@adhdev/daemon-core 0.9.82-rc.49 → 0.9.82-rc.490

package/src/mesh/mesh-delivery-policy.ts

@@ -0,0 +1,284 @@
+import { randomUUID } from 'crypto';
+import { MeshRuntimeStore } from './mesh-runtime-store.js';
+
+/**
+ * Possible delivery statuses for a session delivery record.
+ */
+export type MeshSessionDeliveryStatus =
+    | 'queued'
+    | 'delivering'
+    | 'delivered'
+    | 'acked'
+    | 'completed'
+    | 'failed'
+    | 'expired'
+    | 'cancelled';
+
+/**
+ * Kind of delivery — controls priority and policy handling.
+ */
+export type MeshSessionDeliveryKind =
+    | 'task'
+    | 'followup'
+    | 'approval'
+    | 'recovery'
+    | 'system_notice';
+
+/**
+ * A session delivery decision — what to do when a task arrives for a session.
+ */
+export type MeshDeliveryDecision =
+    | 'immediate'     // Session is idle: deliver now, create an 'acked' delivery record
+    | 'queued'        // Session is busy: hold delivery until session becomes idle
+    | 'rejected';     // Session is terminal or unknown: cannot deliver
+
+export interface MeshDeliveryPolicyResult {
+    decision: MeshDeliveryDecision;
+    reason: string;
+    /** When decision='queued', estimated deliver-after ISO timestamp if known. */
+    deliverAfter?: string;
+    /** Human-readable explanation for coordinator/operator. */
+    message: string;
+}
+
+/**
+ * Session statuses where immediate delivery is allowed.
+ * The session is ready to accept new work.
+ */
+const IMMEDIATE_DELIVERY_STATUSES = new Set([
+    'idle',
+    'waiting_input',
+    'ready',
+]);
+
+/**
+ * Session statuses that indicate the session is busy but still alive.
+ * Delivery is queued rather than attempted immediately.
+ */
+const BUSY_DELIVERY_STATUSES = new Set([
+    'generating',
+    'running',
+    'streaming',
+    'busy',
+    'starting',
+    'initializing',
+    'waiting_approval',
+]);
+
+/**
+ * Session statuses that indicate the session is permanently unavailable.
+ * Delivery should be rejected.
+ */
+const TERMINAL_DELIVERY_STATUSES = new Set([
+    'stopped',
+    'failed',
+    'terminated',
+    'exited',
+    'closed',
+    'deleted',
+    'error',
+]);
+
+/**
+ * Determine whether to deliver immediately, queue, or reject based on session status.
+ *
+ * This is a pure function — it does not write to any store.
+ */
+export function resolveDeliveryDecision(
+    sessionStatus: string | undefined,
+    opts?: {
+        kind?: MeshSessionDeliveryKind;
+        /** When true, busy session immediate injection is allowed (provider-specific capability). */
+        allowBusyInjection?: boolean;
+    },
+): MeshDeliveryPolicyResult {
+    const status = (sessionStatus || '').trim().toLowerCase();
+
+    if (!status) {
+        return {
+            decision: 'rejected',
+            reason: 'unknown_session_status',
+            message: 'Session status is unknown. Delivery rejected (fail-closed). Use mesh_launch_session to start a fresh session.',
+        };
+    }
+
+    if (IMMEDIATE_DELIVERY_STATUSES.has(status)) {
+        return {
+            decision: 'immediate',
+            reason: `session_${status}`,
+            message: `Session is ${status} — delivery allowed immediately.`,
+        };
+    }
+
+    if (BUSY_DELIVERY_STATUSES.has(status)) {
+        if (opts?.allowBusyInjection) {
+            return {
+                decision: 'immediate',
+                reason: `session_${status}_busy_injection_allowed`,
+                message: `Session is ${status} but provider supports busy injection. Delivered immediately.`,
+            };
+        }
+        // approval-kind may be delivered to waiting_approval sessions
+        if (status === 'waiting_approval' && opts?.kind === 'approval') {
+            return {
+                decision: 'immediate',
+                reason: 'session_waiting_approval_approval_message',
+                message: 'Session is waiting for approval — approval message delivered immediately.',
+            };
+        }
+        return {
+            decision: 'queued',
+            reason: `session_${status}_busy`,
+            message: `Session is ${status}. Task queued for delivery when session becomes idle. Do not inject directly into a busy session.`,
+        };
+    }
+
+    if (TERMINAL_DELIVERY_STATUSES.has(status)) {
+        return {
+            decision: 'rejected',
+            reason: `session_${status}_terminal`,
+            message: `Session is ${status} (terminal). Delivery rejected. Launch a new session before dispatching tasks.`,
+        };
+    }
+
+    // Unknown/unrecognized status: fail-closed
+    return {
+        decision: 'rejected',
+        reason: 'unrecognized_session_status',
+        message: `Session status '${sessionStatus}' is not recognized. Delivery rejected (fail-closed). Inspect session state before retrying.`,
+    };
+}
+
+export interface SessionDeliveryRecord {
+    id: string;
+    meshId: string;
+    nodeId?: string;
+    sessionId?: string;
+    providerType?: string;
+    taskId?: string;
+    kind: MeshSessionDeliveryKind;
+    priority: number;
+    message: string;
+    status: MeshSessionDeliveryStatus;
+    deliverAfter?: string;
+    expiresAt?: string;
+    attemptCount: number;
+    sourceCoordinatorSessionId?: string;
+    sourceCoordinatorDaemonId?: string;
+    lastError?: string;
+    createdAt: string;
+    updatedAt: string;
+}
+
+/**
+ * Create a delivery record in the store.
+ */
+export function createSessionDelivery(opts: {
+    meshId: string;
+    nodeId?: string;
+    sessionId?: string;
+    providerType?: string;
+    taskId?: string;
+    kind: MeshSessionDeliveryKind;
+    message: string;
+    status: MeshSessionDeliveryStatus;
+    priority?: number;
+    deliverAfter?: string;
+    expiresAt?: string;
+    sourceCoordinatorSessionId?: string;
+    sourceCoordinatorDaemonId?: string;
+}): SessionDeliveryRecord {
+    const now = new Date().toISOString();
+    const id = randomUUID();
+    const record: SessionDeliveryRecord = {
+        id,
+        meshId: opts.meshId,
+        nodeId: opts.nodeId,
+        sessionId: opts.sessionId,
+        providerType: opts.providerType,
+        taskId: opts.taskId,
+        kind: opts.kind,
+        priority: opts.priority ?? 0,
+        message: opts.message,
+        status: opts.status,
+        deliverAfter: opts.deliverAfter,
+        expiresAt: opts.expiresAt,
+        attemptCount: 0,
+        sourceCoordinatorSessionId: opts.sourceCoordinatorSessionId,
+        sourceCoordinatorDaemonId: opts.sourceCoordinatorDaemonId,
+        createdAt: now,
+        updatedAt: now,
+    };
+    MeshRuntimeStore.getInstance().insertSessionDelivery({
+        id,
+        meshId: opts.meshId,
+        nodeId: opts.nodeId,
+        sessionId: opts.sessionId,
+        providerType: opts.providerType,
+        taskId: opts.taskId,
+        kind: opts.kind,
+        priority: opts.priority ?? 0,
+        message: opts.message,
+        status: opts.status,
+        deliverAfter: opts.deliverAfter,
+        expiresAt: opts.expiresAt,
+        sourceCoordinatorSessionId: opts.sourceCoordinatorSessionId,
+        sourceCoordinatorDaemonId: opts.sourceCoordinatorDaemonId,
+        createdAt: now,
+        updatedAt: now,
+    });
+    return record;
+}
+
+/**
+ * Update the status of a delivery record.
+ */
+export function updateSessionDeliveryStatus(
+    id: string,
+    status: MeshSessionDeliveryStatus,
+    opts?: { lastError?: string; incrementAttempt?: boolean },
+): void {
+    try {
+        MeshRuntimeStore.getInstance().updateSessionDeliveryStatus(id, status, opts);
+    } catch { /* best-effort */ }
+}
+
+/**
+ * Get active (non-terminal) deliveries for a mesh, optionally filtered by session.
+ */
+export function getActiveSessionDeliveries(meshId: string, sessionId?: string) {
+    try {
+        return MeshRuntimeStore.getInstance().getActiveSessionDeliveries(meshId, sessionId);
+    } catch {
+        return [];
+    }
+}
+
+// MESH-COMPLEXITY-AUDIT Part 8-2: the completion-conflict diagnostic
+// (recordCompletionConflict / getRecentCompletionConflicts, backed by
+// mesh_completion_conflicts) was dropped. It recorded WHICH task lost a
+// fingerprint-dedup collision but had no production reader and no role in the
+// no-loss delivery contract — the dedup DECISION lives entirely in the
+// fingerprint match in mesh-event-forwarding.ts, which is unchanged. Removing
+// the side-record does not alter any completion-delivery outcome.
+
+export function __clearSessionDeliveriesForTests(meshId: string): void {
+    MeshRuntimeStore.getInstance().deleteSessionDeliveries(meshId);
+}
+
+/**
+ * Mark all active (queued/delivering/delivered/acked) deliveries for a session as completed or failed.
+ * Called when a task's terminal status is confirmed so delivery records stay in sync.
+ */
+export function markSessionDeliveriesTerminal(
+    meshId: string,
+    sessionId: string,
+    terminalStatus: 'completed' | 'failed',
+): void {
+    try {
+        const active = MeshRuntimeStore.getInstance().getActiveSessionDeliveries(meshId, sessionId);
+        for (const delivery of active) {
+            MeshRuntimeStore.getInstance().updateSessionDeliveryStatus(delivery.id, terminalStatus);
+        }
+    } catch { /* best-effort */ }
+}

package/src/mesh/mesh-event-classify.ts

@@ -0,0 +1,78 @@
+import type { MeshLedgerKind } from './mesh-ledger.js';
+
+// ---------------------------------------------------------------------------
+// Core event injection
+// ---------------------------------------------------------------------------
+
+const MESH_COORDINATOR_EVENTS = new Set([
+    'agent:generating_started',
+    'agent:generating_completed',
+    'agent:waiting_approval',
+    'agent:stopped',
+    'agent:ready',
+    'monitor:no_progress',
+    'refine:accepted',
+    'refine:completed',
+    'refine:failed',
+    'worktree_bootstrap_complete',
+    'worktree_bootstrap_failed',
+    // G3: mission-hygiene nudge. Emitted once when all of a mission's tasks first
+    // become terminal — a "consider closing this mission" hint for the coordinator.
+    // Purely informational: NOT force-injected (no blocked coordinator waits on it)
+    // and NOT an approval; it never drives a mission status transition on its own.
+    'mission_close_candidate',
+]);
+
+export const EVENT_TO_LEDGER_KIND: Record<string, MeshLedgerKind> = {
+    'agent:generating_completed': 'task_completed',
+    'agent:waiting_approval': 'task_approval_needed',
+    'agent:stopped': 'task_failed',
+    'monitor:no_progress': 'task_stalled',
+};
+
+export function isMeshCoordinatorEvent(eventName: unknown): eventName is string {
+    return typeof eventName === 'string' && MESH_COORDINATOR_EVENTS.has(eventName);
+}
+
+// Terminal events that the coordinator is actively blocked waiting on. When the
+// coordinator CLI session dispatches a task (e.g. mesh_send_task) it stays in
+// `generating` until the result arrives — but a generating coordinator queues
+// incoming send_message calls into its adapter's pendingOutboundQueue, which is
+// only flushed on the coordinator's OWN idle transition. That transition can't
+// happen until it receives this very event → deadlock. We force-inject these so
+// they bypass the busy send-guard and land in the PTY while generating.
+export const MESH_FORCE_INJECT_EVENTS: ReadonlySet<string> = new Set([
+    'agent:generating_completed',
+    'agent:stopped',
+    'agent:waiting_approval',
+    'refine:completed',
+    'refine:failed',
+    'worktree_bootstrap_complete',
+    'worktree_bootstrap_failed',
+]);
+
+export function shouldForceInjectMeshEvent(eventName: unknown): boolean {
+    return typeof eventName === 'string' && MESH_FORCE_INJECT_EVENTS.has(eventName);
+}
+
+// APPROVAL-Q1-REALTIME. Approval-kind coordinator events: a worker is blocked on an
+// approval prompt and needs the coordinator to act (mesh_approve). Approval is treated
+// differently from a completion in the reconcile loop's no-idle hold, and the reason is
+// WHERE each event's authoritative state lives:
+//   - A completion's payload (finalSummary / worker result) exists ONLY in the pending
+//     event, so a drain-without-inject loses it forever → it MUST ride the idle-edge hold
+//     until it can land in the coordinator as a real turn.
+//   - An approval's authoritative state is recorded at LEVEL in the ledger the moment the
+//     event is processed (task_approval_needed → mesh_status awaiting_approval, see
+//     onMeshCoordinatorEventForwarded + mesh-active-work). The pending approval event is
+//     therefore only a real-time NUDGE, not the source of truth: it can be delivered to a
+//     busy coordinator's inbox (and dropped) without data loss, because the level state
+//     re-derives it. That is why approval is exempt from the idle-edge hold completions
+//     require, and why a stale/resolved approval nudge can simply be dropped.
+export const MESH_APPROVAL_EVENTS: ReadonlySet<string> = new Set([
+    'agent:waiting_approval',
+]);
+
+export function isMeshApprovalEvent(eventName: unknown): boolean {
+    return typeof eventName === 'string' && MESH_APPROVAL_EVENTS.has(eventName);
+}