myagent-ai 1.9.3 → 1.9.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.9.3",
+  "version": "1.9.4",
   "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/ui/chat/flow_engine.js

@@ -814,6 +814,31 @@ function showToolResultModal(msgIndex, eventId) {
   document.body.appendChild(overlay);
 }
 
+// ══════════════════════════════════════════════════════
+// ── V2 Content Assembler (V2 内容组装) ──
+// ══════════════════════════════════════════════════════
+
+function _assembleV2Content(msg, msgParts) {
+  // Priority 1: V2 reasoning text (user-facing response from v2_reasoning events)
+  if (msg._v2Reasoning && msg._v2Reasoning.trim()) {
+    return msg._v2Reasoning.trim();
+  }
+  // Priority 2: V2 ask user text
+  if (msg._askUser && msg._askUser.trim()) {
+    return msg._askUser.trim();
+  }
+  // Priority 3: V1 text parts (backward compat — non-V2 mode)
+  var textParts = msgParts.filter(function(p) { return p.type === 'text'; });
+  if (textParts.length > 0) {
+    return textParts.map(function(p) { return p.content; }).join('\n\n');
+  }
+  // Priority 4: raw content from message (server-stored response)
+  if (msg.content && msg.content.trim() && msg.content !== '(无回复)') {
+    return msg.content.trim();
+  }
+  return '(无回复)';
+}
+
 // ══════════════════════════════════════════════════════
 // ── Send Message (核心 SSE 流式消息发送) ──
 // ══════════════════════════════════════════════════════
@@ -904,10 +929,17 @@ async function sendMessage() {
     let msgIdx = state.messages.length;
     let sessionIdReceived = sessionId;
     let fullThought = '';
+    let _isV2Mode = false;   // Track whether V2 structured output events are received
+    let _v2RawXml = '';      // In V2 mode, accumulate raw XML from text_delta separately
+    let _v2ReasoningText = ''; // Accumulated V2 reasoning text for content
 
     function flushCurrentText() {
       if (currentText.trim()) {
-        msgParts.push({type: 'text', content: currentText});
+        // In V2 mode, text_delta contains raw XML — do NOT add to msgParts as user-facing text
+        if (!_isV2Mode) {
+          msgParts.push({type: 'text', content: currentText});
+        }
+        // Always accumulate for backward compat content field
       }
       currentText = '';
     }
@@ -950,12 +982,16 @@ async function sendMessage() {
           } else if (evt.type === 'text_delta') {
             // Incremental streaming token
             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 : '');
+            if (_isV2Mode) {
+              // In V2 mode, text_delta contains raw XML — store separately, not as user-facing content
+              _v2RawXml = currentText;
+              // Show V2 reasoning text as streaming content (if available)
+              state.messages[msgIdx]._streamingText = _v2ReasoningText || '';
+            } else {
+              // V1 mode: text_delta IS the user-facing content
+              state.messages[msgIdx]._streamingText = 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) {
@@ -982,7 +1018,7 @@ async function sendMessage() {
             if (state.messages[msgIdx]) {
               state.messages[msgIdx].streaming = false;
               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].content = _assembleV2Content(state.messages[msgIdx], msgParts);
               state.messages[msgIdx]._streamingText = '';
               if (allExecEvents.length > 0) state.messages[msgIdx].exec_events = [...allExecEvents];
             }
@@ -1000,7 +1036,7 @@ async function sendMessage() {
             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') || '';
+            state.messages[msgIdx].content = _assembleV2Content(state.messages[msgIdx], msgParts);
             throttledStreamUpdate(msgIdx);
             // 停止当前轮次的 TTS 播放，防止旧迭代语音与新迭代语音互相打断
             if (ttsManager.enabled && ttsManager._streamActive) {
@@ -1041,7 +1077,10 @@ async function sendMessage() {
                 }
               }
             }
-          // V2 Structured Output Events
+          // V2 Structured Output Events — mark V2 mode
+          } else if (evt.type === 'v2_context') {
+            // V2 context event — confirms we're in V2 mode
+            _isV2Mode = true;
           } else if (evt.type === 'v2_output_parsed') {
             // LLM output was parsed into structured format
             // evt.data contains: {usersays_correct, task_plan, tools_to_call, remember, recall, ask_user, finish}
@@ -1120,11 +1159,22 @@ async function sendMessage() {
             state.messages[msgIdx]._memorySaved = (state.messages[msgIdx]._memorySaved || '') +
               (evt.content ? evt.content.substring(0, 100) : '');
           } else if (evt.type === 'v2_reasoning') {
-            // V2 reasoning text from model (similar to thought_delta)
+            // V2 reasoning text from model — this IS the user-facing content in V2 mode
             if (!state.messages[msgIdx]._v2Reasoning) {
               state.messages[msgIdx]._v2Reasoning = '';
             }
+            // Add separator between multiple reasoning events (e.g., beforecalltext + final text)
+            if (state.messages[msgIdx]._v2Reasoning.length > 0 && evt.content) {
+              state.messages[msgIdx]._v2Reasoning += '\n\n';
+            }
             state.messages[msgIdx]._v2Reasoning += evt.content;
+            _v2ReasoningText = state.messages[msgIdx]._v2Reasoning;
+            // In V2 mode, update streaming text to show reasoning content
+            if (_isV2Mode) {
+              state.messages[msgIdx]._streamingText = _v2ReasoningText;
+              // Also update content for real-time display
+              state.messages[msgIdx].content = _v2ReasoningText;
+            }
             throttledStreamUpdate(msgIdx);
           } else if (evt.type === 'done') {
             flushCurrentText();
@@ -1135,7 +1185,8 @@ async function sendMessage() {
             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)';
+            // Assemble final content: prefer V2 reasoning/ask text over raw XML
+            state.messages[msgIdx].content = _assembleV2Content(state.messages[msgIdx], msgParts);
           } 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 = '';
@@ -1164,7 +1215,8 @@ async function sendMessage() {
       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)';
+      // Assemble final content: prefer V2 reasoning/ask text over raw XML
+      state.messages[msgIdx].content = _assembleV2Content(state.messages[msgIdx], msgParts);
     }
     
     // Task list 已通过 SSE task_list_update 事件实时推送，无需再轮询