agent-relay 1.0.21 → 1.1.0

package/dist/resiliency/health-monitor.js

@@ -0,0 +1,291 @@
+/**
+ * Agent Health Monitor
+ *
+ * Monitors spawned agent processes and ensures they stay alive.
+ * - Periodic health checks (process liveness)
+ * - Auto-restart on crash
+ * - Death detection and logging
+ * - Metrics collection
+ */
+import { EventEmitter } from 'events';
+const DEFAULT_CONFIG = {
+    checkIntervalMs: 5000,
+    responseTimeoutMs: 10000,
+    maxRestarts: 5,
+    restartCooldownMs: 2000,
+    maxConsecutiveFailures: 3,
+};
+export class AgentHealthMonitor extends EventEmitter {
+    agents = new Map();
+    health = new Map();
+    intervalId;
+    config;
+    isRunning = false;
+    constructor(config = {}) {
+        super();
+        this.config = { ...DEFAULT_CONFIG, ...config };
+    }
+    /**
+     * Register an agent for health monitoring
+     */
+    register(agent) {
+        this.agents.set(agent.name, agent);
+        this.health.set(agent.name, {
+            name: agent.name,
+            pid: agent.pid,
+            status: 'healthy',
+            lastHealthCheck: new Date(),
+            lastResponse: new Date(),
+            restartCount: 0,
+            consecutiveFailures: 0,
+            uptime: 0,
+            startedAt: new Date(),
+        });
+        this.emit('registered', { name: agent.name, pid: agent.pid });
+        this.log('info', `Registered agent for health monitoring: ${agent.name} (PID: ${agent.pid})`);
+    }
+    /**
+     * Unregister an agent from health monitoring
+     */
+    unregister(name) {
+        this.agents.delete(name);
+        this.health.delete(name);
+        this.emit('unregistered', { name });
+        this.log('info', `Unregistered agent: ${name}`);
+    }
+    /**
+     * Start the health monitoring loop
+     */
+    start() {
+        if (this.isRunning)
+            return;
+        this.isRunning = true;
+        this.log('info', 'Health monitor started', {
+            checkInterval: this.config.checkIntervalMs,
+            maxRestarts: this.config.maxRestarts,
+        });
+        this.intervalId = setInterval(() => {
+            this.checkAll();
+        }, this.config.checkIntervalMs);
+        // Initial check
+        this.checkAll();
+    }
+    /**
+     * Stop the health monitoring loop
+     */
+    stop() {
+        if (this.intervalId) {
+            clearInterval(this.intervalId);
+            this.intervalId = undefined;
+        }
+        this.isRunning = false;
+        this.log('info', 'Health monitor stopped');
+    }
+    /**
+     * Get health status for all agents
+     */
+    getAll() {
+        return Array.from(this.health.values()).map((h) => ({
+            ...h,
+            uptime: Date.now() - h.startedAt.getTime(),
+        }));
+    }
+    /**
+     * Get health status for a specific agent
+     */
+    get(name) {
+        const health = this.health.get(name);
+        if (health) {
+            return { ...health, uptime: Date.now() - health.startedAt.getTime() };
+        }
+        return undefined;
+    }
+    /**
+     * Check health of all registered agents
+     */
+    async checkAll() {
+        const checks = Array.from(this.agents.entries()).map(([name, agent]) => this.checkAgent(name, agent));
+        await Promise.all(checks);
+    }
+    /**
+     * Check health of a single agent
+     */
+    async checkAgent(name, agent) {
+        const health = this.health.get(name);
+        if (!health)
+            return;
+        health.lastHealthCheck = new Date();
+        try {
+            // First check: Is the process alive?
+            const isAlive = this.isProcessAlive(agent.pid);
+            if (!isAlive) {
+                await this.handleDeath(name, agent, health, 'Process not found');
+                return;
+            }
+            // Second check: Does it respond to health check?
+            if (agent.sendHealthCheck) {
+                const responded = await Promise.race([
+                    agent.sendHealthCheck(),
+                    new Promise((resolve) => setTimeout(() => resolve(false), this.config.responseTimeoutMs)),
+                ]);
+                if (!responded) {
+                    health.consecutiveFailures++;
+                    this.log('warn', `Agent unresponsive: ${name}`, {
+                        failures: health.consecutiveFailures,
+                        max: this.config.maxConsecutiveFailures,
+                    });
+                    if (health.consecutiveFailures >= this.config.maxConsecutiveFailures) {
+                        health.status = 'unresponsive';
+                        await this.handleDeath(name, agent, health, 'Unresponsive after multiple health checks');
+                    }
+                    else {
+                        health.status = 'unresponsive';
+                        this.emit('unhealthy', { name, health });
+                    }
+                    return;
+                }
+            }
+            // Get memory/CPU usage if available
+            try {
+                const usage = await this.getProcessUsage(agent.pid);
+                health.memoryUsage = usage.memory;
+                health.cpuUsage = usage.cpu;
+            }
+            catch {
+                // Ignore usage errors
+            }
+            // All good
+            health.status = 'healthy';
+            health.lastResponse = new Date();
+            health.consecutiveFailures = 0;
+            this.emit('healthy', { name, health });
+        }
+        catch (error) {
+            health.consecutiveFailures++;
+            health.lastError = error instanceof Error ? error.message : String(error);
+            this.log('error', `Health check error for ${name}`, { error: health.lastError });
+            if (health.consecutiveFailures >= this.config.maxConsecutiveFailures) {
+                await this.handleDeath(name, agent, health, health.lastError);
+            }
+        }
+    }
+    /**
+     * Handle agent death - attempt restart or mark as dead
+     */
+    async handleDeath(name, agent, health, reason) {
+        this.log('error', `Agent died: ${name}`, {
+            reason,
+            restartCount: health.restartCount,
+            maxRestarts: this.config.maxRestarts,
+        });
+        this.emit('died', { name, reason, restartCount: health.restartCount });
+        // Check if we should attempt restart
+        if (health.restartCount >= this.config.maxRestarts) {
+            health.status = 'dead';
+            health.lastError = `Exceeded max restarts (${this.config.maxRestarts}): ${reason}`;
+            this.log('error', `Agent permanently dead: ${name}`, { reason: health.lastError });
+            this.emit('permanentlyDead', { name, health });
+            return;
+        }
+        // Attempt restart
+        health.status = 'restarting';
+        health.restartCount++;
+        this.log('info', `Attempting restart ${health.restartCount}/${this.config.maxRestarts}: ${name}`);
+        this.emit('restarting', { name, attempt: health.restartCount });
+        // Wait cooldown
+        await new Promise((resolve) => setTimeout(resolve, this.config.restartCooldownMs));
+        try {
+            await agent.restart();
+            // Update health after successful restart
+            health.status = 'healthy';
+            health.consecutiveFailures = 0;
+            health.startedAt = new Date();
+            health.lastResponse = new Date();
+            health.pid = agent.pid;
+            this.log('info', `Agent restarted successfully: ${name}`, {
+                newPid: agent.pid,
+                attempt: health.restartCount,
+            });
+            this.emit('restarted', { name, pid: agent.pid, attempt: health.restartCount });
+        }
+        catch (error) {
+            health.lastError = error instanceof Error ? error.message : String(error);
+            this.log('error', `Restart failed: ${name}`, { error: health.lastError });
+            this.emit('restartFailed', { name, error: health.lastError });
+            // Recursively try again if under limit
+            if (health.restartCount < this.config.maxRestarts) {
+                await this.handleDeath(name, agent, health, health.lastError);
+            }
+            else {
+                health.status = 'dead';
+                this.emit('permanentlyDead', { name, health });
+            }
+        }
+    }
+    /**
+     * Check if a process is alive by PID
+     */
+    isProcessAlive(pid) {
+        try {
+            // Sending signal 0 checks if process exists without killing it
+            process.kill(pid, 0);
+            return true;
+        }
+        catch {
+            return false;
+        }
+    }
+    /**
+     * Get memory and CPU usage for a process
+     */
+    async getProcessUsage(pid) {
+        const { execSync } = await import('child_process');
+        try {
+            // This works on Linux/Mac
+            const output = execSync(`ps -o rss=,pcpu= -p ${pid}`, { encoding: 'utf8' }).trim();
+            const [rss, cpu] = output.split(/\s+/);
+            return {
+                memory: parseInt(rss, 10) * 1024, // RSS in bytes
+                cpu: parseFloat(cpu),
+            };
+        }
+        catch {
+            return { memory: 0, cpu: 0 };
+        }
+    }
+    /**
+     * Structured logging
+     */
+    log(level, message, context) {
+        const entry = {
+            timestamp: new Date().toISOString(),
+            level,
+            component: 'health-monitor',
+            message,
+            ...context,
+        };
+        this.emit('log', entry);
+        // Also log to console with structure
+        const prefix = `[health-monitor]`;
+        switch (level) {
+            case 'info':
+                console.log(prefix, message, context ? JSON.stringify(context) : '');
+                break;
+            case 'warn':
+                console.warn(prefix, message, context ? JSON.stringify(context) : '');
+                break;
+            case 'error':
+                console.error(prefix, message, context ? JSON.stringify(context) : '');
+                break;
+        }
+    }
+}
+// Singleton instance
+let _monitor = null;
+export function getHealthMonitor(config) {
+    if (!_monitor) {
+        _monitor = new AgentHealthMonitor(config);
+    }
+    return _monitor;
+}
+//# sourceMappingURL=health-monitor.js.map

package/dist/resiliency/health-monitor.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"health-monitor.js","sourceRoot":"","sources":["../../src/resiliency/health-monitor.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,QAAQ,CAAC;AAkCtC,MAAM,cAAc,GAAwB;IAC1C,eAAe,EAAE,IAAI;IACrB,iBAAiB,EAAE,KAAK;IACxB,WAAW,EAAE,CAAC;IACd,iBAAiB,EAAE,IAAI;IACvB,sBAAsB,EAAE,CAAC;CAC1B,CAAC;AAEF,MAAM,OAAO,kBAAmB,SAAQ,YAAY;IAC1C,MAAM,GAAG,IAAI,GAAG,EAAwB,CAAC;IACzC,MAAM,GAAG,IAAI,GAAG,EAAuB,CAAC;IACxC,UAAU,CAAkC;IAC5C,MAAM,CAAsB;IAC5B,SAAS,GAAG,KAAK,CAAC;IAE1B,YAAY,SAAuC,EAAE;QACnD,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,MAAM,GAAG,EAAE,GAAG,cAAc,EAAE,GAAG,MAAM,EAAE,CAAC;IACjD,CAAC;IAED;;OAEG;IACH,QAAQ,CAAC,KAAmB;QAC1B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACnC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE;YAC1B,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,MAAM,EAAE,SAAS;YACjB,eAAe,EAAE,IAAI,IAAI,EAAE;YAC3B,YAAY,EAAE,IAAI,IAAI,EAAE;YACxB,YAAY,EAAE,CAAC;YACf,mBAAmB,EAAE,CAAC;YACtB,MAAM,EAAE,CAAC;YACT,SAAS,EAAE,IAAI,IAAI,EAAE;SACtB,CAAC,CAAC;QAEH,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC;QAC9D,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,2CAA2C,KAAK,CAAC,IAAI,UAAU,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC;IAChG,CAAC;IAED;;OAEG;IACH,UAAU,CAAC,IAAY;QACrB,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QACzB,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QACzB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC;QACpC,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,uBAAuB,IAAI,EAAE,CAAC,CAAC;IAClD,CAAC;IAED;;OAEG;IACH,KAAK;QACH,IAAI,IAAI,CAAC,SAAS;YAAE,OAAO;QAC3B,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC;QAEtB,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,wBAAwB,EAAE;YACzC,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,eAAe;YAC1C,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,WAAW;SACrC,CAAC,CAAC;QAEH,IAAI,CAAC,UAAU,GAAG,WAAW,CAAC,GAAG,EAAE;YACjC,IAAI,CAAC,QAAQ,EAAE,CAAC;QAClB,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;QAEhC,gBAAgB;QAChB,IAAI,CAAC,QAAQ,EAAE,CAAC;IAClB,CAAC;IAED;;OAEG;IACH,IAAI;QACF,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;YACpB,aAAa,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;YAC/B,IAAI,CAAC,UAAU,GAAG,SAAS,CAAC;QAC9B,CAAC;QACD,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC;QACvB,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,wBAAwB,CAAC,CAAC;IAC7C,CAAC;IAED;;OAEG;IACH,MAAM;QACJ,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAClD,GAAG,CAAC;YACJ,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,SAAS,CAAC,OAAO,EAAE;SAC3C,CAAC,CAAC,CAAC;IACN,CAAC;IAED;;OAEG;IACH,GAAG,CAAC,IAAY;QACd,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACrC,IAAI,MAAM,EAAE,CAAC;YACX,OAAO,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,MAAM,CAAC,SAAS,CAAC,OAAO,EAAE,EAAE,CAAC;QACxE,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,QAAQ;QACpB,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,EAAE,CACrE,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,KAAK,CAAC,CAC7B,CAAC;QACF,MAAM,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,UAAU,CAAC,IAAY,EAAE,KAAmB;QACxD,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACrC,IAAI,CAAC,MAAM;YAAE,OAAO;QAEpB,MAAM,CAAC,eAAe,GAAG,IAAI,IAAI,EAAE,CAAC;QAEpC,IAAI,CAAC;YACH,qCAAqC;YACrC,MAAM,OAAO,GAAG,IAAI,CAAC,cAAc,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YAE/C,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,mBAAmB,CAAC,CAAC;gBACjE,OAAO;YACT,CAAC;YAED,iDAAiD;YACjD,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC;gBAC1B,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC;oBACnC,KAAK,CAAC,eAAe,EAAE;oBACvB,IAAI,OAAO,CAAQ,CAAC,OAAO,EAAE,EAAE,CAC7B,UAAU,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAChE;iBACF,CAAC,CAAC;gBAEH,IAAI,CAAC,SAAS,EAAE,CAAC;oBACf,MAAM,CAAC,mBAAmB,EAAE,CAAC;oBAC7B,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,uBAAuB,IAAI,EAAE,EAAE;wBAC9C,QAAQ,EAAE,MAAM,CAAC,mBAAmB;wBACpC,GAAG,EAAE,IAAI,CAAC,MAAM,CAAC,sBAAsB;qBACxC,CAAC,CAAC;oBAEH,IAAI,MAAM,CAAC,mBAAmB,IAAI,IAAI,CAAC,MAAM,CAAC,sBAAsB,EAAE,CAAC;wBACrE,MAAM,CAAC,MAAM,GAAG,cAAc,CAAC;wBAC/B,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,2CAA2C,CAAC,CAAC;oBAC3F,CAAC;yBAAM,CAAC;wBACN,MAAM,CAAC,MAAM,GAAG,cAAc,CAAC;wBAC/B,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;oBAC3C,CAAC;oBACD,OAAO;gBACT,CAAC;YACH,CAAC;YAED,oCAAoC;YACpC,IAAI,CAAC;gBACH,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;gBACpD,MAAM,CAAC,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC;gBAClC,MAAM,CAAC,QAAQ,GAAG,KAAK,CAAC,GAAG,CAAC;YAC9B,CAAC;YAAC,MAAM,CAAC;gBACP,sBAAsB;YACxB,CAAC;YAED,WAAW;YACX,MAAM,CAAC,MAAM,GAAG,SAAS,CAAC;YAC1B,MAAM,CAAC,YAAY,GAAG,IAAI,IAAI,EAAE,CAAC;YACjC,MAAM,CAAC,mBAAmB,GAAG,CAAC,CAAC;YAE/B,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;QACzC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,CAAC,mBAAmB,EAAE,CAAC;YAC7B,MAAM,CAAC,SAAS,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC1E,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,0BAA0B,IAAI,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YAEjF,IAAI,MAAM,CAAC,mBAAmB,IAAI,IAAI,CAAC,MAAM,CAAC,sBAAsB,EAAE,CAAC;gBACrE,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;YAChE,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,WAAW,CACvB,IAAY,EACZ,KAAmB,EACnB,MAAmB,EACnB,MAAc;QAEd,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,eAAe,IAAI,EAAE,EAAE;YACvC,MAAM;YACN,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,WAAW;SACrC,CAAC,CAAC;QAEH,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC;QAEvE,qCAAqC;QACrC,IAAI,MAAM,CAAC,YAAY,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;YACnD,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC;YACvB,MAAM,CAAC,SAAS,GAAG,0BAA0B,IAAI,CAAC,MAAM,CAAC,WAAW,MAAM,MAAM,EAAE,CAAC;YACnF,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,2BAA2B,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YACnF,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;YAC/C,OAAO;QACT,CAAC;QAED,kBAAkB;QAClB,MAAM,CAAC,MAAM,GAAG,YAAY,CAAC;QAC7B,MAAM,CAAC,YAAY,EAAE,CAAC;QAEtB,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,sBAAsB,MAAM,CAAC,YAAY,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW,KAAK,IAAI,EAAE,CAAC,CAAC;QAClG,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC;QAEhE,gBAAgB;QAChB,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC;QAEnF,IAAI,CAAC;YACH,MAAM,KAAK,CAAC,OAAO,EAAE,CAAC;YAEtB,yCAAyC;YACzC,MAAM,CAAC,MAAM,GAAG,SAAS,CAAC;YAC1B,MAAM,CAAC,mBAAmB,GAAG,CAAC,CAAC;YAC/B,MAAM,CAAC,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC;YAC9B,MAAM,CAAC,YAAY,GAAG,IAAI,IAAI,EAAE,CAAC;YACjC,MAAM,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC;YAEvB,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,iCAAiC,IAAI,EAAE,EAAE;gBACxD,MAAM,EAAE,KAAK,CAAC,GAAG;gBACjB,OAAO,EAAE,MAAM,CAAC,YAAY;aAC7B,CAAC,CAAC;YAEH,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,OAAO,EAAE,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC;QACjF,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,CAAC,SAAS,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC1E,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,mBAAmB,IAAI,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YAC1E,IAAI,CAAC,IAAI,CAAC,eAAe,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YAE9D,uCAAuC;YACvC,IAAI,MAAM,CAAC,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;gBAClD,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;YAChE,CAAC;iBAAM,CAAC;gBACN,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC;gBACvB,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;YACjD,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,GAAW;QAChC,IAAI,CAAC;YACH,+DAA+D;YAC/D,OAAO,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;YACrB,OAAO,IAAI,CAAC;QACd,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,eAAe,CAAC,GAAW;QACvC,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,CAAC;QAEnD,IAAI,CAAC;YACH,0BAA0B;YAC1B,MAAM,MAAM,GAAG,QAAQ,CAAC,uBAAuB,GAAG,EAAE,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;YACnF,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;YACvC,OAAO;gBACL,MAAM,EAAE,QAAQ,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,IAAI,EAAE,eAAe;gBACjD,GAAG,EAAE,UAAU,CAAC,GAAG,CAAC;aACrB,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,MAAM,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,CAAC;QAC/B,CAAC;IACH,CAAC;IAED;;OAEG;IACK,GAAG,CACT,KAAgC,EAChC,OAAe,EACf,OAAiC;QAEjC,MAAM,KAAK,GAAG;YACZ,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,KAAK;YACL,SAAS,EAAE,gBAAgB;YAC3B,OAAO;YACP,GAAG,OAAO;SACX,CAAC;QAEF,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;QAExB,qCAAqC;QACrC,MAAM,MAAM,GAAG,kBAAkB,CAAC;QAClC,QAAQ,KAAK,EAAE,CAAC;YACd,KAAK,MAAM;gBACT,OAAO,CAAC,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;gBACrE,MAAM;YACR,KAAK,MAAM;gBACT,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;gBACtE,MAAM;YACR,KAAK,OAAO;gBACV,OAAO,CAAC,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;gBACvE,MAAM;QACV,CAAC;IACH,CAAC;CACF;AAED,qBAAqB;AACrB,IAAI,QAAQ,GAA8B,IAAI,CAAC;AAE/C,MAAM,UAAU,gBAAgB,CAAC,MAAqC;IACpE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,QAAQ,GAAG,IAAI,kBAAkB,CAAC,MAAM,CAAC,CAAC;IAC5C,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC"}

package/dist/resiliency/index.d.ts

@@ -0,0 +1,63 @@
+/**
+ * Agent Resiliency Module
+ *
+ * Provides comprehensive health monitoring, auto-restart, logging,
+ * metrics, and context persistence for agent-relay agents.
+ *
+ * Features:
+ * - Health monitoring with process liveness checks
+ * - Auto-restart on crash with configurable limits
+ * - Prometheus-compatible metrics export
+ * - Structured JSON logging with rotation
+ * - Context persistence across restarts (inspired by Continuous-Claude-v2)
+ * - Provider-specific context injection (Claude hooks, Codex config, Gemini instructions)
+ *
+ * Usage:
+ *
+ * ```ts
+ * import { getSupervisor, metrics, createLogger } from './resiliency';
+ *
+ * // Start the supervisor with context persistence
+ * const supervisor = getSupervisor({
+ *   autoRestart: true,
+ *   maxRestarts: 5,
+ *   contextPersistence: {
+ *     enabled: true,
+ *     autoInjectOnRestart: true,
+ *   },
+ * });
+ * supervisor.start();
+ *
+ * // Add an agent to supervision
+ * supervisor.supervise(
+ *   {
+ *     name: 'worker-1',
+ *     cli: 'claude',
+ *     pid: 12345,
+ *     spawnedAt: new Date(),
+ *     workingDir: '/path/to/repo',
+ *     provider: 'claude', // or 'codex', 'gemini'
+ *   },
+ *   {
+ *     isAlive: () => process.kill(12345, 0),
+ *     kill: (sig) => process.kill(12345, sig),
+ *     restart: async () => { ... },
+ *   }
+ * );
+ *
+ * // Get metrics
+ * console.log(metrics.toPrometheus());
+ * ```
+ *
+ * Context persistence works differently per provider:
+ * - Claude: Uses hooks to inject context into CLAUDE.md
+ * - Codex: Uses config for periodic context refresh via system prompt
+ * - Gemini: Updates system instruction file
+ */
+export { AgentHealthMonitor, getHealthMonitor, type AgentHealth, type AgentProcess, type HealthMonitorConfig, } from './health-monitor.js';
+export { Logger, createLogger, configure as configureLogging, loggers, type LogLevel, type LogEntry, type LoggerConfig, } from './logger.js';
+export { metrics, type AgentMetrics, type SystemMetrics, type MetricPoint } from './metrics.js';
+export { AgentSupervisor, getSupervisor, type SupervisedAgent, type SupervisorConfig, } from './supervisor.js';
+export { ContextPersistence, getContextPersistence, type AgentState, type Decision, type Artifact, type Handoff, type LedgerEntry, } from './context-persistence.js';
+export { createContextHandler, detectProvider, ClaudeContextHandler, CodexContextHandler, GeminiContextHandler, type ProviderType, type ProviderContextConfig, type ClaudeHooksConfig, type CodexContextConfig, } from './provider-context.js';
+//# sourceMappingURL=index.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/resiliency/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAuDG;AAEH,OAAO,EACL,kBAAkB,EAClB,gBAAgB,EAChB,KAAK,WAAW,EAChB,KAAK,YAAY,EACjB,KAAK,mBAAmB,GACzB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EACL,MAAM,EACN,YAAY,EACZ,SAAS,IAAI,gBAAgB,EAC7B,OAAO,EACP,KAAK,QAAQ,EACb,KAAK,QAAQ,EACb,KAAK,YAAY,GAClB,MAAM,aAAa,CAAC;AAErB,OAAO,EAAE,OAAO,EAAE,KAAK,YAAY,EAAE,KAAK,aAAa,EAAE,KAAK,WAAW,EAAE,MAAM,cAAc,CAAC;AAEhG,OAAO,EACL,eAAe,EACf,aAAa,EACb,KAAK,eAAe,EACpB,KAAK,gBAAgB,GACtB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,kBAAkB,EAClB,qBAAqB,EACrB,KAAK,UAAU,EACf,KAAK,QAAQ,EACb,KAAK,QAAQ,EACb,KAAK,OAAO,EACZ,KAAK,WAAW,GACjB,MAAM,0BAA0B,CAAC;AAElC,OAAO,EACL,oBAAoB,EACpB,cAAc,EACd,oBAAoB,EACpB,mBAAmB,EACnB,oBAAoB,EACpB,KAAK,YAAY,EACjB,KAAK,qBAAqB,EAC1B,KAAK,iBAAiB,EACtB,KAAK,kBAAkB,GACxB,MAAM,uBAAuB,CAAC"}

package/dist/resiliency/index.js

@@ -0,0 +1,63 @@
+/**
+ * Agent Resiliency Module
+ *
+ * Provides comprehensive health monitoring, auto-restart, logging,
+ * metrics, and context persistence for agent-relay agents.
+ *
+ * Features:
+ * - Health monitoring with process liveness checks
+ * - Auto-restart on crash with configurable limits
+ * - Prometheus-compatible metrics export
+ * - Structured JSON logging with rotation
+ * - Context persistence across restarts (inspired by Continuous-Claude-v2)
+ * - Provider-specific context injection (Claude hooks, Codex config, Gemini instructions)
+ *
+ * Usage:
+ *
+ * ```ts
+ * import { getSupervisor, metrics, createLogger } from './resiliency';
+ *
+ * // Start the supervisor with context persistence
+ * const supervisor = getSupervisor({
+ *   autoRestart: true,
+ *   maxRestarts: 5,
+ *   contextPersistence: {
+ *     enabled: true,
+ *     autoInjectOnRestart: true,
+ *   },
+ * });
+ * supervisor.start();
+ *
+ * // Add an agent to supervision
+ * supervisor.supervise(
+ *   {
+ *     name: 'worker-1',
+ *     cli: 'claude',
+ *     pid: 12345,
+ *     spawnedAt: new Date(),
+ *     workingDir: '/path/to/repo',
+ *     provider: 'claude', // or 'codex', 'gemini'
+ *   },
+ *   {
+ *     isAlive: () => process.kill(12345, 0),
+ *     kill: (sig) => process.kill(12345, sig),
+ *     restart: async () => { ... },
+ *   }
+ * );
+ *
+ * // Get metrics
+ * console.log(metrics.toPrometheus());
+ * ```
+ *
+ * Context persistence works differently per provider:
+ * - Claude: Uses hooks to inject context into CLAUDE.md
+ * - Codex: Uses config for periodic context refresh via system prompt
+ * - Gemini: Updates system instruction file
+ */
+export { AgentHealthMonitor, getHealthMonitor, } from './health-monitor.js';
+export { Logger, createLogger, configure as configureLogging, loggers, } from './logger.js';
+export { metrics } from './metrics.js';
+export { AgentSupervisor, getSupervisor, } from './supervisor.js';
+export { ContextPersistence, getContextPersistence, } from './context-persistence.js';
+export { createContextHandler, detectProvider, ClaudeContextHandler, CodexContextHandler, GeminiContextHandler, } from './provider-context.js';
+//# sourceMappingURL=index.js.map

package/dist/resiliency/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/resiliency/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAuDG;AAEH,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GAIjB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EACL,MAAM,EACN,YAAY,EACZ,SAAS,IAAI,gBAAgB,EAC7B,OAAO,GAIR,MAAM,aAAa,CAAC;AAErB,OAAO,EAAE,OAAO,EAA2D,MAAM,cAAc,CAAC;AAEhG,OAAO,EACL,eAAe,EACf,aAAa,GAGd,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,kBAAkB,EAClB,qBAAqB,GAMtB,MAAM,0BAA0B,CAAC;AAElC,OAAO,EACL,oBAAoB,EACpB,cAAc,EACd,oBAAoB,EACpB,mBAAmB,EACnB,oBAAoB,GAKrB,MAAM,uBAAuB,CAAC"}

package/dist/resiliency/logger.d.ts

@@ -0,0 +1,114 @@
+/**
+ * Structured Logger
+ *
+ * Provides consistent, structured logging across agent-relay components.
+ * - JSON format for machine parsing
+ * - Log levels with filtering
+ * - Context propagation (correlation IDs, agent names)
+ * - File rotation support
+ */
+import { EventEmitter } from 'events';
+export type LogLevel = 'debug' | 'info' | 'warn' | 'error' | 'fatal';
+export interface LogEntry {
+    timestamp: string;
+    level: LogLevel;
+    component: string;
+    message: string;
+    correlationId?: string;
+    agentName?: string;
+    pid?: number;
+    duration?: number;
+    error?: {
+        name: string;
+        message: string;
+        stack?: string;
+    };
+    [key: string]: unknown;
+}
+export interface LoggerConfig {
+    level: LogLevel;
+    json: boolean;
+    file?: string;
+    maxFileSize?: number;
+    maxFiles?: number;
+    console: boolean;
+}
+export declare class Logger extends EventEmitter {
+    private config;
+    private component;
+    private context;
+    private fileStream?;
+    private currentFileSize;
+    constructor(component: string, config?: Partial<LoggerConfig>);
+    /**
+     * Create a child logger with additional context
+     */
+    child(context: Record<string, unknown>): Logger;
+    /**
+     * Set context that will be included in all log entries
+     */
+    setContext(context: Record<string, unknown>): void;
+    /**
+     * Log at debug level
+     */
+    debug(message: string, context?: Record<string, unknown>): void;
+    /**
+     * Log at info level
+     */
+    info(message: string, context?: Record<string, unknown>): void;
+    /**
+     * Log at warn level
+     */
+    warn(message: string, context?: Record<string, unknown>): void;
+    /**
+     * Log at error level
+     */
+    error(message: string, context?: Record<string, unknown>): void;
+    /**
+     * Log at fatal level
+     */
+    fatal(message: string, context?: Record<string, unknown>): void;
+    /**
+     * Log with timing (returns function to end timing)
+     */
+    time(message: string, context?: Record<string, unknown>): () => void;
+    /**
+     * Log an error with stack trace
+     */
+    logError(error: Error, message?: string, context?: Record<string, unknown>): void;
+    /**
+     * Core log method
+     */
+    private log;
+    /**
+     * Write to console
+     */
+    private writeConsole;
+    /**
+     * Write to file with rotation
+     */
+    private writeFile;
+    /**
+     * Initialize file stream
+     */
+    private initFileStream;
+    /**
+     * Rotate log file
+     */
+    private rotateFile;
+    /**
+     * Close the logger
+     */
+    close(): void;
+}
+export declare function configure(config: Partial<LoggerConfig>): void;
+export declare function createLogger(component: string, config?: Partial<LoggerConfig>): Logger;
+export declare const loggers: {
+    daemon: () => Logger;
+    spawner: () => Logger;
+    router: () => Logger;
+    agent: (name: string) => Logger;
+    health: () => Logger;
+    connection: (id: string) => Logger;
+};
+//# sourceMappingURL=logger.d.ts.map

package/dist/resiliency/logger.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"logger.d.ts","sourceRoot":"","sources":["../../src/resiliency/logger.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,QAAQ,CAAC;AAItC,MAAM,MAAM,QAAQ,GAAG,OAAO,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,OAAO,CAAC;AAErE,MAAM,WAAW,QAAQ;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,QAAQ,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE;QACN,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,CAAC;IACF,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,QAAQ,CAAC;IAChB,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC;CAClB;AA4BD,qBAAa,MAAO,SAAQ,YAAY;IACtC,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,OAAO,CAA+B;IAC9C,OAAO,CAAC,UAAU,CAAC,CAAiB;IACpC,OAAO,CAAC,eAAe,CAAK;gBAEhB,SAAS,EAAE,MAAM,EAAE,MAAM,GAAE,OAAO,CAAC,YAAY,CAAM;IAUjE;;OAEG;IACH,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM;IAM/C;;OAEG;IACH,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAIlD;;OAEG;IACH,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAI/D;;OAEG;IACH,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAI9D;;OAEG;IACH,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAI9D;;OAEG;IACH,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAI/D;;OAEG;IACH,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAI/D;;OAEG;IACH,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI;IAQpE;;OAEG;IACH,QAAQ,CAAC,KAAK,EAAE,KAAK,EAAE,OAAO,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAWjF;;OAEG;IACH,OAAO,CAAC,GAAG;IA4BX;;OAEG;IACH,OAAO,CAAC,YAAY;IAqBpB;;OAEG;IACH,OAAO,CAAC,SAAS;IAkBjB;;OAEG;IACH,OAAO,CAAC,cAAc;IAiBtB;;OAEG;IACH,OAAO,CAAC,UAAU;IA6BlB;;OAEG;IACH,KAAK,IAAI,IAAI;CAMd;AAKD,wBAAgB,SAAS,CAAC,MAAM,EAAE,OAAO,CAAC,YAAY,CAAC,GAAG,IAAI,CAE7D;AAED,wBAAgB,YAAY,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,OAAO,CAAC,YAAY,CAAC,GAAG,MAAM,CAEtF;AAGD,eAAO,MAAM,OAAO;;;;kBAIJ,MAAM;;qBAEH,MAAM;CACxB,CAAC"}