myagent-ai 1.7.1 → 1.7.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.7.1",
+  "version": "1.7.3",
   "description": "本地桌面端执行型AI助手 - Open Interpreter 风格 | Local Desktop Execution-Oriented AI Assistant",
   "main": "main.py",
   "bin": {

package/web/api_server.py

@@ -746,29 +746,41 @@ class ApiServer:
             return ""
 
         base_instruction = (
-            "你当前处于【执行模式】(Execution Mode)。\n"
-            "1. **复杂度分析**：首先评估任务复杂度。对于简单问候或常见问题，直接回答；对于多步骤任务，【必须】先制定计划。\n"
-            "2. **强制规则 - 任务列表**：每次回复【必须】包含 ```tasklist``` 代码块，输出 JSON 格式的任务进度列表。先写纯文本分析，再写 tasklist，最后写 action（如有）。\n"
-            "3. **强制规则 - 单步执行**：每次回复【只能执行一个操作】（一个工具调用或一个代码块）。执行完后等待结果反馈。\n"
+            "你当前处于【执行模式】(Execution Mode)。\n\n"
+            "## 核心规则\n"
+            "1. **任务列表（强制）**：每次回复【必须】包含一个 ```tasklist``` 代码块，内含 JSON 数组格式的任务进度。\n"
+            "   - 格式：```tasklist\\n[{\"text\": \"步骤描述\", \"status\": \"pending\"}]\\n```\n"
+            "   - status 可选值：pending(待执行)、running(进行中)、done(已完成)、blocked(受阻)\n"
+            "   - 首次收到任务时，拆分为多个步骤，全部标记为 pending\n"
+            "   - 每次执行完一个步骤后，更新对应步骤状态为 done，下一个为 running\n"
+            "2. **单步执行（强制）**：每次回复【只能执行一个操作】（一个工具调用、一个代码块或一个技能调用）。\n"
+            "   - 执行完一个操作后停下来，等待结果反馈后再决定下一步\n"
+            "   - 不要一次性执行多个操作\n"
+            "3. **回复格式**：先写纯文本分析/总结 → 再写 ```tasklist``` 更新进度 → 最后写 ```action``` 执行操作（如有）\n"
         )
 
         # 从内存读取当前任务列表
         tasks = self._task_list_store.get(agent_path, [])
         if not tasks:
-            return base_instruction + "请分析需求后制定计划。"
+            return base_instruction + "\n## 当前状态\n暂无任务计划。请先分析用户需求，拆分为具体步骤，然后用 ```tasklist``` 输出计划。"
 
         pending = [f"  - ⏳ {t['text']}" for t in tasks if t.get("status") in ("pending", "running", "blocked")]
         done = [f"  - ✅ {t['text']}" for t in tasks if t.get("status") == "done"]
         running = [f"  - 🔄 {t['text']}" for t in tasks if t.get("status") == "running"]
 
-        context = base_instruction + "\n当前任务进度:\n"
+        context = base_instruction + "\n## 当前任务进度\n"
         if done:
             context += "已完成:\n" + "\n".join(done) + "\n"
         if running:
             context += "进行中:\n" + "\n".join(running) + "\n"
         if pending:
             context += "待执行:\n" + "\n".join(pending) + "\n"
-        context += "\n请在回复中用 ```tasklist``` 更新任务进度（先写文本分析，再写 tasklist，最后写 action）。记住：【每次只能执行一个操作】。"
+        context += (
+            "\n## 下一步\n"
+            "1. 用纯文本简要分析当前进展\n"
+            "2. 用 ```tasklist``` 更新任务进度（标记已完成的步骤为 done，标记当前步骤为 running）\n"
+            "3. 用 ```action``` 执行下一个待执行步骤（每次只执行一个操作）\n"
+        )
         return context
 
     async def handle_chat_page(self, request):
@@ -2613,7 +2625,9 @@ class ApiServer:
         iteration = 0
         # 追踪连续无 action 迭代次数，防止无限重新提示
         _consecutive_no_action = 0
-        _MAX_NO_ACTION_RETRIES = 3
+        _MAX_NO_ACTION_RETRIES = 5  # 提高重试次数，给 LLM 更多机会完成剩余任务
+        # ── 追踪所有流式推送的纯文本（用于刷新后恢复） ──
+        _all_streamed_text_parts = []  # 每轮迭代推送的纯文本片段
 
         while iteration < max_iter:
             iteration += 1
@@ -2682,6 +2696,7 @@ class ApiServer:
                             text_before = remaining[:marker_pos]
                             if text_before.strip():
                                 await _write_sse({"type": "text_delta", "content": text_before})
+                                _all_streamed_text_parts.append(text_before)
                             # 跳过整个开始标记（```action 或 ```tasklist），不要只跳到 ```
                             st["processed_pos"] += marker_pos + len(f"```{block_type}")
                             if block_type == "tasklist":
@@ -2718,7 +2733,9 @@ class ApiServer:
                         # 没有找到标记，流式推送（保留末尾可能的部分标记）
                         safe_end = len(remaining) - _MAX_HOLD
                         if safe_end > 0:
-                            await _write_sse({"type": "text_delta", "content": remaining[:safe_end]})
+                            chunk = remaining[:safe_end]
+                            await _write_sse({"type": "text_delta", "content": chunk})
+                            _all_streamed_text_parts.append(chunk)
                             st["processed_pos"] += safe_end
                             remaining = full_text_so_far[st["processed_pos"]:]
                         else:
@@ -2790,6 +2807,7 @@ class ApiServer:
                         await _stream_text_chunked(remaining, _write_sse, chunk_size=3, delay=0.01)
                     else:
                         await _write_sse({"type": "text_delta", "content": remaining})
+                    _all_streamed_text_parts.append(remaining)
                     st["processed_pos"] = len(full_text)
 
             # Call LLM with streaming — tokens are filtered through _text_delta_callback
@@ -3054,10 +3072,19 @@ class ApiServer:
                 break
 
         # Save assistant response to memory
-        if agent.memory and final_response:
-            agent.memory.add_short_term(session_id=session_id, role="assistant", content=final_response)
-
-        return final_response
+        # ── 优先使用流式累积文本（包含所有迭代的纯文本），回退到 final_response ──
+        saved_response = final_response
+        if not saved_response and _all_streamed_text_parts:
+            saved_response = "\n\n".join(p for p in _all_streamed_text_parts if p.strip())
+        if not saved_response and content:
+            saved_response = content  # 兜底：使用最后一轮的完整输出
+        if agent.memory and saved_response:
+            agent.memory.add_short_term(session_id=session_id, role="assistant", content=saved_response)
+        elif agent.memory:
+            # 即使为空也保存一条，防止刷新后消息丢失
+            agent.memory.add_short_term(session_id=session_id, role="assistant", content="(执行完成，无文本回复)")
+
+        return saved_response or final_response or content or ""
 
     async def _execute_actions_streaming(
         self, agent, action_data: dict, context, write_sse

package/web/ui/chat/chat.css

@@ -1569,6 +1569,20 @@ input,textarea,select{font:inherit}
 .exec-event-result-btn:hover{background:var(--accent-light);color:var(--accent-dark)}
 .exec-event-result-btn svg{width:12px;height:12px}
 
+/* ── Inline Exec Events (Timeline Interleaved) ── */
+.msg-timeline{display:flex;flex-direction:column;gap:6px}
+.inline-exec-event{margin:2px 0;padding:8px 12px;background:var(--bg2);border-left:3px solid var(--border);border-radius:6px;font-size:13px;animation:execEventSlide .3s ease-out}
+.inline-exec-header{display:flex;align-items:center;gap:6px;margin-bottom:4px}
+.inline-exec-icon{font-size:14px}
+.inline-exec-title{font-weight:500;color:var(--text);font-size:12px}
+.inline-exec-meta{color:var(--text3);font-size:11px;margin-left:auto}
+.inline-exec-code{background:var(--bg);padding:6px 8px;border-radius:4px;font-family:'SF Mono','Fira Code','Cascadia Code',monospace;font-size:12px;color:var(--text2);margin:4px 0;max-height:100px;overflow:hidden;cursor:pointer;transition:var(--transition);white-space:pre-wrap;word-break:break-all}
+.inline-exec-code:hover{background:var(--bg3)}
+.inline-exec-code.expanded{max-height:none}
+.inline-exec-summary{color:var(--text2);font-size:12px;margin:4px 0}
+.inline-exec-result-btn{background:none;border:1px solid var(--border);color:var(--text2);font-size:11px;padding:2px 8px;border-radius:4px;cursor:pointer;margin-top:4px;transition:var(--transition)}
+.inline-exec-result-btn:hover{background:var(--bg2);border-color:var(--accent);color:var(--accent)}
+
 /* ── Execution Result Modal ── */
 .exec-result-modal-overlay{position:fixed;inset:0;background:rgba(0,0,0,.5);z-index:1000;display:flex;align-items:center;justify-content:center;animation:fadeIn .15s ease}
 .exec-result-modal{background:var(--bg);border:1px solid var(--border);border-radius:12px;width:min(680px,90vw);max-height:80vh;display:flex;flex-direction:column;box-shadow:0 20px 60px rgba(0,0,0,.25);animation:slideUp .2s ease}
@@ -1605,6 +1619,9 @@ input,textarea,select{font:inherit}
 [data-theme="dark"] .exec-result-modal{background:var(--bg2);border-color:var(--border)}
 [data-theme="dark"] .exec-result-modal-body pre{background:#0a0c10;color:#cdd6f4}
 [data-theme="dark"] .exec-result-info-item{background:var(--bg3)}
+[data-theme="dark"] .inline-exec-event{background:var(--bg3);border-left-color:var(--border)}
+[data-theme="dark"] .inline-exec-code{background:var(--bg)}
+[data-theme="dark"] .inline-exec-result-btn:hover{background:var(--bg4)}
 
   .thought-block {
     background: rgba(0, 0, 0, 0.03);

package/web/ui/chat/chat_main.js

@@ -1680,9 +1680,12 @@ function renderMessages() {
       </div>` : '';
     const ttsIndicator = ttsManager && ttsManager.isPlaying && ttsManager.currentMsgIndex === i ?
       ' <span class="tts-playing-icon"><svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><polygon points="11 5 6 9 2 9 2 15 6 15 11 19 11 5"/><path d="M15.54 8.46a5 5 0 0 1 0 7.07"/></svg></span>' : '';
-    const execEventsHtml = (!isUser && msg.exec_events && msg.exec_events.length > 0)
-      ? renderExecEvents(msg.exec_events, i) : '';
-    const streamingIndicator = msg.streaming && !msg.content && !msg.thought ? `
+
+    // ── Determine rendering mode and streaming indicator ──
+    const hasParts = Array.isArray(msg.parts) && msg.parts.length > 0;
+    const hasStreamingText = msg._streamingText && msg._streamingText.trim();
+    const anyContent = msg.content || msg._streamingText || hasParts;
+    const streamingIndicator = msg.streaming && !anyContent && !msg.thought ? `
       <div class="streaming-indicator">
         <div class="spinner"></div>
         <div class="streaming-dots">
@@ -1690,13 +1693,42 @@ function renderMessages() {
         </div>
         <span style="font-weight:500">Agent 正在思考...</span>
       </div>` : '';
+
+    // ── Timeline rendering for interleaved text + exec events ──
+    let timelineHtml = '';
+    if (hasParts || hasStreamingText) {
+      let partsHtml = '';
+      for (const part of (msg.parts || [])) {
+        if (part.type === 'text' && part.content.trim()) {
+          partsHtml += '<div class="message-bubble">' + renderMarkdown(part.content) + '</div>';
+        } else if (part.type === 'exec') {
+          partsHtml += renderInlineExecEvent(part.data, i);
+        }
+      }
+      if (hasStreamingText) {
+        partsHtml += '<div class="message-bubble">' + renderMarkdown(msg._streamingText) + '</div>';
+      }
+      if (partsHtml) {
+        timelineHtml = '<div class="msg-timeline">' + partsHtml + '</div>';
+      }
+    }
+
+    // Backward compat: single bubble for messages without parts
+    const singleBubbleHtml = (!hasParts && !hasStreamingText)
+      ? ((content || streamingIndicator) ? `<div class="message-bubble">${content}${ttsIndicator}</div>` : '')
+      : '';
+
+    // Exec events panel: only for backward compat (messages without parts loaded from DB)
+    const execEventsHtml = (!isUser && !hasParts && msg.exec_events && msg.exec_events.length > 0)
+      ? renderExecEvents(msg.exec_events, i) : '';
     html += `
       <div class="message-row ${msg.role}">
         <div class="message-avatar">${avatar}</div>
         <div style="flex:1;min-width:0">
           ${reasoningHtml}
           ${thoughtHtml}
-          ${content || streamingIndicator ? `<div class="message-bubble">${content}${ttsIndicator}</div>` : ''}
+          ${timelineHtml}
+          ${singleBubbleHtml}
           ${streamingIndicator}
           ${execEventsHtml}
           ${msg.time ? `<div class="message-time">${formatTime(msg.time)}</div>` : ''}
@@ -2582,6 +2614,7 @@ function insertQuick(text) {
 // ══════════════════════════════════════════════════════
 // ── TTS (Text-to-Speech) Manager ──
 // ══════════════════════════════════════════════════════
+// 支持分段流式播放：文本边生成边朗读，遇到句子边界立即合成播放
 
 // Simple hash function for text caching
 function simpleHash(str) {
@@ -2603,6 +2636,13 @@ const ttsManager = {
   cache: new Map(), // textHash -> blobUrl
   voice: 'zh-CN-XiaoxiaoNeural',
   speed: '+0%',
+  // ── 分段流式状态 ──
+  _streamActive: false,     // 是否正在流式模式
+  _streamBuffer: '',        // 当前缓冲区（积累到句子边界前）
+  _audioQueue: [],          // 待播放的音频 blobUrl 队列
+  _audioPlaying: false,     // 队列是否正在播放
+  _stopRequested: false,    // 是否已请求停止
+  _streamMsgIndex: -1,      // 流式模式对应的消息索引
 
   init() {
     // Load TTS enabled state from localStorage
@@ -2613,15 +2653,25 @@ const ttsManager = {
     this.updateButtonUI();
     // Audio event handlers
     this.audio.addEventListener('ended', () => {
-      this.isPlaying = false;
-      this.currentMsgIndex = -1;
-      this.updatePlayingIndicator();
+      if (this._streamActive) {
+        // 流式模式：播放队列下一段
+        this._playNextInQueue();
+      } else {
+        this.isPlaying = false;
+        this.currentMsgIndex = -1;
+        this.updatePlayingIndicator();
+      }
     });
     this.audio.addEventListener('error', (e) => {
       console.error('TTS audio error:', e);
-      this.isPlaying = false;
-      this.currentMsgIndex = -1;
-      this.updatePlayingIndicator();
+      if (this._streamActive) {
+        // 流式模式：跳过错误段，播放下一段
+        this._playNextInQueue();
+      } else {
+        this.isPlaying = false;
+        this.currentMsgIndex = -1;
+        this.updatePlayingIndicator();
+      }
     });
   },
 
@@ -2652,10 +2702,16 @@ const ttsManager = {
   },
 
   stop() {
+    this._stopRequested = true;
     this.audio.pause();
     this.audio.currentTime = 0;
     this.isPlaying = false;
     this.currentMsgIndex = -1;
+    this._streamActive = false;
+    this._streamBuffer = '';
+    this._audioQueue = [];
+    this._audioPlaying = false;
+    this._streamMsgIndex = -1;
     this.updatePlayingIndicator();
   },
 
@@ -2666,10 +2722,216 @@ const ttsManager = {
     }
   },
 
+  // ════════════════════════════════════════════
+  // ── 分段流式 TTS：text_delta 回调 ──
+  // ════════════════════════════════════════════
+
+  /**
+   * 开始流式 TTS 会话
+   * @param {number} msgIndex - 消息索引
+   */
+  _startStream(msgIndex) {
+    this._stopRequested = false;
+    this._streamActive = true;
+    this._streamBuffer = '';
+    this._audioQueue = [];
+    this._audioPlaying = false;
+    this._streamMsgIndex = msgIndex;
+    this.currentMsgIndex = msgIndex;
+    this.isPlaying = true;
+  },
+
+  /**
+   * 流式推送文本增量
+   * 在 flow_engine.js 的 text_delta 处理中调用
+   * 积累到句子边界时自动触发 TTS 合成
+   * @param {string} delta - 新增文本片段
+   */
+  streamDelta(delta) {
+    if (!this.enabled || !this._streamActive || this._stopRequested) return;
+    if (!delta || !delta.trim()) return;
+
+    this._streamBuffer += delta;
+
+    // 检测句子边界：中文句号/感叹号/问号，英文句号+空格，或换行
+    var boundaryPattern = /[。！？]|\.(?:\s|$)|\n/;
+    var boundaryIdx = -1;
+    for (var i = 0; i < this._streamBuffer.length; i++) {
+      if (boundaryPattern.test(this._streamBuffer[i])) {
+        boundaryIdx = i;
+        break;
+      }
+    }
+
+    // 还没到句子边界，但如果缓冲区已经很长（>200字），强制切分
+    if (boundaryIdx === -1 && this._streamBuffer.length > 200) {
+      // 在最后一个逗号或空格处切分
+      var lastComma = -1;
+      for (var j = 0; j < this._streamBuffer.length; j++) {
+        var ch = this._streamBuffer[j];
+        if (ch === '，' || ch === ',' || ch === '；' || ch === ';' || ch === ' ' || ch === ':') {
+          lastComma = j;
+        }
+      }
+      if (lastComma > 0) {
+        boundaryIdx = lastComma;
+      } else {
+        boundaryIdx = this._streamBuffer.length;
+      }
+    }
+
+    if (boundaryIdx !== -1) {
+      // 提取到边界的文本
+      var sentence = this._streamBuffer.substring(0, boundaryIdx + 1).trim();
+      this._streamBuffer = this._streamBuffer.substring(boundaryIdx + 1);
+
+      if (sentence) {
+        var cleanSentence = this._cleanForStreamTTS(sentence);
+        if (cleanSentence) {
+          this._enqueueTTS(cleanSentence);
+        }
+      }
+    }
+  },
+
+  /**
+   * 刷新剩余缓冲区（流结束时调用）
+   * 将 buffer 中剩余的文本立即合成
+   */
+  streamFlush() {
+    if (!this.enabled || !this._streamActive || this._stopRequested) return;
+    var remaining = this._streamBuffer.trim();
+    this._streamBuffer = '';
+    if (remaining) {
+      var cleanText = this._cleanForStreamTTS(remaining);
+      if (cleanText) {
+        this._enqueueTTS(cleanText);
+      }
+    }
+    // 标记流式阶段结束（队列播完后自动清理状态）
+    this._streamActive = false;
+  },
+
+  /**
+   * 清理文本用于流式 TTS（去 HTML/代码块/执行结果等）
+   */
+  _cleanForStreamTTS(text) {
+    // 去除代码块
+    text = text.replace(/```[\s\S]*?```/g, '');
+    // 去除执行结果标记
+    text = text.replace(/^\s*[✅❌⏰]\s*\[执行结果\].*/gm, '');
+    // 去除 HTML 标签
+    text = text.replace(/<svg[^>]*>[\s\S]*?<\/svg>/gi, '');
+    text = text.replace(/<img[^>]*>/gi, '');
+    text = text.replace(/<br\s*\/?>/gi, '\n');
+    text = text.replace(/<[^>]+>/g, '');
+    // 去除 emoji
+    text = text.replace(/[\u{1F300}-\u{1FAFF}]/gu, '');
+    text = text.replace(/[\u{2600}-\u{27BF}]/gu, '');
+    text = text.replace(/[\u{FE00}-\u{FE0F}]/gu, '');
+    text = text.replace(/[\u{200D}]/gu, '');
+    text = text.replace(/[\u{20E3}]/gu, '');
+    text = text.replace(/[\u{2300}-\u{23FF}]/gu, '');
+    text = text.replace(/[\u{2B50}-\u{2B55}]/gu, '');
+    text = text.replace(/[\u{203C}-\u{3299}]/gu, '');
+    text = text.replace(/[\u{E0020}-\u{E007F}]/gu, '');
+    text = text.replace(/[✅❌⚠️🔄⏰🔒💻🔍📁🧠🌐🛠👋🤖🎯💡🚀👍🎯📊📝🔊🔍💬📌✨✓✗→←↓↑⏹⬇⬆↩]/g, '');
+    // 去除多余换行
+    text = text.replace(/\n{2,}/g, '\n');
+    text = text.trim();
+    return text || null;
+  },
+
+  /**
+   * 将文本加入 TTS 合成队列（异步，不阻塞）
+   */
+  _enqueueTTS(text) {
+    if (this._stopRequested) return;
+    var self = this;
+
+    (async function() {
+      try {
+        var hash = simpleHash(text);
+        var blobUrl = self.cache.get(hash);
+
+        if (!blobUrl) {
+          var resp = await fetch('/api/tts', {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({
+              text: text,
+              voice: self.voice,
+              speed: self.speed,
+            }),
+          });
+
+          if (!resp.ok) {
+            var errData = await resp.json().catch(function() { return {}; });
+            throw new Error(errData.error || 'TTS 请求失败');
+          }
+
+          var blob = await resp.blob();
+          blobUrl = URL.createObjectURL(blob);
+          self.cache.set(hash, blobUrl);
+        }
+
+        if (!self._stopRequested) {
+          self._audioQueue.push(blobUrl);
+          // 如果还没开始播放队列，立即开始
+          if (!self._audioPlaying) {
+            self._playNextInQueue();
+          }
+        }
+      } catch (e) {
+        console.error('TTS stream chunk error:', e);
+      }
+    })();
+  },
+
+  /**
+   * 播放队列中的下一段音频
+   */
+  _playNextInQueue() {
+    if (this._stopRequested) {
+      this.isPlaying = false;
+      this._audioPlaying = false;
+      this.currentMsgIndex = -1;
+      this.updatePlayingIndicator();
+      return;
+    }
+
+    if (this._audioQueue.length === 0) {
+      // 队列空了，检查流式是否已结束
+      if (!this._streamActive) {
+        // 流结束且队列为空 → 播放完成
+        this.isPlaying = false;
+        this._audioPlaying = false;
+        this.currentMsgIndex = -1;
+        this.updatePlayingIndicator();
+      }
+      // 如果流还在继续，等待新的音频入队
+      return;
+    }
+
+    var blobUrl = this._audioQueue.shift();
+    this.audio.src = blobUrl;
+    this._audioPlaying = true;
+
+    var self = this;
+    this.audio.play().catch(function(e) {
+      console.error('TTS play queue error:', e);
+      self._playNextInQueue();
+    });
+  },
+
+  // ════════════════════════════════════════════
+  // ── 完整消息 TTS（非流式，兼容手动点击） ──
+  // ════════════════════════════════════════════
+
   async speak(msgIndex) {
     if (msgIndex < 0 || msgIndex >= state.messages.length) return;
     const msg = state.messages[msgIndex];
-    if (!msg || msg.role !== 'user' && !msg.content) return;
+    if (!msg || msg.role !== 'assistant' && !msg.content) return;
 
     // 跳过命令执行结果（以 [执行结果] 开头的消息）
     var rawText = msg.content.replace(/<[^>]+>/g, '');
@@ -2677,24 +2939,23 @@ const ttsManager = {
 
     // 去除 HTML 标签（msg.content 是 HTML 格式，SVG 图标等会被朗读）
     let text = msg.content
-      .replace(/<svg[^>]*>[\s\S]*?<\/svg>/gi, '')      // 移除 SVG 图标
-      .replace(/<img[^>]*>/gi, '[图片]')                   // 图片替换为文字
-      .replace(/<br\s*\/?>/gi, '\n')                       // <br> 转换为换行
-      .replace(/<[^>]+>/g, '')                              // 移除所有 HTML 标签
-      // emoji 和特殊符号过滤
-      .replace(/[\u{1F300}-\u{1FAFF}]/gu, '')            // 全部 Emoji 范围
-      .replace(/[\u{2600}-\u{27BF}]/gu, '')              // 杂项/装饰符号
-      .replace(/[\u{FE00}-\u{FE0F}]/gu, '')              // 变体选择符
-      .replace(/[\u{200D}]/gu, '')                          // ZWJ 零宽连接符
-      .replace(/[\u{20E3}]/gu, '')                          // 组合符号
-      .replace(/[\u{2300}-\u{23FF}]/gu, '')              // 技术符号
-      .replace(/[\u{2B50}-\u{2B55}]/gu, '')              // 星星等
-      .replace(/[\u{203C}-\u{3299}]/gu, '')              // CJK 符号
-      .replace(/[\u{E0020}-\u{E007F}]/gu, '')            // 标签
-      .replace(/[✅❌⚠️🔄⏰🔒💻🔍📁🧠🌐🛠️👋🤖🎯💡🚀👍🎯📊📝🔊🔍💬📌✨✓✗→←↓↑⏹⬇⬆↩]/g, '') // 常用图标
-      .replace(/```[\s\S]*?```/g, '代码块')               // 代码块替换为文字
-      .replace(/`[^`]+`/g, function(m) { return m.slice(1,-1); }) // 保留内联代码文字但去引号
-      .replace(/\n{2,}/g, '\n')                            // 多余换行
+      .replace(/<svg[^>]*>[\s\S]*?<\/svg>/gi, '')
+      .replace(/<img[^>]*>/gi, '[图片]')
+      .replace(/<br\s*\/?>/gi, '\n')
+      .replace(/<[^>]+>/g, '')
+      .replace(/[\u{1F300}-\u{1FAFF}]/gu, '')
+      .replace(/[\u{2600}-\u{27BF}]/gu, '')
+      .replace(/[\u{FE00}-\u{FE0F}]/gu, '')
+      .replace(/[\u{200D}]/gu, '')
+      .replace(/[\u{20E3}]/gu, '')
+      .replace(/[\u{2300}-\u{23FF}]/gu, '')
+      .replace(/[\u{2B50}-\u{2B55}]/gu, '')
+      .replace(/[\u{203C}-\u{3299}]/gu, '')
+      .replace(/[\u{E0020}-\u{E007F}]/gu, '')
+      .replace(/[✅❌⚠️🔄⏰🔒💻🔍📁🧠🌐🛠👋🤖🎯💡🚀👍🎯📊📝🔊🔍💬📌✨✓✗→←↓↑⏹⬇⬆↩]/g, '')
+      .replace(/```[\s\S]*?```/g, '代码块')
+      .replace(/`[^`]+`/g, function(m) { return m.slice(1,-1); })
+      .replace(/\n{2,}/g, '\n')
       .trim();
 
     if (!text) return;

package/web/ui/chat/flow_engine.js

@@ -350,22 +350,87 @@ function updateStreamingMessage(msgIdx) {
     }
   }
 
-  // Update content bubble
-  let bubble = contentArea.querySelector('.message-bubble');
-  const content = renderMarkdown(msg.content);
-  if (content && !bubble) {
-    // Create bubble
-    bubble = document.createElement('div');
-    bubble.className = 'message-bubble';
-    contentArea.appendChild(bubble);
-  }
-  if (bubble && content) {
-    bubble.innerHTML = content;
+  // Update content - timeline (interleaved text + exec events) or single bubble (backward compat)
+  const hasParts = Array.isArray(msg.parts);
+  if (hasParts) {
+    // ── Timeline rendering for interleaved text + exec events ──
+    let timeline = contentArea.querySelector('.msg-timeline');
+    if (!timeline) {
+      // Remove old single bubble if exists
+      const oldBubble = contentArea.querySelector(':scope > .message-bubble');
+      if (oldBubble) oldBubble.remove();
+      // Create timeline container
+      timeline = document.createElement('div');
+      timeline.className = 'msg-timeline';
+      // Insert after thought blocks or at beginning
+      const allThoughts = contentArea.querySelectorAll(':scope > .thought-block');
+      if (allThoughts.length > 0) {
+        allThoughts[allThoughts.length - 1].insertAdjacentElement('afterend', timeline);
+      } else {
+        contentArea.appendChild(timeline);
+      }
+    }
+
+    // Cache completed parts rendering (only re-render when parts count changes)
+    const partsCount = msg.parts.length;
+    if (!msg._renderedPartsHtml || msg._lastPartsCount !== partsCount) {
+      let html = '';
+      for (const part of msg.parts) {
+        if (part.type === 'text' && part.content.trim()) {
+          html += '<div class="message-bubble">' + renderMarkdown(part.content) + '</div>';
+        } else if (part.type === 'exec') {
+          html += renderInlineExecEvent(part.data, msgIdx);
+        }
+      }
+      msg._renderedPartsHtml = html;
+      msg._lastPartsCount = partsCount;
+    }
+
+    // Build streaming bubble for current in-progress text
+    const streamingText = msg._streamingText || '';
+    const streamingBubbleHtml = streamingText.trim()
+      ? '<div class="message-bubble">' + renderMarkdown(streamingText) + '</div>'
+      : '';
+
+    timeline.innerHTML = msg._renderedPartsHtml + streamingBubbleHtml;
+
+    // Remove exec events panel if present (events are now inline in timeline)
+    const execPanel = contentArea.querySelector('.exec-events-panel');
+    if (execPanel) execPanel.remove();
+  } else {
+    // ── Backward compat: single content bubble + exec events panel ──
+    let bubble = contentArea.querySelector('.message-bubble');
+    const content = renderMarkdown(msg.content);
+    if (content && !bubble) {
+      bubble = document.createElement('div');
+      bubble.className = 'message-bubble';
+      contentArea.appendChild(bubble);
+    }
+    if (bubble && content) {
+      bubble.innerHTML = content;
+    }
+
+    // Exec events panel (only for backward compat messages without parts)
+    if (msg.exec_events && msg.exec_events.length > 0) {
+      let execPanel = contentArea.querySelector('.exec-events-panel');
+      const newExecHtml = renderExecEvents(msg.exec_events, msgIdx);
+      if (execPanel) {
+        execPanel.outerHTML = newExecHtml;
+      } else {
+        const timeEl = contentArea.querySelector('.message-time');
+        if (timeEl) {
+          timeEl.insertAdjacentHTML('beforebegin', newExecHtml);
+        } else {
+          contentArea.insertAdjacentHTML('beforeend', newExecHtml);
+        }
+      }
+    }
   }
 
   // Update streaming indicator
   let indicator = contentArea.querySelector('.streaming-indicator');
-  const streamingIndicator = msg.streaming && !msg.content && !msg.thought ? `
+  const anyContent = msg.content || msg._streamingText || (msg.parts && msg.parts.length > 0);
+  const streamingIndicator = msg.streaming && !anyContent && !msg.thought ? `
       <div class="streaming-indicator">
         <div class="streaming-dots">
           <span class="dot"></span><span class="dot"></span><span class="dot"></span>
@@ -380,23 +445,6 @@ function updateStreamingMessage(msgIdx) {
     indicator.remove();
   }
 
-  // Update exec events panel
-  if (msg.exec_events && msg.exec_events.length > 0) {
-    let execPanel = contentArea.querySelector('.exec-events-panel');
-    const newExecHtml = renderExecEvents(msg.exec_events, msgIdx);
-    if (execPanel) {
-      execPanel.outerHTML = newExecHtml;
-    } else {
-      // Insert before time element or at end
-      const timeEl = contentArea.querySelector('.message-time');
-      if (timeEl) {
-        timeEl.insertAdjacentHTML('beforebegin', newExecHtml);
-      } else {
-        contentArea.insertAdjacentHTML('beforeend', newExecHtml);
-      }
-    }
-  }
-
   // Auto-scroll
   scrollToBottom();
 }
@@ -507,6 +555,52 @@ function toggleExecEventsPanel(header) {
   body.classList.toggle('expanded');
 }
 
+// ══════════════════════════════════════════════════════
+// ── Inline Exec Event (Timeline Card) ──
+// ══════════════════════════════════════════════════════
+
+function renderInlineExecEvent(data, msgIdx) {
+  const iconEmoji = getEventIconEmoji(data);
+  const title = data.title || (data.tool_name || data.skill_name || '执行事件');
+
+  // Build meta text
+  let metaParts = [];
+  if (data.execution_time !== undefined) metaParts.push('耗时 ' + data.execution_time + 's');
+  if (data.language) metaParts.push(escapeHtml(data.language));
+  if (data.tool_name || data.skill_name) metaParts.push(escapeHtml(data.tool_name || data.skill_name));
+  if (data.timed_out) metaParts.push('超时');
+  if (data.exit_code !== undefined) metaParts.push('exit: ' + data.exit_code);
+  const metaText = metaParts.join(' · ');
+
+  // Build body content
+  let bodyHtml = '';
+  // Code preview for code_exec/code_result
+  if (data.code_preview && (data.type === 'code_exec' || data.type === 'code_result')) {
+    bodyHtml += '<div class="inline-exec-code" onclick="showExecResultModal(' + msgIdx + ', ' + data.id + ')" title="点击查看完整结果">' + escapeHtml(data.code_preview) + '</div>';
+  }
+  // Summary for tool_result/skill_result
+  if (data.summary && (data.type === 'tool_result' || data.type === 'skill_result')) {
+    bodyHtml += '<div class="inline-exec-summary">' + escapeHtml(data.summary) + '</div>';
+  }
+  // Result button for code_result
+  if (data.type === 'code_result' && (data.stdout || data.stderr || data.error)) {
+    bodyHtml += '<button class="inline-exec-result-btn" onclick="showExecResultModal(' + msgIdx + ', ' + data.id + ')">查看详情</button>';
+  }
+  // Result button for tool_result/skill_result
+  if ((data.type === 'tool_result' || data.type === 'skill_result') && data.result) {
+    bodyHtml += '<button class="inline-exec-result-btn" onclick="showToolResultModal(' + msgIdx + ', ' + data.id + ')">查看详情</button>';
+  }
+
+  return '<div class="inline-exec-event">' +
+    '<div class="inline-exec-header">' +
+      '<span class="inline-exec-icon">' + iconEmoji + '</span>' +
+      '<span class="inline-exec-title">' + escapeHtml(title) + '</span>' +
+      (metaText ? '<span class="inline-exec-meta">' + metaText + '</span>' : '') +
+    '</div>' +
+    bodyHtml +
+  '</div>';
+}
+
 // ══════════════════════════════════════════════════════
 // ── Execution Result Modal (执行结果弹窗) ──
 // ══════════════════════════════════════════════════════
@@ -702,14 +796,22 @@ async function sendMessage() {
     const reader = resp.body.getReader();
     const decoder = new TextDecoder();
     let buffer = '';
-    let fullResponse = '';
+    let msgParts = [];       // Timeline: [{type:'text', content:'...'}, {type:'exec', data:{...}}]
+    let currentText = '';    // Accumulator for current streaming text segment
+    let allExecEvents = [];  // All exec events (for summary panel at bottom)
     let msgIdx = state.messages.length;
     let sessionIdReceived = sessionId;
-    let execEventsReceived = [];
     let fullThought = '';
-    
+
+    function flushCurrentText() {
+      if (currentText.trim()) {
+        msgParts.push({type: 'text', content: currentText});
+      }
+      currentText = '';
+    }
+
     // Add placeholder for streaming response
-    state.messages.push({ role: 'assistant', content: '', thought: '', time: new Date().toISOString(), streaming: true });
+    state.messages.push({ role: 'assistant', content: '', thought: '', parts: [], time: new Date().toISOString(), streaming: true });
     renderMessages();
     
     while (true) {
@@ -730,14 +832,28 @@ async function sendMessage() {
             // Sync the actual session ID (backend may prefix with agent_path)
             state.activeSessionId = evt.session_id;
           } else if (evt.type === 'text') {
-            fullResponse = evt.content;
+            // Full text event (non-streaming replacement)
+            flushCurrentText();
+            msgParts.push({type: 'text', content: evt.content});
+            state.messages[msgIdx].parts = [...msgParts];
+            state.messages[msgIdx]._streamingText = '';
             state.messages[msgIdx].content = evt.content;
             renderMessages();
           } else if (evt.type === 'text_delta') {
             // Incremental streaming token
-            fullResponse += evt.content;
-            state.messages[msgIdx].content = fullResponse;
+            currentText += evt.content;
+            // Build backward-compat content from all parts + streaming text
+            const allText = msgParts.filter(p => p.type === 'text').map(p => p.content).join('\n\n')
+              + (currentText.trim() ? '\n\n' + currentText : '');
+            state.messages[msgIdx].parts = [...msgParts];
+            state.messages[msgIdx]._streamingText = currentText;
+            state.messages[msgIdx].content = allText;
             throttledStreamUpdate(msgIdx);
+            // ── 分段流式 TTS：推送增量文本 ──
+            if (ttsManager.enabled && !ttsManager._streamActive) {
+              ttsManager._startStream(msgIdx);
+            }
+            ttsManager.streamDelta(evt.content);
           } else if (evt.type === 'thought_delta') {
             // Agent 思考过程增量文本（流式推送，单独显示）
             fullThought += evt.content;
@@ -753,28 +869,39 @@ async function sendMessage() {
             state.messages[msgIdx].thought = fullThought;
             throttledStreamUpdate(msgIdx);
           } else if (evt.type === 'queue_start') {
-            // New message starting from queue
+            // Finalize previous message
+            flushCurrentText();
             if (state.messages[msgIdx]) {
               state.messages[msgIdx].streaming = false;
-              if (execEventsReceived.length > 0) state.messages[msgIdx].exec_events = [...execEventsReceived];
+              state.messages[msgIdx].parts = [...msgParts];
+              state.messages[msgIdx].content = msgParts.filter(p => p.type === 'text').map(p => p.content).join('\n\n') || '(无回复)';
+              state.messages[msgIdx]._streamingText = '';
+              if (allExecEvents.length > 0) state.messages[msgIdx].exec_events = [...allExecEvents];
             }
+            // Start new message
             state.messages.push({ role: 'user', content: evt.message, time: new Date().toISOString() });
             msgIdx = state.messages.length;
-            fullResponse = '';
+            msgParts = [];
+            currentText = '';
+            allExecEvents = [];
             fullThought = '';
-            execEventsReceived = [];
-            state.messages.push({ role: 'assistant', content: '', thought: '', time: new Date().toISOString(), streaming: true });
+            state.messages.push({ role: 'assistant', content: '', thought: '', parts: [], time: new Date().toISOString(), streaming: true });
             renderMessages();
           } else if (evt.type === 'clear_text') {
             // Clear intermediate text from previous agent loop iterations
-            fullResponse = '';
-            state.messages[msgIdx].content = '';
+            flushCurrentText();
+            state.messages[msgIdx].parts = [...msgParts];
+            state.messages[msgIdx]._streamingText = '';
+            state.messages[msgIdx].content = msgParts.filter(p => p.type === 'text').map(p => p.content).join('\n\n') || '';
             throttledStreamUpdate(msgIdx);
           } else if (evt.type === 'exec_event') {
             // Real-time execution event (tool call, code exec, skill result, etc.)
-            execEventsReceived.push(evt.data);
-            // 立即更新消息的 exec_events 并渲染
-            state.messages[msgIdx].exec_events = [...execEventsReceived];
+            flushCurrentText();
+            msgParts.push({type: 'exec', data: evt.data});
+            allExecEvents.push(evt.data);
+            state.messages[msgIdx].parts = [...msgParts];
+            state.messages[msgIdx]._streamingText = '';
+            state.messages[msgIdx].exec_events = [...allExecEvents];
             throttledStreamUpdate(msgIdx);
           } else if (evt.type === 'task_list_update') {
             // 任务列表 JSON 直推更新（exec 模式）
@@ -797,11 +924,15 @@ async function sendMessage() {
               }
             }
           } else if (evt.type === 'done') {
+            flushCurrentText();
             // done 事件提供最终事件列表（可能有去重/合并）
             if (evt.exec_events && evt.exec_events.length > 0) {
-              execEventsReceived = evt.exec_events;
-              state.messages[msgIdx].exec_events = [...execEventsReceived];
+              allExecEvents = evt.exec_events;
             }
+            state.messages[msgIdx].parts = [...msgParts];
+            state.messages[msgIdx]._streamingText = '';
+            state.messages[msgIdx].exec_events = [...allExecEvents];
+            state.messages[msgIdx].content = msgParts.filter(p => p.type === 'text').map(p => p.content).join('\n\n') || '(无回复)';
           } else if (evt.type === 'reasoning_delta') {
             // 模型推理过程增量文本（OpenAI o1/o3/DeepSeek-R1 等推理模型）
             if (!state.messages[msgIdx].reasoning) state.messages[msgIdx].reasoning = '';
@@ -812,22 +943,25 @@ async function sendMessage() {
             state.messages[msgIdx].reasoning = evt.content;
             throttledStreamUpdate(msgIdx);
           } else if (evt.type === 'error') {
-            fullResponse = '❌ ' + evt.error;
-            state.messages[msgIdx].content = fullResponse;
+            flushCurrentText();
+            currentText = '❌ ' + evt.error;
+            msgParts.push({type: 'text', content: currentText});
+            state.messages[msgIdx].parts = [...msgParts];
+            state.messages[msgIdx]._streamingText = '';
+            state.messages[msgIdx].content = msgParts.filter(p => p.type === 'text').map(p => p.content).join('\n\n');
           }
         } catch (e) { /* skip malformed */ }
       }
     }
     
     // Finalize message
+    flushCurrentText();
     if (state.messages[msgIdx]) {
       state.messages[msgIdx].streaming = false;
-      if (execEventsReceived.length > 0) {
-        state.messages[msgIdx].exec_events = execEventsReceived;
-      }
-      if (!state.messages[msgIdx].content) {
-        state.messages[msgIdx].content = '(无回复)';
-      }
+      state.messages[msgIdx].parts = [...msgParts];
+      state.messages[msgIdx]._streamingText = '';
+      state.messages[msgIdx].exec_events = allExecEvents;
+      state.messages[msgIdx].content = msgParts.filter(p => p.type === 'text').map(p => p.content).join('\n\n') || '(无回复)';
     }
     
     // Task list 已通过 SSE task_list_update 事件实时推送，无需再轮询
@@ -843,10 +977,9 @@ async function sendMessage() {
     state.agentSessions[state.activeAgent] = [...state.sessions];
     renderSessions();
 
-    // Auto-play TTS if enabled (skip command execution results)
-    if (ttsManager.enabled && fullResponse && !fullResponse.match(/^\s*[✅❌⏰]\s*\[执行结果\]/m)) {
-      const idx = state.messages.length - 1;
-      ttsManager.speak(idx);
+    // ── 分段流式 TTS：刷新剩余缓冲区 ──
+    if (ttsManager.enabled && ttsManager._streamActive) {
+      ttsManager.streamFlush();
     }
   } catch (e) {
     if (e.name === 'AbortError') {