Merge branch 'dhk-1' into 'master'

feat(roundtable): 第二步席位执行模式(快速/思考/专业问答/多智能体)+ 子智能体任务卡片 + 后台完成跨页通知 + 链条默认模式持久化

See merge request knowledge-base-bj/deerflow-server!124
This commit is contained in:
董宏凯 committed 2026-06-16 06:19:04 +00:00
commit fe0dc8d1a5
48 files changed
+1733 -598

No files matched your search

-265
View File
@@ -1,265 +0,0 @@
# 多智能体协调 API 接口文档
后端路由:`offline-backend-20260512/backend/app/gateway/routers/multi_agent.py`
路由前缀:`/api/multi-agent`
## 概览
提供一个"总控 + 多个子智能体"的协调编排接口。
- **总控(leader)** 智能体:通过 `agent_orchestration` 技能给子智能体派活
- **子智能体(special)**:执行被派到的具体任务,完成后把交付内容广播给其他线程
调用方按下面顺序使用:
1. 调 `POST /api/multi-agent/init` 一次性创建协调智能体 + 给每个子智能体建 `thread`,拿到 `thread_ids` 映射
2. 客户端把 `thread_ids` 保管好,后续每次调 `POST /api/multi-agent/run/stream` 时回传
3. 先调用 `agent_type="leader"` 拿派活清单,再针对清单里每个子智能体调用 `agent_type="special"`
## 1. 初始化多智能体集群
### `POST /api/multi-agent/init`
创建协调智能体并为列表中的每个子智能体建立独立的 `thread`,同时把"职责说明"广播到所有子智能体线程作为初始上下文。
#### 请求体
```json
{
"agent_names": ["taiwan-intelligence-gathering-agent", "us-intelligence-gathering-agent"],
"main_agent_name": "main-controller",
"model": "deepseek-chat"
}
```
| 字段 | 类型 | 必填 | 说明 |
|---|---|---|---|
| `agent_names` | `string[]` | 是 | 已有子智能体的 id 列表(kebab-case,区分大小写不敏感,自动转小写) |
| `main_agent_name` | `string` | 否 | 要创建的协调智能体名,默认 `main-agent`。重名时会复用已存在的同名智能体 |
| `model` | `string` | 否 | 协调智能体使用的模型名。不传则用后端默认模型 |
#### 响应(200)
```json
{
"status": "success",
"main_agent_name": "main-controller",
"thread_ids": {
"main-controller": "thread-uuid-xxx",
"taiwan-intelligence-gathering-agent": "thread-uuid-yyy",
"us-intelligence-gathering-agent": "thread-uuid-zzz"
}
}
```
`thread_ids` 是一个 `{agent_name: thread_id}` 字典。**前端必须把这个对象完整保管下来**,后续调用 stream 接口需要原样回传。
#### 错误
| 状态码 | 触发条件 |
|---|---|
| `400` | `agent_names` 为空 |
| `401` | 未携带 `Authorization` 头 |
| `500` | 协调智能体创建失败且不是"已存在"错误 |
---
## 2. 流式调度
### `POST /api/multi-agent/run/stream`
通过 SSE(Server-Sent Events)流式驱动一个 leader 或 special 智能体跑一轮对话。
#### 请求体
```json
{
"agent_type": "leader",
"d_agent_thread_id": {
"main-controller": "thread-uuid-xxx",
"taiwan-intelligence-gathering-agent": "thread-uuid-yyy"
},
"agent_name": "main-controller",
"new_message": "请分析特朗普的支持率",
"skill_stop_names": [],
"model": "deepseek-chat"
}
```
| 字段 | 类型 | 必填 | 说明 |
|---|---|---|---|
| `agent_type` | `"leader"` \| `"special"` | 否 | 默认 `leader`。`leader` 调用协调智能体,`special` 调用子智能体 |
| `d_agent_thread_id` | `object` | 是 | 由 `/init` 接口返回的 `thread_ids` 映射,原样回传 |
| `agent_name` | `string` | 是 | 本轮要驱动的智能体名(必须存在于 `d_agent_thread_id` 中) |
| `new_message` | `string` | 是 | 这轮发给智能体的用户消息 |
| `skill_stop_names` | `string[]` | 否 | 触发后中断执行的技能名列表(默认空) |
| `model` | `string` | 否 | 本轮 LLM 模型名,默认 `deepseek-chat` |
#### 响应
`Content-Type: text/event-stream` 的 SSE 长连接,按到达顺序产出多种数据:
##### a. 透传上游事件
原样转发底层 `POST /api/threads/{thread_id}/runs/stream` 的事件帧(LangGraph 协议),包含 `event:`、`data:` 等行。前端按 LangGraph SDK 既有方式渲染即可。
##### b. 最后一帧:调度状态
流结束前会额外产出一帧 `data:` 行,结构因 `agent_type` 而异:
**`agent_type == "leader"`:**
```
data: {"status": [["us-intelligence-gathering-agent", "收集美国民众对特朗普的看法"], ["taiwan-intelligence-gathering-agent", "..."]]}
```
`status` 是一个 `[agent_name, task_text]` 二元组数组,代表协调智能体本轮派出去的任务清单。**前端拿到后应当对每个元素再调用一次本接口(`agent_type="special"`)来执行子智能体**。如果数组为空,说明 leader 觉得已有信息足够,不再派活,可以直接展示 leader 的最终文本回答。
**`agent_type == "special"`:**
```
data: {"status": "子智能体xxx完成yyy工作"}
```
代表子智能体完成本轮工作并已把交付内容广播给其他线程。
#### 副作用(重要)
- **leader 调用**:自动把"总控智能体收到用户消息: …"广播到所有非自己的线程;派活后再把"总控智能体给 X 派活, 内容为: …"广播到所有非派活双方的线程
- **special 调用**:完成后把"子智能体 X 完成 Y 工作, 交付内容为: …"广播到所有非自己的线程
这意味着每个线程的对话历史里会逐步累积出全局上下文,无需前端再做合并。
#### 错误
| 状态码 | 触发条件 |
|---|---|
| `400` | `agent_name` 缺失或未在 `d_agent_thread_id` 中 |
| `401` | 未携带 `Authorization` 头 |
| `502` | 上游 `runs/stream` 返回非 200 / SSE 解析失败 |
---
## 3. 前端调用流程示例
```ts
// 1. 拿 token
const loginResp = await fetch("/api/v1/auth/login/username", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ username: "yourname" }),
});
const { access_token } = await loginResp.json();
const authHeaders = {
Authorization: `Bearer ${access_token}`,
"Content-Type": "application/json",
};
// 2. 初始化
const initResp = await fetch("/api/multi-agent/init", {
method: "POST",
headers: authHeaders,
body: JSON.stringify({
agent_names: [
"taiwan-intelligence-gathering-agent",
"us-intelligence-gathering-agent",
],
main_agent_name: "main-controller",
model: "deepseek-chat",
}),
});
const { thread_ids } = await initResp.json();
// 3. 流式调用 leader
async function streamLeader(message: string, model = "deepseek-chat") {
const resp = await fetch("/api/multi-agent/run/stream", {
method: "POST",
headers: authHeaders,
body: JSON.stringify({
agent_type: "leader",
d_agent_thread_id: thread_ids,
agent_name: "main-controller",
new_message: message,
skill_stop_names: [],
model,
}),
});
const reader = resp.body!.getReader();
const decoder = new TextDecoder();
let buf = "";
let dispatched: [string, string][] = [];
while (true) {
const { done, value } = await reader.read();
if (done) break;
buf += decoder.decode(value, { stream: true });
const lines = buf.split("\n");
buf = lines.pop() ?? "";
for (const line of lines) {
if (!line.startsWith("data:")) continue;
try {
const payload = JSON.parse(line.slice(5).trim());
if (Array.isArray(payload.status)) {
dispatched = payload.status;
}
// 其余事件按 LangGraph SDK 协议自行渲染
} catch {
/* 非 JSON 帧,可忽略或交给现有渲染器 */
}
}
}
return dispatched;
}
// 4. 对派活清单逐个调用 special
async function streamSpecial(agentName: string, task: string, model = "deepseek-chat") {
return fetch("/api/multi-agent/run/stream", {
method: "POST",
headers: authHeaders,
body: JSON.stringify({
agent_type: "special",
d_agent_thread_id: thread_ids,
agent_name: agentName,
new_message: task,
skill_stop_names: [],
model,
}),
});
// 同样按 SSE 解析
}
// 5. 编排:先 leader 派活 → 依次执行 special → 再让 leader 汇总
const dispatched = await streamLeader("请分析特朗普的支持率");
for (const [agentName, task] of dispatched) {
await streamSpecial(agentName, task);
}
const next = await streamLeader("根据上面收集到的信息出最终结论");
// 当 leader 返回 status: [] 时,意味着汇报完毕,可结束循环
```
---
## 4. 注意事项
- **`model` 字段的取值**:必须是后端 `config.yaml` 中 `models[]` 已配置的模型名(如 `deepseek-chat`、`glm-4`、`gpt-4o` 等)。传错会导致上游 502
- **`thread_ids` 必须每次回传**:后端不存储这个映射,是无状态接口
- **协调智能体必须装有 `agent_orchestration` 技能**:`/init` 创建时会自动指定 `skills=["agent_orchestration"]`。后端 `skills/` 目录里必须存在该技能定义,否则 leader 无法派活
- **同名协调智能体冲突**:`main_agent_name` 重名时后端不会报错("already exists" 走兼容分支),会复用既有智能体——所以同一个用户多次 init 用同一个名字是安全的,但 SOUL 不会被覆盖
- **SSE 断连**:本接口底层 `on_disconnect: "continue"`,断连后上游 run 会继续跑到结束(不会取消)
- **并发广播**:广播消息使用 `asyncio.gather` 并发更新所有线程 state,性能上不会随子智能体数线性变慢
---
## 5. 相关接口(底层依赖)
本接口内部转发以下 Gateway 原生接口,了解细节可查:
| 接口 | 路由 | 用途 |
|---|---|---|
| 创建 thread | `POST /api/threads` | 为每个智能体建立独立会话 |
| 读 thread state | `GET /api/threads/{id}/state` | 广播前取当前消息列表 |
| 写 thread state | `POST /api/threads/{id}/state` | 把广播消息追加到 messages |
| 取 agent 详情 | `GET /api/agents/{id}` | init 时读取子智能体的 name/description |
| 建 agent | `POST /api/agents` | init 时创建协调智能体 |
| 流式 run | `POST /api/threads/{id}/runs/stream` | leader/special 的实际执行 |
@@ -1,125 +0,0 @@
# 多智能体「后台挂起」—— 当前状态、遗留问题与下次继续指南
> 配套设计文档:`multi-agent-background-run-dev.md`(分 Phase 0-9 的总方案)。
> 本文记录**截至本次会话的实际进展 + 还没解决的遗留问题**,方便下次直接续接。
---
## 一、整体状态:跑通到哪了
「后台挂起」= Step2 把研讨交给后端**进程内驱动真模型**(leader→各席位→共识→报告),
survive 刷新/关页,对话与报告作为一个 **run 写进草稿**(和手动跑同一存取),可在历史回看。
**真网关已在本地验证可跑通**(2026-06-04 日志确认:roundtable-coordinator + 各席位 +
roundtable-report 真调 deepseek、真搜索、跑到 report run success)。但还有几个**内容/产物层面的遗留问题**(见第三节)。
### 已完成(可用)
- ✅ 后端持久化:`roundtable_jobs` 表(迁移 `20260604_01` / `20260605_01` dialogues / `20260605_02` step3)。
- ✅ 编排引擎 `packages/harness/deerflow/agents/roundtable_orchestrator/`(recommend + chain + 报告,纯逻辑可单测)。
- ✅ 控制面 `app/gateway/routers/roundtable_jobs.py`:start/get/by-draft/stream(SSE)/resume/cancel/admin-cleanup。
- ✅ 执行器 `app/gateway/roundtable_job_executor.py`:后台任务 + `set_current_user` + 生命周期 + **每轮把对话作为 `job-<id>` run 写进 `draft.step2.runs[]`(同手动存取)** + done 写 step3。
- ✅ 真网关 `app/gateway/roundtable_inprocess_gateway.py`:`SimpleNamespace(app)` + `start_run` 进程内驱动 agent;复用 `multi_agent._extract_dispatch` 解析派活;checkpointer 读最终 messages。
- ✅ 模拟网关 `roundtable_orchestrator/simulated_gateway.py`(不调真模型,前端联调用)。
- ✅ 前端:`api/roundtable-jobs.ts`、`lib/job-types.ts`、Step2「后台挂起」按钮、历史下拉 loading、`DispatchChainModal`/`DispatchChainJobModal`(流程图 + 对话面板 markdown + 加宽 + 澄清作答 + 取消 + 查看)。
- ✅ 上下文「广播」(inline):席位/总控 prompt 带上「任务背景 + 前序各席位交付 + 别提问」约束(修「席位拿不到关键信息、反问用户、总控瞎编」)。
- ✅ 后端测试:`tests/test_roundtable_jobs.py` / `test_roundtable_orchestrator.py` / `test_roundtable_job_executor.py` / `test_roundtable_jobs_api.py` 全过。
---
## 二、怎么启动 / 验证(每次开工先做)
1. **开真网关**(默认是模拟网关):已写进 `offline-backend-20260512/.env`:
```
ROUNDTABLE_JOB_GATEWAY=inprocess
```
启动日志应出现 `[roundtable-jobs] gateway = inprocess (real models)`。
2. **必须用 `make gateway`(不带 --reload)跑后端!**
`make dev` 的 `--reload` 监视 `backend/` 目录,agent 运行时往 `.deer-flow/` 写文件
(memory.json / checkpoint / 沙箱输出)会触发热重载,**把长时后台任务杀掉**。
(改进项:给 uvicorn 加 `--reload-exclude '.deer-flow/*'`,或始终用 `make gateway`。)
3. **直接查库验证**(最有效,PowerShell 控制台是 GBK,导出到文件看中文):
```bash
cd offline-backend-20260512/backend
# 查最新作业的 dialogues / step3
PYTHONPATH=. uv run python -c "import sqlite3,json; con=sqlite3.connect('.deer-flow/data/deerflow.db'); ..."
```
- 作业表:`.deer-flow/data/deerflow.db` → 表 `roundtable_jobs`(`dialogues`/`step3` 列)。
- 草稿表:`roundtable_drafts` → `step2` 列里应有 `runs[ {id:"job-<jobId>", jobId, jobStatus, step2RoundtableDialogues:[...], step3:{...}} ]`。
4. **前端验证**:后台挂起 → 等跑完 → 从历史加载该草稿 → Step2 应显示后台研讨完整对话(走现有 `normalizeStep2Runs`/`hydrateStep2Run`)。
---
## 三、遗留问题(下次重点,已按优先级排)
### P0 —— 报告 agent 不写 HTML 文件(结果绘制拿不到产物)✅ 已于 2026-06-12 修复并恢复
- **根因(2026-06-12 定位)**:`dc8fca6`(2026-06-11 智能体可见性对齐)给 `services._enforce_agent_access`
加 admin 判断时**裸访问 `request.state`** —— 后台进程内虚拟 request(`SimpleNamespace(app=app)`)
没有 `state` 属性 → `AttributeError` 杀死整个编排(前端真 HTTP 请求有 auth 中间件写的 `state.user`
所以手动路径一直正常)。修复:getattr 全链防御 + 虚拟 request 补空 `state`(含 scheduler 同款),
回归测试 `tests/test_roundtable_inprocess_request.py`。
- **已恢复**:`ROUNDTABLE_BACKGROUND_REPORT_ENABLED = True`。后台收口后自动跑 Step3 双产物:
① `roundtable-summary` 写「方案总结报告.md」(新增 `gateway.run_summary` + `engine._run_step3`,
prompt 对齐 `useStep3Summary.buildSummaryPrompt`)→ ② `roundtable-report` 画「方案总览.html」。
两段故障隔离(一段失败保留另一段产物);step3 形状对齐前端 `DraftStep3Snapshot`
(`summaryReport` 嵌套),写草稿 `draft.step3` 与手动调用同款。测试
`tests/test_roundtable_step3_pipeline.py`。
- **待真机验证**:早期(06-05,回归引入前)曾观察到 deepseek-chat 在后台不调 `write_file`,
当时根因可能就是更早版本的同类访问崩溃,也可能另有 prompt 层问题——恢复后跑一单真模型确认
outputs 下两份产物都落盘;若报告仍不写 HTML,按旧排查清单对比前端 run 参数(run policy 现已
由 `roundtable_run_policy.py` 单源统一,差异面只剩 prompt/上下文)。
### P1 —— 内容质量(本次刚加 inline 广播,待真机验证)
- **已做**:席位/总控 prompt 注入「任务背景 + 前序交付 + 别提问」。
- **待验证**:真机跑一次,看席位是否基于真实上下文产出(不再反问/凭记忆瞎编)。
- **若 inline 不够**:考虑实现**真·thread 广播**(像 `multi_agent._append_thread_message` 那样把
「用户意图 / 派活 / 各席位交付」追加进各 thread 历史)。进程内要避开 HTTP loopback 鉴权,
得用 checkpointer 直接 append human 消息(参考 `threads` 路由 POST `/state` 的内部逻辑)。这块较重,先看 inline 够不够。
### P2 —— 对话「没有步骤(steps)」
- **现象**:手动跑的气泡带工具调用步骤卡(`onPhaseChange` → `appendStepIfNew` → `Step2Dialogue.steps`);
后台对话只存了最终正文,没有 steps。
- **下次怎么做**:后台 `_run_agent_turn` 跑完后,从 checkpoint 的 messages 里抽出 tool_calls,
映射成 `Step2Dialogue.steps`(`{id,key,label,ts,args}`),在执行器 `_to_step2_dialogue` 里带上。
纯展示增强,不影响内容/报告。
### P3 —— 双跑浪费 + --reload 坑(工程健壮性)
- 进 Step2 时前端会自动跑一遍本地研讨(真模型、真线程);点后台挂起后前端那条 run 在后端还会跑完
(`stopOrchestration` 只断客户端、停不了后端 run)→ 和后台作业短暂双跑、双倍 token。
- 优化方向:进 Step2 改成「显式开始」而非自动跑;或后台挂起时真正 cancel 前端那条 run。
- `make dev` 的 `--reload` 会杀后台任务(见第二节);改进项:`--reload-exclude '.deer-flow/*'`。
---
## 四、关键文件清单(下次定位用)
**后端**
| 文件 | 作用 |
|------|------|
| `app/gateway/roundtable_inprocess_gateway.py` | 真网关:进程内驱动 leader/seat/re 编排循环 + `build_report_prompt` + inline 广播(`_seat_message`/`_deliveries_block`) |
| `.../roundtable_orchestrator/progress.py` | 进度落库 + `on_dialogue` 每轮回调 |
| `.../persistence/roundtable_jobs/` | job 表 model/sql |
| `app/gateway/deps.py` | `ROUNDTABLE_JOB_GATEWAY` 选网关 + 注入 executor(draft_store) |
**前端**
| 文件 | 作用 |
|------|------|
| `src/roundtable-planning/api/roundtable-jobs.ts` | 作业 API + SSE |
| `src/roundtable-planning/lib/job-types.ts` | 类型(RoundtableJob/JobDialogue/...) |
| `src/roundtable-planning/components/DispatchChainModal.tsx` / `DispatchChainJobModal.tsx` | 流程图弹窗 + 对话面板 |
| `src/roundtable-planning/pages/RoundtablePlanningPage.tsx` | 后台挂起按钮、历史下拉、**runs[] 多运行模型**(`hydrateStep2Run`/`getStep2SnapshotWithRuns`/`normalizeStep2Runs`) |
| `src/roundtable-planning/api/drafts.ts` | `DraftStep2RunSnapshot`(带 jobId/jobStatus/step3) |
---
## 五、关键架构事实(避免重复踩坑)
1. **真网关进程内驱动**:`start_run(body, thread_id, SimpleNamespace(app=app))` + `set_current_user`,
不走 HTTP(loopback 调 `/api/multi-agent/*` 不可行:`_auth_headers` 只转发用户 token + owner check)。
2. **后台作业首启自建全新线程**(`StartParams.is_resume`),绝不复用前端本地研讨的线程(会撞 409)。
3. **存取统一**:后台作业 = `draft.step2.runs[]` 里一个带 jobId 的 run,和手动跑同一存取/hydrate 路径,
**不要再开 observe/回显分支**(之前试过,被否)。
4. **聊天记录确实存上了**(job.dialogues + draft 的 run);之前"没存上"是测了旧代码 / 看的是前端本地残缺数据。
5. 多 worker / SSE:状态进 DB、SSE 走 DB 轮询;真网关进程内无 loopback。
+161
View File
@@ -0,0 +1,161 @@
# 需求文档 · 问答中心左侧栏改造(0616)
> 整理自 2026-06-16 需求草稿。本批需求统一围绕「问答中心」菜单下 5 个页面入口的**左侧侧边栏**进行改造,目标是让每个业务页面的左栏从「通用导航」变为「贴合该业务场景的快捷入口 + 历史记录」。
>
> 菜单结构参考:`frontend-web/src/core/page-layout/sidebar-menu.ts` 中 `qa-center`(问答中心)分组,含 5 个子项,正好对应下文 5 条需求。
## 0. 背景与共性约定
| 需求 | 菜单项 | id | 现有路由 | 对应页面/组件 |
|------|--------|----|----------|--------------|
| 1 | 通用问答 | `general-qa` | `/page/workspace/chats/new` | Workspace Chat + 左栏(`recent-chat-list.tsx` 等) |
| 2 | 多智能体会商 | `multi-agent-roundtable` | `/page/strategy/qa/multi-agent-roundtable` | `roundtable-planning/pages/RoundtablePlanningPage.tsx` |
| 3 | zzJC支持 | `qa-zzjc` | `/page/strategy/agents/6bf/chats/new` | `pages/AgentChatPage.tsx` |
| 4 | zzXD应用 | `qa-zzxd` | `/page/strategy/agents/10d69686718349688e58a8790cb61024/chats/new` | `pages/AgentChatPage.tsx` |
| 5 | 课题研究 | `qa-research` | `/page/canvas/ai-writing` | `open-canvas/pages/AIWritingPage.tsx` |
**共性约定(适用于全部 5 条,除非单条另有说明):**
- 「更多」一律为可点击入口;点击后跳转到对应的完整列表/管理页(具体目标见各条「待确认」)。
- 列表为空时的兜底:参考各条「现状」中的级联/空态规则;无内容时该区块整体隐藏或显示占位文案(待 UI 设计确认,默认隐藏,与现有 `recent-chat-list.tsx` 一致)。
- 左栏宽度、折叠交互沿用现有侧边栏体系(`SidebarGroup` 系列组件)。
- 所有「智能体」相关数据复用 `core/agents`(`Agent` 类型已含 `is_pinned` / `is_favorite` / `featured_order`,scope 支持 `mine` / `square` / `builtin`)。
---
## 1. 通用问答 — 左侧栏改造
**页面**:`/page/workspace/chats/new`
### 1.1 现状
当前左栏主要展示「最近会话」列表(`recent-chat-list.tsx`,已过滤 `scheduler` / `bootstrap` / `roundtable` 类型线程,置顶会话优先)。
### 1.2 目标布局(自上而下)
左栏从上到下依次为三块:**新建回答按钮 → 常用智能体 → 聊天记录**。
#### ① 顶部:新建回答
- 左栏最顶部放置「**新建回答**」按钮(主操作)。
- 点击 = 开启一次全新的通用问答会话(等价于跳转 `/page/workspace/chats/new`)。
#### ② 常用智能体(最多 3 个 + 更多)
- 展示「常用智能体」卡片/列表项,**最多 3 个**。
- 数据来源采用**级联兜底**逻辑(取到即用,不足再降级):
1. 若用户有**置顶智能体**(`is_pinned = true`)→ 展示置顶的;
2. 否则展示**用户自己的智能体**(`scope = "mine"`);
3. 仍没有 → 展示**默认广场智能体**(`scope = "square"` / `builtin`,即广场默认推荐)。
- 三者按上述优先级取,最终最多取 3 个。
- 下方显示「**更多**」入口。
- 点击某个常用智能体 → 进入该智能体的对话(待确认:是否为该智能体新建会话)。
#### ③ 聊天记录(最多 10 条 + 更多)
- 展示用户的问答聊天记录,**最多 10 条**(沿用现有 `useThreads` + 过滤规则)。
- 下方显示「**更多**」入口。
- 点击「更多」→ **跳转到问答记录页面**(完整聊天记录列表页)。
### 1.3
- 「常用智能体」点击后「打开该智能体主页」
- 「常用智能体」的「更多」跳转目标页 智能体管理 `/page/strategy/agents`。
- 「聊天记录」的「问答记录页面」具体路由http://localhost:5173/#/page/workspace/chats。
- 级联兜底是(置顶组够 3 个就完全不展示我的/广场)如果不够「补足到 3 个
---
## 2. 多智能体会商 — 左侧栏重做
**页面**:`/page/strategy/qa/multi-agent-roundtable`(`RoundtablePlanningPage`)
### 2.1 现状
圆桌会商页已有业务链选择、分析草稿等能力(`roundtable-planning/api/chains.ts`、`hooks/useRoundtableDrafts.ts`)。需求要求**左栏整体重做**。
### 2.2 目标布局(自上而下)
#### ① 顶部:业务链(公共,前 3 个)
- 顶部展示**业务链**,仅取**公共**链条(`listChains("public")`,即所有已发布 `is_public = true` 的链条)。
- 仅展示**前 3 个**。
- 点击某条业务链 → 以该业务链启动一次会商(沿用现有"会商自动串联圆桌"流程,见记忆 `business-chain-mapping`)。
#### ② 下方:分析记录
- 业务链下方展示「**分析记录**」列表。
- 数据来源:圆桌分析草稿/历史(`useRoundtableDrafts` / `listDrafts`)。
- 点击某条分析记录 → 打开对应的历史会商分析。
### 2.3 待确认
- 「业务链」是否需要「更多」入口跳转到业务链条配置页(`/page/strategy/business-chains`)?需求未提及,默认仅展示前 3 个、无更多。 需要更多入口
- 「分析记录」是否需要分页/「更多」?需求未提及,默认展示全部或合理上限 分析记录默认全部吧。
- 公共业务链「前 3 个」的排序口径(创建时间倒序?已有 `listChains` 默认 newest first) 按照现在默认的显示来吧,然后要支持用户单选,如果用户选择后,点击进入多智能体研讨按钮时,直接按照选择的业务链进去执行,如果用户取消勾选,或没有勾选就还是打开弹窗。
---
## 3. zzJC支持 — 左侧栏改造(按类型聚合智能体)
**页面**:`/page/strategy/agents/6bf/chats/new`(当前为单一智能体 `6bf` 的对话页)
### 3.1 现状
当前 zzJC支持直接进入某个固定智能体(id=`6bf`)的对话页(`AgentChatPage`)。
### 3.2 目标布局
- **左栏按「类型」分组**,展示该类型下的智能体列表("类型"待确认,推测为智能体 `tags` / 分类)。
- 智能体分组**下方**展示**聊天记录**。
- 左栏**宽度与通用问答(需求 1)保持一致**("左侧是问答宽")。
### 3.3 Admin 设置入口
- 在 **admin 账号权限下**,页面**右侧顶部显示「设置」图标按钮**(齿轮图标)。
- 点击设置按钮 → 打开设置面板,可对该模块进行配置。
- 非 admin 账号不显示设置按钮(沿用 `sidebar-menu.ts` 中 `adminOnly` 同款角色判断)。
### 3.4 待确认
- 「类型」的确切含义:是智能体的 `tags`,还是 zzJC 业务自定义的一组分类?
- 现 zzJC 入口是单个智能体 `6bf`;改为「按类型展示多个智能体」后,**进入该菜单时默认落地页是什么**(智能体列表?还是仍默认进 `6bf` 对话)?是否需要新建一个 zzJC 聚合页。
- Admin「设置」具体可配置项:配置「该模块纳入哪些类型/哪些智能体」?还是其他(如默认智能体、欢迎语)?需明确设置项清单。
- 「聊天记录」过滤范围:仅 zzJC 相关会话,还是全部会话?
---
## 4. zzXD应用 — 左侧栏改造
**页面**:`/page/strategy/agents/10d69686718349688e58a8790cb61024/chats/new`
### 4.1 需求
**与需求 3(zzJC支持)完全一致**:
- 左栏按类型展示该类型下的智能体 + 下方聊天记录;
- 左栏宽度同通用问答;
- admin 账号下右侧顶部显示设置图标按钮,可设置。
### 4.2 待确认
- 同需求 3 全部待确认项。
- zzXD 与 zzJC 是否共用同一套「按类型 + 设置」组件(仅数据源/默认智能体不同),还是各自独立配置?建议**抽象为同一可复用组件**,通过参数区分。
---
## 5. 课题研究 — 左侧栏改造(研究模板 + 笔记本)
**页面**:`/page/canvas/ai-writing`(`AIWritingPage`)
### 5.1 目标布局(自上而下)
#### ① 上半部分:研究模板
- 左栏上半部分展示「**研究模板**」。
- 研究模板 = **原来的「文章类型」**(`open-canvas/hooks/useArticleTypes.ts` 的 `articleTypes`,即 AI 写作表单里的「文章类型」选择项)。
- 点击某个研究模板 → 以该模板/文章类型发起课题研究(沿用现有 `articleType` 设置逻辑)。
#### ② 下半部分:笔记本
- 左栏下半部分展示「**笔记本**」。
- 笔记本内容 = **「简洁模式」左侧菜单下方展示的笔记本内容**(`useStudioNotebooks`,即 `workspace-nav-studio.tsx` 中渲染的 studio notebooks 列表)。
- 点击某个笔记本 → 打开该笔记本(沿用现有 `/page/workspace/studio/notebooks/{id}` 行为,或作为课题研究的素材来源 `materialSource = 'notebook'`,待确认)。
> 备注:草稿原文写作「简介模式」,结合代码应为「**简洁模式**」(`appearance-settings-page.tsx` 中的全局界面模式:全量 / 简洁;简洁模式下笔记本在左栏菜单展示)。
### 5.2 待确认
- 「研究模板」点击行为:仅预填表单 `articleType`,还是直接开始生成?
- 「笔记本」点击行为:跳转笔记本详情页,还是将笔记本设为当前课题研究的素材来源(`materialSource='notebook'`)?
- 上/下两部分是否各自可折叠、是否需要「更多」入口与数量上限?需求未提及。
---
## 6. 整体待确认与建议
1. **复用性**:需求 3 / 4(zzJC / zzXD)布局完全相同,建议实现为同一套「按类型分组 + 聊天记录 + admin 设置」可复用组件,通过配置区分模块。
2. **左栏宽度统一**:需求 1 / 3 / 4 都要求"问答宽",建议统一左栏宽度常量,避免各页面各写一套。
3. **「更多」目标页**:需求 1 涉及「智能体更多」「聊天记录更多→问答记录页」两个跳转目标需先确定路由是否已存在。
4. **Admin 设置项清单**:需求 3 / 4 的设置面板具体能配置什么,需要产品给出字段清单后才能落地。
5. **入口语义变更**:需求 3 / 4 把原「单一智能体对话」入口改为「按类型聚合」,菜单 `path` 与默认落地页可能需要调整,需同步确认。
@@ -54,7 +54,7 @@ export function PageSidebarDark({ collapsed, onToggleCollapsed }: PageSidebarDar
};
const hasActiveChild = (item: MenuItem): boolean => {
if (item.path && pathname === item.path) {
if (item.path && (pathname === item.path || pathname.startsWith(item.path + "/"))) {
return true;
}
@@ -153,7 +153,7 @@ export function PageSidebarDark({ collapsed, onToggleCollapsed }: PageSidebarDar
inWujieEnv
? item.id === activeTopId
: item.path
? pathname === item.path
? pathname === item.path || pathname.startsWith(item.path + "/")
: hasActiveChild(item);
const handleSidebarItemClick = (item: MenuItem) => {
if (inWujieEnv) {
+5 -3
View File
@@ -42,7 +42,7 @@ type PageSidebarProps = {
type SettingsSection = "account" | "appearance" | "memory" | "tools" | "about";
const hasActiveChild = (item: MenuItem, pathname: string): boolean => {
if (item.path && pathname === item.path) {
if (item.path && (pathname === item.path || pathname.startsWith(item.path + "/"))) {
return true;
}
return item.children?.some((child) => hasActiveChild(child, pathname)) ?? false;
@@ -116,7 +116,9 @@ function SecondaryNavSidebar() {
const secondaryItems = activeTopItem?.children ?? [];
const isItemActive = (item: MenuItem) =>
item.path ? pathname === item.path : hasActiveChild(item, pathname);
item.path
? pathname === item.path || pathname.startsWith(item.path + "/")
: hasActiveChild(item, pathname);
const handleItemClick = (item: MenuItem) => {
const path = item.path ?? findFirstRoutePath(item.children);
@@ -210,7 +212,7 @@ function SecondaryNavSidebar() {
{thirdItems.map((child) => {
const ChildIcon = child.icon;
const childActive = child.path
? pathname === child.path
? pathname === child.path || pathname.startsWith(child.path + "/")
: hasActiveChild(child, pathname);
return (
<button
@@ -0,0 +1,127 @@
"use client";
import { Bot, ChevronRight } from "lucide-react";
import Link from "next/link";
import { usePathname } from "next/navigation";
import { useMemo } from "react";
import {
SidebarGroup,
SidebarGroupContent,
SidebarGroupLabel,
SidebarMenu,
SidebarMenuButton,
SidebarMenuItem,
} from "@/components/ui/sidebar";
import { useAgentsPaginated } from "@/core/agents";
import type { Agent } from "@/core/agents";
import { desensitize } from "@/lib/desensitize";
/** 常用智能体最多展示个数(需求 1.2 ②)。 */
const MAX_FREQUENT = 3;
/** 「更多」跳转目标:智能体管理页(需求 1.3)。 */
const MORE_HREF = "/page/strategy/agents";
/** 打开某个智能体主页(其对话落地页,thread_id=new = 新会话)。 */
function agentHomeHref(agentId: string): string {
return `/page/strategy/agents/${agentId}/chats/new`;
}
/**
* 「常用智能体」侧栏区块(通用问答左栏,全量模式)。
*
* 级联兜底取数(需求 1.2 ② / 1.3)——按优先级补足到最多 3 个、去重:
* 1. 置顶智能体(``is_pinned``,可来自「我的」或「广场」);
* 2. 不足 3 个 → 用「我的」智能体(``scope="mine"``)补足;
* 3. 仍不足 → 用「广场」默认推荐(``scope="square"``)补足。
*
* 取不到任何智能体时整块隐藏(与 ``RecentChatList`` 的空态约定一致)。
*/
export function FrequentAgentsList() {
const pathname = usePathname();
const { agents: mine } = useAgentsPaginated({
scope: "mine",
page: 1,
pageSize: 20,
});
const { agents: square } = useAgentsPaginated({
scope: "square",
page: 1,
pageSize: 20,
});
const frequent = useMemo<Agent[]>(() => {
const out: Agent[] = [];
const seen = new Set<string>();
const add = (agent: Agent | undefined) => {
if (!agent || out.length >= MAX_FREQUENT || seen.has(agent.id)) return;
seen.add(agent.id);
out.push(agent);
};
// 1. 置顶组优先(够 3 个则下面我的/广场都不再补)。
[...mine, ...square].filter((a) => a.is_pinned).forEach(add);
// 2. 我的智能体补足。
mine.forEach(add);
// 3. 广场默认推荐补足。
square.forEach(add);
return out;
}, [mine, square]);
if (frequent.length === 0) {
return null;
}
return (
<SidebarGroup className="px-2 pt-3 mt-2 border-t border-sidebar-border/60">
<SidebarGroupLabel className="text-muted-foreground/70 px-3 mb-1 text-[10px] font-semibold tracking-[0.12em] uppercase select-none">
常用智能体
</SidebarGroupLabel>
<SidebarGroupContent>
<SidebarMenu>
<div className="flex w-full flex-col gap-0.5">
{frequent.map((agent) => {
const href = agentHomeHref(agent.id);
const isActive = pathname === href;
return (
<SidebarMenuItem key={agent.id}>
<SidebarMenuButton
isActive={isActive}
asChild
className={
"rounded-lg h-7 px-3 text-[13px] text-muted-foreground hover:text-foreground hover:bg-sidebar-accent/60 " +
"data-[active=true]:bg-white data-[active=true]:text-foreground data-[active=true]:shadow-[0_1px_2px_rgba(15,23,42,0.05)] " +
"dark:data-[active=true]:bg-slate-700/80 dark:data-[active=true]:text-slate-100 dark:data-[active=true]:shadow-none"
}
>
<Link
className="flex w-full items-center gap-2 whitespace-nowrap"
href={href}
>
<Bot className="size-3.5 shrink-0 text-muted-foreground/70" />
<span className="truncate">{desensitize(agent.name)}</span>
</Link>
</SidebarMenuButton>
</SidebarMenuItem>
);
})}
<SidebarMenuItem>
<SidebarMenuButton
asChild
className="rounded-lg h-7 px-3 text-[12px] text-muted-foreground/70 hover:text-foreground hover:bg-sidebar-accent/60"
>
<Link
className="flex w-full items-center justify-between gap-1"
href={MORE_HREF}
>
<span>更多</span>
<ChevronRight className="size-3.5 shrink-0" />
</Link>
</SidebarMenuButton>
</SidebarMenuItem>
</div>
</SidebarMenu>
</SidebarGroupContent>
</SidebarGroup>
);
}
@@ -1,7 +1,8 @@
import type { Message } from "@langchain/langgraph-sdk";
import type { BaseStream } from "@langchain/langgraph-sdk/react";
import { CheckIcon, SendIcon, XIcon } from "lucide-react";
import { useCallback, useContext, useEffect, useMemo, useState } from "react";
import { useCallback, useContext, useEffect, useMemo, useRef, useState } from "react";
import { useStickToBottomContext } from "use-stick-to-bottom";
import {
Conversation,
@@ -52,6 +53,38 @@ import { SubtaskCard } from "./subtask-card";
export const MESSAGE_LIST_DEFAULT_PADDING_BOTTOM = 160;
export const MESSAGE_LIST_FOLLOWUPS_EXTRA_PADDING_BOTTOM = 80;
/** 用户发送新问题时强制滚到底部(即使用户此前上滑看过历史)。 */
function ScrollToBottomOnUserMessage({
threadId,
messages,
}: {
threadId: string;
messages: Message[];
}) {
const { scrollToBottom } = useStickToBottomContext();
const prevHumanCountRef = useRef<number | null>(null);
useEffect(() => {
prevHumanCountRef.current = null;
}, [threadId]);
useEffect(() => {
const humanCount = messages.reduce(
(count, message) => count + (message.type === "human" ? 1 : 0),
0,
);
const prev = prevHumanCountRef.current;
prevHumanCountRef.current = humanCount;
if (prev !== null && humanCount > prev) {
requestAnimationFrame(() => {
requestAnimationFrame(() => scrollToBottom());
});
}
}, [messages, scrollToBottom]);
return null;
}
export function MessageList({
className,
contentClassName,
@@ -266,6 +299,7 @@ export function MessageList({
<Conversation
className={cn("flex size-full flex-col justify-center", className)}
>
<ScrollToBottomOnUserMessage threadId={threadId} messages={messages} />
<ConversationContent
className={cn(
"mx-auto w-full max-w-(--container-width-lg) gap-8 pt-12",
@@ -2,6 +2,7 @@
import {
ChevronDown,
ChevronRight,
Download,
FileJson,
FileText,
@@ -74,7 +75,17 @@ import { desensitize } from "@/lib/desensitize";
import { isIMEComposing } from "@/lib/ime";
import { cn } from "@/lib/utils";
export function RecentChatList({ compact = false }: { compact?: boolean }) {
export function RecentChatList({
compact = false,
limit,
moreHref,
}: {
compact?: boolean;
/** 最多展示的会话条数;省略则不限制(需求 1.3:通用问答最多 10 条)。 */
limit?: number;
/** 「更多」入口跳转目标;省略则不展示「更多」行。 */
moreHref?: string;
}) {
const { t } = useI18n();
const router = useRouter();
const pathname = usePathname();
@@ -190,6 +201,8 @@ export function RecentChatList({ compact = false }: { compact?: boolean }) {
if (visibleThreads.length === 0) {
return null;
}
const shownThreads =
limit != null ? visibleThreads.slice(0, limit) : visibleThreads;
return (
<>
<SidebarGroup className="px-2 pt-3 mt-2 border-t border-sidebar-border/60">
@@ -225,7 +238,7 @@ export function RecentChatList({ compact = false }: { compact?: boolean }) {
>
<SidebarMenu>
<div className="flex w-full flex-col gap-0.5">
{visibleThreads.map((thread) => {
{shownThreads.map((thread) => {
const isActive = pathOfThread(thread) === pathname;
const pinned = isThreadPinned(thread);
const favorited = isThreadFavorite(thread);
@@ -368,6 +381,22 @@ export function RecentChatList({ compact = false }: { compact?: boolean }) {
</SidebarMenuItem>
);
})}
{moreHref && (
<SidebarMenuItem>
<SidebarMenuButton
asChild
className="rounded-lg h-7 px-3 text-[12px] text-muted-foreground/70 hover:text-foreground hover:bg-sidebar-accent/60"
>
<Link
className="flex w-full items-center justify-between gap-1"
href={moreHref}
>
<span>{t.common.more}</span>
<ChevronRight className="size-3.5 shrink-0" />
</Link>
</SidebarMenuButton>
</SidebarMenuItem>
)}
</div>
</SidebarMenu>
</SidebarGroupContent>
@@ -165,7 +165,7 @@ export function WorkspaceNavMenu() {
Checkpoint 迁移
</DropdownMenuItem>
)}
{isAdmin && (
{isAdmin ? (
<DropdownMenuItem
onClick={() =>
navigate("/page/strategy/admin/user-leaderboard")
@@ -174,6 +174,15 @@ export function WorkspaceNavMenu() {
<TrophyIcon />
用户排行榜
</DropdownMenuItem>
) : (
<DropdownMenuItem
onClick={() =>
navigate("/page/strategy/user-leaderboard")
}
>
<TrophyIcon />
用户排行榜
</DropdownMenuItem>
)}
{isAdmin && (
<DropdownMenuItem
@@ -10,7 +10,9 @@ import {
} from "@/components/ui/sidebar";
import { useTheme } from "@/contexts/ThemeContext";
import { cn } from "@/lib/utils";
import { usePathname } from "next/navigation";
import { FrequentAgentsList } from "./frequent-agents-list";
import { RecentChatList } from "./recent-chat-list";
import { RecentRoundtableList } from "./recent-roundtable-list";
import { RecentWritingList } from "./recent-writing-list";
@@ -25,7 +27,11 @@ export function WorkspaceSidebar({
}: React.ComponentProps<typeof Sidebar> & { compactChrome?: boolean }) {
const { open: isSidebarOpen } = useSidebar();
const { theme } = useTheme();
const pathname = usePathname();
const useWhiteSidebar = compactChrome && theme === "light";
// 通用问答(全量模式)左栏:新建回答 → 常用智能体 → 聊天记录(需求 0616 #1)。
// 仅在通用问答(/page/workspace/chats*)下启用三块式布局;其余 workspace 页保持原样。
const isGeneralQa = pathname.startsWith("/page/workspace/chats");
// 紧凑模式:用「外层定宽容器 + transition-[width]」做展开/收起过渡动画。
// collapsible="none" 的 Sidebar 是普通 in-flow 定宽块(非 absolute,故不受本布局对
@@ -44,7 +50,7 @@ export function WorkspaceSidebar({
variant="sidebar"
collapsible="none"
className={cn(
"h-full w-(--sidebar-width)",
"h-full w-(--sidebar-width) border-r border-border",
useWhiteSidebar &&
"[&_[data-slot=sidebar-inner]]:bg-white [&_[data-slot=sidebar-container]]:bg-white",
className,
@@ -53,7 +59,12 @@ export function WorkspaceSidebar({
>
<SidebarContent>
<WorkspaceNavChatList showCollapseTrigger compact />
<RecentChatList compact />
{isGeneralQa && <FrequentAgentsList />}
<RecentChatList
compact
limit={isGeneralQa ? 10 : undefined}
moreHref={isGeneralQa ? "/page/workspace/chats" : undefined}
/>
</SidebarContent>
<SidebarRail />
</Sidebar>
@@ -43,6 +43,8 @@ export type MenuItem = {
label: string;
icon: ElementType;
path?: string;
/** 非 admin 用户使用的路由(菜单对普通用户可见时生效,与简洁模式公开排行榜一致) */
userPath?: string;
adminOnly?: boolean;
children?: MenuItem[];
};
@@ -81,18 +83,18 @@ export const sidebarMenuItems: MenuItem[] = [
label: "XK中任务",
icon: ClipboardList,
children: [
{
id: "xk-task-jc",
label: "任务JC",
icon: ListTodo,
path: "/page/strategy/task-management/xk-task-jc",
},
{
id: "xk-task-dispatch",
label: "任务调度",
icon: ListTodo,
path: "/page/strategy/task-management/xk-task-dispatch",
},
// {
// id: "xk-task-jc",
// label: "任务JC",
// icon: ListTodo,
// path: "/page/strategy/task-management/xk-task-jc",
// },
// {
// id: "xk-task-dispatch",
// label: "任务调度",
// icon: ListTodo,
// path: "/page/strategy/task-management/xk-task-dispatch",
// },
],
},
{
@@ -125,24 +127,24 @@ export const sidebarMenuItems: MenuItem[] = [
label: "PG中任务",
icon: Gauge,
children: [
{
id: "pg-task-pg",
label: "任务PG",
icon: ListTodo,
path: "/page/strategy/task-management/effect-task-pg",
},
{
id: "pg-task-xdpg",
label: "XDPG",
icon: ListTodo,
path: "/page/strategy/task-management/effect-action-pg",
},
{
id: "pg-task-info-pg",
label: "信息PG",
icon: ListTodo,
path: "/page/strategy/task-management/effect-info-pg",
},
// {
// id: "pg-task-pg",
// label: "任务PG",
// icon: ListTodo,
// path: "/page/strategy/task-management/effect-task-pg",
// },
// {
// id: "pg-task-xdpg",
// label: "XDPG",
// icon: ListTodo,
// path: "/page/strategy/task-management/effect-action-pg",
// },
// {
// id: "pg-task-info-pg",
// label: "信息PG",
// icon: ListTodo,
// path: "/page/strategy/task-management/effect-info-pg",
// },
],
},
{
@@ -150,24 +152,24 @@ export const sidebarMenuItems: MenuItem[] = [
label: "TY中任务",
icon: Radar,
children: [
{
id: "ty-task-xdty",
label: "XDTY任务",
icon: ListTodo,
path: "/page/strategy/task-management/mnty-task-mechanism",
},
{
id: "ty-task-plan",
label: "方案TY任务",
icon: ListTodo,
path: "/page/strategy/task-management/mnty-task-plan",
},
{
id: "ty-task-xxty",
label: "XXTY任务",
icon: ListTodo,
path: "/page/strategy/task-management/mnty-task-zsty",
},
// {
// id: "ty-task-xdty",
// label: "XDTY任务",
// icon: ListTodo,
// path: "/page/strategy/task-management/mnty-task-mechanism",
// },
// {
// id: "ty-task-plan",
// label: "方案TY任务",
// icon: ListTodo,
// path: "/page/strategy/task-management/mnty-task-plan",
// },
// {
// id: "ty-task-xxty",
// label: "XXTY任务",
// icon: ListTodo,
// path: "/page/strategy/task-management/mnty-task-zsty",
// },
],
},
],
@@ -197,6 +199,13 @@ export const sidebarMenuItems: MenuItem[] = [
label: "综合管理",
icon: Settings,
children: [
{
id: "mgmt-user-leaderboard",
label: "用户排行榜",
icon: Trophy,
path: "/page/strategy/admin/user-leaderboard",
userPath: "/page/strategy/user-leaderboard",
},
{ id: "mgmt-agents", label: "智能体管理", icon: Bot, path: "/page/strategy/agents" },
{
id: "mgmt-business-chain-mapping",
@@ -232,13 +241,6 @@ export const sidebarMenuItems: MenuItem[] = [
path: "/page/strategy/admin/users",
adminOnly: true,
},
{
id: "mgmt-user-leaderboard",
label: "用户排行榜",
icon: Trophy,
path: "/page/strategy/admin/user-leaderboard",
adminOnly: true,
},
{
id: "mgmt-admin-scheduled-tasks",
label: "定时任务管理",
@@ -260,12 +262,12 @@ export const sidebarMenuItems: MenuItem[] = [
label: "资源管理",
icon: Database,
children: [
{
id: "resource-network-xwt",
label: "网络xwt资源管理",
icon: Network,
path: "/page/strategy/resource-management/resource-network-actor",
},
// {
// id: "resource-network-xwt",
// label: "网络xwt资源管理",
// icon: Network,
// path: "/page/strategy/resource-management/resource-network-actor",
// },
{
id: "resource-knowledge-native",
label: "问答梳理",
@@ -278,12 +280,12 @@ export const sidebarMenuItems: MenuItem[] = [
icon: Library,
path: "/page/strategy/resource-management/resource-knowledge-base",
},
{
id: "resource-evaluation-index",
label: "作战资源管理",
icon: FileBox,
path: "/page/strategy/resource-management/resource-evaluation-index",
},
// {
// id: "resource-evaluation-index",
// label: "作战资源管理",
// icon: FileBox,
// path: "/page/strategy/resource-management/resource-evaluation-index",
// },
],
},
];
@@ -298,9 +300,9 @@ export const defaultExpandedIds: ReadonlySet<string> = new Set([
"comprehensive-management",
]);
/** 该菜单项自身或任意后代的 path 是否等于当前路由 */
/** 该菜单项自身或任意后代的 path 是否匹配当前路由(支持子路径,如 /knowledge/:id) */
export function menuItemContainsPath(item: MenuItem, pathname: string): boolean {
if (item.path && item.path === pathname) {
if (item.path && (pathname === item.path || pathname.startsWith(item.path + "/"))) {
return true;
}
return item.children?.some((child) => menuItemContainsPath(child, pathname)) ?? false;
@@ -308,7 +310,8 @@ export function menuItemContainsPath(item: MenuItem, pathname: string): boolean
/**
* 在「一级 → 二级 → 三级」菜单中,取当前路由所在的「激活二级项」的三级子项。
* 浅色侧边栏据此在二级右侧展开三级列;无三级则返回空数组(侧边栏保持窄栏)。
* 浅色侧边栏据此在二级右侧展开三级列;深色顶栏据此渲染三级横条。
* 全量模式:三级仅一项时不返回(无选择必要,侧边栏保持窄栏 / 不渲染横条)。
*/
export function getActiveThirdLevelItems(
items: MenuItem[],
@@ -320,16 +323,23 @@ export function getActiveThirdLevelItems(
const activeSecond = secondary.find(
(item) => (item.children?.length ?? 0) > 0 && menuItemContainsPath(item, pathname),
);
return activeSecond?.children ?? [];
const third = activeSecond?.children ?? [];
return third.length <= 1 ? [] : third;
}
/** 按 admin 角色过滤菜单:非 admin 不展示 ``adminOnly: true`` 的项及其子项 */
/** 按 admin 角色过滤菜单:非 admin 不展示 ``adminOnly: true`` 的项;``userPath`` 替换 ``path``。 */
export function filterMenuItemsByRole(items: MenuItem[], isAdmin: boolean): MenuItem[] {
return items
.filter((item) => !item.adminOnly || isAdmin)
.map((item) =>
item.children
? { ...item, children: filterMenuItemsByRole(item.children, isAdmin) }
: item,
);
.map((item) => {
const children = item.children
? filterMenuItemsByRole(item.children, isAdmin)
: undefined;
const base = children ? { ...item, children } : item;
if (!isAdmin && base.userPath) {
const { userPath: _userPath, ...rest } = base;
return { ...rest, path: base.userPath };
}
return base;
});
}
@@ -1,5 +1,5 @@
import { useState } from 'react';
import { ChevronRight, Sparkles, ListChecks, Zap, Loader2, CheckCircle2, CircleStop, Search } from 'lucide-react';
import { ChevronRight, Sparkles, ListChecks, Zap, Loader2, CheckCircle2, CircleStop, Search, AlertTriangle, Shuffle } from 'lucide-react';
import { cn } from '@/lib/utils';
import type { SkillSearchStep } from '../../types/ai-writing';
import { CollapseTransition } from './CollapseTransition';
@@ -16,6 +16,8 @@ const STEP_ICON: Record<SkillSearchStep['type'], typeof Sparkles> = {
select_skills: Sparkles,
invoke_skill: Zap,
skill_result: CheckCircle2,
skill_error: AlertTriangle,
fallback: Shuffle,
stop_early: CircleStop,
};
@@ -25,6 +27,8 @@ const STEP_LABEL: Record<SkillSearchStep['type'], string> = {
select_skills: '筛选相关技能',
invoke_skill: '调用技能',
skill_result: '检索结果',
skill_error: '技能调用失败',
fallback: '改用其他技能',
stop_early: '提前停止',
};
@@ -71,6 +75,9 @@ export function SkillStepsCard({ steps, inProgress }: SkillStepsCardProps) {
const label = STEP_LABEL[step.type] ?? step.type;
const isStopEarly = step.type === 'stop_early';
const isResult = step.type === 'skill_result';
// 容错:技能调用失败标红,回退其它技能标琥珀,提示用户"出错了但已自动改用其他技能"
const isError = step.type === 'skill_error';
const isFallback = step.type === 'fallback';
const isActive = idx === activeIdx;
return (
<li key={idx} className="flex items-start gap-2 text-[11px]">
@@ -79,11 +86,13 @@ export function SkillStepsCard({ steps, inProgress }: SkillStepsCardProps) {
'shrink-0 mt-0.5 inline-flex items-center justify-center size-4 rounded-full text-[10px] font-medium',
isActive
? 'bg-violet-500 text-white'
: isStopEarly
? 'bg-amber-100 dark:bg-amber-900/60 text-amber-600 dark:text-amber-300'
: isResult
? 'bg-emerald-100 dark:bg-emerald-900/60 text-emerald-600 dark:text-emerald-300'
: 'bg-violet-100 dark:bg-violet-900/60 text-violet-600 dark:text-violet-300',
: isError
? 'bg-rose-100 dark:bg-rose-900/60 text-rose-600 dark:text-rose-300'
: isStopEarly || isFallback
? 'bg-amber-100 dark:bg-amber-900/60 text-amber-600 dark:text-amber-300'
: isResult
? 'bg-emerald-100 dark:bg-emerald-900/60 text-emerald-600 dark:text-emerald-300'
: 'bg-violet-100 dark:bg-violet-900/60 text-violet-600 dark:text-violet-300',
)}
>
{isActive ? <Loader2 className="size-2.5 animate-spin" /> : idx + 1}
@@ -92,11 +101,13 @@ export function SkillStepsCard({ steps, inProgress }: SkillStepsCardProps) {
<div
className={cn(
'flex items-center gap-1 font-medium',
isStopEarly
? 'text-amber-700 dark:text-amber-300'
: isResult
? 'text-emerald-700 dark:text-emerald-300'
: 'text-violet-700 dark:text-violet-300',
isError
? 'text-rose-700 dark:text-rose-300'
: isStopEarly || isFallback
? 'text-amber-700 dark:text-amber-300'
: isResult
? 'text-emerald-700 dark:text-emerald-300'
: 'text-violet-700 dark:text-violet-300',
)}
>
<Icon className="size-3 shrink-0" />
@@ -20,6 +20,8 @@ export interface AIMaterial {
// select_skills:LLM 按相关度排序的候选
// invoke_skill:开始调用某个技能去检索
// skill_result:某个技能检索返回了 N 条素材
// skill_error:某个技能调用失败(超时/报错),将改用其他技能(容错)
// fallback:配置技能全部失败,回退到其他技能继续搜集(容错)
// stop_early:累计素材达到上限,停止后续技能
export interface SkillSearchStep {
type:
@@ -28,6 +30,8 @@ export interface SkillSearchStep {
| "select_skills"
| "invoke_skill"
| "skill_result"
| "skill_error"
| "fallback"
| "stop_early";
message: string;
skillNames?: string[];
@@ -133,6 +133,8 @@ export default function LightAppManagementPage() {
const [form, setForm] = useState<FormState>(emptyForm);
const [activeParamPopover, setActiveParamPopover] = useState<string | null>(null);
const [viewMode, setViewMode] = useState<"table" | "cards">("table");
const [categoryFilter, setCategoryFilter] = useState<"all" | LightAppType>("all");
const [mountFilter, setMountFilter] = useState<string>("");
const saving = actions.create.isPending || actions.update.isPending;
@@ -220,7 +222,22 @@ export default function LightAppManagementPage() {
toast.success("已复制地址");
};
const tableApps = useMemo(() => apps, [apps]);
const selectCategory = (cat: "all" | LightAppType) => {
setCategoryFilter(cat);
if (cat !== "iframe") setMountFilter("");
};
const tableApps = useMemo(
() =>
apps.filter((app) => {
if (categoryFilter !== "all" && app.appType !== categoryFilter) return false;
if (categoryFilter === "iframe" && mountFilter && app.mountLocation !== mountFilter) {
return false;
}
return true;
}),
[apps, categoryFilter, mountFilter],
);
return (
<div className="light-app-management-page h-full overflow-y-auto bg-neutral-50 dark:bg-[#0C1836] dark:text-[#C5CFE0]">
@@ -252,7 +269,7 @@ export default function LightAppManagementPage() {
</button>
</div>
<div className="lam-toolbar flex flex-col items-center gap-3.5 rounded-[5px] bg-white p-3 shadow-[0_1px_3px_rgba(0,0,0,0.01)] md:flex-row">
<div className="lam-toolbar flex flex-col items-center gap-3.5 rounded-[5px] bg-white p-3 shadow-[0_1px_3px_rgba(0,0,0,0.01)] md:flex-row md:flex-wrap">
<div className="relative w-full flex-1">
<Search className="absolute left-3.5 top-1/2 h-4 w-4 -translate-y-1/2 text-neutral-400" />
<input
@@ -272,6 +289,41 @@ export default function LightAppManagementPage() {
)}
</div>
<div className="lam-cat-toggle flex shrink-0 items-center space-x-0.5 self-end rounded-[5px] border border-neutral-200/50 bg-neutral-100/60 p-0.5 md:self-auto">
{(
[
{ value: "all", label: "全部" },
{ value: "window_open", label: "新页跳转" },
{ value: "iframe", label: "iframe 嵌套" },
] as const
).map((c) => (
<button
key={c.value}
onClick={() => selectCategory(c.value)}
className={`lam-cat-toggle-btn cursor-pointer rounded-[5px] px-4 py-1.5 text-xs font-bold transition-all ${
categoryFilter === c.value
? "is-active bg-white text-[#F5222D] shadow-sm"
: "text-neutral-500 hover:text-neutral-700"
}`}
>
{c.label}
</button>
))}
</div>
{categoryFilter === "iframe" && (
<div className="lam-mount-filter w-full shrink-0 animate-in fade-in self-end md:w-[180px] md:self-auto">
<Select
className="w-full"
value={mountFilter || undefined}
placeholder="按挂载菜单筛选"
clearable
options={MOUNT_OPTIONS}
onChange={(v) => setMountFilter((v as string) ?? "")}
/>
</div>
)}
<div className="lam-view-toggle flex shrink-0 items-center space-x-0.5 self-end rounded-[5px] border border-neutral-200/50 bg-neutral-100/60 p-0.5 md:self-auto">
<button
onClick={() => setViewMode("table")}
@@ -650,7 +702,7 @@ export default function LightAppManagementPage() {
onClick={() => setDialogOpen(false)}
/>
<div className="lam-dialog relative z-[110] m-4 flex max-h-[90vh] w-full max-w-[580px] flex-col rounded-[5px] border border-[#E8E8E8] bg-white p-6 shadow-[0_8px_32px_rgba(34,34,34,0.16)]">
<div className="lam-dialog relative z-[110] m-4 flex max-h-[90vh] w-full max-w-[900px] flex-col rounded-[5px] border border-[#E8E8E8] bg-white p-6 shadow-[0_8px_32px_rgba(34,34,34,0.16)]">
<div className="lam-dialog-header flex flex-shrink-0 items-center justify-between border-b border-[#E8E8E8] pb-4">
<div className="flex items-center space-x-2.5">
<span className="lam-dialog-icon rounded-[5px] bg-rose-50 p-1.5 text-[#A8071A]">
+3
View File
@@ -17,6 +17,7 @@ import Header from "@/strategy-components/Header";
import OpenCanvasRoutes from "@/open-canvas/routes/OpenCanvasRoutes";
import RoundtableRoutes from "./RoundtableRoutes";
import NotebookRoutes from "./NotebookRoutes";
import { RoundtableJobNotifier } from "@/roundtable-planning/components/RoundtableJobNotifier";
function PageLayout() {
const [collapsed, setCollapsed] = useState(false);
@@ -43,6 +44,7 @@ function PageLayout() {
// 高度链,根布局跟随容器高度。全屏下 100% 解析到视口,与 h-screen 等价;无界等嵌套
// 下容器不占满整屏时,高度才能正确收缩而非锁死整屏。
return (
<RoundtableJobNotifier>
<div className="h-full flex flex-col bg-muted/20">
{pageSidebarAllowed && <Header collapsed={collapsed} />}
<div className="flex flex-1 overflow-hidden">
@@ -97,6 +99,7 @@ function PageLayout() {
</main>
</div>
</div>
</RoundtableJobNotifier>
);
}
@@ -25,6 +25,7 @@ import ChatPage from "./ChatPage";
import ChatsPage from "./ChatsPage";
import AdminUserLeaderboardPage from "./AdminUserLeaderboardPage";
import AdminUserLeaderboardSettingsPage from "./AdminUserLeaderboardSettingsPage";
import PublicLeaderboardPage from "./PublicLeaderboardPage";
import LightAppIframeView from "./LightAppIframeView";
import LightAppListPage from "./LightAppListPage";
import LightAppManagementPage from "./LightAppManagementPage";
@@ -253,6 +254,7 @@ export default function StrategyRoutes() {
)
}
/>
<Route path="user-leaderboard" element={<PublicLeaderboardPage />} />
<Route
path="admin/user-leaderboard"
element={
+6 -4
View File
@@ -70,6 +70,7 @@ export function WorkspaceLayout({
workspaceSidebarHidden = false,
headerHidden = false,
compactWorkspaceSidebar = false,
defaultHistorySidebarOpen,
wrapperOverflowHidden = false,
insetOverflowHidden = false,
pageSidebarDark = false,
@@ -80,6 +81,8 @@ export function WorkspaceLayout({
workspaceSidebarHidden?: boolean;
headerHidden?: boolean;
compactWorkspaceSidebar?: boolean;
/** 全量模式下通用问答等页:聊天记录侧栏默认展开;未指定时紧凑模式仍默认收起。 */
defaultHistorySidebarOpen?: boolean;
wrapperOverflowHidden?: boolean;
insetOverflowHidden?: boolean;
pageSidebarDark?: boolean;
@@ -110,8 +113,7 @@ export function WorkspaceLayout({
}
key={String(compactWorkspaceSidebar)}
data-page-sidebar-collapsed={pageSidebarCollapsed}
// 紧凑模式下历史侧栏默认收起(聊天区更宽),由聊天页右上角图标钮展开
defaultOpen={!compactWorkspaceSidebar}
defaultOpen={defaultHistorySidebarOpen ?? !compactWorkspaceSidebar}
>
{!workspaceSidebarHidden && (
<WorkspaceSidebar compactChrome={compactWorkspaceSidebar} />
@@ -153,7 +155,7 @@ function NewChatEntryRoute({
}
return (
<WorkspaceLayout {...layoutProps}>
<WorkspaceLayout {...layoutProps} defaultHistorySidebarOpen>
<ChatRuntime>
<ChatPage />
</ChatRuntime>
@@ -201,7 +203,7 @@ export default function WorkspaceRoutes({
<Route
path="chats/:thread_id"
element={
<WorkspaceLayout {...layoutProps}>
<WorkspaceLayout {...layoutProps} defaultHistorySidebarOpen>
<ChatRuntime>
<ChatPage />
</ChatRuntime>
@@ -65,22 +65,26 @@
outline: none;
}
.dark .light-app-management-page .lam-view-toggle {
.dark .light-app-management-page .lam-view-toggle,
.dark .light-app-management-page .lam-cat-toggle {
background: var(--color-bg-page);
border-color: var(--color-border);
}
.dark .light-app-management-page .lam-view-toggle-btn.is-active {
.dark .light-app-management-page .lam-view-toggle-btn.is-active,
.dark .light-app-management-page .lam-cat-toggle-btn.is-active {
background: var(--color-bg-card);
color: var(--color-link);
box-shadow: var(--shadow-1);
}
.dark .light-app-management-page .lam-view-toggle-btn:not(.is-active) {
.dark .light-app-management-page .lam-view-toggle-btn:not(.is-active),
.dark .light-app-management-page .lam-cat-toggle-btn:not(.is-active) {
color: var(--color-text-secondary);
}
.dark .light-app-management-page .lam-view-toggle-btn:not(.is-active):hover {
.dark .light-app-management-page .lam-view-toggle-btn:not(.is-active):hover,
.dark .light-app-management-page .lam-cat-toggle-btn:not(.is-active):hover {
color: var(--color-text);
}
@@ -592,7 +596,8 @@
}
/* ── 视图切换激活态 ──────────────────────────────────────────────────── */
[data-theme="lightblue"] .light-app-management-page .lam-view-toggle-btn.is-active {
[data-theme="lightblue"] .light-app-management-page .lam-view-toggle-btn.is-active,
[data-theme="lightblue"] .light-app-management-page .lam-cat-toggle-btn.is-active {
color: #0062D9;
}
@@ -14,6 +14,7 @@ import { apiFetch } from "@/core/api/fetch-client";
import { getBackendBaseURL } from "@/core/config";
import type { BusinessChain, ChainSeat } from "@/roundtable-planning/lib/business-chain";
import type { SeatMode } from "@/roundtable-planning/lib/seat-mode";
/** Fields accepted when creating a chain. */
export interface ChainCreatePayload {
@@ -28,6 +29,8 @@ export interface ChainCreatePayload {
stageGoals?: (string | null)[] | null;
/** Optional coordinator orchestration prompt (dag-only). null/"" = none. */
coordinatorPrompt?: string | null;
/** 默认席位执行模式(flash/thinking/pro/ultra);null = 未设。 */
seatMode?: SeatMode | null;
}
/** Partial update — send only what changed. */
@@ -41,6 +44,8 @@ export interface ChainUpdatePayload {
stageGoals?: (string | null)[] | null;
/** 写给总控的编排提示;send `null`/"" to clear, omit to leave. */
coordinatorPrompt?: string | null;
/** 默认席位执行模式;send `null` to clear, omit to leave. */
seatMode?: SeatMode | null;
/** 发布开关:true 发布为公共 / false 取消发布;omit 不改。仅 owner 生效。 */
isPublic?: boolean;
}
@@ -70,6 +75,7 @@ interface ServerChain {
stages?: string[][] | null;
stage_goals?: (string | null)[] | null;
coordinator_prompt?: string | null;
seat_mode?: string | null;
is_public?: boolean;
is_mine?: boolean;
owner_name?: string | null;
@@ -86,6 +92,7 @@ function toChain(s: ServerChain): BusinessChain {
stages: Array.isArray(s.stages) ? s.stages : null,
stageGoals: Array.isArray(s.stage_goals) ? s.stage_goals : null,
coordinatorPrompt: s.coordinator_prompt ?? null,
seatMode: (s.seat_mode as SeatMode | null) ?? undefined,
isPublic: s.is_public ?? false,
// 后端缺省时按「我的」处理(向后兼容旧响应)。
isMine: s.is_mine ?? true,
@@ -104,6 +111,7 @@ function fromCreate(p: ChainCreatePayload) {
...(p.stages !== undefined ? { stages: p.stages } : {}),
...(p.stageGoals !== undefined ? { stage_goals: p.stageGoals } : {}),
...(p.coordinatorPrompt !== undefined ? { coordinator_prompt: p.coordinatorPrompt } : {}),
...(p.seatMode !== undefined ? { seat_mode: p.seatMode } : {}),
};
}
@@ -117,6 +125,7 @@ function fromUpdate(p: ChainUpdatePayload) {
// Forward stage goals even when explicitly null (clears goals server-side).
if (p.stageGoals !== undefined) out.stage_goals = p.stageGoals;
if (p.coordinatorPrompt !== undefined) out.coordinator_prompt = p.coordinatorPrompt;
if (p.seatMode !== undefined) out.seat_mode = p.seatMode;
if (p.isPublic !== undefined) out.is_public = p.isPublic;
return out;
}
@@ -202,6 +202,15 @@ export interface StreamMultiAgentRequest {
displayName?: string;
newMessage: string;
model?: string;
/**
* 仅 special 用。席位执行模式(快速/思考/专业问答/多智能体问答)派生的运行参数,
* 覆盖后端 `roundtable_run_policy` 的 thinking/reasoning 默认值。省略时后端按 policy
* 默认执行(向后兼容,旧前端不传不破坏)。三者通常一起传(见 `lib/seat-mode.ts`)。
*/
thinkingEnabled?: boolean;
reasoningEffort?: "minimal" | "low" | "medium" | "high";
/** 仅 special 用;ultra 档置真,给席位解锁 `task` 子代理工具。 */
subagentEnabled?: boolean;
skillStopNames?: string[];
/**
* 仅 leader 用。置 true 表示这是「派活后让总控判断继续/综合」的轮次:后端会把各已交付
@@ -444,6 +453,10 @@ export async function streamMultiAgent(
display_name: req.displayName,
new_message: req.newMessage,
model: req.model,
// 席位模式派生参数(仅 special);undefined 时不带上 → 后端按 policy 默认执行。
thinking_enabled: req.thinkingEnabled,
reasoning_effort: req.reasoningEffort,
subagent_enabled: req.subagentEnabled,
skill_stop_names: req.skillStopNames ?? [],
// 仅 leader 用;综合轮让后端注入各席位完整交付正文。省略 → 后端按 false 处理。
synthesis_mode: req.synthesisMode ?? false,
@@ -15,6 +15,7 @@ import { apiFetch } from "@/core/api/fetch-client";
import { getBackendBaseURL } from "@/core/config";
import { getAvatarTypeFor } from "../lib/roundtable-constants";
import { seatModeToRunParams } from "../lib/seat-mode";
import {
isTerminalJobStatus,
type DispatchChainNode,
@@ -75,6 +76,9 @@ function mapJob(raw: Record<string, unknown>): RoundtableJob {
}
function toStartBody(p: StartJobPayload) {
// 席位模式 → thinking/reasoning/subagent,与前端实时路径(/run/stream)同一份映射;
// 缺省(未选模式)时不带这三字段 → 后端按 policy 默认(≈pro),向后兼容。
const seat = p.seatMode ? seatModeToRunParams(p.seatMode) : null;
return {
draftId: p.draftId ?? null,
intent: p.intent,
@@ -86,6 +90,13 @@ function toStartBody(p: StartJobPayload) {
seedMessage: p.seedMessage,
threadIds: p.threadIds ?? null,
coordinatorName: p.coordinatorName ?? "roundtable-coordinator",
...(seat
? {
thinking_enabled: seat.thinking_enabled,
reasoning_effort: seat.reasoning_effort,
subagent_enabled: seat.subagent_enabled,
}
: {}),
};
}
@@ -13,7 +13,7 @@
*
* 详见 `docs/multi-agent-business-chain-dev.md` §6.2 / §7 阶段 3。
*/
import { useEffect, useMemo, useState, type ReactNode } from "react";
import { useEffect, useMemo, useRef, useState, type ReactNode } from "react";
import {
ArrowRightIcon,
CheckIcon,
@@ -44,12 +44,15 @@ export interface BusinessChainPickerProps {
onCancel: () => void;
/** 跳转到业务链条配置页(同时关闭弹窗)。 */
onGoConfig: () => void;
/** 选中链条变化时回调(含其 seatMode),父弹窗据此把执行模式默认设为链条的配置。 */
onSelectedChainChange?: (chain: BusinessChain | null) => void;
}
export function BusinessChainPicker({
onConfirm,
onCancel,
onGoConfig,
onSelectedChainChange,
}: BusinessChainPickerProps) {
const [chains, setChains] = useState<BusinessChain[]>([]);
const [loading, setLoading] = useState(true);
@@ -102,6 +105,15 @@ export function BusinessChainPicker({
() => chains.find((c) => c.id === selectedId) ?? null,
[chains, selectedId],
);
// 选中链条变化 → 通知父弹窗(用 ref 持有回调,只在 selectedChain 真正变化时触发,
// 避免父级每次 render 重置用户已手动改过的执行模式)。
const onSelectedChainChangeRef = useRef(onSelectedChainChange);
useEffect(() => {
onSelectedChainChangeRef.current = onSelectedChainChange;
}, [onSelectedChainChange]);
useEffect(() => {
onSelectedChainChangeRef.current?.(selectedChain);
}, [selectedChain]);
const survivingSeats = useMemo(
() => (selectedChain ? selectedChain.seats.filter((s) => isExisting(s.agent_id)) : []),
[selectedChain, isExisting],
@@ -52,6 +52,8 @@ export interface MessageStepsCardProps {
/** 工具调用阶段的 step 才进入折叠卡;非工具步骤(connecting/streaming/tool_result)归到「思考」段。 */
function isToolStep(step: BubbleStep): boolean {
// `task`(派生子智能体)单独渲染成任务卡片(SubtaskCard),不在步骤折叠卡里重复展示。
if (step.key === "tool_calling:task") return false;
return step.key.startsWith("tool_calling:") && step.key.length > "tool_calling:".length;
}
@@ -34,6 +34,11 @@ import { MessageStepsCard } from "./MessageStepsCard";
import { BusinessChainPicker } from "./BusinessChainPicker";
import type { IntentSummary } from "../api/intent";
import type { SeatSelectionResult } from "../lib/business-chain";
import {
DEFAULT_SEAT_MODE,
SEAT_MODE_OPTIONS,
type SeatMode,
} from "../lib/seat-mode";
import {
filterRecommendCandidates,
listRecommendHistory,
@@ -200,6 +205,9 @@ export function RecommendAgentsDialog({
const navigate = useNavigate();
// 席位来源模式:智能推荐 / 业务链条。默认**业务链条**。
const [mode, setMode] = useState<"recommend" | "chain">("chain");
// 席位执行模式(快速/思考/专业问答/多智能体问答)——推荐 / 业务链条两种来源通用,
// 决定第二步每个子智能体席位的推理深度。默认 pro(= 改造前席位固定行为)。
const [seatMode, setSeatMode] = useState<SeatMode>(DEFAULT_SEAT_MODE);
const [state, setState] = useState<InternalState>(INITIAL_STATE);
const [selectedIds, setSelectedIds] = useState<Set<string>>(new Set());
@@ -515,6 +523,14 @@ export function RecommendAgentsDialog({
return `已选 ${selectedCount} / ${MAX_SELECTED}`;
})();
// 统一在回传前合并席位模式,推荐 / 业务链条两条确认路径共用,保证 seatMode 总被带上。
const handleConfirmWithMode = useCallback(
(result: SeatSelectionResult) => {
onConfirm({ ...result, seatMode });
},
[onConfirm, seatMode],
);
const handleConfirm = useCallback(() => {
if (!canConfirm) return;
// Order: recommended-and-selected first (in recommender order), then
@@ -530,8 +546,8 @@ export function RecommendAgentsDialog({
for (const c of otherCards) {
if (selectedIds.has(c.agent_id)) ordered.push({ agent_id: c.agent_id, name: c.name });
}
onConfirm({ agents: ordered, mode: "recommend" });
}, [canConfirm, recommendedCards, otherCards, selectedIds, onConfirm]);
handleConfirmWithMode({ agents: ordered, mode: "recommend" });
}, [canConfirm, recommendedCards, otherCards, selectedIds, handleConfirmWithMode]);
const handleGoConfig = useCallback(() => {
onCancel();
@@ -595,6 +611,35 @@ export function RecommendAgentsDialog({
</button>
</div>
{/* 席位执行模式:快速 / 思考 / 专业问答 / 多智能体问答。推荐 & 业务链条通用,
决定第二步每个子智能体席位的推理深度(thinking/reasoning/subagent)。 */}
<div className="shrink-0 space-y-1.5">
<div className="flex flex-wrap items-center gap-2">
<span className="text-xs font-medium text-muted-foreground">执行模式</span>
<div className="inline-flex rounded-lg border border-border bg-muted/40 p-0.5 text-xs font-medium">
{SEAT_MODE_OPTIONS.map((opt) => (
<button
key={opt.value}
type="button"
onClick={() => setSeatMode(opt.value)}
title={opt.description}
className={cn(
"rounded-md px-2.5 py-1 transition-colors",
seatMode === opt.value
? "bg-card text-primary shadow-sm"
: "text-muted-foreground hover:text-foreground",
)}
>
{opt.label}
</button>
))}
</div>
</div>
<p className="text-[11px] leading-relaxed text-muted-foreground">
{SEAT_MODE_OPTIONS.find((o) => o.value === seatMode)?.description}
</p>
</div>
{/* ===== 智能推荐模式(现状,零行为变更)===== */}
{mode === "recommend" && (
<>
@@ -750,9 +795,13 @@ export function RecommendAgentsDialog({
{/* ===== 业务链条模式 ===== */}
{mode === "chain" && (
<BusinessChainPicker
onConfirm={onConfirm}
onConfirm={handleConfirmWithMode}
onCancel={onCancel}
onGoConfig={handleGoConfig}
onSelectedChainChange={(chain) => {
// 选中链条若配了默认执行模式 → 把顶部模式选择器默认设为它(用户仍可临时改)。
if (chain?.seatMode) setSeatMode(chain.seatMode);
}}
/>
)}
</DialogContent>
@@ -0,0 +1,125 @@
/**
* 圆桌「后台挂起作业」全工作区跨页完成通知器。
*
* 挂在 `PageLayout`(`/page/*` 的常驻层,导航切页不卸载)里:圆桌页点「后台挂起」后
* 调 `watchJob(jobId)` 注册监听,本 Provider 用 `streamJob` 订阅该作业 SSE;作业跑到
* 终态时弹一个**与「达成共识」同款**的顶部浮窗(bg-primary + Sparkles + bounce),
* 因此用户切到聊天/任务/其它工作区页面也能收到完成提示。
*
* 自带浮窗(不走 sonner)—— 各工作区页各自挂了一个 sonner `<Toaster>`,若这里也用
* sonner 会出现多 Toaster 重复弹;自渲染浮窗由本常驻 Provider 独占,天然不重复。
*
* 不含「刷新恢复」:watchedJobIds 是内存态,整页刷新后丢失(本期范围)。Provider 缺失
* 时(如 embed 路由在 PageRoutes 之外)`watchJob` 为 no-op,调用方不报错。
*/
import {
createContext,
useCallback,
useContext,
useEffect,
useRef,
useState,
type ReactNode,
} from "react";
import { Sparkles } from "lucide-react";
import { streamJob } from "@/roundtable-planning/api/roundtable-jobs";
import type { RoundtableJob } from "@/roundtable-planning/lib/job-types";
interface RoundtableJobNotifierValue {
/** 注册一个后台作业 id,作业完成(done/error)时弹全局提示。重复 id 自动去重。 */
watchJob: (jobId: string) => void;
}
const RoundtableJobNotifierContext = createContext<RoundtableJobNotifierValue>({
watchJob: () => {
/* 无 Provider(如 embed 路由)时 no-op,不报错 */
},
});
export function useRoundtableJobNotifier(): RoundtableJobNotifierValue {
return useContext(RoundtableJobNotifierContext);
}
/** 单个作业的 SSE 监听器(无 UI):到终态回调 onTerminal;传输中断回调 onStreamError。 */
function JobWatcher({
jobId,
onTerminal,
onStreamError,
}: {
jobId: string;
onTerminal: (job: RoundtableJob) => void;
onStreamError: (jobId: string) => void;
}) {
const onTerminalRef = useRef(onTerminal);
const onStreamErrorRef = useRef(onStreamError);
useEffect(() => {
onTerminalRef.current = onTerminal;
onStreamErrorRef.current = onStreamError;
}, [onTerminal, onStreamError]);
useEffect(() => {
const stop = streamJob(jobId, {
onProgress: () => {},
onDone: (job) => onTerminalRef.current(job),
onError: () => onStreamErrorRef.current(jobId),
});
return () => stop();
}, [jobId]);
return null;
}
/** 提示浮窗存活时长(ms)—— 比页内 toast(3.5s) 略长,给跨页用户更多时间看到。 */
const NOTICE_DURATION_MS = 6000;
export function RoundtableJobNotifier({ children }: { children: ReactNode }) {
const [watchedJobIds, setWatchedJobIds] = useState<string[]>([]);
const [notice, setNotice] = useState<string | null>(null);
const watchJob = useCallback((jobId: string) => {
setWatchedJobIds((prev) => (prev.includes(jobId) ? prev : [...prev, jobId]));
}, []);
const handleTerminal = useCallback((job: RoundtableJob) => {
setWatchedJobIds((prev) => prev.filter((id) => id !== job.id));
if (job.status === "done") {
setNotice("🎉 后台研讨已完成!可在「历史记录」查看结论与报告");
} else if (job.status === "error") {
setNotice("❌ 后台任务执行出错,请在「历史记录」中查看");
}
// cancelled:用户主动取消,不提示。
}, []);
const handleStreamError = useCallback((jobId: string) => {
// SSE 传输中断(非作业终态):静默停止监听,不误报完成。
setWatchedJobIds((prev) => prev.filter((id) => id !== jobId));
}, []);
useEffect(() => {
if (!notice) return;
const timer = setTimeout(() => setNotice(null), NOTICE_DURATION_MS);
return () => clearTimeout(timer);
}, [notice]);
return (
<RoundtableJobNotifierContext.Provider value={{ watchJob }}>
{children}
{watchedJobIds.map((id) => (
<JobWatcher
key={id}
jobId={id}
onTerminal={handleTerminal}
onStreamError={handleStreamError}
/>
))}
{notice && (
<div
className="fixed top-4 left-1/2 z-[100] flex -translate-x-1/2 transform items-center gap-2 rounded-xl border border-primary/40 bg-primary px-4 py-2.5 text-xs font-medium text-white shadow-lg animate-bounce"
role="status"
>
<Sparkles className="h-4 w-4 shrink-0" />
<span>{notice}</span>
</div>
)}
</RoundtableJobNotifierContext.Provider>
);
}
@@ -57,6 +57,8 @@ import {
StreamingCursor,
} from "@/roundtable-planning/components/MessageBubble";
import { MessageStepsCard } from "@/roundtable-planning/components/MessageStepsCard";
import { SubtaskCard } from "@/components/workspace/messages/subtask-card";
import { useSubtask } from "@/core/tasks/context";
import { ParallelSeatStrip } from "@/roundtable-planning/components/ParallelSeatStrip";
import { DagFlowMiniPanel } from "@/roundtable-planning/components/DagFlowMiniPanel";
import { useArtifacts } from "@/components/workspace/artifacts";
@@ -909,6 +911,15 @@ export function Step2Message({
resolveAgentName={resolveAgentName}
/>
)}
{/* 「多智能体问答」模式:本席位派生的子智能体任务卡片(复用主聊天 SubtaskCard,
从共享 SubtaskContext 取状态)。同时发起多个 task → 多张卡片纵向堆叠。 */}
{Array.isArray(msg.subtaskIds) && msg.subtaskIds.length > 0 && (
<div className="flex w-full flex-col gap-2">
{msg.subtaskIds.map((taskId) => (
<RoundtableSubtaskCard key={taskId} taskId={taskId} isLoading={isStreaming} />
))}
</div>
)}
{reasoning && (
<ReasoningBlock streaming={isStreaming}>{reasoning}</ReasoningBlock>
)}
@@ -932,6 +943,23 @@ export function Step2Message({
);
}
/**
* 子智能体任务卡片包裹层:子任务内容是**实时态**,存放在共享 SubtaskContext(主聊天同一份);
* 草稿回显 / 历史记录等场景气泡上虽留有 subtaskId,但 context 里已无对应 Subtask —— 此时
* 返回 null 不渲染(而非让 SubtaskCard 的 `useSubtask(id)!` 取到 undefined 崩溃)。
*/
function RoundtableSubtaskCard({
taskId,
isLoading,
}: {
taskId: string;
isLoading: boolean;
}) {
const task = useSubtask(taskId);
if (!task) return null;
return <SubtaskCard taskId={taskId} isLoading={isLoading} />;
}
/**
* 收集气泡里所有"写了文件"的产出路径,按出现顺序去重,聚合到一行文件卡片。
*
@@ -120,6 +120,11 @@ export interface UseDraftPersistenceReturn {
* 发起前 / 点查看前各存一次,保住当时的聊天记录不丢。返回 draft id。
*/
saveDraftNow: (furthestStep?: 1 | 2 | 3) => Promise<string | null>;
/**
* 用**显式传入的 step2 快照**强制存一次(不读 live 状态)。「后台挂起直接跳转」用:
* 跳转会清空 step2,调用方在跳转前冻结快照传进来,保证落库的是挂起那一刻的研讨记录。
*/
saveStep2SnapshotNow: (step2: DraftStep2Snapshot, furthestStep?: 1 | 2 | 3) => Promise<string | null>;
/** 强制刷新草稿列表。 */
refresh: () => Promise<void>;
}
@@ -287,6 +292,51 @@ export function useDraftPersistence(
[currentStep, enqueueSave, refreshList],
);
/**
* 用**显式传入的 step2 快照**强制存一次(不读 live 状态)。
*
* 「后台挂起改直接跳转」用:点挂起后页面立刻 resetToStep1 会清空 step2,而存稿是
* fire-and-forget 在跳转之后才真正落库——若仍走 `saveDraftNow`(其 `getStep2Snapshot`
* 是延迟读取)就会存到被重置后的空 step2。调用方在跳转前同步 `getStep2Snapshot()` 冻结
* 一份快照传进来,这里只认这份,确保存的是挂起那一刻的研讨记录。
*
* 有草稿 id → update(保护重命名,不带 title);无 id → 用当前 step1 + 传入 step2 建档。
*/
const saveStep2SnapshotNow = useCallback(
(step2: DraftStep2Snapshot, furthestStep: 1 | 2 | 3 = 2): Promise<string | null> => {
const run = async (): Promise<string | null> => {
const id = currentDraftIdRef.current;
if (id) {
const updated = await updateDraft(id, { furthestStep, step2 });
if (updated) {
currentDraftIdRef.current = updated.id;
return updated.id;
}
// 服务端已不存在 → 落回新建。
}
const step1 = getStep1Snapshot();
if (!hasSavableContent(step1, step2)) return null;
const created = await createDraft({
...(taskIdRef.current ? { taskId: taskIdRef.current } : {}),
title: deriveDraftName({
intentReady: step1.intentReady,
interactiveMessages: step1.interactiveMessages,
}),
furthestStep,
step1,
step2,
});
currentDraftIdRef.current = created.id;
setCurrentDraftId(created.id);
return created.id;
};
const next = saveQueueRef.current.then(run, run);
saveQueueRef.current = next.catch(() => undefined);
return next as Promise<string | null>;
},
[getStep1Snapshot, hasSavableContent, setCurrentDraftId],
);
const handleManualSave = useCallback(() => {
const furthest = (currentStep as 1 | 2 | 3) || 1;
void (async () => {
@@ -535,6 +585,7 @@ export function useDraftPersistence(
handleCancelRename,
ensureDraft,
saveDraftNow,
saveStep2SnapshotNow,
/** 强制刷新草稿列表(如后台挂起后让「进行中」状态立刻出现)。 */
refresh: refreshList,
};
@@ -27,6 +27,7 @@ import {
type ThreadIdMap,
} from "@/roundtable-planning/api/multi-agent";
import { uploadFiles } from "@/core/uploads/api";
import { useUpdateSubtask } from "@/core/tasks/context";
/** 干预消息携带的上传文件引用(与 streamMultiAgent 的 files 字段同形)。 */
type RunFileRef = NonNullable<StreamMultiAgentRequest["files"]>[number];
@@ -37,12 +38,56 @@ function takeOnce<T>(ref: { current: T | null }): T | undefined {
ref.current = null;
return v;
}
/**
* 从一帧里识别子智能体 custom 事件(task_started/running/completed/failed/cancelled/
* timed_out)。custom 流帧通常是裸 payload 对象 `{type:"task_*", task_id, ...}`;
* 开启 stream_subgraphs 时也可能被包成 `[namespace, payload]`,这里两种都兼容。
*/
function asSubagentTaskEvent(frame: unknown): {
type: string;
task_id: string;
message?: unknown;
result?: unknown;
error?: unknown;
description?: unknown;
} | null {
const cand =
frame && typeof frame === "object" && !Array.isArray(frame)
? (frame as Record<string, unknown>)
: Array.isArray(frame) && frame.length >= 2 && frame[1] && typeof frame[1] === "object"
? (frame[1] as Record<string, unknown>)
: null;
if (!cand) return null;
const type = typeof cand.type === "string" ? cand.type : "";
const taskId = typeof cand.task_id === "string" ? cand.task_id : "";
if (!type.startsWith("task_") || !taskId) return null;
return { type, task_id: taskId, message: cand.message, result: cand.result, error: cand.error, description: cand.description };
}
/** 从 LangGraph 消息 chunk 的 content(string 或结构化 block 数组)里抽纯文本。 */
function extractFrameText(content: unknown): string {
if (typeof content === "string") return content;
if (Array.isArray(content)) {
let out = "";
for (const item of content) {
if (typeof item === "string") out += item;
else if (item && typeof item === "object") {
const block = item as Record<string, unknown>;
if (typeof block.text === "string") out += block.text;
}
}
return out;
}
return "";
}
import {
buildDagStageLeaderTask,
pickDagSeatTask,
type OrchestrationPlan,
} from "@/roundtable-planning/lib/dag-types";
import type { ChainNodeState } from "@/roundtable-planning/lib/job-types";
import { seatModeToRequestFields, type SeatMode } from "@/roundtable-planning/lib/seat-mode";
import {
DEFAULT_SELECTED_AGENTS,
MAX_CYCLES,
@@ -65,6 +110,12 @@ export type Step2Dialogue = {
streaming?: boolean;
/** special 席位气泡对应的 LangGraph thread,用于沙箱预览。 */
agentThreadId?: string;
/**
* 「多智能体问答」模式下,本席位用 `task` 工具派生的子智能体任务 id 列表(=task 工具
* 调用的 tool_call_id)。渲染时据此从共享 SubtaskContext 取每个 Subtask 渲染任务卡片
* (复用主聊天 `SubtaskCard`,见 Step2Panel.Step2Message)。非 ultra 模式恒为空。
*/
subtaskIds?: string[];
steps?: Array<{
id: string;
key: string;
@@ -222,6 +273,12 @@ export interface UseStep2OrchestrationOptions {
currentStep: number;
/** 全局选中模型(顶栏「推理模型」下拉)。 */
selectedModel: string;
/**
* 席位执行模式(快速/思考/专业问答/多智能体问答)。决定第二步**每个子智能体席位**
* 的推理深度(thinking/reasoning_effort/subagent);总控派活不受影响。由「选择参会
* 智能体」弹窗确认时的 `SeatSelectionResult.seatMode` 决定。缺省 `pro`。
*/
selectedMode: SeatMode;
/**
* 编排模式:
* - `recommend`(默认)—— 总控自由派活循环(`runOrchestration`);
@@ -376,6 +433,7 @@ export function useStep2Orchestration(
const {
currentStep,
selectedModel,
selectedMode,
orchestrationMode,
orchestrationPlan,
dagExecMode = "fast",
@@ -507,6 +565,8 @@ export function useStep2Orchestration(
// runOrchestration 的 useCallback deps 大幅收窄,避免每次模型切换、agentConfigs
// 改写都重建一遍循环函数 —— 之前没收窄时 useEffect 会因 deps 变化重复 init。
const selectedModelRef = useRef<string>(selectedModel);
// 席位执行模式同样打到 ref,让席位调用在编排循环里读到最新值而不重建回调。
const selectedModeRef = useRef<SeatMode>(selectedMode);
const agentConfigsRef = useRef<Record<number, AgentConfig>>(agentConfigs);
const agentNameToRoleRef = useRef(agentNameToRole);
const threadIdsRef = useRef<ThreadIdMap | null>(threadIds);
@@ -543,6 +603,9 @@ export function useStep2Orchestration(
useEffect(() => {
selectedModelRef.current = selectedModel;
}, [selectedModel]);
useEffect(() => {
selectedModeRef.current = selectedMode;
}, [selectedMode]);
useEffect(() => {
agentConfigsRef.current = agentConfigs;
}, [agentConfigs]);
@@ -945,6 +1008,137 @@ export function useStep2Orchestration(
[upsertSeatToolCall],
);
// ── 「多智能体问答」子智能体任务卡片(参照主聊天 message-list 的 task 识别)──────────
// ultra 模式下席位用 `task` 工具派生子智能体。从席位流的 messages-tuple 帧里:
// (1) AI chunk 的 tool_calls 里 name==="task" → 建/更新一条 Subtask(in_progress);
// (2) type==="tool" 且 tool_call_id 命中已知 task id → 据「Task Succeeded./failed./
// timed out」前缀更新为 completed/failed(与主聊天 message-list.tsx 同源解析)。
// Subtask 存进共享 SubtaskContext(主聊天同一份),气泡只记 id 列表并渲染 <SubtaskCard>。
const updateSubtask = useUpdateSubtask();
// 已识别的 task tool_call_id —— 用于在收到 tool 结果帧时只认领「子智能体任务」结果,
// 不误把 write_file 等其它工具的结果当成子任务处理。
const knownTaskIdsRef = useRef<Set<string>>(new Set());
/** 把某个子任务 id 记到对应席位气泡上(新引用 → 触发气泡重渲染,卡片随之刷新状态)。 */
const touchBubbleSubtaskId = useCallback((bubbleId: number, taskId: string) => {
setStep2RoundtableDialogues((prev) =>
prev.map((d) => {
if (d.id !== bubbleId) return d;
const ids = d.subtaskIds ?? [];
// 每次都换新数组引用,确保状态(in_progress→completed/failed)变化时卡片重渲染。
return { ...d, subtaskIds: ids.includes(taskId) ? [...ids] : [...ids, taskId] };
}),
);
}, []);
/** 处理一条消息(values 快照里的整条 message,或 messages-tuple 的 chunk)的子任务语义。 */
const ingestSubagentMessage = useCallback(
(msg: Record<string, unknown>, bubbleId: number) => {
// (1) AI 消息:抓 task 工具调用 → 建/更新 in_progress 子任务。
const toolCalls = msg.tool_calls;
if (Array.isArray(toolCalls)) {
for (const tc of toolCalls) {
if (!tc || typeof tc !== "object") continue;
const t = tc as Record<string, unknown>;
if (t.name !== "task") continue;
const id = typeof t.id === "string" ? t.id : "";
if (!id) continue;
const a = (t.args && typeof t.args === "object" ? t.args : {}) as Record<string, unknown>;
updateSubtask({
id,
status: "in_progress",
subagent_type: typeof a.subagent_type === "string" ? a.subagent_type : "",
description: typeof a.description === "string" ? a.description : "",
prompt: typeof a.prompt === "string" ? a.prompt : "",
});
knownTaskIdsRef.current.add(id);
touchBubbleSubtaskId(bubbleId, id);
}
}
// (2) tool 结果消息:命中已知 task id 时按前缀判完成/失败(与主聊天解析一致)。
const type = typeof msg.type === "string" ? msg.type.toLowerCase() : "";
if (type === "tool") {
const taskId = typeof msg.tool_call_id === "string" ? msg.tool_call_id : "";
if (taskId && knownTaskIdsRef.current.has(taskId)) {
const result = extractFrameText(msg.content);
if (result.startsWith("Task Succeeded. Result:")) {
updateSubtask({
id: taskId,
status: "completed",
result: result.split("Task Succeeded. Result:")[1]?.trim(),
});
} else if (result.startsWith("Task failed.")) {
updateSubtask({
id: taskId,
status: "failed",
error: result.split("Task failed.")[1]?.trim(),
});
} else if (result.startsWith("Task timed out")) {
updateSubtask({ id: taskId, status: "failed", error: result });
}
touchBubbleSubtaskId(bubbleId, taskId);
}
}
},
[updateSubtask, touchBubbleSubtaskId],
);
const ingestSubagentTaskFrame = useCallback(
(frame: unknown, bubbleId: number) => {
// custom 事件帧(ultra 席位子智能体进度):task_running 带子智能体最新消息 → 回填
// latestMessage(卡片实时显示「正在调用 X 工具」);完成/失败即时定状态(比 values 更快)。
const ev = asSubagentTaskEvent(frame);
if (ev) {
knownTaskIdsRef.current.add(ev.task_id);
if (ev.type === "task_running") {
// latestMessage 非空才会触发 SubtaskContext setTasks 重渲染(见 useUpdateSubtask)。
updateSubtask({ id: ev.task_id, status: "in_progress", latestMessage: ev.message as never });
} else if (ev.type === "task_started") {
updateSubtask({
id: ev.task_id,
status: "in_progress",
description: typeof ev.description === "string" ? ev.description : undefined,
});
} else if (ev.type === "task_completed") {
updateSubtask({
id: ev.task_id,
status: "completed",
result: typeof ev.result === "string" ? ev.result : extractFrameText(ev.result),
});
} else if (
ev.type === "task_failed" ||
ev.type === "task_cancelled" ||
ev.type === "task_timed_out"
) {
updateSubtask({
id: ev.task_id,
status: "failed",
error: typeof ev.error === "string" ? ev.error : "子智能体任务失败",
});
}
touchBubbleSubtaskId(bubbleId, ev.task_id);
return;
}
// values 快照帧:{messages:[...]} —— 权威来源,按整条消息列表对账(子任务完成/失败)。
if (frame && typeof frame === "object" && !Array.isArray(frame)) {
const msgs = (frame as Record<string, unknown>).messages;
if (Array.isArray(msgs)) {
for (const m of msgs) {
if (m && typeof m === "object") ingestSubagentMessage(m as Record<string, unknown>, bubbleId);
}
}
return;
}
// messages-tuple 帧:[chunk, metadata] —— 增量,派活时即时建卡。
if (!Array.isArray(frame) || frame.length === 0) return;
const chunk = frame[0];
if (!chunk || typeof chunk !== "object") return;
ingestSubagentMessage(chunk as Record<string, unknown>, bubbleId);
},
[ingestSubagentMessage, updateSubtask, touchBubbleSubtaskId],
);
// ─── Dialogue / Speaker helpers ────────────────────────────────────
const appendDialogue = useCallback(
(card: {
@@ -1406,6 +1600,8 @@ export function useStep2Orchestration(
newMessage: fullTask,
peerDeliveries: priorDeliveries,
model: perSeatModel || selectedModelRef.current || undefined,
// 席位执行模式(快速/思考/专业问答/多智能体问答)→ thinking/reasoning/subagent。
...seatModeToRequestFields(selectedModeRef.current),
signal: subAborter.signal,
onTextDelta: (text) =>
a.isParallel ? ingestParallelDelta(a.bubbleId, text) : appendToDialogue(a.bubbleId, text),
@@ -1415,7 +1611,10 @@ export function useStep2Orchestration(
appendStepIfNew(a.bubbleId, label, `${phase}:${toolName ?? ""}`, args);
handleArtifactToolPhase(phase, toolName, args, a.seatThreadId);
},
onUpstreamFrame: (frame) => handleUpstreamFrame(frame, a.seatThreadId),
onUpstreamFrame: (frame) => {
handleUpstreamFrame(frame, a.seatThreadId);
ingestSubagentTaskFrame(frame, a.bubbleId);
},
});
activeAbortersRef.current.delete(subAborter);
if (isStale()) {
@@ -1690,6 +1889,7 @@ export function useStep2Orchestration(
triggerToast,
handleArtifactToolPhase,
handleUpstreamFrame,
ingestSubagentTaskFrame,
resolveSeatThreadId,
],
);
@@ -1829,6 +2029,8 @@ export function useStep2Orchestration(
displayName: seatName,
newMessage: task,
model: perSeatModel || selectedModelRef.current || undefined,
// 席位执行模式 → thinking/reasoning/subagent(与推荐/DAG 路径一致)。
...seatModeToRequestFields(selectedModeRef.current),
signal: subAborter.signal,
onTextDelta: (text) => appendToDialogue(subBubbleId, text),
onPhaseChange: (phase, toolName, args) => {
@@ -1837,7 +2039,10 @@ export function useStep2Orchestration(
appendStepIfNew(subBubbleId, label, `${phase}:${toolName ?? ""}`, args);
handleArtifactToolPhase(phase, toolName, args, seatThreadId);
},
onUpstreamFrame: (frame) => handleUpstreamFrame(frame, seatThreadId),
onUpstreamFrame: (frame) => {
handleUpstreamFrame(frame, seatThreadId);
ingestSubagentTaskFrame(frame, subBubbleId);
},
});
if (isStale()) {
finalizeDialogue(subBubbleId, specRes.content || "(已停止)", "已停止");
@@ -1936,6 +2141,7 @@ export function useStep2Orchestration(
triggerToast,
handleArtifactToolPhase,
handleUpstreamFrame,
ingestSubagentTaskFrame,
resolveSeatThreadId,
],
);
@@ -2163,6 +2369,8 @@ export function useStep2Orchestration(
// 前序各阶段的完整交付 → 供本席位 read_peer_delivery 按需取全文(§12.9)。
peerDeliveries: priorDeliveries,
model: perSeatModel || selectedModelRef.current || undefined,
// 席位执行模式 → thinking/reasoning/subagent(与推荐/链条路径一致)。
...seatModeToRequestFields(selectedModeRef.current),
signal: subAborter.signal,
// §6:并行席位走 ref 全量累积 + 节流 flush(只渲染选中的);单节点
// stage 实际串行,1 路流直接 setState 不卡,沿用 appendToDialogue。
@@ -2174,7 +2382,10 @@ export function useStep2Orchestration(
appendStepIfNew(bubbleId, label, `${phase}:${toolName ?? ""}`, args);
handleArtifactToolPhase(phase, toolName, args, seatThreadId);
},
onUpstreamFrame: (frame) => handleUpstreamFrame(frame, seatThreadId),
onUpstreamFrame: (frame) => {
handleUpstreamFrame(frame, seatThreadId);
ingestSubagentTaskFrame(frame, bubbleId);
},
});
if (isStale()) {
finalizeDialogue(bubbleId, specRes.content || "(已停止)", "已停止");
@@ -2276,6 +2487,7 @@ export function useStep2Orchestration(
triggerToast,
handleArtifactToolPhase,
handleUpstreamFrame,
ingestSubagentTaskFrame,
resolveSeatThreadId,
ingestParallelDelta,
resetParallelStreams,
@@ -10,6 +10,7 @@
*/
import { buildPlanFromStages } from "./dag-types";
import type { SeatMode } from "./seat-mode";
/** 链条里的一个有序席位。`seats[0]` 是第一个发言者。 */
export interface ChainSeat {
@@ -48,6 +49,11 @@ export interface BusinessChain {
* 空 / null = 不注入。
*/
coordinatorPrompt?: string | null;
/**
* 默认席位执行模式(快速/思考/专业问答/多智能体)。选入研讨时弹窗默认用它(仍可临时改)。
* 缺省 = 未设,调用方回退 `DEFAULT_SEAT_MODE`(pro)。
*/
seatMode?: SeatMode;
/** 是否已发布为公共(所有账号可在「公共」筛选查看 / 选入研讨)。 */
isPublic?: boolean;
/** 是否当前用户所有(仅本人可编辑 / 删除 / 发布;公共筛选里别人的链条为 false)。 */
@@ -85,6 +91,12 @@ export interface SeatSelectionResult {
* 缺省 `fast`。
*/
dagExecMode?: "fast" | "sandbox";
/**
* 席位执行模式(快速/思考/专业问答/多智能体问答)——决定第二步**每个子智能体席位**
* 的推理深度(thinking/reasoning_effort/subagent)。三种编排(recommend/chain/dag)通用。
* 缺省由调用方按 `DEFAULT_SEAT_MODE`(pro)处理。
*/
seatMode?: SeatMode;
}
/** 链条席位数硬约束,与圆桌推荐弹窗的 `MIN_SELECTED`/`MAX_SELECTED` 对齐。 */
@@ -132,12 +144,12 @@ export function buildChainSeatSelection(
const agents = plan.stages
.flatMap((st) => st.agentIds)
.map((id) => ({ agent_id: id, name: nameOf(id) }));
return { agents, mode: "dag", chain: chainMeta, plan, dagExecMode: "fast" };
return { agents, mode: "dag", chain: chainMeta, plan, dagExecMode: "fast", seatMode: chain.seatMode };
}
}
const agents = survivingSeats.map((s) => ({ agent_id: s.agent_id, name: nameOf(s.agent_id) }));
return { agents, mode: "chain", chain: chainMeta };
return { agents, mode: "chain", chain: chainMeta, seatMode: chain.seatMode };
}
/**
@@ -9,6 +9,7 @@
*/
import type { OrchestrationPlan } from "./dag-types";
import type { SeatMode } from "./seat-mode";
/** 作业状态机(§2.1)。 */
export type JobStatus =
@@ -120,6 +121,12 @@ export interface StartJobPayload {
/** 预建好的 thread 映射(Phase 3b 真网关用;模拟网关忽略)。 */
threadIds?: Record<string, string> | null;
coordinatorName?: string;
/**
* 席位执行模式(快速/思考/专业问答/多智能体)。后台续跑时席位按此模式执行(与前端实时
* 路径一致);缺省由后端按 policy 默认(≈pro)。在 `toStartBody` 里派生成 thinking/
* reasoning/subagent 三字段发给后端。
*/
seatMode?: SeatMode;
}
/** 终态集合——SSE/轮询判定作业是否结束。 */
@@ -0,0 +1,77 @@
/**
* 圆桌第二步「席位执行模式」共享定义。
*
* 与主聊天输入框(`components/workspace/input-box.tsx`)的四档模式同源:
* 用户在「选择参会智能体」弹窗里选一档,第二步派活的**每个子智能体(special 席位)**
* 都按这一档的推理深度执行。模式只决定 `thinking_enabled / reasoning_effort /
* subagent_enabled` 三个运行参数;席位的工具裁剪仍由后端 `roundtable_run_policy`
* 决定(见 `app/gateway/roundtable_run_policy.py`)。
*
* 总控(leader)派活是短决策任务,**不**受本模式影响(沿用其快速派活策略)。
*
* 映射逻辑刻意对齐主聊天 `core/threads/hooks.ts` 的 mode → context 转换,保持一致体验:
* - flash 快速 —— 不思考、最快
* - thinking 思考 —— 思考 + low
* - pro 专业问答 —— 思考 + medium(= 现有席位默认行为,向后兼容)
* - ultra 多智能体 —— 思考 + high + 席位可派子代理(task 工具)
*/
export type SeatMode = "flash" | "thinking" | "pro" | "ultra";
/** 默认档位:专业问答(pro)—— 与改造前席位的固定行为(thinking + medium)一致。 */
export const DEFAULT_SEAT_MODE: SeatMode = "pro";
/** 席位单轮 run 的派生参数(透传给 `/api/multi-agent/run/stream`,覆盖后端 policy 默认)。 */
export interface SeatRunParams {
thinking_enabled: boolean;
reasoning_effort: "minimal" | "low" | "medium" | "high";
subagent_enabled: boolean;
}
/** 把一个席位模式映射成具体的运行参数。 */
export function seatModeToRunParams(mode: SeatMode): SeatRunParams {
switch (mode) {
case "flash":
return { thinking_enabled: false, reasoning_effort: "minimal", subagent_enabled: false };
case "thinking":
return { thinking_enabled: true, reasoning_effort: "low", subagent_enabled: false };
case "ultra":
return { thinking_enabled: true, reasoning_effort: "high", subagent_enabled: true };
case "pro":
default:
return { thinking_enabled: true, reasoning_effort: "medium", subagent_enabled: false };
}
}
/**
* 把席位模式映射成 `streamMultiAgent` 请求字段(camelCase),方便在席位调用处直接 spread。
* 与 `seatModeToRunParams` 同一份逻辑,只是改成 API 客户端期望的字段名。
*/
export function seatModeToRequestFields(mode: SeatMode): {
thinkingEnabled: boolean;
reasoningEffort: SeatRunParams["reasoning_effort"];
subagentEnabled: boolean;
} {
const p = seatModeToRunParams(mode);
return {
thinkingEnabled: p.thinking_enabled,
reasoningEffort: p.reasoning_effort,
subagentEnabled: p.subagent_enabled,
};
}
/** 弹窗模式选择器用的展示元数据(标签 + 一句话说明),顺序即展示顺序。 */
export const SEAT_MODE_OPTIONS: ReadonlyArray<{
value: SeatMode;
label: string;
description: string;
}> = [
{ value: "flash", label: "快速", description: "快速且高效地完成任务,但可能不够精准" },
{ value: "thinking", label: "思考", description: "思考后再行动,在时间与准确性之间取得平衡" },
{ value: "pro", label: "专业问答", description: "思考、计划再执行,获得更精准的结果,可能需要更多时间" },
{
value: "ultra",
label: "多智能体问答",
description: "在专业问答基础上,每个席位可调用子代理分工协作,能力最强、更慢更耗时",
},
];
@@ -85,6 +85,11 @@ import {
type BusinessChain,
type ChainSeat,
} from "@/roundtable-planning/lib/business-chain";
import {
DEFAULT_SEAT_MODE,
SEAT_MODE_OPTIONS,
type SeatMode,
} from "@/roundtable-planning/lib/seat-mode";
import {
addEmptyStage,
addSeatAsNewStage,
@@ -153,6 +158,8 @@ export default function BusinessChainEditorPage() {
const [description, setDescription] = useState("");
// 写给总控的「编排提示」:告诉总控该怎么派活、做什么(仅分层 DAG 模式注入)。
const [coordinatorPrompt, setCoordinatorPrompt] = useState("");
// 默认席位执行模式(快速/思考/专业问答/多智能体)—— 选入研讨时弹窗默认用它。
const [seatMode, setSeatMode] = useState<SeatMode>(DEFAULT_SEAT_MODE);
// 分层编排是唯一数据源;落库时展平为 seats + 折叠为 stages。
const [stageGroups, setStageGroups] = useState<StageGroups>([]);
const [saving, setSaving] = useState(false);
@@ -277,6 +284,7 @@ export default function BusinessChainEditorPage() {
setTitle(chain.title);
setDescription(chain.description ?? "");
setCoordinatorPrompt(chain.coordinatorPrompt ?? "");
setSeatMode(chain.seatMode ?? DEFAULT_SEAT_MODE);
setStageGroups(chainToStageGroups(chain));
setStep(1);
}, []);
@@ -288,6 +296,7 @@ export default function BusinessChainEditorPage() {
setTitle("");
setDescription("");
setCoordinatorPrompt("");
setSeatMode(DEFAULT_SEAT_MODE);
setStageGroups([]);
setStep(1);
}, []);
@@ -560,6 +569,7 @@ export default function BusinessChainEditorPage() {
stages,
stageGoals,
coordinatorPrompt: coordPrompt,
seatMode,
});
if (!updated) {
toast.error("链条已不存在,可能被删除");
@@ -578,6 +588,7 @@ export default function BusinessChainEditorPage() {
stages,
stageGoals,
coordinatorPrompt: coordPrompt,
seatMode,
});
setCurrentChainId(created.id);
toast.success("已创建业务链条");
@@ -592,6 +603,7 @@ export default function BusinessChainEditorPage() {
title,
description,
coordinatorPrompt,
seatMode,
stageGroups,
seatCountInvalid,
currentChainId,
@@ -699,6 +711,37 @@ export default function BusinessChainEditorPage() {
/>
</div>
{/* 默认执行模式:选入研讨时弹窗默认用它(仍可临时改)。 */}
<div className="space-y-2">
<label className="flex items-center gap-1 text-xs font-semibold text-foreground">
默认执行模式
<span className="text-[10px] font-normal text-muted-foreground">
(选入研讨时默认用它,进研讨前仍可临时切换)
</span>
</label>
<div className="inline-flex flex-wrap gap-0.5 rounded-lg border border-border bg-muted/40 p-0.5 text-xs font-medium">
{SEAT_MODE_OPTIONS.map((opt) => (
<button
key={opt.value}
type="button"
onClick={() => setSeatMode(opt.value)}
title={opt.description}
className={cn(
"rounded-md px-2.5 py-1 transition-colors",
seatMode === opt.value
? "bg-card text-primary shadow-sm"
: "text-muted-foreground hover:text-foreground",
)}
>
{opt.label}
</button>
))}
</div>
<p className="text-[11px] leading-relaxed text-muted-foreground">
{SEAT_MODE_OPTIONS.find((o) => o.value === seatMode)?.description}
</p>
</div>
{/* 总控提示词:label + AI 优化(配色沿用现有 primary)+ 大号等宽编辑区 */}
<div className="space-y-2">
<div className="flex items-center justify-between gap-2">
@@ -100,6 +100,7 @@ import {
buildChainSeatSelection,
type SeatSelectionResult,
} from "@/roundtable-planning/lib/business-chain";
import { DEFAULT_SEAT_MODE, type SeatMode } from "@/roundtable-planning/lib/seat-mode";
import type { OrchestrationPlan } from "@/roundtable-planning/lib/dag-types";
import type { IntentSummary } from "@/roundtable-planning/api/intent";
import {
@@ -108,6 +109,7 @@ import {
startJob,
startJobKeepalive,
} from "@/roundtable-planning/api/roundtable-jobs";
import { useRoundtableJobNotifier } from "@/roundtable-planning/components/RoundtableJobNotifier";
import { getChain } from "@/roundtable-planning/api/chains";
import { filterRecommendCandidates } from "@/roundtable-planning/api/recommend";
import { listAgents } from "@/core/agents";
@@ -169,6 +171,8 @@ function readEmbedHashQueryPrefill(): RoundtablePrefill | null {
export default function RoundtablePlanningPage() {
const { theme } = useTheme();
// 全工作区跨页的后台作业完成通知器(Provider 挂在 PageLayout 常驻层)。
const { watchJob } = useRoundtableJobNotifier();
const isDark = theme === "dark";
const navigate = useNavigate();
const location = useLocation();
@@ -222,6 +226,9 @@ export default function RoundtablePlanningPage() {
// 「导入会商」弹窗(输入分享码把别人分享的会商复制到本账号)。
const [isImportOpen, setIsImportOpen] = useState<boolean>(false);
const [selectedModel, setSelectedModel] = useState<string>("");
// Step 2 席位执行模式(快速/思考/专业问答/多智能体问答)——决定每个子智能体席位的
// 推理深度。由「选择参会智能体」弹窗确认时的 SeatSelectionResult.seatMode 决定。
const [seatMode, setSeatMode] = useState<SeatMode>(DEFAULT_SEAT_MODE);
// Step 2 编排模式:recommend = 总控自由派活;chain = 总控按业务链条顺序派活;
// dag = 分层 DAG(stage 内并行、stage 间串行,配合 orchestrationPlan)。
// 由推荐弹窗确认时的 SeatSelectionResult.mode 决定(见 handleConfirmAgents)。
@@ -337,6 +344,7 @@ export default function RoundtablePlanningPage() {
const step2 = useStep2Orchestration({
currentStep,
selectedModel,
selectedMode: seatMode,
orchestrationMode,
orchestrationPlan,
dagExecMode,
@@ -750,6 +758,8 @@ export default function RoundtablePlanningPage() {
// result.mode 决定 Step 2 编排:recommend = 总控自由派活;chain = 按链条顺序派活;
// dag = 分层并行(来源链条带 stages,携带 result.plan)。
setOrchestrationMode(result.mode);
// 席位执行模式:弹窗未带(旧调用)时回退默认 pro,保证行为可预期。
setSeatMode(result.seatMode ?? DEFAULT_SEAT_MODE);
setChainMeta(result.mode === "recommend" ? null : result.chain ?? null);
setOrchestrationPlan(result.mode === "dag" ? result.plan ?? null : null);
setDagExecMode(result.mode === "dag" ? result.dagExecMode ?? "fast" : "fast");
@@ -1133,12 +1143,34 @@ export default function RoundtablePlanningPage() {
});
}, [resetToStep1]);
// 「挂起中」本地标记:点「后台挂起」后立刻跳转,建作业是 fire-and-forget——从点击到作业
// 真正建好的窗口里,用它让历史记录显示「挂起中」(作业建好后转 queued/running,由后端状态接管)。
// resetToStep1 不卸载本页,故该 state 跨「跳转」存活。
const [pendingSuspendIds, setPendingSuspendIds] = useState<Set<string>>(() => new Set());
const markPendingSuspend = useCallback((draftId: string) => {
setPendingSuspendIds((prev) => {
const next = new Set(prev);
next.add(draftId);
return next;
});
}, []);
const clearPendingSuspend = useCallback((draftId: string) => {
setPendingSuspendIds((prev) => {
if (!prev.has(draftId)) return prev;
const next = new Set(prev);
next.delete(draftId);
return next;
});
}, []);
// 「后台挂起」核心(手动按钮 / 离开页面自动挂起共用):幂等保证草稿存在 →
// 同草稿已有在跑作业则跳过 → 启动后台作业。后端另有「同草稿未终态作业幂等返回」
// 兜底(覆盖跨标签页 / beforeunload 与本函数并发触发),绝不重复建作业。
const startBackgroundJob = useCallback(async (): Promise<boolean> => {
// 返回本次后台作业的 jobId(已在跑则返回既有作业 id),供「完成时提示」监听;
// 未发起(无席位)返回 null。
const startBackgroundJob = useCallback(async (): Promise<string | null> => {
const agents = step2.selectedAgents.map((a) => ({ agentId: a.agent_id, name: a.name }));
if (agents.length === 0) return false;
if (agents.length === 0) return null;
// 后台挂起发起前**立刻强制存一次**当前聊天(保住发起时刻的本地研讨记录,防丢)。
const draftId = await drafts.saveDraftNow(2);
// 防重复:该草稿已有在跑作业就不再新建(避免重复作业 / 前后端双跑)。
@@ -1148,52 +1180,118 @@ export default function RoundtablePlanningPage() {
(existing.status === "running" ||
existing.status === "awaiting_input" ||
existing.status === "queued");
if (!alreadyRunning) {
await startJob({
draftId: draftId ?? undefined,
intent: buildInitialIntent(),
agents,
model: selectedModel || undefined, // 模型沿用第二步选择
// 后台作业引擎支持 dag(后端 _run_dag:stage 串行 / stage 内并行)——
// 直接把当前编排模式透传,业务链的分层 / 并行 / 阶段目标 / 总控提示词全部保留。
orchestrationMode,
chain: chainMeta,
// dag 模式:把整份分层计划带给后端(含 stages / 阶段目标 / 总控提示词 / finalSynthesis)。
// 非 dag 时为 null(后端按 mode 走 chain/recommend)。
orchestrationPlan: orchestrationMode === "dag" ? orchestrationPlan : null,
// 不传 threadIds:后台作业由真网关自建全新线程,绝不复用前端本地研讨的线程
// (那些线程可能有活跃 run,复用会撞 409)。
});
if (alreadyRunning) {
return existing?.id ?? null;
}
return true;
const job = await startJob({
draftId: draftId ?? undefined,
intent: buildInitialIntent(),
agents,
model: selectedModel || undefined, // 模型沿用第二步选择
seatMode, // 席位执行模式沿用第二步选择(后台续跑同样按此模式跑席位)
// 后台作业引擎支持 dag(后端 _run_dag:stage 串行 / stage 内并行)——
// 直接把当前编排模式透传,业务链的分层 / 并行 / 阶段目标 / 总控提示词全部保留。
orchestrationMode,
chain: chainMeta,
// dag 模式:把整份分层计划带给后端(含 stages / 阶段目标 / 总控提示词 / finalSynthesis)。
// 非 dag 时为 null(后端按 mode 走 chain/recommend)。
orchestrationPlan: orchestrationMode === "dag" ? orchestrationPlan : null,
// 不传 threadIds:后台作业由真网关自建全新线程,绝不复用前端本地研讨的线程
// (那些线程可能有活跃 run,复用会撞 409)。
});
return job.id;
}, [
step2,
drafts,
selectedModel,
seatMode,
orchestrationMode,
orchestrationPlan,
chainMeta,
buildInitialIntent,
]);
// 「后台挂起」按钮:把当前研讨交给后端自动跑完 Step2 研讨 + Step3(总结报告 + 结果绘制)。
// 挂起后页面像「新建任务」一样重置回第一步(后台作业继续在历史记录里跑)。
const handleBackgroundSuspend = useCallback(async () => {
if (step2.selectedAgents.length === 0) {
// 「后台挂起」按钮(2026-06-16 改「直接跳转」):点击后**立刻**跳回第一步并提示,挂起动作
// (存稿 → 去重 → 建后台作业)改为 fire-and-forget 在后台完成,不再等接口返回才跳转。
// 历史记录在作业真正建好前显示「挂起中」(pendingSuspendIds),建好后由后端状态(queued→
// running)接管。⚠️ resetToStep1 会清空 step2 等 live 状态,故挂起入参 + step2 快照必须在
// 跳转前同步冻结(getStep2SnapshotWithRuns 返回的是不可变快照对象,reset 替换 state 不影响它)。
const handleBackgroundSuspend = useCallback(() => {
const selected = step2.selectedAgents;
if (selected.length === 0) {
triggerToast("⚠️ 请先选择研讨席位再后台挂起");
return;
}
try {
await startBackgroundJob();
} catch (err) {
triggerToast(`❌ 后台挂起失败:${(err as Error).message}`);
return;
}
// 重置回第一步(同新建任务);后台作业在历史记录里继续跑、可点开看进度。
// —— 跳转前同步冻结所有需要的输入(reset 之后这些 live 状态都会被清空)——
const step2Snapshot = getStep2SnapshotWithRuns();
const agents = selected.map((a) => ({ agentId: a.agent_id, name: a.name }));
const intent = buildInitialIntent();
const model = selectedModel || undefined;
const seatModeSnap = seatMode;
const mode = orchestrationMode;
const chainSnap = chainMeta;
const planSnap = mode === "dag" ? orchestrationPlan : null;
const draftId = drafts.currentDraftId;
// 立刻:标记「挂起中」+ 跳回第一步 + 提示(不等接口)。
if (draftId) markPendingSuspend(draftId);
resetToStep1();
triggerToast("🚀 已转入后台,任务会自动跑完研讨、总结报告与结果绘制;可在「历史记录」查看进度");
void drafts.refresh();
}, [step2, drafts, startBackgroundJob, resetToStep1, triggerToast]);
// 后台 fire-and-forget:用冻结快照存稿 → 去重 → 建作业 → 注册完成通知。失败只提示、不挡跳转。
void (async () => {
let watchId: string | null = null;
try {
const savedId = (await drafts.saveStep2SnapshotNow(step2Snapshot, 2)) ?? draftId ?? null;
// 防重复:该草稿已有未终态作业就不再新建(同 startBackgroundJob 的客户端去重,
// 后端另有「同草稿未终态作业幂等返回」兜底)。
const existing = savedId ? await getJobByDraft(savedId).catch(() => null) : null;
const alreadyRunning =
existing &&
(existing.status === "running" ||
existing.status === "awaiting_input" ||
existing.status === "queued");
if (alreadyRunning) {
watchId = existing?.id ?? null;
} else {
const job = await startJob({
draftId: savedId ?? undefined,
intent,
agents,
model,
seatMode: seatModeSnap,
orchestrationMode: mode,
chain: chainSnap,
orchestrationPlan: planSnap,
});
watchId = job.id;
}
} catch (err) {
triggerToast(`❌ 后台挂起失败:${(err as Error).message}`);
} finally {
// 注册到全工作区通知器,作业完成时**跨页**弹「达成共识同款」提示。
if (watchId) watchJob(watchId);
// 清「挂起中」本地标记:此后历史记录的状态由后端作业(queued→running)权威接管。
if (draftId) clearPendingSuspend(draftId);
void drafts.refresh();
}
})();
}, [
step2,
drafts,
getStep2SnapshotWithRuns,
buildInitialIntent,
selectedModel,
seatMode,
orchestrationMode,
orchestrationPlan,
chainMeta,
resetToStep1,
triggerToast,
watchJob,
markPendingSuspend,
clearPendingSuspend,
]);
// ── 自动后台挂起:研讨进行中离开本页面(路由卸载)或刷新 / 关标签页时,任务自动转后台 ──
// 最新状态进 ref:卸载 cleanup 与 beforeunload 都是「拍快照时刻」的同步回调,
@@ -1223,6 +1321,7 @@ export default function RoundtablePlanningPage() {
intent: buildInitialIntent(),
agents,
model: selectedModel || undefined,
seatMode,
orchestrationMode,
chain: chainMeta,
orchestrationPlan: orchestrationMode === "dag" ? orchestrationPlan : null,
@@ -1347,7 +1446,7 @@ export default function RoundtablePlanningPage() {
)}
{/* HEADER */}
<header className="relative z-[2000] flex shrink-0 items-center justify-between gap-4 border-b border-slate-100 bg-white px-6 py-3 shadow-[0_1px_3px_rgba(0,0,0,0.02)] dark:border-border dark:bg-card dark:shadow-none">
<header className="relative z-[40] flex shrink-0 items-center justify-between gap-4 border-b border-slate-100 bg-white px-6 py-3 shadow-[0_1px_3px_rgba(0,0,0,0.02)] dark:border-border dark:bg-card dark:shadow-none">
<StepNavigation
currentStep={currentStep}
intentReady={step1.intentReady}
@@ -1408,6 +1507,7 @@ export default function RoundtablePlanningPage() {
setIsOpen={drafts.setIsHistoryOpen}
drafts={drafts.drafts}
currentDraftId={drafts.currentDraftId}
pendingSuspendIds={pendingSuspendIds}
renamingDraftId={drafts.renamingDraftId}
renameDraftValue={drafts.renameDraftValue}
setRenameDraftValue={drafts.setRenameDraftValue}
@@ -1733,6 +1833,8 @@ interface HistoryDropdownProps {
setIsOpen: React.Dispatch<React.SetStateAction<boolean>>;
drafts: ReturnType<typeof useDraftPersistence>["drafts"];
currentDraftId: string | null;
/** 正在「挂起中」(已点后台挂起、作业尚未建好)的草稿 id 集合 —— 历史记录显示「挂起中」。 */
pendingSuspendIds: Set<string>;
renamingDraftId: string | null;
renameDraftValue: string;
setRenameDraftValue: React.Dispatch<React.SetStateAction<string>>;
@@ -1755,6 +1857,7 @@ function HistoryDropdown(props: HistoryDropdownProps) {
setIsOpen,
drafts,
currentDraftId,
pendingSuspendIds,
renamingDraftId,
renameDraftValue,
setRenameDraftValue,
@@ -1825,6 +1928,8 @@ function HistoryDropdown(props: HistoryDropdownProps) {
{drafts.map((d) => {
const isCurrent = d.id === currentDraftId;
const isRenaming = d.id === renamingDraftId;
// 「挂起中」:已点后台挂起、作业还没建好(本地标记),或作业刚建好仍排队(queued)。
const isSuspending = pendingSuspendIds.has(d.id) || d.status === "queued";
// 「进行中 / 待澄清」且有 jobId → 点击打开流程图弹窗,而非加载草稿。
const isActiveJob =
(d.status === "running" || d.status === "awaiting_input") && !!d.jobId;
@@ -1862,7 +1967,12 @@ function HistoryDropdown(props: HistoryDropdownProps) {
{d.name}
</span>
<span className="text-[10px] text-slate-400 mt-0.5 flex items-center gap-1 font-mono">
{d.status === "running" ? (
{isSuspending ? (
<span className="inline-flex items-center gap-1 text-sky-500 font-semibold">
<Loader2 className="w-3 h-3 animate-spin" />
挂起中
</span>
) : d.status === "running" ? (
<span className="inline-flex items-center gap-1 text-primary font-semibold">
<Loader2 className="w-3 h-3 animate-spin" />
进行中
+13 -12
View File
@@ -8,7 +8,10 @@ import { NotificationDialog } from "@/components/workspace/notification-dialog";
import { getStoredAuth } from "@/core/auth";
import { isWujieEnv } from "@/core/embed/wujie";
import { useUnreadCount } from "@/core/notifications";
import type { MenuItem } from "@/core/page-layout/sidebar-menu";
import {
getActiveThirdLevelItems,
type MenuItem,
} from "@/core/page-layout/sidebar-menu";
import { useSidebarMenuItems } from "@/core/page-layout/use-sidebar-menu";
import { cn } from "@/lib/utils";
@@ -145,7 +148,7 @@ const Header = ({ collapsed }: HeaderProps) => {
const showLightTopNav = !isDark && !isWujieEnv();
const hasActiveChild = (item: MenuItem): boolean => {
if (item.path && pathname === item.path) {
if (item.path && (pathname === item.path || pathname.startsWith(item.path + "/"))) {
return true;
}
@@ -170,9 +173,7 @@ const Header = ({ collapsed }: HeaderProps) => {
return null;
};
const activeTopItem = visibleMenuItems.find(hasActiveChild);
const activeSecondItem = activeTopItem?.children?.find(hasActiveChild);
const thirdSubNavItems = activeSecondItem?.children ?? [];
const thirdSubNavItems = getActiveThirdLevelItems(visibleMenuItems, pathname);
const leftMenuItems = visibleMenuItems.filter((i) => LEFT_MENU_IDS.includes(i.id));
const rightMenuItems = visibleMenuItems.filter((i) => RIGHT_MENU_IDS.includes(i.id));
@@ -185,7 +186,7 @@ const Header = ({ collapsed }: HeaderProps) => {
};
const renderTopItem = (item: MenuItem) => {
const isSelected = item.path ? pathname === item.path : hasActiveChild(item);
const isSelected = item.path ? (pathname === item.path || pathname.startsWith(item.path + "/")) : hasActiveChild(item);
return (
<button
key={item.id}
@@ -199,7 +200,7 @@ const Header = ({ collapsed }: HeaderProps) => {
};
const renderLightTopItem = (item: MenuItem) => {
const isSelected = item.path ? pathname === item.path : hasActiveChild(item);
const isSelected = item.path ? (pathname === item.path || pathname.startsWith(item.path + "/")) : hasActiveChild(item);
return (
<button
key={item.id}
@@ -294,7 +295,7 @@ const Header = ({ collapsed }: HeaderProps) => {
open={notificationOpen}
onOpenChange={setNotificationOpen}
/>
<header className="strategy-app-header theme-transition relative z-[2101] flex h-16 items-center justify-between px-4">
<header className="strategy-app-header theme-transition relative z-[45] flex h-16 items-center justify-between px-4">
{showTopNav ? (
<>
{/* 左分区(flex-1):菜单靠右贴近中间基座 */}
@@ -310,7 +311,7 @@ const Header = ({ collapsed }: HeaderProps) => {
</div>
{/* 中:CM助手(落在背景图发光基座上,纯文字无图标) */}
<div className="header-center-title">CM助手</div>
<div className="header-center-title">CM业务中心</div>
{/* 右分区(flex-1):菜单靠左贴近中间基座,操作区靠最右 */}
<div className="flex h-full min-w-0 flex-1 items-center justify-between">
@@ -336,7 +337,7 @@ const Header = ({ collapsed }: HeaderProps) => {
<FileText className="h-4 w-4 text-white" />
</div>
<span className="whitespace-nowrap text-base font-semibold text-white">
CM助手
CM业务中心
</span>
</div>
<nav className="light-topnav" aria-label="一级菜单">
@@ -363,7 +364,7 @@ const Header = ({ collapsed }: HeaderProps) => {
</div>
{!collapsed && (
<span className="text-sm font-semibold text-white">
CM助手
CM业务中心
</span>
)}
</div>
@@ -379,7 +380,7 @@ const Header = ({ collapsed }: HeaderProps) => {
<nav className="third-submenu-bar" aria-label="三级菜单">
<div className="third-submenu-list">
{thirdSubNavItems.map((item) => {
const isSelected = item.path ? pathname === item.path : hasActiveChild(item);
const isSelected = item.path ? (pathname === item.path || pathname.startsWith(item.path + "/")) : hasActiveChild(item);
return (
<button
+4 -1
View File
@@ -950,7 +950,10 @@ html.dark .strategy-app-header {
display: flex;
align-items: center;
justify-content: center;
padding: 0 155px;
/* 左右内边距随视口宽度等比缩放,与背景图基座的 background-size:cover 同步,
使两侧菜单在 1080p / 2K / 4K 下都贴齐发光基座边缘。
基准:1920 宽时 155px → 155/1920×100 ≈ 8.07vw。 */
padding: 0 8.07vw;
/* 文字略微上移,贴合基座视觉中心 */
transform: translateY(-6px);
font-family: "Source Han Sans SC", "Noto Sans SC", "Noto Sans S Chinese",
+1 -1
View File
@@ -265,7 +265,7 @@ The AI-writing graph (`packages/harness/deerflow/agents/ai_writing/`) has four s
Seeding mirrors the roundtable pattern: verified `config.yaml` + `SOUL.md` ship in `app/gateway/routers/_ai_writing_seed_assets/<id>/`; `_ai_writing_seed.py::ensure_ai_writing_functional_agents()` copies them to `.deer-flow/agents/` when missing (never overwrites local edits), called from the startup lifespan (before `_sync_legacy_agents`, so the agents appear in `/api/agents` on first boot) and self-heals from `GET /api/ai-writing/sessions` + `GET /api/ai-writing/article-types`. Tests: `tests/test_ai_writing_seed.py`, `tests/test_ai_writing_agent_runner.py`.
**Skill-driven retrieval (检索类型).** `material_source` has exactly two values: `general` (通用检索, default) and `notebook` (我的空间); legacy `knowledge_base`/`skill` are normalized to `general` by the researcher node. General retrieval is orchestrated from the **skills configured on the `ai-writing-researcher` agent** (`config.yaml → skills`, editable from the agent management page): the built-in `knowledge-base-search` skill (知识库检索) runs the native keyword fan-out web/intranet search (the old default path), while any other configured skill searches with a `description`-targeted query; skills run in configured order with early-stop at `max_materials`, and an empty/unresolvable skills config falls back to the built-in knowledge search so retrieval always works. The skill itself is seeded from `_ai_writing_seed_assets/skills/knowledge-base-search/SKILL.md` into `skills/public/` (never overwrites), and a pre-existing researcher `config.yaml` lacking the `skills` key is patched in place with the default (explicit `skills: []` is respected). **PAUSE-1 input append flow**: when the user types natural language into the material-confirm input (`re_search`/legacy `enable_skill_search` + `user_query`), the node matches the most relevant configured skill via `SkillPicker(candidates=…)` (skipped when only one skill is configured), searches with the user's words, and **appends** the new materials to the existing package (URL-dedup, original order preserved, no LLM re-rank/truncation, 100-item hard cap). Tests: `tests/test_ai_writing_material_source.py`.
**Skill-driven retrieval (检索类型).** `material_source` has exactly two values: `general` (通用检索, default) and `notebook` (我的空间); legacy `knowledge_base`/`skill` are normalized to `general` by the researcher node. General retrieval is orchestrated from the **skills configured on the `ai-writing-researcher` agent** (`config.yaml → skills`, editable from the agent management page): the built-in `knowledge-base-search` skill (知识库检索) runs the native keyword fan-out web/intranet search (the old default path), while any other configured skill searches with a `description`-targeted query; skills run in configured order with early-stop at `max_materials`, and an empty/unresolvable skills config falls back to the built-in knowledge search so retrieval always works. The skill itself is seeded from `_ai_writing_seed_assets/skills/knowledge-base-search/SKILL.md` into `skills/public/` (never overwrites), and a pre-existing researcher `config.yaml` lacking the `skills` key is patched in place with the default (explicit `skills: []` is respected). **PAUSE-1 input append flow**: when the user types natural language into the material-confirm input (`re_search`/legacy `enable_skill_search` + `user_query`), the node matches the most relevant configured skill via `SkillPicker(candidates=…)` (skipped when only one skill is configured), searches with the user's words, and **appends** the new materials to the existing package (URL-dedup, original order preserved, no LLM re-rank/truncation, 100-item hard cap). **Skill-failure fault tolerance**: a configured skill that times out / raises / has all its keyword searches fail (vs. legitimately returning 0 results) no longer just silently moves on — the researcher node emits a `skill_search_step` of type `skill_error` (carrying the error `reason`), and if the configured skills collected **no** new materials at all it runs a `fallback` pass over other available skills (`_fallback_candidate_skills`: the built-in `knowledge-base-search` native path first, then other enabled catalog skills not yet tried) instead of erroring out. The frontend `SkillStepsCard` renders `skill_error` (rose) and `fallback` (amber) so the timeline shows "技能报错 → 改用其他技能搜集". Tests: `tests/test_ai_writing_material_source.py`.
### Admin Leaderboard Snapshots (`/api/admin/users/leaderboard*`)
@@ -259,7 +259,14 @@ async def langgraph_runtime(app: FastAPI) -> AsyncGenerator[None, None]:
_app = app # 闭包捕获 app,供网关访问 app.state(checkpointer/thread_store)
gateway_factory = lambda p: InProcessRoundtableGateway( # noqa: E731
_app, user_id=p.user_id, agents=p.agents, model=p.model
_app,
user_id=p.user_id,
agents=p.agents,
model=p.model,
# 席位执行模式沿用 Step2 选择,后台席位 run 据此覆盖 policy 的 thinking/reasoning + 开子代理。
seat_thinking_enabled=p.seat_thinking_enabled,
seat_reasoning_effort=p.seat_reasoning_effort,
seat_subagent_enabled=p.seat_subagent_enabled,
)
logger.info("[roundtable-jobs] gateway = inprocess (real models, default)")
@@ -14,8 +14,10 @@
from __future__ import annotations
import asyncio
import json
import logging
import os
import uuid
from datetime import UTC, datetime
from types import SimpleNamespace
@@ -51,6 +53,33 @@ _SUMMARY_FILE = "方案总结报告.md"
# 这里漏禁了 web_search,导致后台席位 / 报告能联网搜索而前端不能,现由 policy 统一。
# ── 后台作业「按用户限流」(2026-06-16)─────────────────────────────────────────
# 痛点:进 2+ 个后台挂起后,整个网关变慢——建立智能体连接、等待模型响应都慢。
# 根因:① checkpointer 用单连接 aiosqlite(AsyncSqliteSaver,内部一把 asyncio.Lock 串
# 行全进程的 checkpoint 读写);② 后台作业无并发上限——每个作业进程内驱动真实 leader/
# seat/report run,DAG/recommend 的 stage 还用 asyncio.gather 把席位真并行铺开。多个后台
# 作业同时把几十个 agent run 压到同一把 SQLite 锁 + 同一个离线模型端点上,前台实时会商
# 的 checkpoint 读写 / 模型调用只能排队 → 「干啥都慢」。
#
# 解法:给**每个用户**一个并发额度(默认 2),后台每个 agent 轮次(leader/seat/report/
# summary/dashboard)真正开跑前先取一个额度,跑完释放。同一用户的所有后台作业共享这 2 个
# 额度——随便挂多少个作业都能立刻建好,但同一瞬间真正在打模型/写 checkpoint 的后台轮次
# 至多 2 个,不会把共享资源打满。**前台实时路径(multi_agent._leader_run/_special_run、
# 普通聊天)不经过本文件的 `_run_agent_turn`,永不取这个额度,故前台不受限。**
_BACKGROUND_SLOTS_PER_USER = max(1, int(os.getenv("ROUNDTABLE_BACKGROUND_SLOTS_PER_USER", "2") or "2"))
# 按 user_id 缓存 Semaphore。网关全程跑在同一个事件循环上,普通 dict 取/建即可(无跨线程)。
_user_background_sems: dict[str, asyncio.Semaphore] = {}
def _user_background_semaphore(user_id: str) -> asyncio.Semaphore:
"""取该用户的后台并发额度信号量(懒建,每用户一把)。"""
sem = _user_background_sems.get(user_id)
if sem is None:
sem = asyncio.Semaphore(_BACKGROUND_SLOTS_PER_USER)
_user_background_sems[user_id] = sem
return sem
# ── 进程内底层操作 ───────────────────────────────────────────────────────────
@@ -156,6 +185,7 @@ def _tool_calls_since_last_human(messages: list[dict[str, Any]]) -> list[dict[st
async def _run_agent_turn(
app,
*,
user_id: str,
thread_id: str,
agent_name: str,
message: str,
@@ -164,11 +194,16 @@ async def _run_agent_turn(
skill_stop_names: list[str],
thinking_enabled: bool,
reasoning_effort: str,
subagent_enabled: bool = False,
) -> list[dict[str, Any]]:
"""进程内驱动 agent_name 在 thread_id 上跑一轮(发 message),阻塞到完成,返回最终 messages。
镜像 multi_agent ``_run_payload`` 的 body 形状:agent 通过 ``context.agent_name`` 选,
assistant_id 固定 lead_agent;关闭 memory recall / subagent / plan mode(临时研讨任务)。
并发受**该用户的后台额度**约束(``_user_background_semaphore``):真正开跑(start_run +
await run task)前先取额度,跑完释放。同一用户的所有后台作业共享这几个额度,避免 2+
后台挂起把单连接 SQLite checkpointer / 离线模型打满、拖垮前台实时会商。
"""
from app.gateway.routers.thread_runs import RunCreateRequest
from app.gateway.services import start_run
@@ -184,7 +219,8 @@ async def _run_agent_turn(
"reasoning_effort": reasoning_effort,
"thinking_enabled": thinking_enabled,
"is_plan_mode": False,
"subagent_enabled": False,
# ultra 席位才置真 → 解锁 task 子代理工具(席位 policy 未禁 task)。leader 恒 False。
"subagent_enabled": subagent_enabled,
"thread_id": thread_id,
"memory_recall_disabled": True,
# 与 multi_agent._run_payload 对齐:关 builtin 记忆注入,避免单例总控/席位
@@ -199,9 +235,12 @@ async def _run_agent_turn(
# 虚拟 request:带上空 ``state``(auth 中间件在真请求上会写 state.user / state.auth;
# 进程内没有这层,下游凡是读 request.state.X 的代码必须能拿到 None 而不是 AttributeError)。
request = SimpleNamespace(app=app, state=SimpleNamespace(user=None, auth=None))
record = await start_run(body, thread_id, request) # type: ignore[arg-type]
if getattr(record, "task", None) is not None:
await record.task # 阻塞到这轮 agent 跑完,checkpoint 已落盘
# 取该用户的后台额度后再开跑:start_run + await task 是最重的部分(建 agent、调模型、
# 落 checkpoint)。额度满时本轮在此 await 排队,等同用户某个在跑轮次释放额度后再开。
async with _user_background_semaphore(user_id):
record = await start_run(body, thread_id, request) # type: ignore[arg-type]
if getattr(record, "task", None) is not None:
await record.task # 阻塞到这轮 agent 跑完,checkpoint 已落盘
return await _read_thread_messages(app, thread_id)
@@ -262,7 +301,13 @@ def _read_summary_md(app, thread_id: str, user_id: str) -> str | None:
# it all at once」后发出超长 write_file,单次输出被 max_tokens 截断 → 工具调用从未执行、
# run 以 success 收尾但 outputs 下没有 .html → step3.html 回退成开场白文字。
# 自愈:检测产物缺失/截断后,同线程追加纠偏消息(要求 write_file 分段 + append 续写)重试。
_ARTIFACT_RETRY_LIMIT = 2
#
# 取值依据(2026-06-15 修复「后台 HTML 看板只生成一半、概率渲染失败」):模型单轮输出受
# ``config.yaml`` 的 ``max_tokens``(deepseek-chat = 4096)硬截。一份完整看板 HTML 几十 KB
# ≈ 上万 token,单轮写不完;旧值 2(最多 3 轮 ≈ 12k token)对长报告不够 → 落库半截 HTML →
# 渲染失败。调到 5(最多 6 轮 append ≈ 24k token,足够绝大多数看板)显著降低截断概率。
# 循环命中完整产物(``</html>`` / 非空 md / 合法 json)即 break,短报告不会白跑额外轮次。
_ARTIFACT_RETRY_LIMIT = 5
_REPORT_RETRY_MISSING = (
f"检测到 /mnt/user-data/outputs/{_REPORT_FILE} 还没有生成——你上一轮的 write_file 没有执行成功"
@@ -332,11 +377,26 @@ def _build_seat_intro(agents: list[SeatRef]) -> str:
class InProcessRoundtableGateway:
"""进程内真网关。executor 先调 ``init_threads`` 建好线程,再跑编排引擎。"""
def __init__(self, app, *, user_id: str, agents: list[SeatRef], model: str | None = None) -> None:
def __init__(
self,
app,
*,
user_id: str,
agents: list[SeatRef],
model: str | None = None,
seat_thinking_enabled: bool | None = None,
seat_reasoning_effort: str | None = None,
seat_subagent_enabled: bool = False,
) -> None:
self._app = app
self._user_id = user_id
self._agents = list(agents)
self._model = model
# 席位执行模式(沿用 Step2 选择):覆盖 seat policy 的 thinking/reasoning(None → policy 默认),
# subagent 仅 ultra 档为真。只作用 special 席位;leader 派活不受影响。
self._seat_thinking_enabled = seat_thinking_enabled
self._seat_reasoning_effort = seat_reasoning_effort
self._seat_subagent_enabled = seat_subagent_enabled
self._seat_intro = ""
self._first_leader = True
@@ -362,6 +422,7 @@ class InProcessRoundtableGateway:
policy = roundtable_run_policy("leader")
messages = await _run_agent_turn(
self._app,
user_id=self._user_id,
thread_id=thread_ids[coordinator_name],
agent_name=coordinator_name,
message=msg,
@@ -400,16 +461,27 @@ class InProcessRoundtableGateway:
async def run_seat(self, *, thread_ids, agent_id, task, model) -> SeatTurn:
policy = roundtable_run_policy("seat")
# 席位模式覆盖 policy 的 thinking/reasoning(None → policy 默认),与前端 _special_run 一致。
messages = await _run_agent_turn(
self._app,
user_id=self._user_id,
thread_id=thread_ids[agent_id],
agent_name=agent_id,
message=task,
model=model or self._model,
excluded_tools=policy.excluded_tools,
skill_stop_names=policy.skill_stop_names,
thinking_enabled=policy.thinking_enabled,
reasoning_effort=policy.reasoning_effort,
thinking_enabled=(
policy.thinking_enabled
if self._seat_thinking_enabled is None
else self._seat_thinking_enabled
),
reasoning_effort=(
policy.reasoning_effort
if self._seat_reasoning_effort is None
else self._seat_reasoning_effort
),
subagent_enabled=self._seat_subagent_enabled,
)
return SeatTurn(final_text=_latest_ai_text(messages), tool_calls=_tool_calls_since_last_human(messages))
@@ -427,6 +499,7 @@ class InProcessRoundtableGateway:
for attempt in range(1 + _ARTIFACT_RETRY_LIMIT):
messages = await _run_agent_turn(
self._app,
user_id=self._user_id,
thread_id=report_thread_id,
agent_name=REPORT_AGENT_ID,
message=message,
@@ -470,6 +543,7 @@ class InProcessRoundtableGateway:
for attempt in range(1 + _ARTIFACT_RETRY_LIMIT):
messages = await _run_agent_turn(
self._app,
user_id=self._user_id,
thread_id=summary_thread_id,
agent_name=SUMMARY_AGENT_ID,
message=message,
@@ -505,6 +579,7 @@ class InProcessRoundtableGateway:
for attempt in range(1 + _ARTIFACT_RETRY_LIMIT):
messages = await _run_agent_turn(
self._app,
user_id=self._user_id,
thread_id=thread_id,
agent_name=REPORT_AGENT_ID,
message=message,
@@ -179,6 +179,12 @@ class StartParams:
# 是否自动产出 Step3(方案总结报告 + 结果绘制)。默认沿用全局开关(现已恢复 True,
# 见 ROUNDTABLE_BACKGROUND_REPORT_ENABLED)。测试可显式覆盖。
enable_report: bool = ROUNDTABLE_BACKGROUND_REPORT_ENABLED
# 席位执行模式(快速/思考/专业问答/多智能体)派生参数,沿用 Step2 选择。仅 special 席位用:
# 覆盖 roundtable_run_policy("seat") 的 thinking/reasoning(None → policy 默认);
# subagent 仅 ultra 档为真。透传给进程内网关(InProcessRoundtableGateway)的席位 run。
seat_thinking_enabled: bool | None = None
seat_reasoning_effort: str | None = None
seat_subagent_enabled: bool = False
# 网关工厂:给定一次启动参数,返回一个 RoundtableGateway 实例。
@@ -637,6 +637,8 @@ def _run_payload(
skill_stop_names: list[str],
thinking_enabled: bool = True,
reasoning_effort: str = "medium",
subagent_enabled: bool = False,
stream_custom: bool = False,
multitask_strategy: str = "reject",
extra_context: dict[str, Any] | None = None,
files: list[dict[str, Any]] | None = None,
@@ -676,7 +678,9 @@ def _run_payload(
"reasoning_effort": reasoning_effort,
"thinking_enabled": thinking_enabled,
"is_plan_mode": False,
"subagent_enabled": False,
# 默认 False;圆桌「多智能体问答(ultra)」档会让席位 _special_run 传 True,
# 给席位解锁 task 子代理工具(席位 policy 未禁 task,故可用)。
"subagent_enabled": subagent_enabled,
"thread_id": thread_id,
# 圆桌四个 router(intent / recommend / leader / special)都是临时任务,
# 用户个人记忆与本次研讨无关。MemoryMiddleware 的 Hindsight recall 是
@@ -691,7 +695,12 @@ def _run_payload(
# 圆桌是一次性研讨,本就不该吃历史会话记忆,这里统一关注入根治该污染。
"memory_injection_enabled": False,
},
"stream_mode": ["messages-tuple", "values"],
# ultra 席位会派子智能体(task 工具),其 task_started/running/completed 进度走
# **custom** 流(StreamWriter)。仅这种场景加 "custom",让前端任务卡片能显示子智能体
# 实时进度;其余角色不加,避免无谓帧。
"stream_mode": (
["messages-tuple", "values", "custom"] if stream_custom else ["messages-tuple", "values"]
),
"stream_subgraphs": True,
"stream_resumable": True,
"assistant_id": "lead_agent",
@@ -1570,6 +1579,9 @@ async def _special_run(
no_file: bool = False,
peer_deliveries: list[dict[str, Any]] | None = None,
files: list[dict[str, Any]] | None = None,
thinking_enabled: bool | None = None,
reasoning_effort: str | None = None,
subagent_enabled: bool = False,
) -> AsyncIterator[str | dict[str, Any]]:
"""子智能体(special)单轮:执行总控下发的 task 文本,交付 content。
@@ -1584,6 +1596,12 @@ async def _special_run(
``peer_deliveries``:前序席位的**完整交付** ``[{name, content}, ...]``。注入到
run context 的 ``roundtable_peer_deliveries``,供席位用 ``read_peer_delivery`` 工具
**按需**取全文(平时只看广播摘要,§12.9 传摘要按需传全文)。不进 prompt,不耗 token。
``thinking_enabled`` / ``reasoning_effort`` / ``subagent_enabled``:圆桌第二步「席位
执行模式」(快速/思考/专业问答/多智能体问答)派生的覆盖值。**仅 special 席位**接收:
前两者为 None 时回退到 policy 默认(向后兼容,旧前端不传 → 维持 thinking + medium);
``subagent_enabled`` 仅 ultra 档为 True,给席位解锁 ``task`` 子代理工具。**总控
(leader)不受影响**,始终用其快速派活策略。
"""
t0 = time.perf_counter()
thread_id = agent_threads[agent_name]
@@ -1613,8 +1631,13 @@ async def _special_run(
new_message=new_message,
excluded_tools=seat_excluded,
skill_stop_names=merge_skill_stop_names(policy.skill_stop_names, skill_stop_names),
thinking_enabled=policy.thinking_enabled,
reasoning_effort=policy.reasoning_effort,
# 席位执行模式覆盖 policy 的 thinking/reasoning(None → 回退 policy 默认);
# subagent 仅 ultra 档置真。工具裁剪(excluded_tools)仍完全由 policy 决定。
thinking_enabled=(policy.thinking_enabled if thinking_enabled is None else thinking_enabled),
reasoning_effort=(policy.reasoning_effort if reasoning_effort is None else reasoning_effort),
subagent_enabled=subagent_enabled,
# ultra 席位才会派子智能体 → 才需要 custom 流回传子任务实时进度(task_running 等)。
stream_custom=subagent_enabled,
# 干预/重新派活会把同一席位 thread 再跑一次;抢占掉没结束的旧 run,避免 409。
multitask_strategy="rollback",
# 前序席位完整交付 → run context,供 read_peer_delivery 工具按需取全文(§12.9)。
@@ -1756,6 +1779,15 @@ async def run_multi_agent_stream(request: Request, payload: dict = Body(...)) ->
# 不写死任何默认模型;空串会让 lead_agent._resolve_model_name() 回退到
# config.yaml 的 models[0],与 /api/ai-writing 等其它路由一致。
model_name: str = str(payload.get("model") or "")
# 圆桌第二步「席位执行模式」(快速/思考/专业问答/多智能体问答)派生参数。**仅 special
# 席位消费**:覆盖 seat policy 的 thinking/reasoning。thinking_enabled/reasoning_effort
# 缺省(None/空)→ 后端按 policy 默认执行(向后兼容,旧前端不传维持 thinking + medium);
# subagent_enabled 仅 ultra 档为真,给席位解锁 task 子代理工具。leader 不消费这些字段。
_te_raw = payload.get("thinking_enabled")
seat_thinking_enabled: bool | None = bool(_te_raw) if _te_raw is not None else None
_re_raw = payload.get("reasoning_effort")
seat_reasoning_effort: str | None = _re_raw if isinstance(_re_raw, str) and _re_raw else None
seat_subagent_enabled: bool = bool(payload.get("subagent_enabled") or False)
if not agent_name or agent_name not in agent_threads:
raise HTTPException(
@@ -1803,6 +1835,9 @@ async def run_multi_agent_stream(request: Request, payload: dict = Body(...)) ->
no_file=no_file,
peer_deliveries=peer_deliveries,
files=files,
thinking_enabled=seat_thinking_enabled,
reasoning_effort=seat_reasoning_effort,
subagent_enabled=seat_subagent_enabled,
)
async for chunk in gen:
if isinstance(chunk, dict):
@@ -57,6 +57,8 @@ class ChainResponse(BaseModel):
stage_goals: list[str | None] | None = None
# 写给总控的「编排提示」(dag-only):告诉总控怎么派活、做什么。None/空 = 不注入。
coordinator_prompt: str | None = None
# 默认席位执行模式(flash/thinking/pro/ultra);None = 未设,选入研讨时回退默认(pro)。
seat_mode: str | None = None
# 发布状态:True = 公共(所有账号可在「公共」筛选查看 / 选用)。
is_public: bool = False
# 是否当前用户所有(仅本人可编辑 / 删除 / 发布;公共筛选里别人的链条 is_mine=False)。
@@ -83,6 +85,7 @@ class ChainCreateRequest(BaseModel):
stages: list[list[str]] | None = None
stage_goals: list[str | None] | None = None
coordinator_prompt: str | None = Field(default=None, max_length=4000)
seat_mode: str | None = Field(default=None, max_length=16)
class ChainUpdateRequest(BaseModel):
@@ -93,6 +96,7 @@ class ChainUpdateRequest(BaseModel):
stages: list[list[str]] | None = None
stage_goals: list[str | None] | None = None
coordinator_prompt: str | None = Field(default=None, max_length=4000)
seat_mode: str | None = Field(default=None, max_length=16)
# 发布开关(True 发布为公共 / False 取消发布);仅 owner 可改(update 本就 owner 限定)。
is_public: bool | None = None
@@ -224,6 +228,18 @@ def _validate_stage_goals(
)
_VALID_SEAT_MODES = {"flash", "thinking", "pro", "ultra"}
def _validate_seat_mode(seat_mode: str | None) -> None:
"""席位执行模式只能是 flash/thinking/pro/ultra 或 None(未设)。"""
if seat_mode is not None and seat_mode not in _VALID_SEAT_MODES:
raise HTTPException(
status_code=400,
detail=f"非法席位执行模式: {seat_mode}(应为 {'/'.join(sorted(_VALID_SEAT_MODES))} 之一)",
)
# ── routes ───────────────────────────────────────────────────────────────
@@ -249,6 +265,7 @@ async def create_chain(request: Request, body: ChainCreateRequest) -> ChainRespo
_validate_seats(body.seats)
_validate_stages(body.seats, body.stages)
_validate_stage_goals(body.stages, body.stage_goals)
_validate_seat_mode(body.seat_mode)
store = _get_store(request)
user_id = _current_user_id(request)
row = await store.create_chain(
@@ -261,6 +278,7 @@ async def create_chain(request: Request, body: ChainCreateRequest) -> ChainRespo
"stages": body.stages,
"stage_goals": body.stage_goals,
"coordinator_prompt": body.coordinator_prompt,
"seat_mode": body.seat_mode,
},
)
await _enrich_owner_names([row])
@@ -290,6 +308,7 @@ async def update_chain(request: Request, chain_id: str, body: ChainUpdateRequest
_validate_stages(body.seats, body.stages)
# Goals align to stages; reject goals sent without a layering, enforce length/size.
_validate_stage_goals(body.stages, body.stage_goals)
_validate_seat_mode(body.seat_mode)
store = _get_store(request)
user_id = _current_user_id(request)
# Distinguish "omitted" from "explicitly sent"; only forward provided fields.
@@ -68,6 +68,12 @@ class StartJobRequest(BaseModel):
# 预建好的 thread 映射(Phase 3b 真网关用;模拟网关忽略)。
threadIds: dict[str, str] | None = None
coordinatorName: str = "roundtable-coordinator"
# 席位执行模式(快速/思考/专业问答/多智能体)派生参数,由前端 toStartBody 下发。
# 仅 special 席位用:覆盖 roundtable_run_policy("seat") 的 thinking/reasoning;
# None/缺省 → 后端按 policy 默认(≈pro),向后兼容旧前端。subagent 仅 ultra 档为真。
thinking_enabled: bool | None = None
reasoning_effort: str | None = None
subagent_enabled: bool = False
class ResumeJobRequest(BaseModel):
@@ -219,6 +225,9 @@ async def start_job(request: Request, body: StartJobRequest) -> RoundtableJobRes
draft_id=body.draftId,
chain=body.chain,
orchestration_plan=body.orchestrationPlan,
seat_thinking_enabled=body.thinking_enabled,
seat_reasoning_effort=body.reasoning_effort,
seat_subagent_enabled=body.subagent_enabled,
)
)
return _to_response(row)
+12 -12
View File
@@ -10,18 +10,18 @@ gateway:
port: 8001
models:
- name: deepseek-chat
display_name: DeepSeek Chat
use: langchain_openai:ChatOpenAI
model: deepseek-chat
api_key: sk-3fee562f834a4c08bb0d471ef44cb445
base_url: https://api.deepseek.com
request_timeout: 600.0
max_retries: 2
max_tokens: 4096
temperature: 1.0
supports_vision: false
supports_thinking: true
# - name: deepseek-chat
# display_name: DeepSeek Chat
# use: langchain_openai:ChatOpenAI
# model: deepseek-chat
# api_key: sk-e1c22b7931dc446f8b67e57878edd105
# base_url: https://api.deepseek.com
# request_timeout: 600.0
# max_retries: 2
# max_tokens: 4096
# temperature: 1.0
# supports_vision: false
# supports_thinking: true
# - name: zai-org/GLM-5-FP8
# display_name: zai-org/GLM-5-FP8
@@ -141,6 +141,30 @@ def _load_configured_skills() -> list:
return resolved or [_fallback_knowledge_skill()]
def _fallback_candidate_skills(tried: set[str]) -> list:
"""容错兜底:从技能目录里挑出**尚未尝试过**的可用技能,内置「知识库检索」优先。
当配置技能调用全部失败、且没收集到任何新素材时,researcher 用这些技能再试一轮,
而不是直接报错。内置知识库检索(原生多关键词并发,最稳)排在最前,其后是技能目录里
其它已启用的技能。技能目录加载失败时只回退内置知识库检索,保证兜底链永不断。
"""
candidates: list = []
seen: set[str] = set(tried)
if KNOWLEDGE_SEARCH_SKILL_NAME not in seen:
candidates.append(_fallback_knowledge_skill())
seen.add(KNOWLEDGE_SEARCH_SKILL_NAME)
try:
from deerflow.skills.storage import get_or_new_skill_storage
for s in get_or_new_skill_storage().load_skills(enabled_only=True):
if getattr(s, "name", None) and s.name not in seen:
candidates.append(s)
seen.add(s.name)
except Exception:
logger.warning("researcher: 兜底加载技能目录失败", exc_info=True)
return candidates
def _resolve_keyword_count(raw: object) -> int:
"""把用户传入的检索关键词数量规整到 [1, _MAX_KEYWORD_COUNT],非法值回退默认 6。"""
try:
@@ -511,24 +535,35 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
added += 1
return added
async def _timed_search(skill_name: str, query: str, limit: int) -> list[SearchResult]:
async def _timed_search(skill_name: str, query: str, limit: int) -> tuple[list[SearchResult], str]:
# 加超时兜底:单个技能检索最多等 _SKILL_SEARCH_TIMEOUT_SECONDS 秒,
# 超时不再阻塞后续技能,避免整段过程看起来"冻结"。
# 返回 (结果, 错误详情):错误详情非空 = 该技能调用失败(触发兜底)。
try:
return await asyncio.wait_for(
results = await asyncio.wait_for(
provider.search(query, max_results=limit),
timeout=_SKILL_SEARCH_TIMEOUT_SECONDS,
) or []
)
return (results or []), ""
except asyncio.TimeoutError:
msg = f"检索超时(>{_SKILL_SEARCH_TIMEOUT_SECONDS}s)"
logger.warning("skill[%s] search timed out after %ss", skill_name, _SKILL_SEARCH_TIMEOUT_SECONDS)
search_errors.append(f"skill[{skill_name}]: 检索超时(>{_SKILL_SEARCH_TIMEOUT_SECONDS}s)")
return []
search_errors.append(f"skill[{skill_name}]: {msg}")
return [], msg
except Exception as e:
logger.warning("skill[%s] search failed: %s", skill_name, e)
search_errors.append(f"skill[{skill_name}]: {e}")
return []
return [], str(e)
for sk in ordered:
def _skill_display(sk) -> str:
return _KNOWLEDGE_SKILL_DISPLAY if sk.name == KNOWLEDGE_SEARCH_SKILL_NAME else sk.name
async def _execute_skill(sk) -> tuple[int, str]:
"""运行单个技能检索,返回 (新增素材数, 错误详情)。
错误详情非空 = 该技能整体调用失败(超时 / 抛异常 / 全部检索词失败),
调用方据此 emit ``skill_error`` 步骤并触发"改用其它技能"的兜底逻辑。
"""
is_knowledge = sk.name == KNOWLEDGE_SEARCH_SKILL_NAME
if is_knowledge and not is_user_query_search:
# 原生知识库检索:对全部检索词并发检索(旧版默认路径)。
@@ -557,13 +592,20 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
gathered = await asyncio.gather(*[_search_one(kw) for kw in search_keywords], return_exceptions=True)
added = 0
errs: list[str] = []
for results in gathered:
if isinstance(results, Exception):
logger.warning("search failed: %s", results)
search_errors.append(f"{type(results).__name__}: {results}")
errs.append(f"{type(results).__name__}: {results}")
continue
added += _collect(results)
search_errors.extend(getattr(provider, "errors", []))
search_errors.extend(errs)
# 全部检索词都抛异常且 0 新增 → 视为该技能整体调用失败,触发兜底
error_detail = ""
if added == 0 and errs and len(errs) == total:
error_detail = ";".join(dict.fromkeys(errs))[:200]
return added, error_detail
else:
# 定向技能检索:user_query 模式优先用用户原话;非知识库技能再
# 附上 description 让 query 贴合该技能的领域。
@@ -576,14 +618,35 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
"message": f"使用技能「{sk.name}」检索:{skill_query[:60]}",
"skill_name": sk.name,
})
added = _collect(await _timed_search(sk.name, skill_query, per_skill))
results, error_detail = await _timed_search(sk.name, skill_query, per_skill)
return _collect(results), error_detail
_emit_skill_step({
"type": "skill_result",
"message": f"技能「{_KNOWLEDGE_SKILL_DISPLAY if is_knowledge else sk.name}」检索到 {added} 条新素材",
"skill_name": sk.name,
"matched_count": added,
})
# 失败技能名(去重后用于兜底提示):记录配置技能里调用失败的那些
failed_skill_names: list[str] = []
tried_names: set[str] = set()
def _emit_skill_outcome(sk, added: int, error_detail: str) -> None:
"""单个技能跑完后,按成功 / 失败 emit 对应步骤事件。"""
if error_detail:
failed_skill_names.append(_skill_display(sk))
_emit_skill_step({
"type": "skill_error",
"message": f"技能「{_skill_display(sk)}」调用失败,尝试使用其他技能检索",
"skill_name": sk.name,
"reason": error_detail,
})
else:
_emit_skill_step({
"type": "skill_result",
"message": f"技能「{_skill_display(sk)}」检索到 {added} 条新素材",
"skill_name": sk.name,
"matched_count": added,
})
for sk in ordered:
added, error_detail = await _execute_skill(sk)
tried_names.add(sk.name)
_emit_skill_outcome(sk, added, error_detail)
# 早停:累积素材已达上限
if len(new_results) >= max_materials:
@@ -594,6 +657,27 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
})
break
# ── 容错兜底:配置技能调用失败且没收集到任何新素材 → 改用其它可用技能继续检索 ──
# (不直接报错;前端时间线靠 skill_error / fallback 步骤展示"技能报错,改用其它技能搜集")
if not new_results and failed_skill_names:
fallback_skills = _fallback_candidate_skills(tried_names)
if fallback_skills:
_emit_skill_step({
"type": "fallback",
"message": (
f"已配置技能({'、'.join(dict.fromkeys(failed_skill_names))})检索失败,"
"改用其他技能继续搜集素材"
),
"skill_names": [_skill_display(s) for s in fallback_skills],
})
for sk in fallback_skills:
added, error_detail = await _execute_skill(sk)
tried_names.add(sk.name)
_emit_skill_outcome(sk, added, error_detail)
if new_results:
# 兜底已拿到素材即止,不再无谓尝试后续技能
break
# 合并去重
# - 首次检索 / 普通 re_search(无输入,用户只想重出):旧素材丢弃,覆盖式替换
# - 素材不足求助 supplement_search:保留旧素材 + 追加新检索结果
@@ -0,0 +1,50 @@
"""Add seat_mode column to roundtable_chains.
Per-chain「默认席位执行模式」: an optional short string (flash/thinking/pro/ultra, see
frontend ``lib/seat-mode.ts``) the user picks in the chain editor. When the chain is
selected into a roundtable discussion the picker defaults to it (still overridable per
run). NULL = unset → callers fall back to the default (pro). VARCHAR(16) on MySQL, plain
String elsewhere — matching ``String(16)`` on ``RoundtableChainRow.seat_mode``.
``create_all`` adds the column for brand-new DBs; the startup ``_ensure_orm_columns_sync``
sweep adds it to already-existing tables (sqlite/mysql/postgres). This revision keeps the
alembic graph consistent with the rest of the chain columns and documents the change.
Revision ID: 20260615_01
Revises: 20260612_01
Create Date: 2026-06-15
"""
from __future__ import annotations
import sqlalchemy as sa
from alembic import op
from sqlalchemy import inspect
revision = "20260615_01"
down_revision = "20260612_01"
branch_labels = None
depends_on = None
def _has_table(table_name: str) -> bool:
inspector = inspect(op.get_bind())
return table_name in inspector.get_table_names()
def _has_column(table_name: str, column_name: str) -> bool:
if not _has_table(table_name):
return False
inspector = inspect(op.get_bind())
return any(col["name"] == column_name for col in inspector.get_columns(table_name))
def upgrade() -> None:
if _has_table("roundtable_chains") and not _has_column("roundtable_chains", "seat_mode"):
op.add_column("roundtable_chains", sa.Column("seat_mode", sa.String(length=16), nullable=True))
def downgrade() -> None:
if _has_column("roundtable_chains", "seat_mode"):
op.drop_column("roundtable_chains", "seat_mode")
@@ -49,6 +49,9 @@ class RoundtableChainRow(Base):
# 怎么派活、做什么(整体编排策略 / 风格 / 约束)。**仅分层 DAG 模式生效**:注入到总控
# 每阶段派活提示(``buildDagStageLeaderTask``)。NULL = 不注入。纯文本,存为 LONGTEXT。
coordinator_prompt: Mapped[str | None] = mapped_column(PortableLongText, nullable=True)
# 可选「默认席位执行模式」:flash/thinking/pro/ultra(见前端 lib/seat-mode.ts)。选入研讨时
# 弹窗默认用它(仍可临时改)。NULL = 未设,调用方回退默认(pro)。短串,VARCHAR(16)。
seat_mode: Mapped[str | None] = mapped_column(String(16), nullable=True)
# 发布开关:True = 公共(所有账号可在「公共」筛选里查看 / 选用);False = 仅本人可见。
# 编辑 / 删除始终只限 owner;发布只是放开「查看 + 选入研讨」。索引以支持「公共」列表筛选。
is_public: Mapped[bool] = mapped_column(
@@ -56,6 +56,8 @@ class RoundtableChainRepository:
"stage_goals": _loads(getattr(row, "stage_goals", None)),
# 总控编排提示(写给总控的派活指引,dag-only);None/空 = 不注入。纯文本,不 json。
"coordinator_prompt": getattr(row, "coordinator_prompt", None),
# 默认席位执行模式(flash/thinking/pro/ultra);None = 未设。
"seat_mode": getattr(row, "seat_mode", None),
# 发布状态 + 是否本人所有(前端据此显示「公共」徽标、决定能否编辑 / 发布)。
"is_public": bool(getattr(row, "is_public", False)),
"is_mine": (current_user_id is None) or (row.user_id == current_user_id),
@@ -110,6 +112,7 @@ class RoundtableChainRepository:
stages=_dumps(data.get("stages")),
stage_goals=_dumps(data.get("stage_goals")),
coordinator_prompt=(data.get("coordinator_prompt") or None),
seat_mode=(data.get("seat_mode") or None),
is_public=bool(data.get("is_public") or False),
created_at=now,
updated_at=now,
@@ -131,6 +134,7 @@ class RoundtableChainRepository:
stages=_UNSET,
stage_goals=_UNSET,
coordinator_prompt=_UNSET,
seat_mode=_UNSET,
is_public=_UNSET,
) -> dict[str, Any] | None:
async with self._sf() as session:
@@ -149,6 +153,8 @@ class RoundtableChainRepository:
row.stage_goals = _dumps(stage_goals)
if coordinator_prompt is not _UNSET:
row.coordinator_prompt = coordinator_prompt or None
if seat_mode is not _UNSET:
row.seat_mode = seat_mode or None
if is_public is not _UNSET and is_public is not None:
row.is_public = bool(is_public)
row.updated_at = datetime.now(UTC)
@@ -82,6 +82,10 @@ CALL _add_column_if_missing('roundtable_drafts', 'task_id', 'VARCHAR(128) NULL D
-- ORM: RoundtableJobRow.orchestration_plan = PortableLongText(LONGTEXT), nullable。
CALL _add_column_if_missing('roundtable_jobs', 'orchestration_plan', 'LONGTEXT NULL DEFAULT NULL');
-- seat_mode: 业务链条「默认席位执行模式」(flash/thinking/pro/ultra),对应 migration 20260615_01。
-- 选入研讨时弹窗默认用它(仍可临时改)。ORM: RoundtableChainRow.seat_mode = String(16), nullable。
CALL _add_column_if_missing('roundtable_chains', 'seat_mode', 'VARCHAR(16) NULL DEFAULT NULL');
-- ---- 清理临时存储过程 --------------------------------------------------------
DROP PROCEDURE IF EXISTS _add_column_if_missing;