From 35b26ea3efdb45d74acbad574f16c1f17f36beb9 Mon Sep 17 00:00:00 2001
From: Suzumiya <suzmii@qq.com>
Date: Tue, 15 Sep 2026 19:37:16 +0800
Subject: [PATCH] =?UTF-8?q?=E5=AF=B9=E8=AF=9D=E9=9D=A2=E6=9D=BF=EF=BC=9A?=
 =?UTF-8?q?=E6=96=B0=E5=A2=9E=E9=BB=98=E8=AE=A4=E6=8A=98=E5=8F=A0=E7=9A=84?=
 =?UTF-8?q?=E3=80=8C=E6=80=9D=E8=80=83=E8=BF=87=E7=A8=8B=E3=80=8D=E5=9D=97?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- App.tsx：直连回合按回合累积过程/步骤行（相邻重复去重、最多 40 行），每次回合更新时收集，新回合开始时清空
- App.tsx：把累积结果作为 processSteps 传给对话面板
- ProjectSupervisorView：新增可选属性 processSteps；在消息流里渲染默认折叠的「思考过程」块（details 默认收起，data-testid=agent-thinking）
- styles.css：思考过程块与步骤列表的最小样式（左右与消息内容对齐 16px）
- 说明（限制）：Codex app-server 0.147/0.149 只推结构性 item、不推 reasoning 文本（codex_app_server.rs 对 reasoning 类型是明确忽略并有单测钉住），所以这里展示的是我们真实能拿到的过程行（正在思考中 / 正在执行命令：x / 正在写入文件：y），不是模型内部推理原文；策划/设计链原有的 designReasoning 折叠块不受影响
---
 apps/ai-game-creator-shell/src/App.tsx        | 23 +++++++++++++++++++
 .../ProjectSupervisorView.tsx                 | 16 +++++++++++++
 apps/ai-game-creator-shell/src/styles.css     | 20 ++++++++++++++++
 3 files changed, 59 insertions(+)

diff --git a/apps/ai-game-creator-shell/src/App.tsx b/apps/ai-game-creator-shell/src/App.tsx
index 703ebcaf7..ff5e9dcdf 100644
--- a/apps/ai-game-creator-shell/src/App.tsx
+++ b/apps/ai-game-creator-shell/src/App.tsx
@@ -823,6 +823,26 @@ export function App({
   const directCodexConversationTurnSequenceRef = useRef(0);
   // 工具调用卡片：按 **id** 归并（实时增量 + 回读历史共用一份），同一 id 只渲染一次。
   // 用 ref 做写入基准，避免同一批事件里多条增量互相覆盖。
+  // 直连回合的"思考过程"行：按回合累积过程/中间态文本（去重相邻重复、最多 40 行）。
+  const [directTurnSteps, setDirectTurnSteps] = useState<string[]>([]);
+  const directTurnStepsRef = useRef<string[]>([]);
+  function pushDirectTurnStep(text: string) {
+    const line = text.trim();
+    if (!line) {
+      return;
+    }
+    const current = directTurnStepsRef.current;
+    if (current[current.length - 1] === line) {
+      return;
+    }
+    const next = [...current, line].slice(-40);
+    directTurnStepsRef.current = next;
+    setDirectTurnSteps(next);
+  }
+  function clearDirectTurnSteps() {
+    directTurnStepsRef.current = [];
+    setDirectTurnSteps([]);
+  }
   const [directToolCalls, setDirectToolCalls] = useState<
     GameCreatorDirectToolCall[]
   >([]);
@@ -892,6 +912,7 @@ export function App({
   }
 
   function resetDirectCodexTurn() {
+    clearDirectTurnSteps();
     activeDirectCodexTurnRef.current = null;
     lastDirectCodexActivityRef.current = null;
     setDirectCodexProgress('');
@@ -1850,6 +1871,7 @@ export function App({
             ? payload.updatedAt
             : Date.now();
         const processDetail = directCodexProcessDetail(payload);
+        pushDirectTurnStep(processDetail);
         if (payload.status === 'failed') {
           activeDirectCodexTurnRef.current = null;
           lastDirectCodexActivityRef.current = null;
@@ -12181,6 +12203,7 @@ export function App({
               : projectSupervisorTransientReply
         }
         designReasoning={planningV2Reasoning}
+        processSteps={directTurnSteps}
         visibleMessages={visibleMessages}
         visibleProfessionalAgentCards={visibleProfessionalAgentCards}
         showProfessionalCollaboration={
diff --git a/apps/ai-game-creator-shell/src/features/project-workspace/ProjectSupervisorView.tsx b/apps/ai-game-creator-shell/src/features/project-workspace/ProjectSupervisorView.tsx
index 3970a1c1a..2ebc0f320 100644
--- a/apps/ai-game-creator-shell/src/features/project-workspace/ProjectSupervisorView.tsx
+++ b/apps/ai-game-creator-shell/src/features/project-workspace/ProjectSupervisorView.tsx
@@ -156,6 +156,8 @@ type ProjectSupervisorViewProps = RuntimePanelProps & {
   showProfessionalCollaboration?: boolean;
   transientReply: string;
   designReasoning?: string;
+  /** 直连回合累积的过程/步骤行（真 reasoning 文本当前拿不到，见文档限制）。 */
+  processSteps?: string[];
   visibleMessages: ChatMessage[];
   visibleProfessionalAgentCards: AgentStatusCard[];
   walletEntry?: ReactNode;
@@ -220,6 +222,7 @@ export function ProjectSupervisorView({
   showProfessionalCollaboration = true,
   transientReply,
   designReasoning = '',
+  processSteps = [],
   visibleMessages,
   visibleProfessionalAgentCards,
   walletEntry,
@@ -454,6 +457,19 @@ export function ProjectSupervisorView({
               className="message-tool-call"
             />
           ) : null}
+          {directCodex && processSteps.length > 0 ? (
+            <details
+              className="design-agent-reasoning project-supervisor-thinking"
+              data-testid="agent-thinking"
+            >
+              <summary>思考过程</summary>
+              <ol className="project-supervisor-thinking-steps">
+                {processSteps.map((step, index) => (
+                  <li key={`${index}-${step}`}>{step}</li>
+                ))}
+              </ol>
+            </details>
+          ) : null}
           {designReasoning ? (
             <details className="design-agent-reasoning">
               <summary>显示思考过程</summary>
diff --git a/apps/ai-game-creator-shell/src/styles.css b/apps/ai-game-creator-shell/src/styles.css
index 021a7bd1e..f79a82f5f 100644
--- a/apps/ai-game-creator-shell/src/styles.css
+++ b/apps/ai-game-creator-shell/src/styles.css
@@ -11877,3 +11877,23 @@ button.design-workspace-tree__entry:hover,
   > .project-supervisor-process-card {
   margin: 0 16px 8px;
 }
+
+/* 直连回合的「思考过程」块：默认折叠（<details> 默认收起），展开后按步骤列过程行。
+   注意：Codex app-server 当前不推 reasoning 文本，这里展示的是我们真实拿到的过程行。 */
+.project-supervisor-thinking {
+  margin: 0 16px 8px;
+}
+
+.project-supervisor-thinking > summary {
+  cursor: pointer;
+  color: var(--platform-text-soft);
+  font-size: 12px;
+}
+
+.project-supervisor-thinking-steps {
+  margin: 6px 0 0;
+  padding-left: 18px;
+  color: var(--platform-text-soft);
+  font-size: 12px;
+  line-height: 1.6;
+}