@bolloon/bolloon-agent 0.1.35 → 0.1.37

package/src/heartbeat/StartupVerifier.ts

@@ -1,223 +0,0 @@
-/**
- * StartupVerifier - 启动自检验证器
- * 在服务启动时验证所有关键组件
- */
-
-import type { StartupReport, StartupCheck } from './types.js';
-
-export class StartupVerifier {
-  private checks: Map<string, StartupCheck> = new Map();
-
-  /**
-   * 添加检查项
-   */
-  addCheck(name: string): void {
-    this.checks.set(name, { name, status: 'pending' });
-  }
-
-  /**
-   * 标记检查开始
-   */
-  startCheck(name: string): void {
-    const check = this.checks.get(name);
-    if (check) {
-      check.status = 'running';
-      check.startTime = Date.now();
-    }
-  }
-
-  /**
-   * 标记检查完成
-   */
-  completeCheck(name: string, passed: boolean, message?: string): void {
-    const check = this.checks.get(name);
-    if (check) {
-      check.status = passed ? 'passed' : 'failed';
-      check.message = message;
-      check.duration_ms = Date.now() - (check.startTime || Date.now());
-    }
-  }
-
-  /**
-   * 运行所有自检
-   */
-  async verify(): Promise<StartupReport> {
-    const startTime = Date.now();
-    const critical_failures: string[] = [];
-
-    // 1. 检查 Node.js 版本
-    this.addCheck('node_version');
-    this.startCheck('node_version');
-    const nodeVersion = process.version;
-    const majorVersion = parseInt(nodeVersion.replace('v', '').split('.')[0], 10);
-    if (majorVersion >= 18) {
-      this.completeCheck('node_version', true, `Node.js ${nodeVersion}`);
-    } else {
-      this.completeCheck('node_version', false, `Node.js ${nodeVersion} too old, need >= 18`);
-      critical_failures.push('Node.js version too old');
-    }
-
-    // 2. 检查必要目录
-    this.addCheck('required_directories');
-    this.startCheck('required_directories');
-    try {
-      const fs = await import('fs/promises');
-      const paths = [
-        process.env.HOME ? `${process.env.HOME}/.bolloon` : '/tmp/.bolloon',
-        '/tmp'
-      ];
-      for (const p of paths) {
-        await fs.access(p);
-      }
-      this.completeCheck('required_directories', true, 'All required directories accessible');
-    } catch (err: any) {
-      this.completeCheck('required_directories', false, err.message);
-      critical_failures.push('Cannot access required directories');
-    }
-
-    // 3. 检查 LLM 配置
-    this.addCheck('llm_config');
-    this.startCheck('llm_config');
-    const hasLlmKey = !!(
-      process.env.OPENAI_API_KEY ||
-      process.env.ANTHROPIC_API_KEY ||
-      process.env.MINIMAX_API_KEY ||
-      process.env.OPENROUTER_API_KEY ||
-      process.env.GEMINI_API_KEY ||
-      process.env.OLLAMA_BASE_URL
-    );
-    if (hasLlmKey) {
-      this.completeCheck('llm_config', true, 'LLM API key found');
-    } else {
-      this.completeCheck('llm_config', false, 'No LLM API key configured (optional)');
-      // LLM 配置不是致命的，只是警告
-    }
-
-    // 4. 检查 P2P 依赖
-    this.addCheck('p2p_modules');
-    this.startCheck('p2p_modules');
-    try {
-      await import('@diap/sdk');
-      this.completeCheck('p2p_modules', true, '@diap/sdk available');
-    } catch (err: any) {
-      this.completeCheck('p2p_modules', false, `@diap/sdk not available: ${err.message}`);
-      critical_failures.push('@diap/sdk module not found');
-    }
-
-    // 5. 检查 IPFS 连接
-    this.addCheck('ipfs_connection');
-    this.startCheck('ipfs_connection');
-    try {
-      const ipfsRes = await fetch('http://127.0.0.1:5001/api/v0/id', {
-        method: 'POST',
-        signal: AbortSignal.timeout(3000)
-      });
-      if (ipfsRes.ok) {
-        this.completeCheck('ipfs_connection', true, 'IPFS local node accessible');
-      } else {
-        this.completeCheck('ipfs_connection', false, 'IPFS returned error');
-      }
-    } catch (err: any) {
-      this.completeCheck('ipfs_connection', false, 'IPFS not accessible (optional)');
-    }
-
-    // 6. 检查内存
-    this.addCheck('memory');
-    this.startCheck('memory');
-    const usage = process.memoryUsage();
-    const heapPercent = (usage.heapUsed / usage.heapTotal) * 100;
-    if (heapPercent < 90) {
-      this.completeCheck('memory', true, `Heap usage ${heapPercent.toFixed(1)}%`);
-    } else {
-      this.completeCheck('memory', false, `Heap usage critical: ${heapPercent.toFixed(1)}%`);
-      critical_failures.push('Memory usage too high at startup');
-    }
-
-    // 7. 检查网络
-    this.addCheck('network');
-    this.startCheck('network');
-    try {
-      const res = await fetch('https://www.google.com/generate_204', {
-        signal: AbortSignal.timeout(5000)
-      });
-      this.completeCheck('network', true, 'Internet connection available');
-    } catch {
-      this.completeCheck('network', false, 'No internet connection (optional)');
-    }
-
-    const totalDuration = Date.now() - startTime;
-
-    return {
-      success: critical_failures.length === 0,
-      checks: Array.from(this.checks.values()),
-      total_duration_ms: totalDuration,
-      critical_failures
-    };
-  }
-
-  /**
-   * 打印检查报告
-   */
-  printReport(report: StartupReport): void {
-    console.log('\n━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
-    console.log('🚀 Bolloon 启动自检报告');
-    console.log('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
-
-    for (const check of report.checks) {
-      const icon = check.status === 'passed' ? '✅' :
-                   check.status === 'failed' ? '❌' :
-                   check.status === 'running' ? '⏳' : '⭕';
-      const duration = check.duration_ms ? ` (${check.duration_ms}ms)` : '';
-      console.log(`  ${icon} ${check.name}${duration}`);
-      if (check.message) {
-        console.log(`     ${check.message}`);
-      }
-    }
-
-    console.log('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
-    console.log(`总计耗时: ${report.total_duration_ms}ms`);
-
-    if (report.success) {
-      console.log('✅ 所有关键检查通过！');
-    } else {
-      console.log('❌ 关键检查失败:');
-      for (const failure of report.critical_failures) {
-        console.log(`   · ${failure}`);
-      }
-    }
-    console.log('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n');
-  }
-
-  /**
-   * 快速验证（不打印）
-   */
-  async quickCheck(): Promise<boolean> {
-    const report = await this.verify();
-    return report.success;
-  }
-
-  /**
-   * 获取检查状态
-   */
-  getChecks(): StartupCheck[] {
-    return Array.from(this.checks.values());
-  }
-}
-
-// 导出便捷函数
-let startupVerifierInstance: StartupVerifier | null = null;
-
-export async function runStartupVerification(): Promise<StartupReport> {
-  const verifier = new StartupVerifier();
-  startupVerifierInstance = verifier;
-  const report = await verifier.verify();
-  verifier.printReport(report);
-  return report;
-}
-
-export function getStartupVerifier(): StartupVerifier {
-  if (!startupVerifierInstance) {
-    startupVerifierInstance = new StartupVerifier();
-  }
-  return startupVerifierInstance;
-}

package/src/heartbeat/Watchdog.ts

@@ -1,208 +0,0 @@
-/**
- * Watchdog - 看门狗机制
- * 检测进程卡死、无响应等情况
- */
-
-import type { WatchdogConfig, WatchdogState, RestartStrategy, RestartLevel } from './types.js';
-
-export interface WatchdogCallbacks {
-  onRestart?: (level: RestartLevel, reason: string) => void;
-  onLog?: (message: string) => void;
-  onHealthCheckFailed?: (failures: number) => void;
-}
-
-export class Watchdog {
-  private config: Required<WatchdogConfig>;
-  private state: WatchdogState;
-  private checkIntervalId: ReturnType<typeof setInterval> | null = null;
-  private restartStrategies: Map<RestartLevel, () => void | Promise<void>> = new Map();
-  private restartCount: number = 0;
-  private callbacks: WatchdogCallbacks;
-
-  constructor(config: Partial<WatchdogConfig> = {}, callbacks: WatchdogCallbacks = {}) {
-    this.config = {
-      silentThresholdMs: config.silentThresholdMs || 300000, // 5 分钟无日志
-      maxConsecutiveFailures: config.maxConsecutiveFailures || 3,
-      checkIntervalMs: config.checkIntervalMs || 30000 // 30 秒检查一次
-    };
-    this.state = {
-      lastActivityTime: Date.now(),
-      consecutiveFailures: 0,
-      isTriggered: false
-    };
-    this.callbacks = callbacks;
-  }
-
-  /**
-   * 注册重启策略
-   */
-  registerRestartStrategy(level: RestartLevel, action: () => void | Promise<void>): void {
-    this.restartStrategies.set(level, action);
-  }
-
-  /**
-   * 记录活动（调用后更新 lastActivityTime）
-   * 2026-06-10: 加 5s 去抖, 防 broadcast / SSE 高频路径炸 log + CPU
-   * - lastActivityTime 始终更新 (cheap)
-   * - 但 onLog 回调只在距上次 ≥5s 时触发, 避免每秒几十次 console.log
-   */
-  private _lastLogAt = 0;
-  recordActivity(component?: string): void {
-    this.state.lastActivityTime = Date.now();
-    if (component) {
-      const now = Date.now();
-      if (now - this._lastLogAt >= 5000) {
-        this._lastLogAt = now;
-        this.callbacks.onLog?.(`[Watchdog] Activity from ${component}`);
-      }
-    }
-  }
-
-  /**
-   * 记录日志（同时触发活动检测）
-   */
-  log(message: string): void {
-    this.recordActivity();
-    console.log(`[Watchdog] ${message}`);
-    this.callbacks.onLog?.(message);
-  }
-
-  /**
-   * 开始看门狗监控
-   */
-  start(): void {
-    if (this.checkIntervalId) return;
-
-    console.log(`[Watchdog] Started - silent threshold: ${this.config.silentThresholdMs}ms, check interval: ${this.config.checkIntervalMs}ms`);
-
-    this.checkIntervalId = setInterval(() => {
-      this.check();
-    }, this.config.checkIntervalMs);
-  }
-
-  /**
-   * 停止看门狗
-   */
-  stop(): void {
-    if (this.checkIntervalId) {
-      clearInterval(this.checkIntervalId);
-      this.checkIntervalId = null;
-    }
-    console.log('[Watchdog] Stopped');
-  }
-
-  /**
-   * 报告健康检查结果
-   */
-  reportHealthCheck(success: boolean, details?: string): void {
-    if (success) {
-      this.state.consecutiveFailures = 0;
-      this.recordActivity('health_check');
-    } else {
-      this.state.consecutiveFailures++;
-      console.log(`[Watchdog] Health check failed (${this.state.consecutiveFailures}/${this.config.maxConsecutiveFailures}): ${details || 'unknown'}`);
-      this.callbacks.onHealthCheckFailed?.(this.state.consecutiveFailures);
-
-      if (this.state.consecutiveFailures >= this.config.maxConsecutiveFailures) {
-        this.triggerRestart(1, `Health check failed ${this.state.consecutiveFailures} times`);
-      }
-    }
-  }
-
-  /**
-   * 获取当前状态
-   */
-  getState(): WatchdogState & { uptime_ms: number; restartCount: number } {
-    return {
-      ...this.state,
-      uptime_ms: Date.now() - this.state.lastActivityTime,
-      restartCount: this.restartCount
-    };
-  }
-
-  /**
-   * 执行检查
-   */
-  private check(): void {
-    const now = Date.now();
-    const silentTime = now - this.state.lastActivityTime;
-
-    // 检查是否沉默太久
-    if (silentTime > this.config.silentThresholdMs) {
-      console.warn(`[Watchdog] No activity for ${Math.round(silentTime / 1000)}s`);
-      this.triggerRestart(1, `No activity for ${Math.round(silentTime / 1000)}s`);
-      return;
-    }
-
-    // 检查内存使用: 用绝对阈值, 不看 heapUsed/heapTotal 比例 (V8 内部比例不可靠, 经常 80-95% 误报)
-    const usage = process.memoryUsage();
-    const heapUsedMB = usage.heapUsed / 1024 / 1024;
-    const rssMB = usage.rss / 1024 / 1024;
-    // 1.2GB heap 或 1.5GB RSS 才是真危险
-    if (heapUsedMB > 1224 || rssMB > 1536) {
-      console.warn(`[Watchdog] Memory usage critical: heapUsed=${heapUsedMB.toFixed(0)}MB, rss=${rssMB.toFixed(0)}MB`);
-      this.triggerRestart(1, `Memory usage heap=${heapUsedMB.toFixed(0)}MB rss=${rssMB.toFixed(0)}MB`);
-    }
-  }
-
-  /**
-   * 触发重启
-   */
-  private triggerRestart(level: RestartLevel, reason: string): void {
-    if (this.state.isTriggered) {
-      console.log(`[Watchdog] Already triggered, ignoring restart request: ${reason}`);
-      return;
-    }
-
-    this.state.isTriggered = true;
-    this.state.triggerReason = reason;
-    this.restartCount++;
-
-    console.error(`[Watchdog] 🔴 TRIGGERED: ${reason} (restart #${this.restartCount})`);
-
-    const strategy = this.restartStrategies.get(level);
-    if (strategy) {
-      try {
-        this.callbacks.onRestart?.(level, reason);
-        const result = strategy();
-        if (result instanceof Promise) {
-          result.catch(err => {
-            console.error('[Watchdog] Restart action failed:', err);
-          });
-        }
-      } catch (err) {
-        console.error('[Watchdog] Failed to execute restart strategy:', err);
-      }
-    } else {
-      console.warn(`[Watchdog] No restart strategy registered for level ${level}`);
-      this.callbacks.onRestart?.(level, reason);
-    }
-  }
-
-  /**
-   * 重置触发状态（允许再次触发）
-   */
-  reset(): void {
-    this.state.isTriggered = false;
-    this.state.triggerReason = undefined;
-    this.state.lastActivityTime = Date.now();
-  }
-}
-
-// 全局实例
-let watchdogInstance: Watchdog | null = null;
-
-export function getWatchdog(): Watchdog {
-  if (!watchdogInstance) {
-    watchdogInstance = new Watchdog();
-  }
-  return watchdogInstance;
-}
-
-export function createWatchdog(
-  config?: Partial<WatchdogConfig>,
-  callbacks?: WatchdogCallbacks
-): Watchdog {
-  watchdogInstance = new Watchdog(config, callbacks);
-  return watchdogInstance;
-}

package/src/heartbeat/index.ts

@@ -1,109 +0,0 @@
-/**
- * Bolloon 24h Heartbeat Module
- * 24 小时稳定运行心跳闭环
- */
-
-export * from './types.js';
-export * from './HealthMonitor.js';
-export * from './Watchdog.js';
-export * from './DaemonManager.js';
-export * from './StartupVerifier.js';
-export * from './self-improve-bus.js';
-
-import { createHealthMonitor, getHealthMonitor } from './HealthMonitor.js';
-import { createWatchdog, getWatchdog } from './Watchdog.js';
-import { createDaemonManager, getDaemonManager } from './DaemonManager.js';
-import { StartupVerifier, runStartupVerification, getStartupVerifier } from './StartupVerifier.js';
-import type { HealthStatus } from './types.js';
-
-/**
- * 初始化完整的 24h 心跳系统
- */
-export async function initialize24hHeartbeat(): Promise<{
-  healthMonitor: ReturnType<typeof createHealthMonitor>;
-  watchdog: ReturnType<typeof createWatchdog>;
-  daemonManager: ReturnType<typeof createDaemonManager>;
-  startupVerifier: StartupVerifier;
-}> {
-  // 1. 启动自检
-  const startupReport = await runStartupVerification();
-  if (!startupReport.success) {
-    console.warn('[24h] Startup verification failed, continuing anyway...');
-  }
-
-  // 2. 初始化健康监控
-  const healthMonitor = createHealthMonitor({
-    onStatusChange: (status: HealthStatus) => {
-      if (status.status !== 'healthy') {
-        console.warn(`[24h] Health status: ${status.status}`);
-        for (const rec of status.recommendations || []) {
-          console.warn(`[24h]   → ${rec}`);
-        }
-      }
-    }
-  });
-
-  // 3. 初始化看门狗
-  const watchdog = createWatchdog(
-    { silentThresholdMs: 300000, maxConsecutiveFailures: 3, checkIntervalMs: 30000 },
-    {
-      onLog: (msg) => console.log(`[24h-Watchdog] ${msg}`),
-      onHealthCheckFailed: (failures) => console.warn(`[24h] Health check failures: ${failures}`),
-      onRestart: (level, reason) => {
-        console.error(`[24h] Restart triggered (level ${level}): ${reason}`);
-      }
-    }
-  );
-
-  // 4. 连接健康监控和看门狗
-  healthMonitor.startPeriodicCheck(60000, (status) => {
-    const hasError = Object.values(status.checks).some(c => c.status === 'error');
-    watchdog.reportHealthCheck(!hasError, status.status);
-  });
-
-  // 5. 初始化守护进程管理器
-  const daemonManager = createDaemonManager({
-    maxRestarts: 5,
-    restartDelayMs: 5000
-  });
-
-  // 6. 注册看门狗重启策略
-  watchdog.registerRestartStrategy(1 as const, () => {
-    console.log('[24h] Level 1 restart: Graceful restart...');
-    // 可以添加进程重启逻辑
-  });
-
-  watchdog.registerRestartStrategy(2 as const, () => {
-    console.log('[24h] Level 2 restart: Process restart...');
-    // 可以添加进程重启逻辑
-  });
-
-  watchdog.registerRestartStrategy(3 as const, () => {
-    console.log('[24h] Level 3 restart: System notification...');
-    // 可以发送通知
-  });
-
-  // 7. 启动看门狗
-  watchdog.start();
-
-  console.log('[24h] 24h heartbeat system initialized');
-  console.log(`[24h] Health checks: every 60s`);
-  console.log(`[24h] Watchdog checks: every 30s`);
-  console.log(`[24h] Max restarts: 5`);
-
-  return { healthMonitor, watchdog, daemonManager, startupVerifier: getStartupVerifier() };
-}
-
-/**
- * 获取健康状态 JSON
- */
-export async function getHealthStatus(): Promise<HealthStatus> {
-  const healthMonitor = getHealthMonitor();
-  return healthMonitor.check();
-}
-
-// Re-export singleton getters for convenience
-export { getHealthMonitor, createHealthMonitor };
-export { getWatchdog, createWatchdog };
-export { getDaemonManager, createDaemonManager };
-export { getStartupVerifier };

package/src/heartbeat/self-improve-bus.ts

@@ -1,110 +0,0 @@
-/**
- * Self-Improve Event Bus
- *
- * 心跳事件 → 自改触发器 (解耦 watchdog)
- *
- * 设计原则:
- *   - Watchdog 负责保活 (重启进程), 不知道"自改"是什么
- *   - Self-Improve Bus 监听"信号事件" (CI 失败, 任务连续失败, 静默超时)
- *   - 信号达到阈值 + 通过冷却期 → 触发 runSelfImproveLoop
- *   - 触发时通过 SSE 广播给前端, 用户能在 UI 里看到
- *
- * 关键不变量:
- *   1. 心跳**不**直接调自改 - 通过 emit() 异步触发
- *   2. 触发频率受 SELF_IMPROVE_COOLDOWN_MS 限制
- *   3. 同类事件 24 小时内只触发 1 次
- *   4. 触发后不阻塞健康检查
- */
-
-import { SELF_IMPROVE_COOLDOWN_MS } from '../agents/shell-guard.js';
-
-export type SelfImproveEvent =
-  | { kind: 'ci-failed'; details: string }
-  | { kind: 'task-failures'; details: string }
-  | { kind: 'silent-timeout'; details: string }
-  | { kind: 'user-requested'; details: string };
-
-interface EventRecord {
-  at: number;
-  count: number;
-}
-
-const EVENT_COOLDOWN_MS = 24 * 60 * 60 * 1000; // 24 小时内同类型只触发 1 次
-
-const eventHistory: Map<SelfImproveEvent['kind'], EventRecord> = new Map();
-let lastTriggerAt: number | null = null;
-
-type Listener = (event: SelfImproveEvent, goal: string) => void | Promise<void>;
-const listeners: Set<Listener> = new Set();
-
-/**
- * 订阅自改触发事件
- */
-export function onSelfImproveTrigger(fn: Listener): () => void {
-  listeners.add(fn);
-  return () => { listeners.delete(fn); };
-}
-
-/**
- * 心跳事件 → 自改总线
- *
- * @returns { triggered: boolean, reason?: string }
- */
-export function reportSelfImproveEvent(event: SelfImproveEvent): { triggered: boolean; reason?: string } {
-  // 1. 24 小时同类事件冷却
-  const prev = eventHistory.get(event.kind);
-  if (prev && Date.now() - prev.at < EVENT_COOLDOWN_MS) {
-    return { triggered: false, reason: `同类事件 ${event.kind} 在 24h 内已记录过, 跳过` };
-  }
-
-  // 2. 累加计数
-  eventHistory.set(event.kind, {
-    at: Date.now(),
-    count: (prev?.count || 0) + 1
-  });
-
-  // 3. 自改循环冷却
-  if (lastTriggerAt && Date.now() - lastTriggerAt < SELF_IMPROVE_COOLDOWN_MS) {
-    const waitHrs = Math.ceil((SELF_IMPROVE_COOLDOWN_MS - (Date.now() - lastTriggerAt)) / 3600000);
-    return { triggered: false, reason: `自改冷却中, 还需要约 ${waitHrs} 小时` };
-  }
-
-  // 4. 触发
-  lastTriggerAt = Date.now();
-  const goal = `信号事件: ${event.kind} - ${event.details}`;
-
-  console.log(`[self-improve-bus] 🚀 触发自改循环: ${goal}`);
-
-  // 异步触发所有 listener, 不阻塞调用方
-  Promise.resolve().then(async () => {
-    for (const listener of listeners) {
-      try {
-        await listener(event, goal);
-      } catch (err) {
-        console.error(`[self-improve-bus] listener 失败:`, err);
-      }
-    }
-  });
-
-  return { triggered: true };
-}
-
-/**
- * 获取当前事件历史 (供调试 / UI 显示)
- */
-export function getEventHistory(): Array<{ kind: string; at: string; count: number }> {
-  return Array.from(eventHistory.entries()).map(([kind, rec]) => ({
-    kind,
-    at: new Date(rec.at).toISOString(),
-    count: rec.count
-  }));
-}
-
-/**
- * 强制重置 (仅供调试)
- */
-export function resetSelfImproveBus(): void {
-  eventHistory.clear();
-  lastTriggerAt = null;
-  console.log('[self-improve-bus] 已重置');
-}