diff --git a/README.md b/README.md
index f9d7c32..32cf091 100644
--- a/README.md
+++ b/README.md
@@ -124,25 +124,80 @@ QuantDinger’s agents don’t start from scratch every time. The backend includ
 - **Where it lives**: Local SQLite files under `backend_api_python/data/memory/` (privacy-first).
 
 ```mermaid
-flowchart TD
-  A[POST /api/analysis/multi] --> B[AnalysisService]
-  B --> C[AgentCoordinator]
-  C --> D[Build context: price/kline/news/indicators]
+flowchart TB
+    %% ===== 🌐 Entry Layer =====
+    subgraph Entry["🌐 API Entry"]
+        A["📡 POST /api/analysis/multi"]
+        A2["🔄 POST /api/analysis/reflect"]
+    end
 
-  subgraph Agents[Agents]
-    E[Analysts + Researchers + Trader]
-  end
+    %% ===== ⚙️ Service Layer =====
+    subgraph Service["⚙️ Service Orchestration"]
+        B[AnalysisService]
+        C[AgentCoordinator]
+        D["📊 Build Context<br/>price · kline · news · indicators"]
+    end
 
-  C --> E
-  E -->|get_memories()| M[(SQLite: role_memory.db)]
-  M -->|Top-K memories| E
-  E -->|memory_prompt injected into system prompt| LLM[LLMService (OpenRouter)]
+    %% ===== 🤖 Multi-Agent Workflow =====
+    subgraph Agents["🤖 Multi-Agent Workflow"]
 
-  C --> R[ReflectionService.record_analysis]
-  R --> RR[(SQLite: reflection_records.db)]
-  W[ReflectionWorker (optional)] --> RR
-  W -->|auto-verify + learn| M
-  A2[POST /api/analysis/reflect] -->|manual learn| M
+        subgraph P1["📈 Phase 1 · Analysis (Parallel)"]
+            E1["🔍 MarketAnalyst<br/><i>Technical</i>"]
+            E2["📑 FundamentalAnalyst<br/><i>Fundamentals</i>"]
+            E3["📰 NewsAnalyst<br/><i>News & Events</i>"]
+            E4["💭 SentimentAnalyst<br/><i>Market Mood</i>"]
+            E5["⚠️ RiskAnalyst<br/><i>Risk Assessment</i>"]
+        end
+
+        subgraph P2["🎯 Phase 2 · Debate (Parallel)"]
+            F1["🐂 BullResearcher<br/><i>Bullish Case</i>"]
+            F2["🐻 BearResearcher<br/><i>Bearish Case</i>"]
+        end
+
+        subgraph P3["💹 Phase 3 · Decision"]
+            G["🎰 TraderAgent<br/><i>Final Verdict → BUY / SELL / HOLD</i>"]
+        end
+
+    end
+
+    %% ===== 🧠 Memory Layer =====
+    subgraph Memory["🧠 Local SQLite Memory (data/memory/)"]
+        M1[("market_analyst")]
+        M2[("fundamental")]
+        M3[("news_analyst")]
+        M4[("sentiment")]
+        M5[("risk_analyst")]
+        M6[("bull_researcher")]
+        M7[("bear_researcher")]
+        M8[("trader_agent")]
+    end
+
+    %% ===== 🔄 Reflection Loop =====
+    subgraph Reflect["🔄 Reflection Loop (Optional)"]
+        R[ReflectionService]
+        RR[("reflection_records.db")]
+        W["⏰ ReflectionWorker"]
+    end
+
+    %% ===== Main Flow =====
+    A --> B --> C --> D
+    D --> P1 --> P2 --> P3
+
+    %% ===== Memory Read/Write =====
+    E1 <-.-> M1
+    E2 <-.-> M2
+    E3 <-.-> M3
+    E4 <-.-> M4
+    E5 <-.-> M5
+    F1 <-.-> M6
+    F2 <-.-> M7
+    G <-.-> M8
+
+    %% ===== Reflection Flow =====
+    C --> R --> RR
+    W --> RR
+    W -.->|"verify + learn"| M8
+    A2 -.->|"manual review"| M8
 ```
 
 **Retrieval ranking (simplified)**:
diff --git a/README_CN.md b/README_CN.md
index b1d40eb..e59914a 100644
--- a/README_CN.md
+++ b/README_CN.md
@@ -126,57 +126,80 @@ QuantDinger 的多智能体不是“每次从零开始”。它内置了一个**
 #### 逻辑图（从请求到记忆闭环）
 
 ```mermaid
-flowchart TD
-  A[POST /api/analysis/multi] --> B[AnalysisService]
-  B --> C[AgentCoordinator]
+flowchart TB
+    %% ===== 🌐 入口层 =====
+    subgraph Entry["🌐 API 入口"]
+        A["📡 POST /api/analysis/multi"]
+        A2["🔄 POST /api/analysis/reflect"]
+    end
 
-  C --> D[构建上下文: price/kline/news/indicators]
+    %% ===== ⚙️ 服务编排层 =====
+    subgraph Service["⚙️ 服务编排"]
+        B[AnalysisService]
+        C[AgentCoordinator]
+        D["📊 构建上下文<br/>price · kline · news · indicators"]
+    end
 
-  subgraph Agents[多智能体并行/串行工作流]
-    E1[MarketAnalyst]
-    E2[FundamentalAnalyst]
-    E3[NewsAnalyst]
-    E4[SentimentAnalyst]
-    E5[RiskAnalyst]
-    F1[BullResearcher]
-    F2[BearResearcher]
-    G[TraderAgent]
-  end
+    %% ===== 🤖 多智能体工作流 =====
+    subgraph Agents["🤖 多智能体工作流"]
 
-  C -->|Phase 1 并行| E1
-  C -->|Phase 1 并行| E2
-  C -->|Phase 1 并行| E3
-  C -->|Phase 1 并行| E4
-  C -->|Phase 1 并行| E5
-  C -->|Phase 2 并行| F1
-  C -->|Phase 2 并行| F2
-  C -->|Phase 3| G
+        subgraph P1["📈 Phase 1 · 多维分析（并行）"]
+            E1["🔍 MarketAnalyst<br/><i>技术面分析</i>"]
+            E2["📑 FundamentalAnalyst<br/><i>基本面分析</i>"]
+            E3["📰 NewsAnalyst<br/><i>新闻舆情</i>"]
+            E4["💭 SentimentAnalyst<br/><i>市场情绪</i>"]
+            E5["⚠️ RiskAnalyst<br/><i>风险评估</i>"]
+        end
 
-  subgraph MemDB[本地 SQLite 记忆库（按角色拆分）]
-    M1[(data/memory/market_analyst_memory.db)]
-    M2[(data/memory/fundamental_analyst_memory.db)]
-    M3[(data/memory/news_analyst_memory.db)]
-    M4[(data/memory/sentiment_analyst_memory.db)]
-    M5[(data/memory/risk_analyst_memory.db)]
-    M6[(data/memory/bull_researcher_memory.db)]
-    M7[(data/memory/bear_researcher_memory.db)]
-    M8[(data/memory/trader_agent_memory.db)]
-  end
+        subgraph P2["🎯 Phase 2 · 多空博弈（并行）"]
+            F1["🐂 BullResearcher<br/><i>看多论据</i>"]
+            F2["🐻 BearResearcher<br/><i>看空论据</i>"]
+        end
 
-  E1 <-->|get_memories / add_memory| M1
-  E2 <-->|get_memories / add_memory| M2
-  E3 <-->|get_memories / add_memory| M3
-  E4 <-->|get_memories / add_memory| M4
-  E5 <-->|get_memories / add_memory| M5
-  F1 <-->|get_memories / add_memory| M6
-  F2 <-->|get_memories / add_memory| M7
-  G  <-->|get_memories / add_memory| M8
+        subgraph P3["💹 Phase 3 · 交易决策"]
+            G["🎰 TraderAgent<br/><i>综合研判 → BUY / SELL / HOLD</i>"]
+        end
 
-  C --> R[ReflectionService.record_analysis]
-  R --> RR[(data/memory/reflection_records.db)]
-  W[ReflectionWorker（可选，定时）] --> RR
-  W -->|到期验证 + 写回经验| M8
-  A2[POST /api/analysis/reflect（手动复盘）] -->|reflect_and_learn| M8
+    end
+
+    %% ===== 🧠 记忆层 =====
+    subgraph Memory["🧠 本地记忆库 SQLite（data/memory/）"]
+        M1[("market_analyst")]
+        M2[("fundamental")]
+        M3[("news_analyst")]
+        M4[("sentiment")]
+        M5[("risk_analyst")]
+        M6[("bull_researcher")]
+        M7[("bear_researcher")]
+        M8[("trader_agent")]
+    end
+
+    %% ===== 🔄 反思闭环 =====
+    subgraph Reflect["🔄 反思闭环（可选）"]
+        R[ReflectionService]
+        RR[("reflection_records.db")]
+        W["⏰ ReflectionWorker"]
+    end
+
+    %% ===== 主流程 =====
+    A --> B --> C --> D
+    D --> P1 --> P2 --> P3
+
+    %% ===== 记忆读写 =====
+    E1 <-.-> M1
+    E2 <-.-> M2
+    E3 <-.-> M3
+    E4 <-.-> M4
+    E5 <-.-> M5
+    F1 <-.-> M6
+    F2 <-.-> M7
+    G <-.-> M8
+
+    %% ===== 反思流程 =====
+    C --> R --> RR
+    W --> RR
+    W -.->|"验证 + 学习"| M8
+    A2 -.->|"手动复盘"| M8
 ```
 
 #### 1) 记忆是如何“注入提示词”的？
diff --git a/README_JA.md b/README_JA.md
index a5515d3..e523358 100644
--- a/README_JA.md
+++ b/README_JA.md
@@ -126,22 +126,80 @@ QuantDinger のエージェントは毎回「ゼロから」ではありませ
 #### フロー図（リクエスト → 記憶閉ループ）
 
 ```mermaid
-flowchart TD
-  A[POST /api/analysis/multi] --> B[AnalysisService]
-  B --> C[AgentCoordinator]
-  C --> D[コンテキスト構築: price/kline/news/indicators]
+flowchart TB
+    %% ===== 🌐 入口層 =====
+    subgraph Entry["🌐 API エントリー"]
+        A["📡 POST /api/analysis/multi"]
+        A2["🔄 POST /api/analysis/reflect"]
+    end
 
-  subgraph Agents[Agents]
-    E[Analysts + Researchers + Trader]
-  end
+    %% ===== ⚙️ サービス層 =====
+    subgraph Service["⚙️ サービス編成"]
+        B[AnalysisService]
+        C[AgentCoordinator]
+        D["📊 コンテキスト構築<br/>price · kline · news · indicators"]
+    end
 
-  C --> E
-  E -->|get_memories / add_memory| M[(SQLite: data/memory/*_memory.db)]
-  C --> R[ReflectionService.record_analysis]
-  R --> RR[(SQLite: data/memory/reflection_records.db)]
-  W[ReflectionWorker（任意）] --> RR
-  W -->|期限到来を検証し学習| M
-  A2[POST /api/analysis/reflect] -->|手動で学習| M
+    %% ===== 🤖 マルチエージェントワークフロー =====
+    subgraph Agents["🤖 マルチエージェントワークフロー"]
+
+        subgraph P1["📈 Phase 1 · 多次元分析（並列）"]
+            E1["🔍 MarketAnalyst<br/><i>テクニカル</i>"]
+            E2["📑 FundamentalAnalyst<br/><i>ファンダメンタル</i>"]
+            E3["📰 NewsAnalyst<br/><i>ニュース</i>"]
+            E4["💭 SentimentAnalyst<br/><i>センチメント</i>"]
+            E5["⚠️ RiskAnalyst<br/><i>リスク評価</i>"]
+        end
+
+        subgraph P2["🎯 Phase 2 · 強気弱気論争（並列）"]
+            F1["🐂 BullResearcher<br/><i>強気の論点</i>"]
+            F2["🐻 BearResearcher<br/><i>弱気の論点</i>"]
+        end
+
+        subgraph P3["💹 Phase 3 · 取引判断"]
+            G["🎰 TraderAgent<br/><i>総合判定 → BUY / SELL / HOLD</i>"]
+        end
+
+    end
+
+    %% ===== 🧠 記憶層 =====
+    subgraph Memory["🧠 ローカル SQLite 記憶庫（data/memory/）"]
+        M1[("market_analyst")]
+        M2[("fundamental")]
+        M3[("news_analyst")]
+        M4[("sentiment")]
+        M5[("risk_analyst")]
+        M6[("bull_researcher")]
+        M7[("bear_researcher")]
+        M8[("trader_agent")]
+    end
+
+    %% ===== 🔄 反省ループ =====
+    subgraph Reflect["🔄 反省ループ（オプション）"]
+        R[ReflectionService]
+        RR[("reflection_records.db")]
+        W["⏰ ReflectionWorker"]
+    end
+
+    %% ===== メインフロー =====
+    A --> B --> C --> D
+    D --> P1 --> P2 --> P3
+
+    %% ===== 記憶読み書き =====
+    E1 <-.-> M1
+    E2 <-.-> M2
+    E3 <-.-> M3
+    E4 <-.-> M4
+    E5 <-.-> M5
+    F1 <-.-> M6
+    F2 <-.-> M7
+    G <-.-> M8
+
+    %% ===== 反省フロー =====
+    C --> R --> RR
+    W --> RR
+    W -.->|"検証 + 学習"| M8
+    A2 -.->|"手動復習"| M8
 ```
 
 #### 検索ランキング（簡略）
diff --git a/README_KO.md b/README_KO.md
index e6cf0b7..d221e56 100644
--- a/README_KO.md
+++ b/README_KO.md
@@ -126,22 +126,80 @@ QuantDinger의 에이전트는 매번 “처음부터” 시작하지 않습니
 #### 흐름도(요청 → 메모리 폐루프)
 
 ```mermaid
-flowchart TD
-  A[POST /api/analysis/multi] --> B[AnalysisService]
-  B --> C[AgentCoordinator]
-  C --> D[컨텍스트 구성: price/kline/news/indicators]
+flowchart TB
+    %% ===== 🌐 진입 레이어 =====
+    subgraph Entry["🌐 API 진입점"]
+        A["📡 POST /api/analysis/multi"]
+        A2["🔄 POST /api/analysis/reflect"]
+    end
 
-  subgraph Agents[Agents]
-    E[Analysts + Researchers + Trader]
-  end
+    %% ===== ⚙️ 서비스 레이어 =====
+    subgraph Service["⚙️ 서비스 오케스트레이션"]
+        B[AnalysisService]
+        C[AgentCoordinator]
+        D["📊 컨텍스트 구성<br/>price · kline · news · indicators"]
+    end
 
-  C --> E
-  E -->|get_memories / add_memory| M[(SQLite: data/memory/*_memory.db)]
-  C --> R[ReflectionService.record_analysis]
-  R --> RR[(SQLite: data/memory/reflection_records.db)]
-  W[ReflectionWorker(옵션)] --> RR
-  W -->|만기 검증 + 학습 반영| M
-  A2[POST /api/analysis/reflect] -->|수동 학습| M
+    %% ===== 🤖 멀티에이전트 워크플로우 =====
+    subgraph Agents["🤖 멀티에이전트 워크플로우"]
+
+        subgraph P1["📈 Phase 1 · 다차원 분석（병렬）"]
+            E1["🔍 MarketAnalyst<br/><i>기술적 분석</i>"]
+            E2["📑 FundamentalAnalyst<br/><i>펀더멘털</i>"]
+            E3["📰 NewsAnalyst<br/><i>뉴스</i>"]
+            E4["💭 SentimentAnalyst<br/><i>심리</i>"]
+            E5["⚠️ RiskAnalyst<br/><i>리스크 평가</i>"]
+        end
+
+        subgraph P2["🎯 Phase 2 · 강세/약세 논쟁（병렬）"]
+            F1["🐂 BullResearcher<br/><i>강세 논거</i>"]
+            F2["🐻 BearResearcher<br/><i>약세 논거</i>"]
+        end
+
+        subgraph P3["💹 Phase 3 · 거래 결정"]
+            G["🎰 TraderAgent<br/><i>종합 판단 → BUY / SELL / HOLD</i>"]
+        end
+
+    end
+
+    %% ===== 🧠 메모리 레이어 =====
+    subgraph Memory["🧠 로컬 SQLite 메모리（data/memory/）"]
+        M1[("market_analyst")]
+        M2[("fundamental")]
+        M3[("news_analyst")]
+        M4[("sentiment")]
+        M5[("risk_analyst")]
+        M6[("bull_researcher")]
+        M7[("bear_researcher")]
+        M8[("trader_agent")]
+    end
+
+    %% ===== 🔄 리플렉션 루프 =====
+    subgraph Reflect["🔄 리플렉션 루프（옵션）"]
+        R[ReflectionService]
+        RR[("reflection_records.db")]
+        W["⏰ ReflectionWorker"]
+    end
+
+    %% ===== 메인 흐름 =====
+    A --> B --> C --> D
+    D --> P1 --> P2 --> P3
+
+    %% ===== 메모리 읽기/쓰기 =====
+    E1 <-.-> M1
+    E2 <-.-> M2
+    E3 <-.-> M3
+    E4 <-.-> M4
+    E5 <-.-> M5
+    F1 <-.-> M6
+    F2 <-.-> M7
+    G <-.-> M8
+
+    %% ===== 리플렉션 흐름 =====
+    C --> R --> RR
+    W --> RR
+    W -.->|"검증 + 학습"| M8
+    A2 -.->|"수동 회고"| M8
 ```
 
 #### 검색 랭킹(간략)
diff --git a/README_TW.md b/README_TW.md
index 3ea6d34..8d47e9e 100644
--- a/README_TW.md
+++ b/README_TW.md
@@ -126,50 +126,80 @@ QuantDinger 的多智能體不是「每次從零開始」。後端內建**本地
 #### 邏輯圖（從請求到記憶閉環）
 
 ```mermaid
-flowchart TD
-  A[POST /api/analysis/multi] --> B[AnalysisService]
-  B --> C[AgentCoordinator]
+flowchart TB
+    %% ===== 🌐 入口層 =====
+    subgraph Entry["🌐 API 入口"]
+        A["📡 POST /api/analysis/multi"]
+        A2["🔄 POST /api/analysis/reflect"]
+    end
 
-  C --> D[建構上下文: price/kline/news/indicators]
+    %% ===== ⚙️ 服務編排層 =====
+    subgraph Service["⚙️ 服務編排"]
+        B[AnalysisService]
+        C[AgentCoordinator]
+        D["📊 建構上下文<br/>price · kline · news · indicators"]
+    end
 
-  subgraph Agents[多智能體並行/串行工作流]
-    E1[MarketAnalyst]
-    E2[FundamentalAnalyst]
-    E3[NewsAnalyst]
-    E4[SentimentAnalyst]
-    E5[RiskAnalyst]
-    F1[BullResearcher]
-    F2[BearResearcher]
-    G[TraderAgent]
-  end
+    %% ===== 🤖 多智能體工作流 =====
+    subgraph Agents["🤖 多智能體工作流"]
 
-  C -->|Phase 1 並行| E1
-  C -->|Phase 1 並行| E2
-  C -->|Phase 1 並行| E3
-  C -->|Phase 1 並行| E4
-  C -->|Phase 1 並行| E5
-  C -->|Phase 2 並行| F1
-  C -->|Phase 2 並行| F2
-  C -->|Phase 3| G
+        subgraph P1["📈 Phase 1 · 多維分析（並行）"]
+            E1["🔍 MarketAnalyst<br/><i>技術面分析</i>"]
+            E2["📑 FundamentalAnalyst<br/><i>基本面分析</i>"]
+            E3["📰 NewsAnalyst<br/><i>新聞輿情</i>"]
+            E4["💭 SentimentAnalyst<br/><i>市場情緒</i>"]
+            E5["⚠️ RiskAnalyst<br/><i>風險評估</i>"]
+        end
 
-  subgraph MemDB[本地 SQLite 記憶庫（按角色拆分）]
-    M1[(data/memory/*_memory.db)]
-  end
+        subgraph P2["🎯 Phase 2 · 多空博弈（並行）"]
+            F1["🐂 BullResearcher<br/><i>看多論據</i>"]
+            F2["🐻 BearResearcher<br/><i>看空論據</i>"]
+        end
 
-  E1 <-->|get_memories / add_memory| M1
-  E2 <-->|get_memories / add_memory| M1
-  E3 <-->|get_memories / add_memory| M1
-  E4 <-->|get_memories / add_memory| M1
-  E5 <-->|get_memories / add_memory| M1
-  F1 <-->|get_memories / add_memory| M1
-  F2 <-->|get_memories / add_memory| M1
-  G  <-->|get_memories / add_memory| M1
+        subgraph P3["💹 Phase 3 · 交易決策"]
+            G["🎰 TraderAgent<br/><i>綜合研判 → BUY / SELL / HOLD</i>"]
+        end
 
-  C --> R[ReflectionService.record_analysis]
-  R --> RR[(data/memory/reflection_records.db)]
-  W[ReflectionWorker（可選，定時）] --> RR
-  W -->|到期驗證 + 寫回經驗| M1
-  A2[POST /api/analysis/reflect（手動復盤）] -->|reflect_and_learn| M1
+    end
+
+    %% ===== 🧠 記憶層 =====
+    subgraph Memory["🧠 本地記憶庫 SQLite（data/memory/）"]
+        M1[("market_analyst")]
+        M2[("fundamental")]
+        M3[("news_analyst")]
+        M4[("sentiment")]
+        M5[("risk_analyst")]
+        M6[("bull_researcher")]
+        M7[("bear_researcher")]
+        M8[("trader_agent")]
+    end
+
+    %% ===== 🔄 反思閉環 =====
+    subgraph Reflect["🔄 反思閉環（可選）"]
+        R[ReflectionService]
+        RR[("reflection_records.db")]
+        W["⏰ ReflectionWorker"]
+    end
+
+    %% ===== 主流程 =====
+    A --> B --> C --> D
+    D --> P1 --> P2 --> P3
+
+    %% ===== 記憶讀寫 =====
+    E1 <-.-> M1
+    E2 <-.-> M2
+    E3 <-.-> M3
+    E4 <-.-> M4
+    E5 <-.-> M5
+    F1 <-.-> M6
+    F2 <-.-> M7
+    G <-.-> M8
+
+    %% ===== 反思流程 =====
+    C --> R --> RR
+    W --> RR
+    W -.->|"驗證 + 學習"| M8
+    A2 -.->|"手動復盤"| M8
 ```
 
 #### 檢索排序（簡化）