代码修改

frontend-web/src/core/threads/hooks.ts

@@ -30,11 +30,20 @@ export type ToolEndEvent = {
   data: unknown;
 };
 
-export type ThreadStreamOptions = {
-  threadId?: string | null | undefined;
-  context: LocalSettings["context"];
-  isMock?: boolean;
-  assistantId?: string;
+export type ThreadStreamOptions = {
+  threadId?: string | null | undefined;
+  context: LocalSettings["context"];
+  isMock?: boolean;
+  assistantId?: string;
+  /**
+   * Per-page run behaviour.  Most chats keep the SDK defaults; long-running
+   * workspace flows can opt into a resumable stream without changing global
+   * chat semantics.
+   */
+  runOptions?: {
+    onDisconnect?: "cancel" | "continue";
+    multitaskStrategy?: "reject" | "rollback" | "interrupt" | "enqueue";
+  };
   /** 额外写到新建 thread 的 metadata（与 agent_id 合并）。
    *  任务深链用 `{ taskId }` 把对话挂到任务下（1:N），列表按 metadata.taskId 过滤。 */
   threadMetadata?: Record<string, unknown>;
@@ -405,9 +414,10 @@ function lastHumanFromChatModelInput(input: unknown): string {
 export function useThreadStream({
   threadId,
   context,
-  isMock,
-  assistantId,
-  threadMetadata,
+  isMock,
+  assistantId,
+  runOptions,
+  threadMetadata,
   onStart,
   onFinish,
   onToolEnd,
@@ -1037,11 +1047,13 @@ export function useThreadStream({
           {
             messages: submittedMessages,
           },
-          {
-            threadId: threadId,
-            streamSubgraphs: true,
-            streamResumable: true,
-            config: {
+          {
+            threadId: threadId,
+            streamSubgraphs: true,
+            streamResumable: true,
+            onDisconnect: runOptions?.onDisconnect,
+            multitaskStrategy: runOptions?.multitaskStrategy,
+            config: {
               recursion_limit: 1000,
             },
             context: {
@@ -1076,7 +1088,7 @@ export function useThreadStream({
         sendInFlightRef.current = false;
       }
     },
-    [thread, _handleOnStart, t.uploads.uploadingFiles, context, localSettings.display.ragModeEnabled, queryClient],
+    [thread, _handleOnStart, t.uploads.uploadingFiles, context, localSettings.display.ragModeEnabled, queryClient, runOptions],
   );
 
   // Merge thread with optimistic messages for display

frontend-web/src/position-roundtable/README.md

@@ -0,0 +1,196 @@
+# 岗位多智能体会商页面开发说明
+
+## 1. 页面定位
+
+岗位多智能体会商将一个业务链条拆成多个“岗位视角”的单智能体问答：每个席位仍使用既有的 LangGraph 对话线程完成流式生成，岗位会商服务只负责保存任务快照、冻结后的链条、节点状态与线程索引。
+
+因此它不是新的多智能体调度图，也不创建后台会商任务。页面的核心约束是：
+
+- 一条岗位会商会话对应一个任务和一份冻结的业务链配置；
+- 一个业务链席位对应一个 `PositionRoundtableNode` 和一条直接对话线程；
+- 节点写出 Markdown 交付物后才能被视为完成并解锁后序；
+- 所有岗位交付完成后，测试分析岗可以依次进入“方案总结”和“行动规划”两个收口智能体；
+- 页面中的文件预览沙箱只有一个实例，不会为多个节点重复打开或同时流式渲染。
+
+前端入口位于 `src/pages/StrategyRoutes.tsx`：`qa/position-roundtable`。
+
+## 2. URL 与会话恢复
+
+推荐访问地址：
+
+```text
+/#/page/strategy/qa/position-roundtable?taskId=<任务ID>&sessionId=<会商会话ID>&positionId=<岗位ID>
+```
+
+| 参数 | 是否必填 | 作用 |
+| --- | --- | --- |
+| `taskId` | 否 | 外部任务 ID。缺失时使用本地 mock 任务，便于演示和开发。 |
+| `sessionId` | 否 | 直接恢复指定岗位会商会话；缺失时页面会按 `taskId` 查找最近一条未归档会话。 |
+| `positionId` | 否 | 初始打开的岗位，同时同步顶部岗位选择器。无效值回退到 `intelligence`。 |
+| `new=1` | 否 | 强制不恢复历史会话，以新会话开始。 |
+
+当前岗位 ID 定义在 `lib/positions.ts`：
+
+| ID | 页面岗位 | 职责 |
+| --- | --- | --- |
+| `intelligence` | 测试分析岗 | 完成任务意图识别、业务链确认，以及收口智能体入口。 |
+| `planning` | 组织测试岗 | 承接上游信息并完成组织侧分析。 |
+| `solution` | 方案设计岗 | 产出方案侧 Markdown 交付物。 |
+| `evaluation` | 综合测试岗 | 汇总评估并给出综合建议。 |
+
+切换岗位时，`PositionRoundtablePage` 会关闭当前沙箱、清空当前节点/收口智能体选择，并写回 `positionId`。不要在子组件中直接修改 URL，否则容易留下上一岗位的预览状态。
+
+## 3. 页面结构与职责
+
+```text
+PositionRoundtablePage
+├─ 左侧：任务分析
+│  ├─ 任务卡片（名称、简介、来源、下发时间）
+│  ├─ 意图分析
+│  └─ 内嵌业务链流程图
+├─ 中间：当前岗位的对话区
+│  ├─ 测试分析岗：Step1Panel + PositionIntentInput
+│  ├─ 普通席位：PositionNodeChat + PositionAgentInput
+│  └─ 收口阶段：PositionBuiltinAgentChat + PositionAgentInput
+├─ 右侧：智能体配置 / 交付产物抽屉
+└─ 文件预览：唯一的 ArtifactFileDetail 沙箱实例
+```
+
+主要文件：
+
+| 文件 | 责任 |
+| --- | --- |
+| `pages/PositionRoundtablePage.tsx` | 页面状态机、URL 同步、任务与会话恢复、岗位/节点切换、唯一沙箱预览。 |
+| `api/sessions.ts` | 岗位会商 API 的前端类型、snake_case/camelCase 映射和请求函数。 |
+| `components/PositionNodeChat.tsx` | 一个普通业务链节点的一对一对话、线程绑定、上游上下文注入与交付完成回写。 |
+| `components/PositionBuiltinAgentChat.tsx` | 方案总结、行动规划的直接对话和“准备 → 线程就绪 → 发送 → 归档”流程。 |
+| `components/PositionIntentInput.tsx` | 仅测试分析岗使用的意图输入框。样式、工具栏和按钮均与其他页面隔离。 |
+| `components/PositionAgentInput.tsx` | 普通节点与收口智能体共用的岗位专用输入框边界。 |
+| `components/PositionAgentSwitcher.tsx` | 测试分析岗中两个收口智能体的全名切换按钮。 |
+| `lib/positions.ts` | 前端岗位目录和稳定岗位 ID。 |
+
+## 4. 状态流转
+
+```text
+意图识别
+  └─ 确认意图并启动业务链
+       ├─ 冻结 taskSnapshot / intentSnapshot / chainSnapshot
+       ├─ 按业务链席位创建节点
+       └─ intelligence 岗节点自动标记完成
+
+普通业务链节点
+  └─ 直接问答并生成 Markdown
+       ├─ 写入 LangGraph 线程
+       ├─ 回写 node.threadId
+       ├─ 保存 latestAnswer 与 artifactManifest
+       └─ Markdown 交付成功后解锁后续阶段
+
+全部普通节点已交付
+  └─ 方案总结（测试分析岗）
+       └─ 行动规划（测试分析岗）
+```
+
+### 4.1 意图识别与业务链启动
+
+测试分析岗复用 `Step1Panel` 的消息和状态逻辑；页面通过 `inputArea` 插槽替换为 `PositionIntentInput`，因此不会改动普通会商 Step 1 的输入样式。
+
+`handleActivateChain` 启动前必须同时满足：
+
+1. 选择了业务链；
+2. 业务链至少有一个席位；
+3. 每个席位已配置智能体和岗位；
+4. `intent.intentReady` 存在结构化意图结果。
+
+启动成功后会调用：
+
+1. `createPositionSession`（首次启动时）；
+2. `savePositionIntent`；
+3. `activatePositionSession`。
+
+后端会冻结链条快照并根据链条阶段创建节点。已启动后意图不可再修改；需要重新识别时必须新建会话。
+
+### 4.2 普通节点问答
+
+`PositionNodeChat` 在发送前调用 `getPositionNodeContext` 获取受长度控制的任务、意图和已交付上游内容，再以 `task_external_context` 作为隐藏上下文发送给当前智能体。
+
+首次创建线程时，`onStart` 会调用 `bindPositionNodeThread`。组件还监听 SDK 实际线程 ID，处理“线程创建后、绑定写入前刷新页面”的竞态。
+
+流结束后，组件会读取该线程的产物列表并调用 `completePositionNodeTurn`。普通节点只有交付 Markdown 文件才会变为 `done`；单纯继续追问只会保存回答，不会误解锁下游节点。
+
+### 4.3 收口智能体
+
+收口阶段只有测试分析岗可见：
+
+- 方案总结：全部岗位交付完成后可开始；
+- 行动规划：方案总结完成后可开始。
+
+`PositionBuiltinAgentChat` 不使用后台任务。每次发送先调用 `preparePositionSummary` 或 `preparePositionActionPlan`，从服务端得到冻结的输入提示词、来源版本和可复用线程 ID；待 `useThreadStream` 真正绑定到该线程后才发送首条消息。这个两步提交避免了界面先出现用户气泡、实际却没有启动流的竞态。
+
+完成回调会调用 `completePositionSummary` 或 `completePositionActionPlan`。服务端使用 `source_revisions` 判断上游是否被重做，以避免将过期总结当作当前交付物。
+
+## 5. 文件预览沙箱
+
+页面只维护一个 `sandboxArtifact`。`PositionNodeChat` 和 `PositionBuiltinAgentChat` 通过 `onPreviewArtifact` 上报预览目标；父页面统一渲染 `ArtifactFileDetail`。
+
+设计要点：
+
+- 文件正在由 `write_file` 流式生成时，优先显示源码；完成后交由预览组件切换 Markdown 预览；
+- `sandboxThreadSnapshot` 有 160ms 合并节流，只保留当前打开文件所在线程的最新快照，避免流式时频繁 setState；
+- 用户手动关闭后，`closedSandboxKeyRef` 与 `suppressedAutoPreviewThreadIdsRef` 阻止同一线程立即把沙箱重新打开；
+- 切岗位、切节点、切收口阶段、新建会话和切换历史会话都会先调用 `closeSandbox()`。
+
+不要在节点组件中直接操作全局 `ArtifactsContext` 或创建第二个 `ArtifactFileDetail`，否则会出现预览重复、关闭失效和更新深度异常。
+
+## 6. 输入框隔离约定
+
+岗位会商有两类输入框，均应只在 `src/position-roundtable/components/` 内调整：
+
+| 组件 | 使用范围 | 调整原则 |
+| --- | --- | --- |
+| `PositionIntentInput` | 测试分析岗意图识别 | 完整自有布局；发送和阶段按钮均在同一输入框内。 |
+| `PositionAgentInput` | 普通岗位节点、方案总结、行动规划 | 目前包装全局 `InputBox`，保留全部对话能力，同时提供 `position-roundtable-agent-input` 样式边界。 |
+
+禁止直接修改 `components/workspace/input-box` 或默认 `Step1Panel` 的输入样式来满足岗位会商需求。这会影响普通工作区、其他业务和原始多智能体会商。
+
+## 7. API 契约
+
+前端基地址：`/api/position-roundtable`。消息内容和真实产物仍由既有 LangGraph/API 保存，以下接口只保存岗位会商索引与快照。
+
+| 方法 | 路径 | 用途 |
+| --- | --- | --- |
+| `GET` | `/sessions?external_task_id=` | 按任务查询会话，用于恢复最近会话和历史记录。 |
+| `POST` | `/sessions` | 创建初始会话并保存任务快照。 |
+| `GET` | `/sessions/{id}` | 恢复会话、节点和冻结快照。 |
+| `PUT` | `/sessions/{id}/task` | 更新未冻结的任务快照。 |
+| `PUT` | `/sessions/{id}/intent` | 保存意图线程 ID 和结构化意图。 |
+| `POST` | `/sessions/{id}/activate` | 冻结业务链并创建节点。 |
+| `POST` | `/sessions/{id}/archive` | 归档会话，后续仅允许查看。 |
+| `POST` | `/sessions/{id}/nodes/{nodeKey}/thread` | 绑定普通节点的 LangGraph 线程。 |
+| `POST` | `/sessions/{id}/nodes/{nodeKey}/context` | 取得当前轮隐藏交接上下文。 |
+| `POST` | `/sessions/{id}/nodes/{nodeKey}/complete-turn` | 保存回答、产物清单、上游版本并推进状态。 |
+| `POST` | `/sessions/{id}/summary/*` | 准备、发送和完成方案总结。 |
+| `POST` | `/sessions/{id}/action-plan/*` | 准备、发送和完成行动规划。 |
+
+后端实现在 `offline-backend-20260512/backend/app/gateway/routers/position_roundtable.py`。
+
+## 8. 开发与排查清单
+
+1. 新增岗位时：仅补充 `POSITION_OPTIONS`，并确保业务链席位的 `position_id` 使用稳定 ID；不要以显示名称作持久化键。
+2. 新增收口智能体时：同时补充前端 `PositionBuiltinAgentKind`、后端 prepare/complete API、解锁条件和 UI 切换项。
+3. 变更节点完成条件时：前后端必须一致。当前普通节点的完成条件是 Markdown 交付物，而不是“有一条回答”。
+4. 改预览逻辑时：始终通过父页面的 `onPreviewArtifact` / `closeSandbox` 入口；不要在子组件里新增沙箱状态。
+5. 变更输入框视觉时：只改对应的岗位专用输入组件，不要向全局 `InputBox` 添加岗位样式。
+6. 验证最少覆盖：新建会话、刷新恢复、切岗位、单节点多智能体切换、生成 Markdown、关闭后再次对话、历史会话、方案总结、行动规划。
+
+## 9. 构建验证
+
+```powershell
+cd frontend-web
+pnpm build
+```
+
+提交前还应执行：
+
+```powershell
+git diff --check
+```

frontend-web/src/position-roundtable/api/sessions.ts

@@ -7,6 +7,10 @@
 import { apiFetch } from "@/core/api/fetch-client";
 import { getBackendBaseURL } from "@/core/config";
 
+/**
+ * 节点状态由岗位会商服务推进；前端只据此决定可否发送、展示进度，
+ * 不应在组件内自行把节点改成 done 或 ready。
+ */
 export type PositionNodeStatus =
   | "locked"
   | "ready"
@@ -16,6 +20,12 @@ export type PositionNodeStatus =
   | "rejected"
   | "error";
 
+/**
+ * 冻结业务链中的一个席位。
+ *
+ * `threadId` 指向真实的 LangGraph 对话；节点表只承担“席位 ↔ 线程”索引、
+ * 上游版本与交付清单的职责，绝不保存完整聊天消息。
+ */
 export interface PositionRoundtableNode {
   id: string;
   sessionId: string;
@@ -37,6 +47,12 @@ export interface PositionRoundtableNode {
   updatedAt?: string;
 }
 
+/**
+ * 岗位会商会话的可恢复快照。
+ *
+ * 任务、意图与链条在启动后被分别冻结，避免外部任务或链条配置发生变化时
+ * 改写已经开始的协同过程。`nodes` 只在详情接口中携带。
+ */
 export interface PositionRoundtableSession {
   id: string;
   externalTaskId: string | null;
@@ -104,6 +120,10 @@ async function readJson<T>(response: Response, label: string): Promise<T> {
   return response.json() as Promise<T>;
 }
 
+/**
+ * API 使用 snake_case，页面使用 camelCase。所有字段映射集中在这里，
+ * 组件禁止直接依赖后端字段名，避免 API 演进时遗漏状态字段。
+ */
 function toNode(node: ServerNode): PositionRoundtableNode {
   return {
     id: node.id,
@@ -127,6 +147,7 @@ function toNode(node: ServerNode): PositionRoundtableNode {
   };
 }
 
+/** 同上：会话快照在这一层完成默认值填充与字段转换。 */
 function toSession(session: ServerSession): PositionRoundtableSession {
   const result: PositionRoundtableSession = {
     id: session.id,
@@ -222,6 +243,7 @@ export async function activatePositionSession(
   sessionId: string,
   chainId: string,
 ): Promise<PositionRoundtableSession> {
+  // 启动即冻结链条快照并创建节点；后续页面不得用当前编辑中的链条替换它。
   const result = await readJson<ServerSession>(
     await apiFetch(`${BASE()}/sessions/${encodeURIComponent(sessionId)}/activate`, {
       method: "POST",
@@ -262,7 +284,12 @@ export async function bindPositionNodeThread(
   return toNode(result);
 }
 
-/** Fetch the bounded hidden task/intent/upstream handoff before one node turn. */
+/**
+ * 在普通节点发送前取得受长度限制的隐藏交接上下文。
+ *
+ * 该上下文会作为 `task_external_context` 发送给 LangGraph，而不是渲染在聊天
+ * 消息列表中；这样既能让节点承接上游，又不会把整段交接材料重复展示给用户。
+ */
 export async function getPositionNodeContext(
   sessionId: string,
   nodeKey: string,
@@ -285,6 +312,7 @@ export async function completePositionNodeTurn(
     upstreamRevision?: Record<string, unknown> | null;
   },
 ): Promise<PositionRoundtableNode> {
+  // 后端依据 artifactManifest 判断是否有 Markdown 交付物，并在需要时解锁下游。
   const result = await readJson<ServerNode>(
     await apiFetch(
       `${BASE()}/sessions/${encodeURIComponent(sessionId)}/nodes/${encodeURIComponent(nodeKey)}/complete-turn`,
@@ -358,6 +386,7 @@ async function prepareBuiltinRun(
   kind: BuiltinKind,
   message = "",
 ): Promise<PositionBuiltinRunPreparation> {
+  // prepare 阶段固定本轮的上游版本与隐藏提示词，避免总结/行动规划引用漂移数据。
   const path = kind === "summary" ? "summary/prepare" : "action-plan/prepare";
   const result = await readJson<ServerBuiltinRunPreparation>(
     await apiFetch(`${BASE()}/sessions/${encodeURIComponent(sessionId)}/${path}`, {
@@ -381,6 +410,7 @@ async function completeBuiltinRun(
     model?: string;
   },
 ): Promise<PositionRoundtableSession> {
+  // complete 阶段会校验 sourceRevisions；上游已变化时不能把旧结果覆盖为最新交付。
   const path = kind === "summary" ? "summary/complete" : "action-plan/complete";
   const result = await readJson<ServerSession>(
     await apiFetch(`${BASE()}/sessions/${encodeURIComponent(sessionId)}/${path}`, {

frontend-web/src/position-roundtable/components/PositionAgentInput.tsx

@@ -0,0 +1,25 @@
+import type { ComponentProps } from "react";
+
+import { InputBox } from "@/components/workspace/input-box";
+import { cn } from "@/lib/utils";
+
+/**
+ * 岗位会商的岗位智能体输入框。
+ *
+ * 目前保留 InputBox 的完整交互能力；后续本页面需要独立调整视觉或工具栏时，
+ * 只改这个组件即可，不会影响普通对话、工作区或其他业务页面。
+ */
+export function PositionAgentInput({
+  className,
+  ...props
+}: ComponentProps<typeof InputBox>) {
+  // Keep the page-specific class on the wrapper rather than adding a variant
+  // to InputBox. Node chat and built-in chat can evolve together while every
+  // other workspace keeps the global composer untouched.
+  return (
+    <InputBox
+      {...props}
+      className={cn("position-roundtable-agent-input w-full", className)}
+    />
+  );
+}

frontend-web/src/position-roundtable/components/PositionBuiltinAgentChat.tsx

@@ -4,7 +4,6 @@ import { Bot, CheckCircle2, FileText, LockKeyhole, Sparkles, Target } from "luci
 import { toast } from "sonner";
 
 import type { PromptInputMessage } from "@/components/ai-elements/prompt-input";
-import { InputBox } from "@/components/workspace/input-box";
 import { MessageList } from "@/components/workspace/messages";
 import { ThreadContext } from "@/components/workspace/messages/context";
 import { useAgent } from "@/core/agents";
@@ -19,6 +18,7 @@ import {
   PositionBuiltinAgentTabs,
   type PositionBuiltinAgentSwitchOption,
 } from "@/position-roundtable/components/PositionAgentSwitcher";
+import { PositionAgentInput } from "@/position-roundtable/components/PositionAgentInput";
 import {
   completePositionActionPlan,
   completePositionSummary,
@@ -76,6 +76,12 @@ interface PositionBuiltinAgentChatProps {
 
 const SUMMARY_AGENT_ID = "roundtable-summary";
 const ACTION_PLAN_AGENT_ID = "position-action-planner";
+// 方案总结 / 行动规划仍使用普通的 thread 流；首次自动提交期间若组件因
+// thread 绑定而短暂重渲染，保留服务端运行并允许恢复 SSE，避免首轮被取消。
+const POSITION_BUILTIN_RUN_OPTIONS = {
+  onDisconnect: "continue",
+  multitaskStrategy: "rollback",
+} as const;
 
 function builtinAgentId(kind: PositionBuiltinAgentKind) {
   return kind === "summary" ? SUMMARY_AGENT_ID : ACTION_PLAN_AGENT_ID;
@@ -166,40 +172,6 @@ function artifactDisplayName(pathOrUrl: string): string {
   return path.split("/").filter(Boolean).at(-1) || "Markdown 报告";
 }
 
-function buildWriteFileVirtualUrl(path: string, messageId: string, toolCallId: string): string {
-  return new URL(
-    `write-file:${path}?message_id=${encodeURIComponent(
-      messageId,
-    )}&tool_call_id=${encodeURIComponent(toolCallId)}`,
-  ).toString();
-}
-
-function latestStreamingWriteFile(
-  messages: AgentThreadState["messages"],
-): { path: string; name: string; mimeType: string | null } | null {
-  for (let messageIndex = messages.length - 1; messageIndex >= 0; messageIndex -= 1) {
-    const message = messages[messageIndex] as (typeof messages)[number] & {
-      id?: string;
-      tool_calls?: Array<{ id?: string; name?: string; args?: Record<string, unknown> }>;
-    };
-    if (message.type !== "ai" || !message.id || !Array.isArray(message.tool_calls)) continue;
-    for (let toolIndex = message.tool_calls.length - 1; toolIndex >= 0; toolIndex -= 1) {
-      const toolCall = message.tool_calls[toolIndex];
-      if (toolCall?.name !== "write_file" || !toolCall.id) continue;
-      const rawPath = toolCall.args?.path;
-      if (typeof rawPath !== "string" || !rawPath.trim()) continue;
-      const content = toolCall.args?.content;
-      if (typeof content !== "string" || content.length === 0) continue;
-      return {
-        path: buildWriteFileVirtualUrl(rawPath, message.id, toolCall.id),
-        name: artifactDisplayName(rawPath),
-        mimeType: rawPath.toLowerCase().endsWith(".md") ? "text/markdown" : null,
-      };
-    }
-  }
-  return null;
-}
-
 function latestAssistantText(messages: AgentThreadState["messages"]): string {
   for (let index = messages.length - 1; index >= 0; index -= 1) {
     const message = messages[index];
@@ -261,7 +233,6 @@ export function PositionBuiltinAgentChat({
   const [pendingSubmission, setPendingSubmission] =
     useState<PendingBuiltinSubmission | null>(null);
   const pendingSubmissionRef = useRef(false);
-  const autoPreviewedArtifactKeyRef = useRef<string | null>(null);
   const handledAutoStartRequestRef = useRef<number | undefined>(undefined);
 
   useEffect(() => {
@@ -272,7 +243,6 @@ export function PositionBuiltinAgentChat({
     pendingRunRef.current = null;
     pendingSubmissionRef.current = false;
     setPendingSubmission(null);
-    autoPreviewedArtifactKeyRef.current = null;
     handledAutoStartRequestRef.current = undefined;
   }, [fallbackAgentId, kind, persistedThreadId, session?.id]);
 
@@ -294,6 +264,7 @@ export function PositionBuiltinAgentChat({
   const [thread, sendMessage, , actualThreadId] = useThreadStream({
     threadId: boundThreadId,
     assistantId,
+    runOptions: POSITION_BUILTIN_RUN_OPTIONS,
     context: {
       ...settings.context,
       agent_id: assistantId,
@@ -314,8 +285,8 @@ export function PositionBuiltinAgentChat({
     onFinish: (state) => {
       const pending = pendingRunRef.current;
       if (!pending || !session) return;
-      const latestAnswer = latestAssistantText(state.messages ?? []);
-      if (!latestAnswer) return;
+      const content = latestAssistantText(state.messages ?? []);
+      if (!content) return;
       pendingRunRef.current = null;
       void (async () => {
         let artifactManifest: Array<Record<string, unknown>> = [];
@@ -323,23 +294,22 @@ export function PositionBuiltinAgentChat({
           const artifacts = await listSessionArtifacts(pending.threadId);
           artifactManifest = manifestFromArtifacts(artifacts.files);
         } catch {
-          // Keep the text turn usable even if the artifact endpoint is briefly
-          // unavailable; the completion API will surface a clearer validation
-          // message if an output file is genuinely missing.
+          // Completion still receives the visible answer; it will give a
+          // clearer error if a required deliverable is genuinely missing.
         }
         try {
           const next =
             kind === "summary"
               ? await completePositionSummary(session.id, {
                   threadId: pending.threadId,
-                  content: latestAnswer,
+                  content,
                   artifactManifest,
                   sourceRevisions: pending.sourceRevisions,
                   model: pending.model,
                 })
               : await completePositionActionPlan(session.id, {
                   threadId: pending.threadId,
-                  content: latestAnswer,
+                  content,
                   artifactManifest,
                   sourceRevisions: pending.sourceRevisions,
                   model: pending.model,
@@ -374,25 +344,6 @@ export function PositionBuiltinAgentChat({
     onThreadSnapshot?.({ threadId: activeThreadId, thread });
   }, [activeThreadId, onThreadSnapshot, thread]);
 
-  const streamingArtifact = useMemo(
-    () => (thread.isLoading ? latestStreamingWriteFile(thread.messages) : null),
-    [thread.isLoading, thread.messages],
-  );
-
-  useEffect(() => {
-    if (!thread.isLoading || !streamingArtifact) return;
-    const artifactKey = `${activeThreadId}:${streamingArtifact.path}`;
-    if (autoPreviewedArtifactKeyRef.current === artifactKey) return;
-    autoPreviewedArtifactKeyRef.current = artifactKey;
-    onPreviewArtifact?.({
-      threadId: activeThreadId,
-      path: streamingArtifact.path,
-      name: streamingArtifact.name,
-      mimeType: streamingArtifact.mimeType,
-      auto: true,
-    });
-  }, [activeThreadId, onPreviewArtifact, streamingArtifact, thread.isLoading]);
-
   const ensureBuiltinThread = useCallback(
     async (preparedThreadId: string | null, preparedAgentId: string) => {
       if (preparedThreadId) {
@@ -437,10 +388,9 @@ export function PositionBuiltinAgentChat({
           sourceRevisions: preparation.sourceRevisions,
           model: selectedModel || undefined,
         };
-        // `useThreadStream` is still attached to the previous (or empty)
-        // thread during this render. Queue the first turn and dispatch it
-        // only after the hook has rebound to the prepared thread; otherwise
-        // the UI can show the optimistic user bubble without opening a run.
+        // The standard chat stream must first bind to the prepared thread.
+        // The effect below then submits the same visible message a manual
+        // click would send, so this flow has no separate presentation layer.
         setPendingSubmission({
           threadId: builtinThread.threadId,
           agentId: preparation.agentId,
@@ -449,9 +399,10 @@ export function PositionBuiltinAgentChat({
           sourceRevisions: preparation.sourceRevisions,
         });
       } catch (error) {
+        const message = error instanceof Error ? error.message : String(error);
         pendingRunRef.current = null;
         pendingSubmissionRef.current = false;
-        toast.error(`发送失败：${error instanceof Error ? error.message : String(error)}`);
+        toast.error(`发送失败：${message}`);
       }
     },
     [
@@ -538,6 +489,9 @@ export function PositionBuiltinAgentChat({
     ) {
       return;
     }
+    // The request key changes only after an explicit page action. Keeping the
+    // handled key locally prevents route/session re-renders from auto-sending
+    // the first question again.
     handledAutoStartRequestRef.current = autoStartRequestKey;
     void handleSubmit({ text: defaultStartText(kind), files: [] });
   }, [autoStartRequestKey, canWrite, handleSubmit, kind, snapshotDone, thread.isLoading]);
@@ -655,7 +609,7 @@ export function PositionBuiltinAgentChat({
         )}
 
         <div className="shrink-0 border-t border-slate-100 px-4 py-3 dark:border-[#2b8ffe]/20">
-          <InputBox
+          <PositionAgentInput
             className="w-full"
             isNewThread={!boundThreadId}
             threadId={displayThreadId}

frontend-web/src/position-roundtable/components/PositionIntentInput.tsx

@@ -0,0 +1,240 @@
+import { useRef } from "react";
+import {
+  FileText,
+  Loader2,
+  Paperclip,
+  Pause,
+  Repeat,
+  Save,
+  Send,
+  Smile,
+  Sparkles,
+  X,
+} from "lucide-react";
+
+import { AutoGrowTextarea } from "@/components/ui/auto-grow-textarea";
+import type { Step1InputAreaProps } from "@/roundtable-planning/components/Step1Panel";
+
+/** Keep attachment metadata presentation local to this position-only composer. */
+function formatFileSize(bytes: number): string {
+  if (bytes < 1024) return `${bytes} B`;
+  if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)} KB`;
+  return `${(bytes / (1024 * 1024)).toFixed(1)} MB`;
+}
+
+/**
+ * 岗位会商的意图理解输入框。
+ *
+ * 它只服务岗位会商页，通过 Step1Panel 的输入插槽接入；视觉调整不会回流到
+ * 原有多智能体会商 Step 1 页面。
+ */
+export function PositionIntentInput({
+  value,
+  onValueChange,
+  attachedFiles,
+  onAddFiles,
+  onRemoveFile,
+  charactersLeft,
+  isSending,
+  canSend,
+  pendingClarification,
+  onSend,
+  onStop,
+  onFollowUp,
+  onSkip,
+  onEnterRoundtable,
+  enterButtonLabel,
+  intentReady,
+  footerAction,
+  hideEnterRoundtable,
+  onManualSave,
+  onTriggerToast,
+}: Step1InputAreaProps) {
+  // Files are owned by Step1Panel. This ref only opens the native picker and
+  // forwards its selection through onAddFiles, so replacement layouts do not
+  // accidentally create a second attachment state.
+  const fileInputRef = useRef<HTMLInputElement>(null);
+
+  return (
+    <div className="position-roundtable-intent-input shrink-0 border-t border-slate-200 bg-[#f7faff] px-4 py-3 dark:border-[#2b8ffe]/35 dark:bg-[#071a38]">
+      <div className="relative overflow-hidden rounded-[4px] border border-slate-300 bg-white shadow-[0_2px_10px_rgba(15,23,42,0.08)] transition-all focus-within:border-primary focus-within:ring-2 focus-within:ring-[var(--brand-200)]/45 dark:border-[#2b8ffe]/55 dark:bg-[#0d2855] dark:shadow-none dark:focus-within:border-[#56baff] dark:focus-within:ring-[#2b8ffe]/25">
+        {attachedFiles.length > 0 && (
+          <div className="mx-3 mb-2 flex flex-wrap gap-1.5 border-b border-slate-100 pb-2 pt-2.5 dark:border-[#2b8ffe]/25">
+            {attachedFiles.map((file, index) => (
+              <span
+                key={`${file.name}-${file.size}-${index}`}
+                className="inline-flex max-w-[200px] items-center gap-1.5 rounded-[4px] border border-slate-200 bg-slate-50 py-1 pl-2 pr-1 text-[11px] text-slate-600 dark:border-[#2b8ffe]/30 dark:bg-[#102f63] dark:text-[#c4dcff]"
+              >
+                <FileText className="size-3 shrink-0 text-primary" />
+                <span className="truncate font-medium" title={file.name}>
+                  {file.name}
+                </span>
+                <span className="shrink-0 text-slate-400">{formatFileSize(file.size)}</span>
+                <button
+                  type="button"
+                  onClick={() => onRemoveFile(index)}
+                  className="shrink-0 rounded-[3px] p-0.5 text-slate-400 transition hover:bg-slate-200 hover:text-slate-600 dark:text-[#8aaee0] dark:hover:bg-[#21447c] dark:hover:text-white"
+                  title="移除"
+                >
+                  <X className="size-3" />
+                </button>
+              </span>
+            ))}
+          </div>
+        )}
+
+        <AutoGrowTextarea
+          id="input-text-area"
+          value={value}
+          onChange={(event) => onValueChange(event.target.value)}
+          onKeyDown={(event) => {
+            if (event.key === "Enter" && !event.shiftKey) {
+              event.preventDefault();
+              onSend();
+            }
+          }}
+          placeholder={
+            pendingClarification
+              ? "也可在此输入自定义回答（Enter 发送 / Shift+Enter 换行）"
+              : "请输入补充信息 / 回答问题 / 继续澄清（Enter 发送 / Shift+Enter 换行）"
+          }
+          className="min-h-[74px] w-full max-h-80 bg-transparent px-3 py-2.5 text-sm font-normal leading-relaxed text-slate-700 placeholder:text-slate-400 focus:outline-none dark:text-[#e7f0ff] dark:placeholder:text-[#7e9bc5]"
+          maxHeight={320}
+          maxLength={500}
+        />
+
+        <div className="hidden">
+          <div className="flex items-center gap-1.5">
+            <button
+              type="button"
+              onClick={() => onTriggerToast("表情包功能仍在微调状态（仅供演示）")}
+              aria-label="打开表情"
+              className="rounded-[3px] p-1.5 transition hover:bg-white hover:text-primary dark:hover:bg-[#16386e] dark:hover:text-[#73c8ff]"
+            >
+              <Smile className="size-4" />
+            </button>
+            <input
+              ref={fileInputRef}
+              type="file"
+              multiple
+              className="hidden"
+              onChange={(event) => {
+                onAddFiles(Array.from(event.target.files ?? []));
+                event.target.value = "";
+              }}
+            />
+            <button
+              type="button"
+              onClick={() => fileInputRef.current?.click()}
+              title="上传参考文件（智能体将读取文件内容辅助理解任务）"
+              className="rounded-[3px] p-1.5 transition hover:bg-white hover:text-primary dark:hover:bg-[#16386e] dark:hover:text-[#73c8ff]"
+            >
+              <Paperclip className="size-4" />
+            </button>
+          </div>
+          <span className="font-mono text-[10px] text-slate-400">{charactersLeft}/500</span>
+        </div>
+        {/*
+          All stage actions intentionally stay inside the same composer frame.
+          Parent pages can still hide optional actions by their stable ids
+          without changing this component's layout or global input styles.
+        */}
+        <div className="flex flex-wrap items-center gap-3 border-t border-slate-100 bg-white px-3 py-2.5 dark:border-[#2b8ffe]/25 dark:bg-[#0b234c]">
+          <div className="flex shrink-0 items-center">
+            <input
+              ref={fileInputRef}
+              type="file"
+              multiple
+              className="hidden"
+              onChange={(event) => {
+                onAddFiles(Array.from(event.target.files ?? []));
+                event.target.value = "";
+              }}
+            />
+            <button
+              type="button"
+              onClick={() => fileInputRef.current?.click()}
+              title="上传参考文件"
+              aria-label="上传参考文件"
+              className="flex size-10 items-center justify-center rounded-[4px] border border-slate-200 bg-white text-slate-500 transition hover:border-primary/45 hover:text-primary dark:border-[#2b8ffe]/35 dark:bg-[#102f63] dark:text-[#b9d3f7] dark:hover:bg-[#1a417b] dark:hover:text-[#73c8ff]"
+            >
+              <Paperclip className="size-4" />
+            </button>
+          </div>
+          <div className="ml-auto flex min-w-0 flex-wrap items-center justify-end gap-1.5">
+          <button
+            id="send-message-btn"
+            type="button"
+            onClick={onSend}
+            disabled={!canSend}
+            className="flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] bg-primary px-3 text-xs font-bold text-white shadow-[0_2px_8px_rgba(245,34,45,0.15)] transition-all active:scale-[0.98] hover:bg-[var(--brand-900)] disabled:cursor-not-allowed disabled:bg-slate-300 dark:bg-[#1478e8] dark:hover:bg-[#2995ff] dark:disabled:bg-[#28476f]"
+          >
+            {isSending ? <Loader2 className="size-3 animate-spin" /> : <Send className="size-3" />}
+            {isSending ? "发送中" : "发送"}
+          </button>
+
+          {isSending && (
+            <button
+              id="pause-intent-btn"
+              type="button"
+              onClick={onStop}
+              className="flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] border border-amber-200/40 bg-amber-100 px-3 text-xs font-bold text-amber-800 shadow-sm transition active:scale-[0.98] hover:bg-amber-200 dark:border-amber-300/30 dark:bg-amber-400/15 dark:text-amber-200 dark:hover:bg-amber-400/25"
+            >
+              <Pause className="size-3" />
+              暂停
+            </button>
+          )}
+
+          <button
+            id="suggest-followup-btn"
+            type="button"
+            onClick={onFollowUp}
+            className="flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] border border-slate-200 bg-white px-2.5 text-xs font-bold text-slate-700 transition hover:bg-slate-50 dark:border-[#2b8ffe]/35 dark:bg-[#102f63] dark:text-[#b9d3f7] dark:hover:bg-[#1a417b]"
+          >
+            <Repeat className="size-3 text-slate-500" />
+            继续追问
+          </button>
+
+          <button
+            id="skip-question-btn"
+            type="button"
+            onClick={onSkip}
+            className="flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] border border-slate-200 bg-white px-2.5 text-xs font-bold text-slate-700 transition hover:bg-slate-50 dark:border-[#2b8ffe]/35 dark:bg-[#102f63] dark:text-[#b9d3f7] dark:hover:bg-[#1a417b]"
+          >
+            <X className="size-3 text-slate-500" strokeWidth={2.5} />
+            跳过此项
+          </button>
+
+          {footerAction ??
+            (!hideEnterRoundtable && (
+              <button
+                id="enter-roundtable-btn"
+                type="button"
+                onClick={onEnterRoundtable}
+                disabled={!intentReady}
+                title={intentReady ? undefined : "请先完成任务意图澄清"}
+                className={`flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] px-3 text-xs font-black transition active:scale-[0.98] shadow-md shadow-emerald-500/10 ${
+                  intentReady
+                    ? "cursor-pointer bg-emerald-600 text-white hover:bg-emerald-700"
+                    : "cursor-not-allowed bg-emerald-600/40 text-white"
+                }`}
+              >
+                <Sparkles className="size-3.5 text-emerald-100" />
+                {enterButtonLabel}
+              </button>
+            ))}
+          <button
+          id="save-draft-btn"
+          type="button"
+          onClick={onManualSave}
+          className="flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] border border-slate-200 bg-white px-2.5 text-xs font-bold text-slate-700 transition hover:bg-slate-50 dark:border-[#2b8ffe]/35 dark:bg-[#102f63] dark:text-[#b9d3f7] dark:hover:bg-[#1a417b]"
+        >
+          <Save className="size-3 text-slate-500" />
+          保存草稿
+          </button>
+          </div>
+      </div>
+      </div>
+    </div>
+  );
+}

frontend-web/src/position-roundtable/components/PositionNodeChat.tsx

@@ -1,10 +1,9 @@
 import type { BaseStream } from "@langchain/langgraph-sdk/react";
 import { useCallback, useEffect, useMemo, useRef, useState } from "react";
-import { Bot, FileText, LockKeyhole } from "lucide-react";
+import { Bot, FileText, LockKeyhole, Settings } from "lucide-react";
 import { toast } from "sonner";
-import { Select as TSelect } from "tdesign-react";
+import { Button as TButton, Select as TSelect } from "tdesign-react";
 import type { PromptInputMessage } from "@/components/ai-elements/prompt-input";
-import { InputBox } from "@/components/workspace/input-box";
 import { ThreadContext } from "@/components/workspace/messages/context";
 import { MessageList } from "@/components/workspace/messages";
 import { useAgent } from "@/core/agents";
@@ -26,6 +25,7 @@ import {
   type PositionBuiltinAgentKind,
   type PositionBuiltinAgentSwitchOption,
 } from "@/position-roundtable/components/PositionAgentSwitcher";
+import { PositionAgentInput } from "@/position-roundtable/components/PositionAgentInput";
 
 interface PositionNodeSwitchOption {
   node: PositionRoundtableNode;
@@ -45,6 +45,8 @@ interface PositionNodeChatProps {
   onSelectPositionNode?: (node: PositionRoundtableNode) => void;
   builtinAgentOptions?: PositionBuiltinAgentSwitchOption[];
   onSelectBuiltinAgent?: (kind: PositionBuiltinAgentKind) => void;
+  /** Rendered in the header so the wide same-position selector never overlaps it. */
+  onOpenConfiguration?: () => void;
   onNodeUpdate: (node: PositionRoundtableNode) => void;
   /** The workspace collapses the configuration panel on the first real turn. */
   onConversationStart?: () => void;
@@ -183,7 +185,7 @@ function latestStreamingWriteFile(
  * Direct single-agent chat for one frozen business-chain node.
  *
  * It deliberately uses the regular `useThreadStream` + `MessageList` +
- * `InputBox` stack. The position Session merely records which existing thread
+ * `PositionAgentInput` stack. The position Session merely records which existing thread
  * belongs to the node; it is not a second message persistence mechanism.
  */
 export function PositionNodeChat({
@@ -198,6 +200,7 @@ export function PositionNodeChat({
   onSelectPositionNode,
   builtinAgentOptions,
   onSelectBuiltinAgent,
+  onOpenConfiguration,
   onNodeUpdate,
   onConversationStart,
   onPreviewArtifact,
@@ -233,6 +236,8 @@ export function PositionNodeChat({
 
   const bindCreatedThread = useCallback(
     (threadId: string) => {
+      // Persist as soon as the SDK exposes an id. A refresh after this point
+      // can restore the exact seat thread instead of starting a second chat.
       boundThreadIdRef.current = threadId;
       setBoundThreadId(threadId);
       void bindPositionNodeThread(sessionId, node.nodeKey, threadId)
@@ -265,6 +270,9 @@ export function PositionNodeChat({
     },
     onStart: bindCreatedThread,
     onFinish: (state) => {
+      // A turn is not considered delivered by the UI alone. The persisted
+      // completion call records the answer, file manifest and upstream version
+      // so the service can decide whether this seat unlocks its downstream.
       const threadId = boundThreadIdRef.current;
       const latestAi = [...(state.messages ?? [])]
         .reverse()
@@ -364,6 +372,8 @@ export function PositionNodeChat({
 
   useEffect(() => {
     if (!thread.isLoading || !streamingArtifact) return;
+    // One write_file call should open at most one preview. The parent further
+    // guards manual close state, because it owns the page-wide singleton sandbox.
     const artifactKey = `${activeThreadId}:${streamingArtifact.path}`;
     if (autoPreviewedArtifactKeyRef.current === artifactKey) return;
     autoPreviewedArtifactKeyRef.current = artifactKey;
@@ -400,6 +410,9 @@ export function PositionNodeChat({
       }
       try {
         onConversationStart?.();
+        // Fetch handoff immediately before sending rather than when the page
+        // mounts. This guarantees the prompt sees the newest completed upstream
+        // delivery and captures the matching revision for completion checks.
         const taskExternalContext = await getPositionNodeContext(sessionId, node.nodeKey);
         const upstreamRevision = taskExternalContext.upstream_revision;
         upstreamRevisionRef.current =
@@ -491,6 +504,17 @@ export function PositionNodeChat({
               onSelect={onSelectBuiltinAgent}
             />
           )}
+          {onOpenConfiguration && (
+            <TButton
+              size="small"
+              variant="outline"
+              shape="circle"
+              className="!h-8 !w-8 !min-w-8 !shrink-0 !p-0 !bg-white/90 dark:!border-[#2b8ffe]/55 dark:!bg-[#082d5f]/90"
+              icon={<Settings className="!size-4 !shrink-0" />}
+              onClick={onOpenConfiguration}
+              title="打开右侧配置"
+            />
+          )}
         </header>
 
         {nodeArtifacts.length > 0 && (
@@ -555,7 +579,7 @@ export function PositionNodeChat({
         )}
 
         <div className="shrink-0 border-t border-slate-100 px-4 py-3 dark:border-[#2b8ffe]/20">
-          <InputBox
+          <PositionAgentInput
             className="w-full"
             isNewThread={!boundThreadId}
             threadId={displayThreadId}

frontend-web/src/position-roundtable/pages/PositionRoundtablePage.tsx

@@ -73,6 +73,7 @@ import { cn } from "@/lib/utils";
 import expandLogo from "@/assets/position-roundtable/expand-logo.svg";
 import { listChains } from "@/roundtable-planning/api/chains";
 import { Step1Panel } from "@/roundtable-planning/components/Step1Panel";
+import { PositionIntentInput } from "@/position-roundtable/components/PositionIntentInput";
 import { PositionNodeChat } from "@/position-roundtable/components/PositionNodeChat";
 import { getAvatar } from "@/roundtable-planning/components/Avatars";
 import { useStep1Intent, type InteractiveMessage } from "@/roundtable-planning/hooks/useStep1Intent";
@@ -144,6 +145,16 @@ type SandboxThreadSnapshot = {
   signature: string;
 };
 
+/**
+ * URL parameters are externally supplied, so never cast the raw value into
+ * page state without checking it against the position directory first.
+ */
+function resolvePositionId(value: string | null): PositionId {
+  return POSITIONS.some((position) => position.id === value)
+    ? (value as PositionId)
+    : DEFAULT_POSITION_ID;
+}
+
 function normalizePreviewArtifactPath(path: string): string {
   let normalized = path.trim();
   if (normalized.startsWith("write-file:")) {
@@ -479,7 +490,8 @@ export default function PositionRoundtablePage({
   const [searchParams, setSearchParams] = useSearchParams();
   const { theme } = useTheme();
   const isDark = theme === "dark";
-  const [positionId, setPositionId] = useState<PositionId>(DEFAULT_POSITION_ID);
+  const positionIdFromUrl = resolvePositionId(searchParams.get("positionId"));
+  const [positionId, setPositionId] = useState<PositionId>(() => positionIdFromUrl);
   const taskIdFromUrl = searchParams.get("taskId") ?? DEFAULT_MOCK_POSITION_TASK_ID;
   const searchParamsString = searchParams.toString();
   const fallbackTask = useMemo(() => getMockPositionTask(taskIdFromUrl), [taskIdFromUrl]);
@@ -495,6 +507,9 @@ export default function PositionRoundtablePage({
   const [historyMenuOpen, setHistoryMenuOpen] = useState(false);
   const [historySessions, setHistorySessions] = useState<PositionRoundtableSession[]>([]);
   const [historyLoading, setHistoryLoading] = useState(false);
+  // The page owns exactly one artifact preview. Node and built-in chat panels
+  // only report preview intents upward; this prevents two sandboxes from
+  // rendering the same streaming file and avoids close/reopen feedback loops.
   const artifactsCtx = useArtifactsOptional();
   const artifactsCtxRef = useRef(artifactsCtx);
   const [sandboxArtifact, setSandboxArtifact] = useState<PositionArtifactPreview | null>(null);
@@ -1112,6 +1127,9 @@ export default function PositionRoundtablePage({
   }, [intent]);
 
   const handleActivateChain = useCallback(async () => {
+    // Activation is the transaction boundary between editable intent discovery
+    // and a frozen execution graph. Validate all local configuration first so
+    // users receive a clear configuration error instead of a failed node chat.
     if (startingChain) return;
     if (!selectedChain) {
       toast.warning("请先选择业务链条");
@@ -1239,6 +1257,8 @@ export default function PositionRoundtablePage({
   );
 
   const handlePreviewArtifact = useCallback((artifact: PositionArtifactPreview) => {
+    // Manual opens clear the suppression marker; auto opens respect it so a
+    // user-closed sandbox is not reopened by the next streaming state update.
     if (!artifact.threadId || !artifact.path) {
       toast.warning("该文件缺少沙箱预览所需的信息");
       return;
@@ -1382,6 +1402,8 @@ export default function PositionRoundtablePage({
   }, []);
   const handleSelectBuiltinStage = useCallback(
     (kind: ClosureStageKind) => {
+      // The closing agents are part of the testing-analysis workspace only;
+      // ordinary positions keep a single business-chain node conversation.
       if (positionId !== DEFAULT_POSITION_ID) {
         toast.info("收口智能体仅在测试分析岗中开放使用。");
         return;
@@ -1412,6 +1434,18 @@ export default function PositionRoundtablePage({
     },
     [closeSandbox],
   );
+  useEffect(() => {
+    // Support both an initial deep link and browser URL changes. Reset every
+    // view-specific selection so no node, drawer or sandbox leaks across jobs.
+    if (positionIdFromUrl === positionId) return;
+    closeSandbox();
+    setPositionId(positionIdFromUrl);
+    setActiveBuiltinKind(null);
+    setSelectedAgentId(null);
+    setActiveNodeKey(null);
+    setRightPanelOpen(positionIdFromUrl === DEFAULT_POSITION_ID);
+    setLeftPanelCollapsed(false);
+  }, [closeSandbox, positionId, positionIdFromUrl]);
   const handlePositionChange = useCallback(
     (value: unknown) => {
       // 右侧详情只属于刚才的岗位；切岗后回到该岗位的智能体列表，
@@ -1425,8 +1459,11 @@ export default function PositionRoundtablePage({
       setActiveNodeKey(null);
       setRightPanelOpen(nextPositionId === DEFAULT_POSITION_ID);
       setLeftPanelCollapsed(false);
+      const next = new URLSearchParams(searchParamsString);
+      next.set("positionId", nextPositionId);
+      setSearchParams(next, { replace: true });
     },
-    [closeSandbox],
+    [closeSandbox, searchParamsString, setSearchParams],
   );
   const renderIntentChatPanel = (chainStarted: boolean) => (
     <div className="contents">
@@ -1438,6 +1475,8 @@ export default function PositionRoundtablePage({
           </TButton>
         </div>
       )}
+      {/* Step1Panel retains intent state; only the composer is replaced so this
+          workspace can evolve its visual language without changing other Step 1 pages. */}
       <Step1Panel
         headerLeftSlot={
           <div className="flex min-w-0 items-center gap-2">
@@ -1484,7 +1523,7 @@ export default function PositionRoundtablePage({
               type="button"
               onClick={() => handleSelectBuiltinStage(summaryDone ? "action_plan" : "summary")}
               className={cn(
-                "flex min-h-[44px] items-center gap-1.5 rounded-xl px-4 py-2.5 text-xs font-black text-white shadow-md transition active:scale-95",
+                "flex h-10 shrink-0 items-center gap-1.5 rounded-[4px] px-3 text-xs font-black text-white shadow-md transition active:scale-[0.98]",
                 summaryDone
                   ? "bg-sky-600 shadow-sky-500/10 hover:bg-sky-700"
                   : "bg-emerald-600 shadow-emerald-500/10 hover:bg-emerald-700",
@@ -1499,6 +1538,7 @@ export default function PositionRoundtablePage({
             </button>
           ) : undefined
         }
+        inputArea={(inputProps) => <PositionIntentInput {...inputProps} />}
         onManualSave={() => undefined}
         onTriggerToast={triggerToast}
         submitIntentClarifications={intent.submitIntentClarifications}
@@ -1654,7 +1694,7 @@ export default function PositionRoundtablePage({
         <aside
           aria-hidden={leftSidebarHidden || undefined}
           className={cn(
-            "custom-scrollbar relative flex min-h-0 shrink-0 flex-col gap-3 overflow-y-auto rounded-2xl border border-slate-200 bg-white shadow-sm transition-[width,opacity,transform,padding,border-width] duration-300 ease-[cubic-bezier(0.22,1,0.36,1)] motion-reduce:transition-none dark:rounded-none dark:border-0 dark:bg-transparent dark:text-[#c5cfe0] dark:shadow-none",
+            "position-roundtable-sidebar-scrollbar custom-scrollbar relative flex min-h-0 shrink-0 flex-col gap-3 overflow-y-auto rounded-2xl border border-slate-200 bg-white shadow-sm transition-[width,opacity,transform,padding,border-width] duration-300 ease-[cubic-bezier(0.22,1,0.36,1)] motion-reduce:transition-none dark:rounded-none dark:border-0 dark:bg-transparent dark:text-[#c5cfe0] dark:shadow-none",
             leftSidebarHidden
               ? "w-0 translate-x-4 !border-0 !p-0 opacity-0 pointer-events-none"
               : "w-[360px] translate-x-0 p-3 opacity-100 dark:p-0",
@@ -1714,7 +1754,10 @@ export default function PositionRoundtablePage({
         )}
 
         <section className="relative flex min-w-[420px] flex-1 flex-col overflow-hidden rounded-[4px] border border-slate-200 bg-white shadow-sm dark:border-[#2b8ffe]/45 dark:bg-[#032650]/80 dark:shadow-none dark:backdrop-blur-md">
-          {!isIntelligenceDeliveryOverview && !rightPanelOpen && !sandboxOpen && (
+          {!isIntelligenceDeliveryOverview &&
+            !rightPanelOpen &&
+            !sandboxOpen &&
+            !(positionSession && activeNode && !activeBuiltinKind) && (
             <TButton
               size="small"
               variant="outline"
@@ -1759,6 +1802,9 @@ export default function PositionRoundtablePage({
                 onSelectPositionNode={handleSelectPositionNode}
                 builtinAgentOptions={showClosureAgents ? builtinAgentOptions : undefined}
                 onSelectBuiltinAgent={showClosureAgents ? handleSelectBuiltinStage : undefined}
+                onOpenConfiguration={
+                  !rightPanelOpen && !sandboxOpen ? () => setRightPanelOpen(true) : undefined
+                }
                 onNodeUpdate={handleNodeUpdate}
                 onConversationStart={() => setRightPanelOpen(false)}
                 onPreviewArtifact={handlePreviewArtifact}

frontend-web/src/roundtable-planning/components/Step1Panel.tsx

@@ -65,6 +65,35 @@ import type {
 } from "@/roundtable-planning/api/multi-agent";
 import type { IntentSummary } from "@/roundtable-planning/api/intent";
 
+/**
+ * Step 1 输入区的受控数据与行为契约。
+ *
+ * 默认输入区继续由 Step1Panel 渲染；嵌入式页面可通过 `inputArea` 复用相同的
+ * 发送、附件和阶段按钮行为，同时拥有完全独立的 DOM 与视觉样式。
+ */
+export interface Step1InputAreaProps {
+  value: string;
+  onValueChange: (value: string) => void;
+  attachedFiles: File[];
+  onAddFiles: (files: File[]) => void;
+  onRemoveFile: (index: number) => void;
+  charactersLeft: number;
+  isSending: boolean;
+  canSend: boolean;
+  pendingClarification: boolean;
+  onSend: () => void;
+  onStop: () => void;
+  onFollowUp: () => void;
+  onSkip: () => void;
+  onEnterRoundtable: () => void;
+  enterButtonLabel: string;
+  intentReady: object | null;
+  footerAction?: React.ReactNode;
+  hideEnterRoundtable: boolean;
+  onManualSave: () => void;
+  onTriggerToast: (message: string) => void;
+}
+
 export interface Step1PanelProps {
   // ─── State ────────────────────────────────────────────────────
   /**
@@ -94,6 +123,11 @@ export interface Step1PanelProps {
   footerAction?: React.ReactNode;
   /** 已进入后续工作流时，隐藏默认的「进入研讨」动作。 */
   hideEnterRoundtable?: boolean;
+  /**
+   * 仅替换输入区的渲染插槽。消息列表、意图状态及附件状态仍由 Step1Panel
+   * 统一管理，因此岗位会商定制样式时不会分叉或复制 Step 1 的业务逻辑。
+   */
+  inputArea?: (props: Step1InputAreaProps) => React.ReactNode;
   onManualSave: () => void;
   onTriggerToast: (msg: string) => void;
   submitIntentClarifications: (
@@ -122,6 +156,7 @@ export function Step1Panel(props: Step1PanelProps) {
     enterButtonLabel = "进入多智能体研讨 >",
     footerAction,
     hideEnterRoundtable = false,
+    inputArea,
     onManualSave,
     onTriggerToast,
     submitIntentClarifications,
@@ -144,8 +179,7 @@ export function Step1Panel(props: Step1PanelProps) {
 
   const charactersLeft = inputText.length;
 
-  const handlePickFiles = (e: React.ChangeEvent<HTMLInputElement>) => {
-    const picked = Array.from(e.target.files ?? []);
+  const addAttachedFiles = (picked: File[]) => {
     if (picked.length > 0) {
       // 去重（按 名称+大小），追加到已选列表。
       setAttachedFiles((prev) => {
@@ -153,6 +187,10 @@ export function Step1Panel(props: Step1PanelProps) {
         return [...prev, ...picked.filter((f) => !seen.has(`${f.name}:${f.size}`))];
       });
     }
+  };
+
+  const handlePickFiles = (e: React.ChangeEvent<HTMLInputElement>) => {
+    addAttachedFiles(Array.from(e.target.files ?? []));
     // 清空 input 值，允许再次选同名文件。
     e.target.value = "";
   };
@@ -216,6 +254,30 @@ export function Step1Panel(props: Step1PanelProps) {
       </div>
 
       {/* Bottom input area */}
+      {inputArea ? (
+        inputArea({
+          value: inputText,
+          onValueChange: setInputText,
+          attachedFiles,
+          onAddFiles: addAttachedFiles,
+          onRemoveFile: removeAttachedFile,
+          charactersLeft,
+          isSending: isIntentSending,
+          canSend: !isIntentSending && Boolean(intentThreadId),
+          pendingClarification: Boolean(pendingIntentClarificationMsgId),
+          onSend: handleSend,
+          onStop: onStopIntent,
+          onFollowUp,
+          onSkip: onSkipQuestion,
+          onEnterRoundtable,
+          enterButtonLabel,
+          intentReady,
+          footerAction,
+          hideEnterRoundtable,
+          onManualSave,
+          onTriggerToast,
+        })
+      ) : (
       <div className="shrink-0 p-4 border-t border-slate-100 bg-[#fcfcfd]">
         <div className="relative border border-slate-200 focus-within:border-primary focus-within:ring-2 focus-within:ring-[var(--brand-200)]/50 transition-all rounded-xl overflow-hidden bg-white px-3 py-2.5">
           {/* 已选附件预览（点发送时随消息上传） */}
@@ -366,6 +428,7 @@ export function Step1Panel(props: Step1PanelProps) {
           </button>
         </div>
       </div>
+      )}
     </div>
   );
 }

frontend-web/src/styles/globals.css

@@ -616,6 +616,11 @@
       padding-box;
   }
 
+  /* 岗位会商左侧总面板：仅收窄滚动条宽度，沿用全局既有的滑块颜色、圆角和轨道样式。 */
+  .position-roundtable-sidebar-scrollbar::-webkit-scrollbar {
+    width: 8px;
+  }
+
   .workspace-shell {
     background: #ffffff;
   }