macro-agent 0.2.5 → 0.2.6

package/dist/monitor/health-check-service.d.ts

@@ -0,0 +1,143 @@
+/**
+ * HealthCheckService - Per-coordinator health check timer service
+ *
+ * Runs periodic health checks for worker agents under each coordinator:
+ * - Detects stalled agents (no activity + not processing)
+ * - Tracks consecutive failures before escalating
+ * - Emits STALE_AGENT signals for coordinator remediation
+ * - Cleans up zombie sessions periodically
+ *
+ * @module monitor/health-check-service
+ * @see s-5yhx Phase B: Monitor Active Behaviors
+ */
+import type { EventStore } from "../store/event-store.js";
+import type { SessionMapper } from "../acp/session-mapper.js";
+import type { MessageRouter } from "../router/message-router.js";
+import type { AgentId, Timestamp } from "../store/types/index.js";
+import { type StalledAgent } from "./stall-detector.js";
+/**
+ * Configuration for health check service
+ */
+export interface HealthCheckConfig {
+    /**
+     * Interval between health checks (ms)
+     * Default: 300000 (5 minutes)
+     */
+    intervalMs: number;
+    /**
+     * How long without activity before an agent is considered stalled (ms)
+     * Default: 600000 (10 minutes)
+     */
+    stalledThresholdMs: number;
+    /**
+     * How long with no progress before GUPP violation (ms)
+     * Default: 1800000 (30 minutes)
+     */
+    guppThresholdMs: number;
+    /**
+     * How many consecutive failures before escalating
+     * Default: 3
+     */
+    consecutiveFailuresBeforeEscalate: number;
+}
+/**
+ * Default configuration values
+ */
+export declare const DEFAULT_HEALTH_CHECK_CONFIG: HealthCheckConfig;
+/**
+ * Health state for a single worker
+ */
+export interface WorkerHealthState {
+    agentId: AgentId;
+    lastActivityAt: Timestamp;
+    consecutiveFailures: number;
+    lastCheckAt: Timestamp;
+    status: "healthy" | "warning" | "stalled";
+}
+/**
+ * Health state for a coordinator and its workers
+ */
+export interface CoordinatorHealthState {
+    coordinatorId: AgentId;
+    workers: Map<AgentId, WorkerHealthState>;
+    lastCheckAt: Timestamp;
+    nextCheckAt: Timestamp;
+    isRunning: boolean;
+}
+/**
+ * Result of a health check
+ */
+export interface HealthCheckResult {
+    coordinatorId: AgentId;
+    checkedAt: Timestamp;
+    stalledAgents: StalledAgent[];
+    escalatedAgents: AgentId[];
+    zombiesCleanedUp: number;
+}
+/**
+ * Service that runs periodic health checks for coordinator workers
+ */
+export declare class HealthCheckService {
+    private readonly eventStore;
+    private readonly sessionMapper;
+    private readonly messageRouter;
+    private readonly config;
+    private readonly stallDetector;
+    /** Timer handles per coordinator */
+    private timers;
+    /** Health state per coordinator */
+    private healthStates;
+    /** Last zombie cleanup time */
+    private lastZombieCleanup;
+    constructor(eventStore: EventStore, sessionMapper: SessionMapper, messageRouter: MessageRouter, config?: Partial<HealthCheckConfig>);
+    /**
+     * Start health check monitoring for a coordinator
+     *
+     * @param coordinatorId - The coordinator to monitor
+     * @param config - Optional config overrides for this coordinator
+     */
+    startForCoordinator(coordinatorId: AgentId, config?: Partial<HealthCheckConfig>): void;
+    /**
+     * Stop health check monitoring for a coordinator
+     *
+     * @param coordinatorId - The coordinator to stop monitoring
+     */
+    stopForCoordinator(coordinatorId: AgentId): void;
+    /**
+     * Stop all health check timers
+     */
+    stopAll(): void;
+    /**
+     * Manually trigger a health check for a coordinator
+     *
+     * @param coordinatorId - The coordinator to check
+     * @returns Health check result
+     */
+    checkNow(coordinatorId: AgentId): Promise<HealthCheckResult>;
+    /**
+     * Get the current health state for a coordinator
+     *
+     * @param coordinatorId - The coordinator to get state for
+     * @returns Health state or undefined if not monitoring
+     */
+    getHealthState(coordinatorId: AgentId): CoordinatorHealthState | undefined;
+    /**
+     * Get all monitored coordinators
+     *
+     * @returns Array of coordinator IDs being monitored
+     */
+    getMonitoredCoordinators(): AgentId[];
+    /**
+     * Get the current configuration
+     */
+    getConfig(): HealthCheckConfig;
+    /**
+     * Run a health check for a coordinator
+     */
+    private runHealthCheck;
+    /**
+     * Emit a STALE_AGENT signal via MessageRouter
+     */
+    private emitStaleAgentSignal;
+}
+//# sourceMappingURL=health-check-service.d.ts.map

package/dist/monitor/health-check-service.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"health-check-service.d.ts","sourceRoot":"","sources":["../../src/monitor/health-check-service.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAC;AACjE,OAAO,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,yBAAyB,CAAC;AAClE,OAAO,EAAiB,KAAK,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAMvE;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC;;;OAGG;IACH,UAAU,EAAE,MAAM,CAAC;IAEnB;;;OAGG;IACH,kBAAkB,EAAE,MAAM,CAAC;IAE3B;;;OAGG;IACH,eAAe,EAAE,MAAM,CAAC;IAExB;;;OAGG;IACH,iCAAiC,EAAE,MAAM,CAAC;CAC3C;AAED;;GAEG;AACH,eAAO,MAAM,2BAA2B,EAAE,iBAKzC,CAAC;AAEF;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,OAAO,EAAE,OAAO,CAAC;IACjB,cAAc,EAAE,SAAS,CAAC;IAC1B,mBAAmB,EAAE,MAAM,CAAC;IAC5B,WAAW,EAAE,SAAS,CAAC;IACvB,MAAM,EAAE,SAAS,GAAG,SAAS,GAAG,SAAS,CAAC;CAC3C;AAED;;GAEG;AACH,MAAM,WAAW,sBAAsB;IACrC,aAAa,EAAE,OAAO,CAAC;IACvB,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;IACzC,WAAW,EAAE,SAAS,CAAC;IACvB,WAAW,EAAE,SAAS,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,EAAE,SAAS,CAAC;IACrB,aAAa,EAAE,YAAY,EAAE,CAAC;IAC9B,eAAe,EAAE,OAAO,EAAE,CAAC;IAC3B,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAMD;;GAEG;AACH,qBAAa,kBAAkB;IAC7B,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAa;IACxC,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAgB;IAC9C,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAgB;IAC9C,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAoB;IAC3C,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAgB;IAE9C,oCAAoC;IACpC,OAAO,CAAC,MAAM,CAA2D;IAEzE,mCAAmC;IACnC,OAAO,CAAC,YAAY,CAAmD;IAEvE,+BAA+B;IAC/B,OAAO,CAAC,iBAAiB,CAAgB;gBAGvC,UAAU,EAAE,UAAU,EACtB,aAAa,EAAE,aAAa,EAC5B,aAAa,EAAE,aAAa,EAC5B,MAAM,GAAE,OAAO,CAAC,iBAAiB,CAAM;IAWzC;;;;;OAKG;IACH,mBAAmB,CACjB,aAAa,EAAE,OAAO,EACtB,MAAM,CAAC,EAAE,OAAO,CAAC,iBAAiB,CAAC,GAClC,IAAI;IAqCP;;;;OAIG;IACH,kBAAkB,CAAC,aAAa,EAAE,OAAO,GAAG,IAAI;IAahD;;OAEG;IACH,OAAO,IAAI,IAAI;IAWf;;;;;OAKG;IACG,QAAQ,CAAC,aAAa,EAAE,OAAO,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAIlE;;;;;OAKG;IACH,cAAc,CAAC,aAAa,EAAE,OAAO,GAAG,sBAAsB,GAAG,SAAS;IAI1E;;;;OAIG;IACH,wBAAwB,IAAI,OAAO,EAAE;IAIrC;;OAEG;IACH,SAAS,IAAI,iBAAiB;IAQ9B;;OAEG;YACW,cAAc;IAgF5B;;OAEG;IACH,OAAO,CAAC,oBAAoB;CAqB7B"}

package/dist/monitor/health-check-service.js

@@ -0,0 +1,240 @@
+/**
+ * HealthCheckService - Per-coordinator health check timer service
+ *
+ * Runs periodic health checks for worker agents under each coordinator:
+ * - Detects stalled agents (no activity + not processing)
+ * - Tracks consecutive failures before escalating
+ * - Emits STALE_AGENT signals for coordinator remediation
+ * - Cleans up zombie sessions periodically
+ *
+ * @module monitor/health-check-service
+ * @see s-5yhx Phase B: Monitor Active Behaviors
+ */
+import { StallDetector } from "./stall-detector.js";
+/**
+ * Default configuration values
+ */
+export const DEFAULT_HEALTH_CHECK_CONFIG = {
+    intervalMs: 5 * 60 * 1000, // 5 minutes
+    stalledThresholdMs: 10 * 60 * 1000, // 10 minutes
+    guppThresholdMs: 30 * 60 * 1000, // 30 minutes
+    consecutiveFailuresBeforeEscalate: 3,
+};
+// =============================================================================
+// HealthCheckService Class
+// =============================================================================
+/**
+ * Service that runs periodic health checks for coordinator workers
+ */
+export class HealthCheckService {
+    eventStore;
+    sessionMapper;
+    messageRouter;
+    config;
+    stallDetector;
+    /** Timer handles per coordinator */
+    timers = new Map();
+    /** Health state per coordinator */
+    healthStates = new Map();
+    /** Last zombie cleanup time */
+    lastZombieCleanup = 0;
+    constructor(eventStore, sessionMapper, messageRouter, config = {}) {
+        this.eventStore = eventStore;
+        this.sessionMapper = sessionMapper;
+        this.messageRouter = messageRouter;
+        this.config = { ...DEFAULT_HEALTH_CHECK_CONFIG, ...config };
+        this.stallDetector = new StallDetector(eventStore, sessionMapper, {
+            stalledThresholdMs: this.config.stalledThresholdMs,
+        });
+    }
+    /**
+     * Start health check monitoring for a coordinator
+     *
+     * @param coordinatorId - The coordinator to monitor
+     * @param config - Optional config overrides for this coordinator
+     */
+    startForCoordinator(coordinatorId, config) {
+        // Stop existing timer if any
+        this.stopForCoordinator(coordinatorId);
+        const effectiveConfig = { ...this.config, ...config };
+        // Initialize health state
+        const now = Date.now();
+        this.healthStates.set(coordinatorId, {
+            coordinatorId,
+            workers: new Map(),
+            lastCheckAt: now,
+            nextCheckAt: now + effectiveConfig.intervalMs,
+            isRunning: true,
+        });
+        // Start periodic health check timer
+        const timer = setInterval(() => {
+            this.runHealthCheck(coordinatorId).catch((error) => {
+                console.error(`[HealthCheckService] Error running health check for ${coordinatorId}:`, error);
+            });
+        }, effectiveConfig.intervalMs);
+        this.timers.set(coordinatorId, timer);
+        // Run initial check immediately
+        this.runHealthCheck(coordinatorId).catch((error) => {
+            console.error(`[HealthCheckService] Error running initial health check for ${coordinatorId}:`, error);
+        });
+    }
+    /**
+     * Stop health check monitoring for a coordinator
+     *
+     * @param coordinatorId - The coordinator to stop monitoring
+     */
+    stopForCoordinator(coordinatorId) {
+        const timer = this.timers.get(coordinatorId);
+        if (timer) {
+            clearInterval(timer);
+            this.timers.delete(coordinatorId);
+        }
+        const state = this.healthStates.get(coordinatorId);
+        if (state) {
+            state.isRunning = false;
+        }
+    }
+    /**
+     * Stop all health check timers
+     */
+    stopAll() {
+        for (const [coordinatorId, timer] of this.timers) {
+            clearInterval(timer);
+            const state = this.healthStates.get(coordinatorId);
+            if (state) {
+                state.isRunning = false;
+            }
+        }
+        this.timers.clear();
+    }
+    /**
+     * Manually trigger a health check for a coordinator
+     *
+     * @param coordinatorId - The coordinator to check
+     * @returns Health check result
+     */
+    async checkNow(coordinatorId) {
+        return this.runHealthCheck(coordinatorId);
+    }
+    /**
+     * Get the current health state for a coordinator
+     *
+     * @param coordinatorId - The coordinator to get state for
+     * @returns Health state or undefined if not monitoring
+     */
+    getHealthState(coordinatorId) {
+        return this.healthStates.get(coordinatorId);
+    }
+    /**
+     * Get all monitored coordinators
+     *
+     * @returns Array of coordinator IDs being monitored
+     */
+    getMonitoredCoordinators() {
+        return Array.from(this.timers.keys());
+    }
+    /**
+     * Get the current configuration
+     */
+    getConfig() {
+        return { ...this.config };
+    }
+    // ===========================================================================
+    // Private Methods
+    // ===========================================================================
+    /**
+     * Run a health check for a coordinator
+     */
+    async runHealthCheck(coordinatorId) {
+        const now = Date.now();
+        const escalatedAgents = [];
+        // Get or create health state
+        let state = this.healthStates.get(coordinatorId);
+        if (!state) {
+            state = {
+                coordinatorId,
+                workers: new Map(),
+                lastCheckAt: now,
+                nextCheckAt: now + this.config.intervalMs,
+                isRunning: true,
+            };
+            this.healthStates.set(coordinatorId, state);
+        }
+        // Detect stalled agents
+        const stalledAgents = this.stallDetector.detectStalled(coordinatorId);
+        // Update worker health states and check for escalation
+        for (const stalled of stalledAgents) {
+            const workerState = state.workers.get(stalled.agentId) ?? {
+                agentId: stalled.agentId,
+                lastActivityAt: stalled.lastActivityAt,
+                consecutiveFailures: 0,
+                lastCheckAt: now,
+                status: "healthy",
+            };
+            // Increment failure counter
+            workerState.consecutiveFailures++;
+            workerState.lastCheckAt = now;
+            workerState.lastActivityAt = stalled.lastActivityAt;
+            // Determine status
+            if (workerState.consecutiveFailures >= this.config.consecutiveFailuresBeforeEscalate) {
+                workerState.status = "stalled";
+                escalatedAgents.push(stalled.agentId);
+                // Emit STALE_AGENT signal
+                this.emitStaleAgentSignal(stalled);
+            }
+            else {
+                workerState.status = "warning";
+            }
+            state.workers.set(stalled.agentId, workerState);
+        }
+        // Clear failure counters for healthy workers
+        const stalledIds = new Set(stalledAgents.map((s) => s.agentId));
+        for (const [workerId, workerState] of state.workers) {
+            if (!stalledIds.has(workerId)) {
+                workerState.consecutiveFailures = 0;
+                workerState.status = "healthy";
+                workerState.lastCheckAt = now;
+            }
+        }
+        // Periodically clean up zombies (every 5 checks)
+        let zombiesCleanedUp = 0;
+        if (now - this.lastZombieCleanup > this.config.intervalMs * 5) {
+            const zombies = this.stallDetector.cleanupZombies();
+            zombiesCleanedUp = zombies.length;
+            this.lastZombieCleanup = now;
+        }
+        // Update state
+        state.lastCheckAt = now;
+        state.nextCheckAt = now + this.config.intervalMs;
+        return {
+            coordinatorId,
+            checkedAt: now,
+            stalledAgents,
+            escalatedAgents,
+            zombiesCleanedUp,
+        };
+    }
+    /**
+     * Emit a STALE_AGENT signal via MessageRouter
+     */
+    emitStaleAgentSignal(stalled) {
+        try {
+            this.messageRouter.emitStatus({
+                from: { agent_id: stalled.coordinatorId },
+                status_type: "checkpoint",
+                summary: `Worker ${stalled.agentId} is stalled`,
+                details: {
+                    signal: "STALE_AGENT",
+                    workerId: stalled.agentId,
+                    taskId: stalled.assignedTaskId,
+                    lastActivity: stalled.lastActivityAt,
+                    durationMs: stalled.stalledDurationMs,
+                },
+            });
+        }
+        catch (error) {
+            console.error(`[HealthCheckService] Error emitting STALE_AGENT signal:`, error);
+        }
+    }
+}
+//# sourceMappingURL=health-check-service.js.map

package/dist/monitor/health-check-service.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"health-check-service.js","sourceRoot":"","sources":["../../src/monitor/health-check-service.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAMH,OAAO,EAAE,aAAa,EAAqB,MAAM,qBAAqB,CAAC;AAmCvE;;GAEG;AACH,MAAM,CAAC,MAAM,2BAA2B,GAAsB;IAC5D,UAAU,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI,EAAE,YAAY;IACvC,kBAAkB,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,EAAE,aAAa;IACjD,eAAe,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,EAAE,aAAa;IAC9C,iCAAiC,EAAE,CAAC;CACrC,CAAC;AAmCF,gFAAgF;AAChF,2BAA2B;AAC3B,gFAAgF;AAEhF;;GAEG;AACH,MAAM,OAAO,kBAAkB;IACZ,UAAU,CAAa;IACvB,aAAa,CAAgB;IAC7B,aAAa,CAAgB;IAC7B,MAAM,CAAoB;IAC1B,aAAa,CAAgB;IAE9C,oCAAoC;IAC5B,MAAM,GAAiD,IAAI,GAAG,EAAE,CAAC;IAEzE,mCAAmC;IAC3B,YAAY,GAAyC,IAAI,GAAG,EAAE,CAAC;IAEvE,+BAA+B;IACvB,iBAAiB,GAAc,CAAC,CAAC;IAEzC,YACE,UAAsB,EACtB,aAA4B,EAC5B,aAA4B,EAC5B,SAAqC,EAAE;QAEvC,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,MAAM,GAAG,EAAE,GAAG,2BAA2B,EAAE,GAAG,MAAM,EAAE,CAAC;QAC5D,IAAI,CAAC,aAAa,GAAG,IAAI,aAAa,CAAC,UAAU,EAAE,aAAa,EAAE;YAChE,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAkB;SACnD,CAAC,CAAC;IACL,CAAC;IAED;;;;;OAKG;IACH,mBAAmB,CACjB,aAAsB,EACtB,MAAmC;QAEnC,6BAA6B;QAC7B,IAAI,CAAC,kBAAkB,CAAC,aAAa,CAAC,CAAC;QAEvC,MAAM,eAAe,GAAG,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,MAAM,EAAE,CAAC;QAEtD,0BAA0B;QAC1B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,aAAa,EAAE;YACnC,aAAa;YACb,OAAO,EAAE,IAAI,GAAG,EAAE;YAClB,WAAW,EAAE,GAAG;YAChB,WAAW,EAAE,GAAG,GAAG,eAAe,CAAC,UAAU;YAC7C,SAAS,EAAE,IAAI;SAChB,CAAC,CAAC;QAEH,oCAAoC;QACpC,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,EAAE;YAC7B,IAAI,CAAC,cAAc,CAAC,aAAa,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE;gBACjD,OAAO,CAAC,KAAK,CACX,uDAAuD,aAAa,GAAG,EACvE,KAAK,CACN,CAAC;YACJ,CAAC,CAAC,CAAC;QACL,CAAC,EAAE,eAAe,CAAC,UAAU,CAAC,CAAC;QAE/B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAEtC,gCAAgC;QAChC,IAAI,CAAC,cAAc,CAAC,aAAa,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE;YACjD,OAAO,CAAC,KAAK,CACX,+DAA+D,aAAa,GAAG,EAC/E,KAAK,CACN,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC;IAED;;;;OAIG;IACH,kBAAkB,CAAC,aAAsB;QACvC,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QAC7C,IAAI,KAAK,EAAE,CAAC;YACV,aAAa,CAAC,KAAK,CAAC,CAAC;YACrB,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC;QACpC,CAAC;QAED,MAAM,KAAK,GAAG,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QACnD,IAAI,KAAK,EAAE,CAAC;YACV,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC;QAC1B,CAAC;IACH,CAAC;IAED;;OAEG;IACH,OAAO;QACL,KAAK,MAAM,CAAC,aAAa,EAAE,KAAK,CAAC,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YACjD,aAAa,CAAC,KAAK,CAAC,CAAC;YACrB,MAAM,KAAK,GAAG,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;YACnD,IAAI,KAAK,EAAE,CAAC;gBACV,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC;YAC1B,CAAC;QACH,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;IACtB,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,QAAQ,CAAC,aAAsB;QACnC,OAAO,IAAI,CAAC,cAAc,CAAC,aAAa,CAAC,CAAC;IAC5C,CAAC;IAED;;;;;OAKG;IACH,cAAc,CAAC,aAAsB;QACnC,OAAO,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;IAC9C,CAAC;IAED;;;;OAIG;IACH,wBAAwB;QACtB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IACxC,CAAC;IAED;;OAEG;IACH,SAAS;QACP,OAAO,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;IAC5B,CAAC;IAED,8EAA8E;IAC9E,kBAAkB;IAClB,8EAA8E;IAE9E;;OAEG;IACK,KAAK,CAAC,cAAc,CAAC,aAAsB;QACjD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,MAAM,eAAe,GAAc,EAAE,CAAC;QAEtC,6BAA6B;QAC7B,IAAI,KAAK,GAAG,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QACjD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,KAAK,GAAG;gBACN,aAAa;gBACb,OAAO,EAAE,IAAI,GAAG,EAAE;gBAClB,WAAW,EAAE,GAAG;gBAChB,WAAW,EAAE,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU;gBACzC,SAAS,EAAE,IAAI;aAChB,CAAC;YACF,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAC9C,CAAC;QAED,wBAAwB;QACxB,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC,aAAa,CAAC,aAAa,CAAC,CAAC;QAEtE,uDAAuD;QACvD,KAAK,MAAM,OAAO,IAAI,aAAa,EAAE,CAAC;YACpC,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI;gBACxD,OAAO,EAAE,OAAO,CAAC,OAAO;gBACxB,cAAc,EAAE,OAAO,CAAC,cAAc;gBACtC,mBAAmB,EAAE,CAAC;gBACtB,WAAW,EAAE,GAAG;gBAChB,MAAM,EAAE,SAAkB;aAC3B,CAAC;YAEF,4BAA4B;YAC5B,WAAW,CAAC,mBAAmB,EAAE,CAAC;YAClC,WAAW,CAAC,WAAW,GAAG,GAAG,CAAC;YAC9B,WAAW,CAAC,cAAc,GAAG,OAAO,CAAC,cAAc,CAAC;YAEpD,mBAAmB;YACnB,IAAI,WAAW,CAAC,mBAAmB,IAAI,IAAI,CAAC,MAAM,CAAC,iCAAiC,EAAE,CAAC;gBACrF,WAAW,CAAC,MAAM,GAAG,SAAS,CAAC;gBAC/B,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;gBAEtC,0BAA0B;gBAC1B,IAAI,CAAC,oBAAoB,CAAC,OAAO,CAAC,CAAC;YACrC,CAAC;iBAAM,CAAC;gBACN,WAAW,CAAC,MAAM,GAAG,SAAS,CAAC;YACjC,CAAC;YAED,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;QAClD,CAAC;QAED,6CAA6C;QAC7C,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QAChE,KAAK,MAAM,CAAC,QAAQ,EAAE,WAAW,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YACpD,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC9B,WAAW,CAAC,mBAAmB,GAAG,CAAC,CAAC;gBACpC,WAAW,CAAC,MAAM,GAAG,SAAS,CAAC;gBAC/B,WAAW,CAAC,WAAW,GAAG,GAAG,CAAC;YAChC,CAAC;QACH,CAAC;QAED,iDAAiD;QACjD,IAAI,gBAAgB,GAAG,CAAC,CAAC;QACzB,IAAI,GAAG,GAAG,IAAI,CAAC,iBAAiB,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC;YAC9D,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,CAAC,cAAc,EAAE,CAAC;YACpD,gBAAgB,GAAG,OAAO,CAAC,MAAM,CAAC;YAClC,IAAI,CAAC,iBAAiB,GAAG,GAAG,CAAC;QAC/B,CAAC;QAED,eAAe;QACf,KAAK,CAAC,WAAW,GAAG,GAAG,CAAC;QACxB,KAAK,CAAC,WAAW,GAAG,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC;QAEjD,OAAO;YACL,aAAa;YACb,SAAS,EAAE,GAAG;YACd,aAAa;YACb,eAAe;YACf,gBAAgB;SACjB,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,oBAAoB,CAAC,OAAqB;QAChD,IAAI,CAAC;YACH,IAAI,CAAC,aAAa,CAAC,UAAU,CAAC;gBAC5B,IAAI,EAAE,EAAE,QAAQ,EAAE,OAAO,CAAC,aAAa,EAAE;gBACzC,WAAW,EAAE,YAAY;gBACzB,OAAO,EAAE,UAAU,OAAO,CAAC,OAAO,aAAa;gBAC/C,OAAO,EAAE;oBACP,MAAM,EAAE,aAAa;oBACrB,QAAQ,EAAE,OAAO,CAAC,OAAO;oBACzB,MAAM,EAAE,OAAO,CAAC,cAAc;oBAC9B,YAAY,EAAE,OAAO,CAAC,cAAc;oBACpC,UAAU,EAAE,OAAO,CAAC,iBAAiB;iBACtC;aACF,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CACX,yDAAyD,EACzD,KAAK,CACN,CAAC;QACJ,CAAC;IACH,CAAC;CACF"}

package/dist/monitor/index.d.ts

@@ -0,0 +1,14 @@
+/**
+ * Monitor module - Health monitoring for multi-agent orchestration
+ *
+ * Provides:
+ * - Stall detection for inactive agents
+ * - Zombie cleanup for orphaned sessions
+ * - Health check service for periodic monitoring
+ *
+ * @module monitor
+ * @see s-5yhx Phase B: Monitor Active Behaviors
+ */
+export { StallDetector, type StalledAgent, type ZombieAgent, type StallDetectorConfig, DEFAULT_STALL_DETECTOR_CONFIG, } from "./stall-detector.js";
+export { HealthCheckService, type HealthCheckConfig, type HealthCheckResult, type CoordinatorHealthState, type WorkerHealthState, DEFAULT_HEALTH_CHECK_CONFIG, } from "./health-check-service.js";
+//# sourceMappingURL=index.d.ts.map

package/dist/monitor/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/monitor/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EACL,aAAa,EACb,KAAK,YAAY,EACjB,KAAK,WAAW,EAChB,KAAK,mBAAmB,EACxB,6BAA6B,GAC9B,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EACL,kBAAkB,EAClB,KAAK,iBAAiB,EACtB,KAAK,iBAAiB,EACtB,KAAK,sBAAsB,EAC3B,KAAK,iBAAiB,EACtB,2BAA2B,GAC5B,MAAM,2BAA2B,CAAC"}

package/dist/monitor/index.js

@@ -0,0 +1,14 @@
+/**
+ * Monitor module - Health monitoring for multi-agent orchestration
+ *
+ * Provides:
+ * - Stall detection for inactive agents
+ * - Zombie cleanup for orphaned sessions
+ * - Health check service for periodic monitoring
+ *
+ * @module monitor
+ * @see s-5yhx Phase B: Monitor Active Behaviors
+ */
+export { StallDetector, DEFAULT_STALL_DETECTOR_CONFIG, } from "./stall-detector.js";
+export { HealthCheckService, DEFAULT_HEALTH_CHECK_CONFIG, } from "./health-check-service.js";
+//# sourceMappingURL=index.js.map

package/dist/monitor/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/monitor/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EACL,aAAa,EAIb,6BAA6B,GAC9B,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EACL,kBAAkB,EAKlB,2BAA2B,GAC5B,MAAM,2BAA2B,CAAC"}

package/dist/monitor/stall-detector.d.ts

@@ -0,0 +1,109 @@
+/**
+ * StallDetector - Detects stalled agents and cleans up zombies
+ *
+ * A stalled agent is one that:
+ * - Has no recent activity (lastActivityAt is stale)
+ * - Is not actively processing a prompt (isProcessing is false)
+ *
+ * A zombie agent is one that:
+ * - Has state 'stopped' in EventStore
+ * - But still has a session mapping in SessionMapper
+ *
+ * @module monitor/stall-detector
+ * @see s-5yhx Phase B: Monitor Active Behaviors
+ */
+import type { EventStore } from "../store/event-store.js";
+import type { SessionMapper } from "../acp/session-mapper.js";
+import type { AgentId, TaskId, Timestamp } from "../store/types/index.js";
+/**
+ * Information about a stalled agent
+ */
+export interface StalledAgent {
+    /** The stalled agent's ID */
+    agentId: AgentId;
+    /** The coordinator this worker belongs to */
+    coordinatorId: AgentId;
+    /** Last time the agent had activity */
+    lastActivityAt: Timestamp;
+    /** How long the agent has been stalled (ms) */
+    stalledDurationMs: number;
+    /** Task the agent was assigned to (if any) */
+    assignedTaskId?: TaskId;
+    /** Session processing status */
+    sessionStatus: "processing" | "idle" | "unknown";
+}
+/**
+ * Information about a zombie agent
+ */
+export interface ZombieAgent {
+    /** The zombie agent's ID */
+    agentId: AgentId;
+    /** When the agent stopped */
+    stoppedAt: Timestamp;
+    /** The orphaned ACP session ID */
+    acpSessionId: string;
+    /** Why the agent stopped */
+    reason: "done_called" | "parent_terminated" | "unknown";
+}
+/**
+ * Configuration for stall detection
+ */
+export interface StallDetectorConfig {
+    /**
+     * How long without activity before an agent is considered stalled (ms)
+     * Default: 600000 (10 minutes)
+     */
+    stalledThresholdMs: number;
+}
+/**
+ * Default configuration values
+ */
+export declare const DEFAULT_STALL_DETECTOR_CONFIG: StallDetectorConfig;
+/**
+ * Detects stalled agents and cleans up zombie sessions
+ */
+export declare class StallDetector {
+    private readonly eventStore;
+    private readonly sessionMapper;
+    private readonly config;
+    constructor(eventStore: EventStore, sessionMapper: SessionMapper, config?: Partial<StallDetectorConfig>);
+    /**
+     * Detect stalled workers under a coordinator
+     *
+     * A worker is considered stalled if:
+     * 1. It's in 'running' state
+     * 2. Its lastActivityAt is older than stalledThresholdMs
+     * 3. Its session is NOT currently processing a prompt
+     *
+     * @param coordinatorId - The coordinator whose workers to check
+     * @returns Array of stalled agents
+     */
+    detectStalled(coordinatorId: AgentId): StalledAgent[];
+    /**
+     * Detect all stalled agents across all coordinators
+     *
+     * @returns Array of all stalled agents
+     */
+    detectAllStalled(): StalledAgent[];
+    /**
+     * Detect and clean up zombie agents
+     *
+     * A zombie is an agent that:
+     * - Is in 'stopped' state in EventStore
+     * - Still has a session mapping in SessionMapper
+     *
+     * This can happen if an agent terminates but the session cleanup fails.
+     *
+     * @returns Array of cleaned up zombie agents
+     */
+    cleanupZombies(): ZombieAgent[];
+    /**
+     * Get the current configuration
+     */
+    getConfig(): StallDetectorConfig;
+    /**
+     * Update the configuration
+     */
+    updateConfig(config: Partial<StallDetectorConfig>): void;
+}
+//# sourceMappingURL=stall-detector.d.ts.map

package/dist/monitor/stall-detector.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stall-detector.d.ts","sourceRoot":"","sources":["../../src/monitor/stall-detector.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,yBAAyB,CAAC;AAM1E;;GAEG;AACH,MAAM,WAAW,YAAY;IAC3B,6BAA6B;IAC7B,OAAO,EAAE,OAAO,CAAC;IAEjB,6CAA6C;IAC7C,aAAa,EAAE,OAAO,CAAC;IAEvB,uCAAuC;IACvC,cAAc,EAAE,SAAS,CAAC;IAE1B,+CAA+C;IAC/C,iBAAiB,EAAE,MAAM,CAAC;IAE1B,8CAA8C;IAC9C,cAAc,CAAC,EAAE,MAAM,CAAC;IAExB,gCAAgC;IAChC,aAAa,EAAE,YAAY,GAAG,MAAM,GAAG,SAAS,CAAC;CAClD;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,4BAA4B;IAC5B,OAAO,EAAE,OAAO,CAAC;IAEjB,6BAA6B;IAC7B,SAAS,EAAE,SAAS,CAAC;IAErB,kCAAkC;IAClC,YAAY,EAAE,MAAM,CAAC;IAErB,4BAA4B;IAC5B,MAAM,EAAE,aAAa,GAAG,mBAAmB,GAAG,SAAS,CAAC;CACzD;AAED;;GAEG;AACH,MAAM,WAAW,mBAAmB;IAClC;;;OAGG;IACH,kBAAkB,EAAE,MAAM,CAAC;CAC5B;AAED;;GAEG;AACH,eAAO,MAAM,6BAA6B,EAAE,mBAE3C,CAAC;AAMF;;GAEG;AACH,qBAAa,aAAa;IACxB,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAa;IACxC,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAgB;IAC9C,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAsB;gBAG3C,UAAU,EAAE,UAAU,EACtB,aAAa,EAAE,aAAa,EAC5B,MAAM,GAAE,OAAO,CAAC,mBAAmB,CAAM;IAO3C;;;;;;;;;;OAUG;IACH,aAAa,CAAC,aAAa,EAAE,OAAO,GAAG,YAAY,EAAE;IAqCrD;;;;OAIG;IACH,gBAAgB,IAAI,YAAY,EAAE;IAiBlC;;;;;;;;;;OAUG;IACH,cAAc,IAAI,WAAW,EAAE;IAwC/B;;OAEG;IACH,SAAS,IAAI,mBAAmB;IAIhC;;OAEG;IACH,YAAY,CAAC,MAAM,EAAE,OAAO,CAAC,mBAAmB,CAAC,GAAG,IAAI;CAGzD"}