+    if (!Array.isArray(value)) return [];
+    const seen = new Set<string>();
+    return value
+        .map(tag => typeof tag === 'string' ? tag.trim() : '')
+        .filter(Boolean)
+        .filter(tag => {
+            if (seen.has(tag)) return false;
+            seen.add(tag);
+            return true;
+        });
+}
+
+function firstProviderPriority(policy: unknown): string | undefined {
+    const raw = policy && typeof policy === 'object' && !Array.isArray(policy)
+        ? (policy as Record<string, unknown>).providerPriority
+        : undefined;
+    if (!Array.isArray(raw)) return undefined;
+    return raw.find(type => typeof type === 'string' && type.trim())?.trim();
+}
+
+export function buildMeshNodeCapabilityTags(
+    node: { capabilities?: unknown; policy?: unknown; isLocalWorktree?: unknown; worktreeBranch?: unknown } | undefined,
+    providerType?: string,
+): string[] {
+    const provider = typeof providerType === 'string' && providerType.trim()
+        ? providerType.trim()
+        : firstProviderPriority(node?.policy);
+    const worktreeBranch = typeof node?.worktreeBranch === 'string' && node.worktreeBranch.trim()
+        ? node.worktreeBranch.trim()
+        : null;
+    return normalizeMeshCapabilityTags([
+        ...(Array.isArray(node?.capabilities) ? node.capabilities : []),
+        `os=${process.platform}`,
+        `arch=${process.arch}`,
+        ...(provider ? [`provider=${provider}`] : []),
+        // Worktree nodes automatically expose a "worktree=<branch>" tag so that
+        // mesh_enqueue_task with required_tags: ["worktree=<branch>"] routes
+        // only to the matching worktree node.
+        ...(node?.isLocalWorktree === true && worktreeBranch ? [`worktree=${worktreeBranch}`] : []),
+    ]);
+}
+
+export function nodeSatisfiesRequiredTags(requiredTags: unknown, capabilityTags: unknown): boolean {
+    const required = normalizeMeshCapabilityTags(requiredTags);
+    if (required.length === 0) return true;
+    const available = new Set(normalizeMeshCapabilityTags(capabilityTags));
+    return required.every(tag => available.has(tag));
+}
+
+function withQueueLock<T>(_meshId: string, fn: () => T): T {
+    return MeshRuntimeStore.getInstance().transaction(fn);
 }
 
 function readQueue(meshId: string): MeshWorkQueueEntry[] {
-    const path = getQueuePath(meshId);
-    if (!existsSync(path)) return [];
-    try {
-        const content = readFileSync(path, 'utf-8');
-        return JSON.parse(content) as MeshWorkQueueEntry[];
-    } catch {
-        return [];
-    }
+    return MeshRuntimeStore.getInstance().getQueueEntries(meshId);
 }
 
 function writeQueue(meshId: string, queue: MeshWorkQueueEntry[]): void {
-    const path = getQueuePath(meshId);
-    writeFileSync(path, JSON.stringify(queue, null, 2), 'utf-8');
+    MeshRuntimeStore.getInstance().replaceQueue(meshId, queue);
+}
+
+function normalizeDependsOn(value: unknown): string[] {
+    if (!Array.isArray(value)) return [];
+    const seen = new Set<string>();
+    return value
+        .map(id => typeof id === 'string' ? id.trim() : '')
+        .filter(Boolean)
+        .filter(id => {
+            if (seen.has(id)) return false;
+            seen.add(id);
+            return true;
+        });
+}
+
+/**
+ * M1: detect dependency cycles before enqueue. Walks the dependency graph of
+ * existing queue entries plus the new task's edges. Fail-closed: a cycle
+ * rejects the enqueue entirely. Synchronous and bounded by queue size.
+ */
+export function assertNoDependencyCycle(meshId: string, newTaskId: string, dependsOn: string[]): void {
+    if (dependsOn.length === 0) return;
+    if (dependsOn.includes(newTaskId)) {
+        throw new Error(`dependency_cycle_detected: task '${newTaskId}' cannot depend on itself`);
+    }
+    const adjacency = new Map<string, string[]>();
+    for (const entry of readQueue(meshId)) {
+        adjacency.set(entry.id, normalizeDependsOn(entry.dependsOn));
+    }
+    adjacency.set(newTaskId, dependsOn);
+    // DFS from the new task: if we can reach newTaskId again, the edges form a cycle.
+    const stack = [...dependsOn];
+    const visited = new Set<string>();
+    while (stack.length > 0) {
+        const current = stack.pop()!;
+        if (current === newTaskId) {
+            throw new Error(`dependency_cycle_detected: task '${newTaskId}' is part of a dependency cycle via '${dependsOn.join(', ')}'`);
+        }
+        if (visited.has(current)) continue;
+        visited.add(current);
+        stack.push(...(adjacency.get(current) ?? []));
+    }
 }
 
 /**
@@ -72,75 +228,119 @@ function writeQueue(meshId: string, queue: MeshWorkQueueEntry[]): void {
 export function enqueueTask(
     meshId: string,
     message: string,
-    opts?: { targetNodeId?: string; targetSessionId?: string }
+    opts?: {
+        targetNodeId?: string;
+        targetSessionId?: string;
+        taskMode?: MeshTaskMode | string;
+        requiredTags?: string[];
+        /** M1: tasks that must complete before this one is claimable. */
+        dependsOn?: string[];
+        /** M1/M3: mission this task belongs to. */
+        missionId?: string;
+        /** Explicit task id for batch/template flows (M5). Random UUID when omitted. */
+        id?: string;
+    } & MeshQueueMutationOptions,
 ): 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;
+    requireMeshHostQueueOwner(opts);
+    const modeValidation = validateMeshTaskModeRequest(opts?.taskMode, message);
+    if (!modeValidation.valid) {
+        throw new Error(`live_debug_readonly_guardrail_violation: forbidden operations (${modeValidation.violations.join(', ')})`);
+    }
+    const id = typeof opts?.id === 'string' && opts.id.trim() ? opts.id.trim() : randomUUID();
+    const dependsOn = normalizeDependsOn(opts?.dependsOn);
+    return withQueueLock(meshId, () => {
+        if (MeshRuntimeStore.getInstance().findQueueEntryById(meshId, id)) {
+            throw new Error(`duplicate_task_id: task '${id}' already exists in mesh '${meshId}'`);
+        }
+        assertNoDependencyCycle(meshId, id, dependsOn);
+        const entry: MeshWorkQueueEntry = {
+            id,
+            meshId,
+            message,
+            status: 'pending',
+            taskMode: modeValidation.taskMode,
+            targetNodeId: opts?.targetNodeId,
+            targetSessionId: opts?.targetSessionId,
+            requiredTags: normalizeMeshCapabilityTags(opts?.requiredTags),
+            ...(dependsOn.length > 0 ? { dependsOn } : {}),
+            ...(typeof opts?.missionId === 'string' && opts.missionId.trim() ? { missionId: opts.missionId.trim() } : {}),
+            createdAt: new Date().toISOString(),
+            updatedAt: new Date().toISOString(),
+        };
+        MeshRuntimeStore.getInstance().insertQueueEntry(entry);
+        return entry;
+    });
 }
 
 /**
  * Get all tasks in the queue, optionally filtered by status.
  */
 export function getQueue(meshId: string, opts?: { status?: MeshTaskStatus[] }): MeshWorkQueueEntry[] {
-    let queue = readQueue(meshId);
-    if (opts?.status?.length) {
-        const statuses = new Set(opts.status);
-        queue = queue.filter(q => statuses.has(q.status));
-    }
-    return queue;
+    return MeshRuntimeStore.getInstance().getQueueEntries(meshId, opts?.status?.length ? opts.status : undefined);
+}
+
+export function getMeshQueueRevision(meshId: string): string {
+    return MeshRuntimeStore.getInstance().getQueueRevision(meshId);
 }
 
 /**
  * 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);
-    }
+export function claimNextTask(meshId: string, nodeId: string, sessionId: string, capabilityTags?: string[]): MeshWorkQueueEntry | null {
+    return MeshRuntimeStore.getInstance().claimNextQueueTask(meshId, nodeId, sessionId, capabilityTags);
+}
 
-    if (targetIdx === -1) return null;
+// ─── M1: Dependency Failure Propagation ─────────
 
-    const entry = queue[targetIdx];
-    entry.status = 'assigned';
-    entry.assignedNodeId = nodeId;
-    entry.assignedSessionId = sessionId;
-    entry.dispatchTimestamp = new Date().toISOString();
-    entry.updatedAt = new Date().toISOString();
+export type DependencyFailurePolicy = 'block' | 'cancel';
 
-    writeQueue(meshId, queue);
-    return entry;
+function resolveDependencyFailurePolicy(meshId: string): DependencyFailurePolicy {
+    try {
+        const policy = (getMesh(meshId)?.policy ?? {}) as Record<string, unknown>;
+        return policy.onDependencyFailure === 'cancel' ? 'cancel' : 'block';
+    } catch {
+        return 'block';
+    }
 }
 
+/**
+ * Apply the mesh's onDependencyFailure policy to pending dependents of a task
+ * that just reached a failed/cancelled terminal state.
+ *
+ * - 'block' (default): dependents stay pending with blockedReason
+ *   "dependency_failed:<taskId>" so an operator can requeue/cancel them.
+ * - 'cancel': dependents are cancelled (cascading to their own dependents).
+ *
+ * Must be called inside the queue lock of the triggering transition.
+ */
+function propagateDependencyFailure(meshId: string, failedTaskId: string): void {
+    const policy = resolveDependencyFailurePolicy(meshId);
+    const store = MeshRuntimeStore.getInstance();
+    const frontier = [failedTaskId];
+    const seen = new Set<string>(frontier);
+    while (frontier.length > 0) {
+        const currentId = frontier.pop()!;
+        const dependents = store.getQueueEntries(meshId, ['pending'])
+            .filter(entry => Array.isArray(entry.dependsOn) && entry.dependsOn.includes(currentId));
+        for (const dependent of dependents) {
+            if (seen.has(dependent.id)) continue;
+            seen.add(dependent.id);
+            if (policy === 'cancel') {
+                dependent.status = 'cancelled';
+                dependent.cancelledAt = new Date().toISOString();
+                dependent.cancelReason = `dependency_failed:${currentId}`;
+                store.updateQueueEntry(dependent);
+                frontier.push(dependent.id); // cascade to transitive dependents
+            } else {
+                dependent.blockedReason = `dependency_failed:${currentId}`;
+                store.updateQueueEntry(dependent);
+            }
+        }
+    }
+}
+
+const DEPENDENCY_FAILURE_TERMINALS = new Set<MeshTaskStatus>(['failed', 'cancelled']);
+
 /**
  * Update the status of a specific task.
  * Used when a session completes, fails, or stalls.
@@ -149,15 +349,17 @@ export function updateTaskStatus(
     meshId: string,
     taskId: string,
     status: MeshTaskStatus,
+    opts?: MeshQueueMutationOptions,
 ): 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];
+    requireMeshHostQueueOwner(opts);
+    return withQueueLock(meshId, () => {
+        const entry = MeshRuntimeStore.getInstance().findQueueEntryById(meshId, taskId);
+        if (!entry) return null;
+        entry.status = status;
+        MeshRuntimeStore.getInstance().updateQueueEntry(entry);
+        if (DEPENDENCY_FAILURE_TERMINALS.has(status)) propagateDependencyFailure(meshId, taskId);
+        return entry;
+    });
 }
 
 export function recordTaskAutoLaunch(
@@ -165,17 +367,14 @@ 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 entry = MeshRuntimeStore.getInstance().findQueueEntryById(meshId, taskId);
+        if (!entry) return null;
+        const now = new Date().toISOString();
+        entry.autoLaunch = { ...autoLaunch, updatedAt: now };
+        MeshRuntimeStore.getInstance().updateQueueEntry(entry);
+        return entry;
+    });
 }
 
 /**
@@ -184,25 +383,31 @@ export function recordTaskAutoLaunch(
 export function cancelTask(
     meshId: string,
     taskId: string,
-    opts?: { reason?: string },
+    opts?: { reason?: string } & MeshQueueMutationOptions,
 ): 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];
+    requireMeshHostQueueOwner(opts);
+    return withQueueLock(meshId, () => {
+        const entry = MeshRuntimeStore.getInstance().findQueueEntryById(meshId, taskId);
+        if (!entry) return null;
+        const now = new Date().toISOString();
+        entry.status = 'cancelled';
+        entry.cancelledAt = now;
+        if (opts?.reason) entry.cancelReason = opts.reason;
+        MeshRuntimeStore.getInstance().updateQueueEntry(entry);
+        propagateDependencyFailure(meshId, taskId);
+        return entry;
+    });
 }
 
 /**
  * Return a queue task to pending for retry. By default, dead session targeting
  * and assigned ownership are cleared so stale assignments do not strand again.
  */
+export type RequeueResult =
+    | { status: 'requeued'; entry: MeshWorkQueueEntry }
+    | { status: 'failed_max_retries'; entry: MeshWorkQueueEntry; maxRetries: number; requeueCount: number }
+    | { status: 'not_found' };
+
 export function requeueTask(
     meshId: string,
     taskId: string,
@@ -212,29 +417,48 @@ export function requeueTask(
         targetSessionId?: string;
         clearTargetNode?: boolean;
         clearTargetSession?: boolean;
-    },
+        /**
+         * Override the retry cap for this call. Use only for explicit operator actions.
+         * If true, the task is requeued even when requeueCount >= maxRetries.
+         */
+        force?: boolean;
+        /** Per-task retry cap override. Falls back to mesh policy maxTaskRetries (default 1). */
+        maxRetries?: number;
+    } & MeshQueueMutationOptions,
 ): 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;
+    requireMeshHostQueueOwner(opts);
+    return withQueueLock(meshId, () => {
+        const entry = MeshRuntimeStore.getInstance().findQueueEntryById(meshId, taskId);
+        if (!entry) return null;
+        const currentCount = entry.requeueCount || 0;
+        const maxRetries = opts?.maxRetries ?? entry.maxRetries ?? 1;
+        if (!opts?.force && currentCount >= maxRetries) {
+            // Auto-fail: cap exceeded without explicit force override.
+            entry.status = 'failed';
+            entry.cancelReason = `max_retries_exceeded: requeued ${currentCount} time(s), limit is ${maxRetries}`;
+            entry.updatedAt = new Date().toISOString();
+            MeshRuntimeStore.getInstance().updateQueueEntry(entry);
+            propagateDependencyFailure(meshId, taskId);
+            return entry;
+        }
+        entry.status = 'pending';
+        // Operator requeue clears a dependency-failure block — the operator is
+        // explicitly overriding the held-back state.
+        delete entry.blockedReason;
+        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.requeuedAt = new Date().toISOString();
+        entry.requeueCount = currentCount + 1;
+        if (opts?.reason) entry.requeueReason = opts.reason;
+        MeshRuntimeStore.getInstance().updateQueueEntry(entry);
+        return entry;
+    });
 }
 
 /**
@@ -244,29 +468,43 @@ export function updateSessionTaskStatus(
     meshId: string,
     sessionId: string,
     status: MeshTaskStatus,
+    opts?: { occurredAt?: string },
 ): 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;
-            }
-        }
-    }
-    if (bestIdx === -1) return null;
+    return withQueueLock(meshId, () => {
+        const occurredAtIso = opts?.occurredAt ? new Date(opts.occurredAt).toISOString() : undefined;
+        const entry = MeshRuntimeStore.getInstance().findAssignedBySession(meshId, sessionId, occurredAtIso);
+        if (!entry) return null;
+        entry.status = status;
+        MeshRuntimeStore.getInstance().updateQueueEntry(entry);
+        if (DEPENDENCY_FAILURE_TERMINALS.has(status)) propagateDependencyFailure(meshId, entry.id);
+        return entry;
+    });
+}
+
+/**
+ * M1-3: true when at least one pending task is waiting on the given task.
+ * Used by the completion event path to decide whether to wake the queue.
+ */
+export function hasPendingDependents(meshId: string, taskId: string): boolean {
+    return MeshRuntimeStore.getInstance().getQueueEntries(meshId, ['pending'])
+        .some(entry => Array.isArray(entry.dependsOn) && entry.dependsOn.includes(taskId));
+}
 
-    queue[bestIdx].status = status;
-    queue[bestIdx].updatedAt = new Date().toISOString();
-    writeQueue(meshId, queue);
-    return queue[bestIdx];
+/**
+ * M1-4: view-time dependency state for a task — unmet dependency ids and
+ * whether the task is currently claimable from a dependency standpoint.
+ * Not stored (truth stays in task statuses).
+ */
+export function describeTaskDependencyState(
+    entry: Pick<MeshWorkQueueEntry, 'dependsOn' | 'blockedReason'>,
+    statusById: Map<string, MeshTaskStatus | string>,
+): { waitingOn: string[]; dependenciesSatisfied: boolean } {
+    const deps = Array.isArray(entry.dependsOn) ? entry.dependsOn : [];
+    const waitingOn = deps.filter(depId => statusById.get(depId) !== 'completed');
+    return {
+        waitingOn,
+        dependenciesSatisfied: waitingOn.length === 0 && !entry.blockedReason,
+    };
 }
 
 export interface MeshWorkQueueStats {
@@ -294,14 +532,16 @@ export interface MeshWorkQueueStats {
  * Return aggregate queue statistics for the given mesh.
  */
 export function getMeshQueueStats(meshId: string): MeshWorkQueueStats {
-    const queue = readQueue(meshId);
-    const pending = queue.filter(q => q.status === 'pending').length;
-    const assigned = queue.filter(q => q.status === 'assigned').length;
-    const completed = queue.filter(q => q.status === 'completed').length;
-    const failed = queue.filter(q => q.status === 'failed').length;
-    const cancelled = queue.filter(q => q.status === 'cancelled').length;
+    const rows = MeshRuntimeStore.getInstance().getQueueStatsByStatus(meshId);
+    const counts: Record<string, number> = {};
+    for (const r of rows) counts[r.status] = r.count;
+    const pending = counts['pending'] ?? 0;
+    const assigned = counts['assigned'] ?? 0;
+    const completed = counts['completed'] ?? 0;
+    const failed = counts['failed'] ?? 0;
+    const cancelled = counts['cancelled'] ?? 0;
     return {
-        total: queue.length,
+        total: pending + assigned + completed + failed + cancelled,
         active: pending + assigned,
         historical: completed + failed + cancelled,
         pending,
@@ -309,22 +549,106 @@ export function getMeshQueueStats(meshId: string): MeshWorkQueueStats {
         completed,
         failed,
         cancelled,
-        activeCounts: {
-            pending,
-            assigned,
-        },
-        historicalCounts: {
-            completed,
-            failed,
-            cancelled,
-        },
-        activeAssignments: queue
-            .filter(q => q.status === 'assigned')
-            .map(q => ({
-                id: q.id,
-                nodeId: q.assignedNodeId,
-                sessionId: q.assignedSessionId,
-                message: q.message,
-            })),
+        activeCounts: { pending, assigned },
+        historicalCounts: { completed, failed, cancelled },
+        activeAssignments: MeshRuntimeStore.getInstance().getActiveAssignmentDetails(meshId),
     };
 }
+
+export function __replaceMeshQueueForTests(meshId: string, queue: MeshWorkQueueEntry[]): void {
+    MeshRuntimeStore.getInstance().transaction(() => {
+        MeshRuntimeStore.getInstance().replaceQueue(meshId, queue);
+    });
+}
+
+export function __clearMeshQueueForTests(meshId: string): void {
+    MeshRuntimeStore.getInstance().deleteQueue(meshId);
+}
+
+export function __clearDirectDispatchesForTests(meshId: string): void {
+    MeshRuntimeStore.getInstance().deleteDirectDispatches(meshId);
+}
+
+export function __resetMeshRuntimeStoreForTests(): void {
+    MeshRuntimeStore.resetForTests();
+}
+
+// ── Direct Dispatch Tracking ─────────────────────────────────────────────────
+// Persists direct (non-queue) task dispatches so buildMeshActiveWork can read
+// active work from MeshRuntimeStore instead of scanning ledger JSONL entries.
+
+export type DirectDispatchRecord = ReturnType<MeshRuntimeStore['getActiveDirectDispatches']>[number];
+
+export function insertDirectDispatch(
+    meshId: string,
+    data: {
+        taskId: string;
+        nodeId?: string;
+        sessionId?: string;
+        providerType?: string;
+        message: string;
+        taskMode?: string;
+        via: string;
+        dispatchedToIdleSession?: boolean;
+        dispatchedAt: string;
+    },
+): void {
+    try {
+        MeshRuntimeStore.getInstance().insertDirectDispatch({ ...data, meshId });
+    } catch (e: any) {
+        process.stderr.write(`[adhdev-mesh] insertDirectDispatch failed for task ${data.taskId}: ${e?.message || e}\n`);
+    }
+}
+
+export function getActiveDirectDispatches(meshId: string): DirectDispatchRecord[] {
+    try {
+        return MeshRuntimeStore.getInstance().getActiveDirectDispatches(meshId);
+    } catch {
+        return [];
+    }
+}
+
+export function updateDirectDispatchStatus(
+    meshId: string,
+    sessionId: string,
+    status: 'acked' | 'completed' | 'failed' | 'stale',
+): void {
+    try {
+        MeshRuntimeStore.getInstance().updateDirectDispatchStatus(meshId, sessionId, status);
+    } catch { /* best-effort */ }
+}
+
+export function cleanupTerminalDirectDispatches(olderThanMs = 7 * 24 * 60 * 60_000): void {
+    try {
+        MeshRuntimeStore.getInstance().cleanupTerminalDirectDispatches(olderThanMs);
+    } catch { /* best-effort */ }
+}
+
+export function markStaleDirectDispatches(meshId: string, olderThanMs = 60 * 60_000): void {
+    try {
+        MeshRuntimeStore.getInstance().markStaleDirectDispatches(meshId, olderThanMs);
+    } catch { /* best-effort */ }
+}
+
+export type MeshToolCallRateResult = { rateLimitExceeded: boolean; callsInWindow: number; advisory: string | null };
+
+/**
+ * Record a coordinator tool call and return a rate-limit advisory when the
+ * call rate for that tool exceeds the allowed threshold.
+ *
+ * Defaults: 10-second sliding window, max 5 calls before advisory is raised.
+ * Returns { rateLimitExceeded: false } on any store error so callers are not blocked.
+ */
+export function recordMeshToolCall(opts: {
+    meshId: string;
+    tool: string;
+    sessionId?: string | null;
+    windowMs?: number;
+    maxCalls?: number;
+}): MeshToolCallRateResult {
+    try {
+        return MeshRuntimeStore.getInstance().recordMeshToolCall(opts);
+    } catch {
+        return { rateLimitExceeded: false, callsInWindow: 0, advisory: null };
+    }
+}