.claude/worktrees/dazzling-leakey

@@ -1 +1 @@
-Subproject commit 26f91ef0e47bc70526accb11dc43d2c20ea07a55
+Subproject commit 4f219bd74ef1d1c40f309bc5a2ec8c395111fd48

frontend/components/analysis/AnalysisForm.tsx

@@ -52,6 +52,7 @@ const formSchema = z.object({
     .regex(/^\d{4}-\d{2}-\d{2}$/, "日期格式必須為 YYYY-MM-DD"),
   analysts: z.array(z.string()).min(1, "請至少選擇一位分析師"),
   research_depth: z.number().int().min(1).max(5),
+  analysis_mode: z.enum(["fast", "deep"]).optional().default("deep"),
   quick_think_llm: z.string().min(1, "請選擇快速思維模型"),
   deep_think_llm: z.string().min(1, "請選擇深層思維模型"),
   embedding_model: z.string().min(1, "請選擇嵌入式模型"),
@@ -111,6 +112,7 @@ export function AnalysisForm({ onSubmit, loading = false }: AnalysisFormProps) {
       analysis_date: format(new Date(), "yyyy-MM-dd"),
       analysts: ["market", "social", "news", "fundamentals"], // 預設全選
       research_depth: 3, // 預設中等層級
+      analysis_mode: "deep", // 預設深層分析
       market_type: "us", // 預設美股
       quick_think_llm: "gpt-5-mini",
       deep_think_llm: "gpt-5-mini",
@@ -473,6 +475,35 @@ export function AnalysisForm({ onSubmit, loading = false }: AnalysisFormProps) {
                 />
               </div>
 
+              {/* 分析模式行 */}
+              <div className="md:col-span-2 grid grid-cols-1 md:grid-cols-2 gap-6">
+                <FormField
+                  control={form.control}
+                  name="analysis_mode"
+                  render={({ field }) => (
+                    <FormItem>
+                      <FormLabel>分析模式</FormLabel>
+                      <Select
+                        onValueChange={field.onChange}
+                        defaultValue={field.value}
+                      >
+                        <FormControl>
+                          <SelectTrigger>
+                            <SelectValue placeholder="選擇分析模式" />
+                          </SelectTrigger>
+                        </FormControl>
+                        <SelectContent>
+                          <SelectItem value="fast">快速分析 (~15-25 分鐘)</SelectItem>
+                          <SelectItem value="deep">深層分析 (~1 小時)</SelectItem>
+                        </SelectContent>
+                      </Select>
+                      <FormDescription>快速模式跳過辯論，深層模式包含投資和風險辯論</FormDescription>
+                      <FormMessage />
+                    </FormItem>
+                  )}
+                />
+              </div>
+
               {/* 第二行：研究深度、快速思維模型、深層思維模型、嵌入式模型（4列） */}
               <div className="md:col-span-2 grid grid-cols-1 md:grid-cols-4 gap-6">
                 <FormField

tradingagents/agents/analysts/fundamentals_analyst.py

@@ -1,7 +1,7 @@
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 import time
 import json
-from tradingagents.agents.utils.agent_utils import get_fundamentals, get_balance_sheet, get_cashflow, get_income_statement, get_insider_sentiment, get_insider_transactions, filter_messages_for_analyst
+from tradingagents.agents.utils.agent_utils import get_fundamentals, get_balance_sheet, get_cashflow, get_income_statement, get_insider_sentiment, get_insider_transactions
 from tradingagents.agents.utils.prompts import get_fundamentals_analyst_prompt, get_agent_role_instruction, get_context_message
 from tradingagents.dataflows.config import get_config
 
@@ -60,8 +60,7 @@ def create_fundamentals_analyst(llm, language: str = "zh-TW"):
 
         chain = prompt | llm.bind_tools(tools)
 
-        tool_names = {t.name for t in tools}
-        result = chain.invoke(filter_messages_for_analyst(state["messages"], tool_names))
+        result = chain.invoke(state["messages"])
 
         # Report logic: only save report when LLM gives final response
         report = state.get("fundamentals_report", "")

tradingagents/agents/analysts/market_analyst.py

@@ -1,7 +1,7 @@
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 import time
 import json
-from tradingagents.agents.utils.agent_utils import get_stock_data, get_indicators, filter_messages_for_analyst
+from tradingagents.agents.utils.agent_utils import get_stock_data, get_indicators
 from tradingagents.agents.utils.prompts import get_language_instruction, get_agent_role_instruction, get_context_message
 from tradingagents.dataflows.config import get_config
 
@@ -140,8 +140,7 @@ Please provide a professional, precise, and actionable technical analysis report
 
         chain = prompt | llm.bind_tools(tools)
 
-        tool_names = {t.name for t in tools}
-        result = chain.invoke(filter_messages_for_analyst(state["messages"], tool_names))
+        result = chain.invoke(state["messages"])
 
         # Report logic: only save report when LLM gives final response
         report = state.get("market_report", "")

tradingagents/agents/analysts/news_analyst.py

@@ -1,7 +1,7 @@
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 import time
 import json
-from tradingagents.agents.utils.agent_utils import get_news, get_global_news, filter_messages_for_analyst
+from tradingagents.agents.utils.agent_utils import get_news, get_global_news
 from tradingagents.agents.utils.prompts import get_news_analyst_prompt, get_agent_role_instruction, get_context_message
 from tradingagents.dataflows.config import get_config
 
@@ -57,8 +57,7 @@ def create_news_analyst(llm, language: str = "zh-TW"):
         prompt = prompt.partial(tool_names=", ".join([tool.name for tool in tools]))
 
         chain = prompt | llm.bind_tools(tools)
-        tool_names = {t.name for t in tools}
-        result = chain.invoke(filter_messages_for_analyst(state["messages"], tool_names))
+        result = chain.invoke(state["messages"])
 
         # Report logic: only save report when LLM gives final response
         report = state.get("news_report", "")

tradingagents/agents/analysts/social_media_analyst.py

@@ -1,7 +1,7 @@
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 import time
 import json
-from tradingagents.agents.utils.agent_utils import get_news, filter_messages_for_analyst
+from tradingagents.agents.utils.agent_utils import get_news
 from tradingagents.agents.utils.prompts import get_social_analyst_prompt, get_agent_role_instruction, get_context_message
 from tradingagents.dataflows.config import get_config
 
@@ -57,8 +57,7 @@ def create_social_media_analyst(llm, language: str = "zh-TW"):
 
         chain = prompt | llm.bind_tools(tools)
 
-        tool_names = {t.name for t in tools}
-        result = chain.invoke(filter_messages_for_analyst(state["messages"], tool_names))
+        result = chain.invoke(state["messages"])
 
         # Report logic: only save report when LLM gives final response
         report = state.get("sentiment_report", "")

tradingagents/agents/utils/agent_utils.py

@@ -1,4 +1,4 @@
-from langchain_core.messages import HumanMessage, RemoveMessage, AIMessage, ToolMessage
+from langchain_core.messages import HumanMessage, RemoveMessage
 
 # 從獨立的工具程式檔案匯入工具
 from tradingagents.agents.utils.core_stock_tools import (
@@ -20,42 +20,6 @@ from tradingagents.agents.utils.news_data_tools import (
     get_global_news
 )
 
-def filter_messages_for_analyst(messages, analyst_tool_names):
-    """
-    Filter shared state messages to only include conversations relevant to this analyst.
-
-    In parallel analyst execution, all analysts share state["messages"]. When analyst A
-    makes tool_use calls, those get merged into shared state. When analyst B runs its
-    second iteration, it sees orphaned tool_use blocks from analyst A without matching
-    tool_results, causing anthropic.BadRequestError (400).
-
-    This function keeps only:
-    - Non-tool messages (HumanMessage etc.)
-    - AIMessages whose tool_calls ALL belong to this analyst's tools
-    - ToolMessages whose tool_call_id matches this analyst's tool calls
-    """
-    analyst_tool_call_ids = set()
-    for msg in messages:
-        if isinstance(msg, AIMessage) and msg.tool_calls:
-            for tc in msg.tool_calls:
-                if tc["name"] in analyst_tool_names:
-                    analyst_tool_call_ids.add(tc["id"])
-
-    result = []
-    for msg in messages:
-        if isinstance(msg, AIMessage) and msg.tool_calls:
-            if all(tc["name"] in analyst_tool_names for tc in msg.tool_calls):
-                result.append(msg)
-            # Skip AI messages containing tool_calls from other parallel analysts
-        elif isinstance(msg, ToolMessage):
-            if msg.tool_call_id in analyst_tool_call_ids:
-                result.append(msg)
-        else:
-            result.append(msg)
-
-    return result
-
-
 def create_msg_delete():
     """
     建立一個刪除訊息的函式。