@zhin.js/agent 0.1.0 → 0.1.2

package/src/zhin-agent/index.ts

@@ -1,633 +0,0 @@
-/**
- * ZhinAgent — 全局持久 AI 大脑
- *
- * 核心能力：
- *   1. 全局单例，应用生命周期内常驻
- *   2. Skill 感知：两级过滤 Skill → Tool
- *   3. 双层记忆：per-scene（对话上下文）+ per-user（长期偏好）
- *   4. 任务规划：复杂请求自动分解为子步骤
- *   5. 多模态输出：结构化 OutputElement[]
- *   6. 智能路径选择：纯闲聊走轻量路径，工具请求走完整路径
- *   7. 用户画像：跨会话个性化记忆
- *   8. 速率限制：防止单用户过度消耗资源
- *   9. 流式输出：onChunk 回调实时推送部分文本
- *  10. 情绪感知：根据用户语气调整回复风格
- *  11. 主动跟进：schedule_followup 定时回查
- *  12. 多模态输入：图片/音频直接传给视觉模型
- */
-
-import { Logger } from '@zhin.js/core';
-import type { Tool, ToolContext, SkillFeature, AIProvider, AgentTool, ChatMessage, ContentPart } from '@zhin.js/core';
-import { createAgent } from '@zhin.js/ai';
-import { SessionManager, createMemorySessionManager } from '@zhin.js/ai';
-import type { ContextManager } from '@zhin.js/ai';
-import { ConversationMemory } from '@zhin.js/ai';
-import type { OutputElement } from '@zhin.js/ai';
-import { parseOutput } from '@zhin.js/ai';
-import type { ModelRegistry } from '@zhin.js/ai';
-import { UserProfileStore } from '../user-profile.js';
-import { RateLimiter } from '@zhin.js/ai';
-import { detectTone } from '@zhin.js/ai';
-import { FollowUpManager, type FollowUpSender } from '../follow-up.js';
-import { SubagentManager, type SubagentResultSender } from '../subagent.js';
-import {
-  pruneHistoryForContext,
-  DEFAULT_CONTEXT_TOKENS,
-} from '@zhin.js/ai';
-import { triggerAIHook, createAIHookEvent } from '../hooks.js';
-
-// ── Sub-modules ─────────────────────────────────────────────────────
-import {
-  type ZhinAgentConfig,
-  type OnChunkCallback,
-  DEFAULT_CONFIG,
-  KEYWORD_TRIGGERS,
-} from './config.js';
-import { applyExecPolicyToTools } from './exec-policy.js';
-import { collectRelevantTools } from './tool-collector.js';
-import {
-  buildEnhancedPersona,
-  buildContextHint,
-  buildRichSystemPrompt,
-  buildUserMessageWithHistory,
-} from './prompt.js';
-import {
-  createChatHistoryTool,
-  createUserProfileTool,
-  createScheduleFollowUpTool,
-  createSpawnTaskTool,
-} from './builtin-tools.js';
-
-// Re-export public types for backward compat
-export type { ZhinAgentConfig, OnChunkCallback } from './config.js';
-
-const logger = new Logger(null, 'ZhinAgent');
-const now = () => performance.now();
-
-/** Strip `<think>…</think>` blocks that some reasoning models embed in content. */
-function stripThinkBlocks(text: string): string {
-  return text.replace(/<think>[\s\S]*?<\/think>\s*/g, '').trim();
-}
-
-// ============================================================================
-// ZhinAgent
-// ============================================================================
-
-export class ZhinAgent {
-  private provider: AIProvider;
-  private config: Required<ZhinAgentConfig>;
-  private skillRegistry: SkillFeature | null = null;
-  private sessions: SessionManager;
-  private contextManager: ContextManager | null = null;
-  private memory: ConversationMemory;
-  private externalTools: Map<string, AgentTool> = new Map();
-  private userProfiles: UserProfileStore;
-  private rateLimiter: RateLimiter;
-  private followUps: FollowUpManager;
-  private subagentManager: SubagentManager | null = null;
-  private bootstrapContext: string = '';
-  private activeSkillsContext: string = '';
-  private skillsSummaryXML: string = '';
-  private modelRegistry: ModelRegistry | null = null;
-
-  constructor(provider: AIProvider, config?: ZhinAgentConfig) {
-    this.provider = provider;
-    this.config = { ...DEFAULT_CONFIG, ...config } as Required<ZhinAgentConfig>;
-    this.sessions = createMemorySessionManager();
-    this.memory = new ConversationMemory({
-      minTopicRounds: this.config.minTopicRounds,
-      slidingWindowSize: this.config.slidingWindowSize,
-      topicChangeThreshold: this.config.topicChangeThreshold,
-    });
-    this.memory.setProvider(provider);
-    this.userProfiles = new UserProfileStore();
-    this.rateLimiter = new RateLimiter(this.config.rateLimit);
-    this.followUps = new FollowUpManager();
-  }
-
-  // ── DI setters ──────────────────────────────────────────────────────
-
-  setSkillRegistry(registry: SkillFeature): void {
-    this.skillRegistry = registry;
-    logger.debug(`SkillRegistry connected (${registry.size} skills)`);
-  }
-
-  setSessionManager(manager: SessionManager): void {
-    this.sessions.dispose();
-    this.sessions = manager;
-  }
-
-  setContextManager(manager: ContextManager): void {
-    this.contextManager = manager;
-    manager.setAIProvider(this.provider);
-  }
-
-  setModelRegistry(registry: ModelRegistry): void {
-    this.modelRegistry = registry;
-  }
-
-  upgradeMemoryToDatabase(msgModel: any, sumModel: any): void {
-    this.memory.upgradeToDatabase(msgModel, sumModel);
-  }
-
-  upgradeProfilesToDatabase(model: any): void {
-    this.userProfiles.upgradeToDatabase(model);
-  }
-
-  upgradeFollowUpsToDatabase(model: any): void {
-    this.followUps.upgradeToDatabase(model);
-  }
-
-  setFollowUpSender(sender: FollowUpSender): void {
-    this.followUps.setSender(sender);
-  }
-
-  async restoreFollowUps(): Promise<number> {
-    return this.followUps.restore();
-  }
-
-  initSubagentManager(createTools: () => AgentTool[]): void {
-    this.subagentManager = new SubagentManager({
-      provider: this.provider,
-      workspace: process.cwd(),
-      createTools,
-      maxIterations: this.config.maxSubagentIterations,
-      execPolicyConfig: this.config,
-    });
-    logger.debug('SubagentManager initialized');
-  }
-
-  setSubagentSender(sender: SubagentResultSender): void {
-    if (this.subagentManager) {
-      this.subagentManager.setSender(sender);
-    }
-  }
-
-  getSubagentManager(): SubagentManager | null {
-    return this.subagentManager;
-  }
-
-  getUserProfiles(): UserProfileStore {
-    return this.userProfiles;
-  }
-
-  /** 根据任务类型选择最合适的模型，优先使用 config 显式指定，其次 ModelRegistry，回退到 provider.models[0] */
-  private resolveModel(task: 'chat' | 'vision' | 'tool_call' | 'summary' = 'chat', preferred?: string): string {
-    return this.resolveModelCandidates(task, preferred)[0];
-  }
-
-  /**
-   * 返回按优先级排序的候选模型列表（用于自动降级）。
-   * 第一个是最优选择，失败后依次尝试后续模型。
-   */
-  private resolveModelCandidates(task: 'chat' | 'vision' | 'tool_call' | 'summary' = 'chat', preferred?: string): string[] {
-    const candidates: string[] = [];
-
-    // 1. 用户显式指定 / 配置指定优先
-    if (preferred) candidates.push(preferred);
-    if (task === 'chat' && this.config.chatModel && !candidates.includes(this.config.chatModel)) {
-      candidates.push(this.config.chatModel);
-    }
-    if (task === 'vision' && this.config.visionModel && !candidates.includes(this.config.visionModel)) {
-      candidates.push(this.config.visionModel);
-    }
-
-    // 2. ModelRegistry 自动排序的候选列表
-    if (this.modelRegistry) {
-      for (const id of this.modelRegistry.selectModels(this.provider.name, task, 5)) {
-        if (!candidates.includes(id)) candidates.push(id);
-      }
-    }
-
-    // 3. 兜底: provider.models[0]
-    const fallback = this.provider.models[0];
-    if (fallback && !candidates.includes(fallback)) candidates.push(fallback);
-
-    return candidates;
-  }
-
-  registerTool(tool: AgentTool): () => void {
-    this.externalTools.set(tool.name, tool);
-    return () => { this.externalTools.delete(tool.name); };
-  }
-
-  setBootstrapContext(context: string): void {
-    this.bootstrapContext = context;
-    logger.debug(`Bootstrap context set (${context.length} chars)`);
-  }
-
-  setActiveSkillsContext(content: string): void {
-    this.activeSkillsContext = content || '';
-  }
-
-  setSkillsSummaryXML(xml: string): void {
-    this.skillsSummaryXML = xml || '';
-  }
-
-  // ── Core processing ─────────────────────────────────────────────────
-
-  async process(
-    content: string,
-    context: ToolContext,
-    externalTools: Tool[] = [],
-    onChunk?: OnChunkCallback,
-  ): Promise<OutputElement[]> {
-    const t0 = now();
-    const { senderId, sceneId, platform } = context;
-    const sessionId = SessionManager.generateId(platform || '', senderId || '', sceneId);
-    const userId = senderId || 'unknown';
-
-    // 0. Rate limit
-    const rateCheck = this.rateLimiter.check(userId);
-    if (!rateCheck.allowed) {
-      logger.debug(`[速率限制] 用户 ${userId} 被限制: ${rateCheck.message}`);
-      return parseOutput(rateCheck.message || '请稍后再试');
-    }
-
-    triggerAIHook(createAIHookEvent('message', 'received', sessionId, {
-      userId,
-      content,
-      platform: platform || '',
-    })).catch(() => {});
-
-    // 1. Collect tools
-    const tFilter = now();
-    const allTools = collectRelevantTools(content, context, externalTools, {
-      config: this.config,
-      skillRegistry: this.skillRegistry,
-      externalRegistered: this.externalTools,
-    });
-
-    // Inject context-aware built-in tools on keyword match
-    if (KEYWORD_TRIGGERS.chatHistory.test(content)) {
-      allTools.push(createChatHistoryTool(sessionId, this.memory));
-    }
-    if (KEYWORD_TRIGGERS.userProfile.test(content)) {
-      allTools.push(createUserProfileTool(userId, this.userProfiles));
-    }
-    if (KEYWORD_TRIGGERS.scheduleFollowUp.test(content)) {
-      allTools.push(createScheduleFollowUpTool(sessionId, context, this.followUps));
-    }
-    if (this.subagentManager && KEYWORD_TRIGGERS.spawnTask.test(content)) {
-      allTools.push(createSpawnTaskTool(context, this.subagentManager));
-    }
-
-    const filterMs = (now() - tFilter).toFixed(0);
-
-    // 2. History + profile
-    const tMem = now();
-    let historyMessages = await this.buildHistoryMessages(sessionId);
-
-    const contextTokens = this.config.contextTokens ?? DEFAULT_CONTEXT_TOKENS;
-    const maxHistoryShare = this.config.maxHistoryShare ?? 0.5;
-    const pruneResult = pruneHistoryForContext({
-      messages: historyMessages,
-      maxContextTokens: contextTokens,
-      maxHistoryShare,
-    });
-    historyMessages = pruneResult.messages;
-    if (pruneResult.droppedCount > 0) {
-      logger.debug(`[上下文窗口] 丢弃 ${pruneResult.droppedCount} 条历史消息 (${pruneResult.droppedTokens} tokens)`);
-    }
-
-    const memMs = (now() - tMem).toFixed(0);
-
-    // 2.5 Profile + tone
-    const profileSummary = await this.userProfiles.buildProfileSummary(userId);
-    const toneHint = this.config.toneAwareness ? detectTone(content).hint : '';
-    const personaEnhanced = buildEnhancedPersona(this.config, profileSummary, toneHint);
-
-    // 3. No tools → chat path
-    if (allTools.length === 0) {
-      logger.info(`[System Prompt] chat-path: ${personaEnhanced.length} chars ≈ ${Math.ceil(personaEnhanced.length / 2.5)} tokens`);
-      logger.debug(`[闲聊路径] 过滤=${filterMs}ms, 记忆=${memMs}ms (${historyMessages.length}条), 0 工具`);
-      const tLLM = now();
-      const reply = await this.streamChatWithHistory(content, personaEnhanced, historyMessages, onChunk);
-      const llmMs = (now() - tLLM).toFixed(0);
-      logger.info(`[闲聊路径] 过滤=${filterMs}ms, 记忆=${memMs}ms, LLM=${llmMs}ms, 总=${(now() - t0).toFixed(0)}ms`);
-      await this.saveToSession(sessionId, content, reply, sceneId);
-      return parseOutput(reply);
-    }
-
-    logger.debug(`[工具路径] 过滤=${filterMs}ms, 记忆=${memMs}ms, ${allTools.length} 工具 (${allTools.map(t => t.name).join(', ')})`);
-
-    // 4. Pre-executable tools
-    const preExecTools: AgentTool[] = [];
-    for (const tool of allTools) {
-      if (tool.preExecutable) preExecTools.push(tool);
-    }
-
-    // 5. Pre-execution
-    let preData = '';
-    if (preExecTools.length > 0) {
-      const tPre = now();
-      logger.debug(`预执行: ${preExecTools.map(t => t.name).join(', ')}`);
-      const results = await Promise.allSettled(
-        preExecTools.map(async (tool) => {
-          const result = await Promise.race([
-            tool.execute({}),
-            new Promise<never>((_, rej) =>
-              setTimeout(() => rej(new Error('超时')), this.config.preExecTimeout)),
-          ]);
-          return { name: tool.name, result };
-        }),
-      );
-      for (const r of results) {
-        if (r.status === 'fulfilled') {
-          let s = typeof r.value.result === 'string' ? r.value.result : JSON.stringify(r.value.result);
-          if (s.length > 500) {
-            s = s.slice(0, 500) + `\n... (truncated, ${s.length} chars total)`;
-          }
-          preData += `\n【${r.value.name}】${s}`;
-        }
-      }
-      logger.debug(`预执行耗时: ${(now() - tPre).toFixed(0)}ms`);
-    }
-
-    // 6. Path selection
-    let reply: string;
-    const hasNonPreExecTools = allTools.some(t => !t.preExecutable);
-
-    if (!hasNonPreExecTools && preData) {
-      // Fast path
-      const tLLM = now();
-      const prompt = `${personaEnhanced}
-
-Pre-fetched data (from user's question):
-${preData}
-
-Answer the user's question based on the data above. Be clear and concise; use emoji when appropriate.`;
-      logger.info(`[System Prompt] fast-path: ${prompt.length} chars ≈ ${Math.ceil(prompt.length / 2.5)} tokens`);
-      reply = await this.streamChatWithHistory(content, prompt, historyMessages, onChunk);
-      logger.info(`[快速路径] 过滤=${filterMs}ms, 记忆=${memMs}ms, LLM=${(now() - tLLM).toFixed(0)}ms, 总=${(now() - t0).toFixed(0)}ms`);
-    } else {
-      // Agent path
-      const tAgent = now();
-      logger.debug(`Agent 路径: ${allTools.length} 个工具`);
-      const contextHint = buildContextHint(context, content);
-
-      const richPrompt = buildRichSystemPrompt({
-        config: this.config,
-        skillRegistry: this.skillRegistry,
-        skillsSummaryXML: this.skillsSummaryXML,
-        activeSkillsContext: this.activeSkillsContext,
-        bootstrapContext: this.bootstrapContext,
-      });
-      const systemPrompt = `${richPrompt}
-${contextHint}
-${preData ? `\nPre-fetched data:\n${preData}\n` : ''}`;
-
-      const promptChars = systemPrompt.length;
-      const estimatedTokens = Math.ceil(promptChars / 2.5);
-      logger.info(`[System Prompt] ${promptChars} chars ≈ ${estimatedTokens} tokens`);
-      logger.debug(`[System Prompt Preview]\n${systemPrompt.slice(0, 500)}...\n---END PREVIEW---`);
-
-      const agentTools = applyExecPolicyToTools(this.config, allTools);
-
-      // Adaptive maxIterations: boost when skills are active (multi-step skill flows)
-      const SKILL_ITERATION_BOOST = 3;
-      const hasSkillActivation = agentTools.some(t => t.name === 'activate_skill' || t.name === 'install_skill');
-      const effectiveMaxIterations = hasSkillActivation
-        ? this.config.maxIterations + SKILL_ITERATION_BOOST
-        : this.config.maxIterations;
-
-      const chatCandidates = this.resolveModelCandidates('chat');
-      const agent = createAgent(this.provider, {
-        model: chatCandidates[0],
-        modelFallbacks: chatCandidates.slice(1),
-        systemPrompt,
-        tools: agentTools,
-        maxIterations: effectiveMaxIterations,
-      });
-
-      const userMessageWithHistory = buildUserMessageWithHistory(historyMessages, content);
-      const result = await agent.run(userMessageWithHistory, []);
-      reply = stripThinkBlocks(result.content) || this.fallbackFormat(result.toolCalls);
-      logger.info(`[Agent 路径] 过滤=${filterMs}ms, 记忆=${memMs}ms, Agent=${(now() - tAgent).toFixed(0)}ms, 总=${(now() - t0).toFixed(0)}ms`);
-    }
-
-    await this.saveToSession(sessionId, content, reply, sceneId);
-
-    triggerAIHook(createAIHookEvent('message', 'sent', sessionId, {
-      userId,
-      content: reply,
-      platform: platform || '',
-    })).catch(() => {});
-
-    return parseOutput(reply);
-  }
-
-  async processMultimodal(
-    parts: ContentPart[],
-    context: ToolContext,
-    onChunk?: OnChunkCallback,
-  ): Promise<OutputElement[]> {
-    const { senderId, sceneId, platform } = context;
-    const sessionId = SessionManager.generateId(platform || '', senderId || '', sceneId);
-    const userId = senderId || 'unknown';
-
-    const rateCheck = this.rateLimiter.check(userId);
-    if (!rateCheck.allowed) {
-      return parseOutput(rateCheck.message || '请稍后再试');
-    }
-
-    const historyMessages = await this.buildHistoryMessages(sessionId);
-    const profileSummary = await this.userProfiles.buildProfileSummary(userId);
-    const personaEnhanced = buildEnhancedPersona(this.config, profileSummary, '');
-
-    // Build text summary describing the multimodal content
-    const textFragments: string[] = [];
-    const llmParts: ContentPart[] = [];
-
-    /** Full multimodal ContentPart union (core/ai may export a narrower type in some builds) */
-    type MultimodalPart =
-      | ContentPart
-      | { type: 'video_url'; video_url: { url: string } }
-      | { type: 'face'; face: { id: string; text?: string } };
-
-    for (const p of parts as MultimodalPart[]) {
-      switch (p.type) {
-        case 'text':
-          textFragments.push(p.text);
-          llmParts.push(p);
-          break;
-        case 'image_url':
-          textFragments.push('[图片]');
-          llmParts.push(p);
-          break;
-        case 'video_url':
-          textFragments.push('[视频]');
-          // Most LLMs don't support video natively; describe it as a URL for context
-          llmParts.push({ type: 'text', text: `[用户发送了一个视频: ${p.video_url.url}]` });
-          break;
-        case 'audio':
-          textFragments.push('[音频]');
-          llmParts.push(p);
-          break;
-        case 'face':
-          textFragments.push(p.face.text || `[表情:${p.face.id}]`);
-          llmParts.push({ type: 'text', text: p.face.text ? `[表情: ${p.face.text}]` : `[表情ID: ${p.face.id}]` });
-          break;
-      }
-    }
-
-    const textContent = textFragments.join(' ') || '[多模态消息]';
-    const visionCandidates = this.resolveModelCandidates('vision');
-
-    const messages: ChatMessage[] = [
-      { role: 'system', content: personaEnhanced },
-      ...historyMessages,
-      { role: 'user', content: llmParts },
-    ];
-
-    let reply = '';
-    for (let i = 0; i < visionCandidates.length; i++) {
-      const visionModel = visionCandidates[i];
-      try {
-        reply = '';
-        for await (const chunk of this.provider.chatStream({ model: visionModel, messages })) {
-          const delta = chunk.choices?.[0]?.delta;
-          if (!delta) continue;
-          const text = typeof delta.content === 'string' ? delta.content : '';
-          if (text) {
-            reply += text;
-            if (onChunk) onChunk(text, reply);
-          }
-        }
-        reply = stripThinkBlocks(reply);
-        if (!reply) {
-          logger.warn(`[processMultimodal] ${visionModel} 流式为空，尝试非流式`);
-          const response = await this.provider.chat({ model: visionModel, messages });
-          const msg = response.choices[0]?.message?.content;
-          reply = stripThinkBlocks(typeof msg === 'string' ? msg : '');
-        }
-        if (reply) break; // 成功，退出循环
-      } catch (err) {
-        const isLast = i === visionCandidates.length - 1;
-        if (isLast) {
-          try {
-            const response = await this.provider.chat({ model: visionModel, messages });
-            const msg = response.choices[0]?.message?.content;
-            reply = stripThinkBlocks(typeof msg === 'string' ? msg : '');
-          } catch { /* all candidates exhausted */ }
-        } else {
-          logger.warn(`[processMultimodal] ${visionModel} 失败，降级到 ${visionCandidates[i + 1]}: ${(err as Error).message}`);
-        }
-      }
-    }
-
-    if (!reply) reply = '抱歉，我无法理解这条消息。';
-    await this.saveToSession(sessionId, textContent, reply, sceneId);
-    return parseOutput(reply);
-  }
-
-  // ── Internal helpers ────────────────────────────────────────────────
-
-  private async buildHistoryMessages(sessionId: string): Promise<ChatMessage[]> {
-    return this.memory.buildContext(sessionId);
-  }
-
-  private async streamChatWithHistory(
-    content: string,
-    systemPrompt: string,
-    history: ChatMessage[],
-    onChunk?: OnChunkCallback,
-  ): Promise<string> {
-    const candidates = this.resolveModelCandidates('chat');
-    const userContent = history.length > 0
-      ? buildUserMessageWithHistory(history, content)
-      : content;
-    const messages: ChatMessage[] = [
-      { role: 'system', content: systemPrompt },
-      { role: 'user', content: userContent },
-    ];
-
-    for (let i = 0; i < candidates.length; i++) {
-      const model = candidates[i];
-      try {
-        let result = '';
-        for await (const chunk of this.provider.chatStream({ model, messages })) {
-          const delta = chunk.choices?.[0]?.delta;
-          if (!delta) continue;
-          const text = typeof delta.content === 'string' ? delta.content : '';
-          if (text) {
-            result += text;
-            if (onChunk) onChunk(text, result);
-          }
-        }
-        result = stripThinkBlocks(result);
-        if (result) return result;
-        // Streaming returned empty content — try non-streaming with same model
-        logger.warn(`[streamChat] ${model} 流式响应为空，尝试非流式`);
-        const response = await this.provider.chat({ model, messages });
-        const msg = response.choices[0]?.message?.content;
-        result = stripThinkBlocks(typeof msg === 'string' ? msg : '');
-        if (result) {
-          if (onChunk) onChunk(result, result);
-          return result;
-        }
-      } catch (err) {
-        const isLast = i === candidates.length - 1;
-        if (isLast) {
-          // No more candidates — try non-streaming as final attempt
-          try {
-            const response = await this.provider.chat({ model, messages });
-            const msg = response.choices[0]?.message?.content;
-            let result = stripThinkBlocks(typeof msg === 'string' ? msg : '');
-            if (onChunk && result) onChunk(result, result);
-            return result;
-          } catch {
-            return '';
-          }
-        }
-        logger.warn(`[streamChat] ${model} 失败，降级到 ${candidates[i + 1]}: ${(err as Error).message}`);
-      }
-    }
-    return '';
-  }
-
-  private async saveToSession(
-    sessionId: string,
-    userContent: string,
-    assistantContent: string,
-    sceneId?: string,
-  ): Promise<void> {
-    await this.memory.saveRound(sessionId, userContent, assistantContent);
-    await this.sessions.addMessage(sessionId, { role: 'user', content: userContent });
-    await this.sessions.addMessage(sessionId, { role: 'assistant', content: assistantContent });
-    if (this.contextManager && sceneId) {
-      this.contextManager.autoSummarizeIfNeeded(sceneId).catch(() => {});
-    }
-  }
-
-  private fallbackFormat(toolCalls: { tool: string; args: any; result: any }[]): string {
-    if (toolCalls.length === 0) return 'Done.';
-    const userFacing = toolCalls.filter(tc => tc.tool !== 'activate_skill');
-    if (userFacing.length === 0) {
-      return '技能已激活但未能完成后续操作，请重试或换一种方式描述你的需求。';
-    }
-    return userFacing.map(tc => {
-      const s = typeof tc.result === 'string' ? tc.result : JSON.stringify(tc.result, null, 2);
-      return `【${tc.tool}】\n${s}`;
-    }).join('\n\n');
-  }
-
-  // ── Lifecycle ───────────────────────────────────────────────────────
-
-  isReady(): boolean {
-    return true;
-  }
-
-  dispose(): void {
-    this.memory.dispose();
-    this.sessions.dispose();
-    this.externalTools.clear();
-    this.userProfiles.dispose();
-    this.rateLimiter.dispose();
-    this.followUps.dispose();
-    if (this.subagentManager) {
-      this.subagentManager.dispose();
-      this.subagentManager = null;
-    }
-  }
-}