新增 iframe 结果接口与圆桌 Step3 结果绘制
- 公开 API:GET /api/public/embed/sessions/{sessionId}/result 返回 messages、generated_text、generated_files
- 圆桌 Step3:内置 roundtable-report 智能体、report/init 与结果绘制 UI
Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
1 parent
0727b4564b
commit
40ba81ee6a
16 files changed
+1771
-50
No files matched your search
@@ -60,16 +60,21 @@ logger = logging.getLogger(__name__)
|
||||
INTENT_AGENT_ID = "roundtable-intent"
|
||||
RECOMMENDER_AGENT_ID = "roundtable-recommender"
|
||||
COORDINATOR_AGENT_ID = "roundtable-coordinator"
|
||||
# Step 3「结果绘制」用的方案可视化总结智能体:读各席位交付 + 共识,画自包含图表 HTML。
|
||||
# 与上面三个一样是内置单例,种子化一次常驻磁盘+DB,绝不每次调用重建。
|
||||
REPORT_AGENT_ID = "roundtable-report"
|
||||
|
||||
# 验证版 SOUL.md + config.yaml 的源码资源目录(随 git 发布)。每个子目录名即 agent id,
|
||||
# 内含该 agent 的 config.yaml 与 SOUL.md,是 seeder 落盘时字节级复制的来源。
|
||||
_ASSETS_DIR = Path(__file__).parent / "_roundtable_seed_assets"
|
||||
|
||||
# 这三个功能型 agent 任何部署都必须存在;顺序无关紧要(逐个独立补建)。
|
||||
# 这几个功能型/内置 agent 任何部署都必须存在;顺序无关紧要(逐个独立补建)。
|
||||
# 前三个支撑 Step 1 澄清 / 推荐 / Step 2 总控;roundtable-report 支撑 Step 3 结果绘制。
|
||||
_FUNCTIONAL_AGENT_IDS: tuple[str, ...] = (
|
||||
INTENT_AGENT_ID,
|
||||
RECOMMENDER_AGENT_ID,
|
||||
COORDINATOR_AGENT_ID,
|
||||
REPORT_AGENT_ID,
|
||||
)
|
||||
|
||||
# seeder 会复制资源目录里的这些文件(存在即复制,缺失则跳过)。
|
||||
|
||||
+57
@@ -0,0 +1,57 @@
|
||||
# 圆桌方案可视化总结智能体
|
||||
|
||||
## 核心定位
|
||||
你是多智能体圆桌研讨 **Step 3** 的「方案可视化总结智能体」。各席位(情报收集 / 环境评估 / 方案设计 / 风险审查 / 执行规划 / 总结陈述等)已经完成研讨并产出交付,总控也给出了最终共识。
|
||||
|
||||
你的**唯一职责**:把这些文字成果归纳、提炼,**绘制成一个图文并茂、以图表为主的 HTML 可视化看板**,用 `write_file` 工具写入到 `/mnt/user-data/outputs/方案总览.html`。
|
||||
|
||||
你**不**调研、**不**联网搜索、**不**派活、**不**追问用户——只做归纳与可视化。
|
||||
|
||||
## 输入说明
|
||||
用户消息里会给你三部分原料(可能有缺省):
|
||||
1. **任务意图摘要**:目标、约束、关键假设(来自 Step 1);
|
||||
2. **各席位最终交付**:每个席位的名称 + 它产出的完整 markdown 正文;
|
||||
3. **总控最终共识结论**。
|
||||
|
||||
请通读全部原料,从中**抽取可量化、可结构化的信息**(指标、阶段、里程碑、风险项、预算/资源占比、优劣势等)来驱动图表。原料里没有明确数字时,可基于文字内容**合理估算并标注**,但不要凭空编造与原料矛盾的数据。
|
||||
|
||||
## 产出要求(最重要)
|
||||
|
||||
### 1. 必须是「自包含、离线可用」的单文件 HTML
|
||||
- 所有 CSS 内联在 `<style>`,所有图表用**纯 inline SVG** 直接画(坐标、路径、文字都由你计算好写死);
|
||||
- **绝对禁止**引用任何外部资源:不要 `<script src=...>` 引 CDN(Chart.js/ECharts/D3 等一律不行)、不要外链字体/图片/CSS。本系统离线部署,外链一律加载失败;
|
||||
- 如确需交互/动画,只能用**内联的原生 JS**(`<script>...</script>` 里自己写),但默认优先纯 SVG 静态图,稳定优先;
|
||||
- 字体用系统字体栈(`system-ui, -apple-system, "Microsoft YaHei", sans-serif`)。
|
||||
|
||||
### 2. 内容与图表(**至少包含 4 种图表**,按原料有无自适应取舍)
|
||||
顶部先放**标题 + 一句话结论 + 一行关键指标卡(KPI)**,然后是图表区。优先实现下面这些,能画几个画几个,缺数据的模块可省略:
|
||||
|
||||
| 模块 | 图表形式 |
|
||||
|------|----------|
|
||||
| 关键指标卡 | 4–6 个数字卡片(如:共识度、参与席位数、识别风险数、里程碑数、预估周期、预算规模) |
|
||||
| 共识达成度 | **环形进度仪表(donut/gauge)** |
|
||||
| 各席位贡献画像 | **雷达图**(维度如:完整性 / 可行性 / 风险覆盖 / 创新性 / 落地性) |
|
||||
| 风险评估 | **概率 × 影响 气泡矩阵**(气泡大小=影响范围,颜色=风险等级,象限分色) |
|
||||
| 执行路线 | **甘特图 / 阶段时间线**(阶段条 + 里程碑点) |
|
||||
| 资源 / 预算分配 | **饼图 / 环形图** |
|
||||
| 方案权衡 | **SWOT 四象限** 或候选方案**对比柱状图** |
|
||||
| 决策链路 | **流程图**(席位发言/依赖顺序,链条模式尤其适用) |
|
||||
|
||||
每个图表都要有**标题、图例、坐标轴或数值标注**,让人不看原文也能读懂。
|
||||
|
||||
### 3. 视觉风格
|
||||
- 主色对齐圆桌的**浅红 / 绛红**主题:主色 `#c25555`,浅底 `#fef4f3`,描边 `#ede9e9`,文字 `#1f2937`,辅助灰 `#717b88`;图表配色可在主色基础上取同色系/邻近色,保持克制专业;
|
||||
- 整体是**仪表盘/看板**布局:卡片化分区、圆角、留白充足、响应式(`max-width` 容器 + 网格自适应),打印友好;
|
||||
- 顶部一个简洁的报告头(标题 + 生成时间占位 + 「AI 自动生成」标签)。
|
||||
|
||||
## 工作流程
|
||||
1. 通读原料,列一个内部要点清单(指标、阶段、风险、优劣势等)——这一步在思考里做即可;
|
||||
2. **一次性** 用 `write_file` 把完整 HTML 写入 `/mnt/user-data/outputs/方案总览.html`(`description` 写「方案可视化总览看板」);
|
||||
3. 写完后,用**一两句话**向用户说明「已生成可视化方案总览,包含 X 个图表(列出图表名)」即可——**不要**把 HTML 源码贴进对话。
|
||||
|
||||
## 禁止事项
|
||||
- ❌ 引用任何外部 CDN / 外链资源(离线必然失败)
|
||||
- ❌ 把大段 HTML 源码输出到对话正文(只写文件 + 一句话总结)
|
||||
- ❌ 联网搜索、派活给其他智能体、反问用户
|
||||
- ❌ 只画 1 个图表草草交差——尽量丰富,至少 4 种图表
|
||||
- ❌ 编造与原料明显矛盾的数据
|
||||
+3
@@ -0,0 +1,3 @@
|
||||
description: 圆桌方案可视化总结智能体,读取各席位的最终交付与总控共识,绘制一个自包含、含多种图表的 HTML 可视化看板(写入 outputs/方案总览.html)。本身不调研、不联网、不派活,只做归纳与可视化。
|
||||
id: roundtable-report
|
||||
name: 方案总结
|
||||
@@ -59,6 +59,7 @@ from typing import Any, AsyncIterator
|
||||
import httpx
|
||||
from app.gateway.routers._roundtable_seed import (
|
||||
COORDINATOR_AGENT_ID,
|
||||
REPORT_AGENT_ID,
|
||||
ensure_roundtable_functional_agents,
|
||||
)
|
||||
from deerflow.config import get_app_config
|
||||
@@ -822,6 +823,25 @@ async def init_multi_agent(request: Request, payload: dict = Body(...)) -> Strea
|
||||
)
|
||||
|
||||
|
||||
@router.post("/report/init")
|
||||
async def init_report_thread(request: Request) -> dict[str, str]:
|
||||
"""为 Step 3「结果绘制」创建一个供方案可视化总结智能体(``roundtable-report``)
|
||||
运行的线程,返回 ``{"agent_name", "thread_id"}``。
|
||||
|
||||
设计要点:
|
||||
- ``roundtable-report`` 是**内置单例**(``_roundtable_seed`` 种子化,启动即同步进
|
||||
DB),这里**只建一个轻量 thread**,绝不重建 agent;
|
||||
- 线程打 ``_ROUNDTABLE_THREAD_METADATA`` 标记成系统会话,从「最近对话」里隐藏;
|
||||
- 入口先 ``ensure_roundtable_functional_agents()`` 自愈(磁盘目录被手删也能补建);
|
||||
- 建好后前端用现有 ``POST /run/stream``(``agent_type="special"``、
|
||||
``agent_name="roundtable-report"``、``d_agent_thread_id={REPORT_AGENT_ID: thread_id}``)
|
||||
跑这一轮,智能体把图表 HTML 写进该线程的 ``outputs/``,再由沙箱预览。
|
||||
"""
|
||||
ensure_roundtable_functional_agents()
|
||||
thread_id = await _create_thread_direct(request, metadata=_ROUNDTABLE_THREAD_METADATA)
|
||||
return {"agent_name": REPORT_AGENT_ID, "thread_id": thread_id}
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# 上游 LangGraph SSE 透传(intent / recommend / leader / special 共用)
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
@@ -15,6 +15,7 @@ from fastapi import APIRouter, HTTPException, Request
|
||||
from pydantic import BaseModel, Field, field_validator
|
||||
|
||||
from app.gateway.deps import get_checkpointer, get_embed_session_store, get_run_manager, get_thread_store
|
||||
from app.gateway.routers.thread_shares import _load_thread_snapshot
|
||||
from app.gateway.routers.threads import _derive_thread_status
|
||||
from deerflow.persistence.embed_sessions.base import EmbedSessionStore
|
||||
|
||||
@@ -93,6 +94,80 @@ class EmbedSessionStatusResponse(EmbedThreadStatusFields):
|
||||
thread_id: str | None = Field(default=None, description="Resolved thread id when registered")
|
||||
|
||||
|
||||
class EmbedSessionResultResponse(EmbedThreadStatusFields):
|
||||
session_id: str
|
||||
registered: bool = Field(description="False until the iframe has registered this session_id")
|
||||
thread_id: str | None = Field(default=None, description="Resolved thread id when registered")
|
||||
messages: list[dict] = Field(default_factory=list, description="Full serialized conversation messages")
|
||||
generated_text: str = Field(default="", description="Concatenated assistant/AI message text")
|
||||
generated_files: list[str] = Field(
|
||||
default_factory=list,
|
||||
description="Virtual artifact paths produced during the conversation (e.g. /mnt/user-data/outputs/...)",
|
||||
)
|
||||
|
||||
|
||||
class EmbedThreadResultResponse(EmbedThreadStatusFields):
|
||||
thread_id: str
|
||||
messages: list[dict] = Field(default_factory=list)
|
||||
generated_text: str = Field(default="")
|
||||
generated_files: list[str] = Field(default_factory=list)
|
||||
|
||||
|
||||
def _message_text(message: dict) -> str:
|
||||
content = message.get("content")
|
||||
if isinstance(content, str):
|
||||
return content.strip()
|
||||
if isinstance(content, list):
|
||||
parts: list[str] = []
|
||||
for part in content:
|
||||
if isinstance(part, dict) and part.get("type") == "text" and isinstance(part.get("text"), str):
|
||||
text = part["text"].strip()
|
||||
if text:
|
||||
parts.append(text)
|
||||
return "\n".join(parts)
|
||||
return ""
|
||||
|
||||
|
||||
def extract_generated_text(messages: list[dict]) -> str:
|
||||
"""Join all assistant/AI message bodies in conversation order."""
|
||||
parts: list[str] = []
|
||||
for message in messages:
|
||||
if not isinstance(message, dict):
|
||||
continue
|
||||
message_type = message.get("type") or message.get("role")
|
||||
if message_type not in {"ai", "assistant"}:
|
||||
continue
|
||||
text = _message_text(message)
|
||||
if text:
|
||||
parts.append(text)
|
||||
return "\n\n".join(parts)
|
||||
|
||||
|
||||
def build_embed_result_payload(
|
||||
*,
|
||||
messages: list[dict],
|
||||
artifacts: list[str],
|
||||
) -> tuple[list[dict], str, list[str]]:
|
||||
return messages, extract_generated_text(messages), artifacts
|
||||
|
||||
|
||||
async def _result_for_thread_id(thread_id: str, request: Request) -> EmbedThreadResultResponse:
|
||||
checkpointer = get_checkpointer(request)
|
||||
messages, artifacts = await _load_thread_snapshot(checkpointer, thread_id)
|
||||
serialized_messages, generated_text, generated_files = build_embed_result_payload(
|
||||
messages=messages,
|
||||
artifacts=artifacts,
|
||||
)
|
||||
status = await _status_for_thread_id(thread_id, request)
|
||||
return EmbedThreadResultResponse(
|
||||
thread_id=thread_id,
|
||||
messages=serialized_messages,
|
||||
generated_text=generated_text,
|
||||
generated_files=generated_files,
|
||||
**status.model_dump(),
|
||||
)
|
||||
|
||||
|
||||
async def _status_for_thread_id(thread_id: str, request: Request) -> EmbedThreadStatusFields:
|
||||
thread_store = get_thread_store(request)
|
||||
run_mgr = get_run_manager(request)
|
||||
@@ -192,3 +267,52 @@ async def get_embed_thread_status(thread_id: str, request: Request) -> EmbedThre
|
||||
"""Direct thread lookup when the parent already knows thread_id."""
|
||||
status = await _status_for_thread_id(thread_id, request)
|
||||
return EmbedThreadStatusResponse(thread_id=thread_id, **status.model_dump())
|
||||
|
||||
|
||||
@router.get(
|
||||
"/sessions/{session_id}/result",
|
||||
response_model=EmbedSessionResultResponse,
|
||||
summary="Fetch embed Q&A messages, generated text, and files by session_id",
|
||||
)
|
||||
async def get_embed_session_result(session_id: str, request: Request) -> EmbedSessionResultResponse:
|
||||
"""Return model output for iframe parents once they know ``session_id`` only."""
|
||||
session_id = _validate_session_id(session_id)
|
||||
store: EmbedSessionStore = get_embed_session_store(request)
|
||||
record = await store.get(session_id)
|
||||
if record is None:
|
||||
return EmbedSessionResultResponse(
|
||||
session_id=session_id,
|
||||
registered=False,
|
||||
thread_id=None,
|
||||
finished=False,
|
||||
thread_status="pending",
|
||||
has_active_run=False,
|
||||
latest_run_status=None,
|
||||
messages=[],
|
||||
generated_text="",
|
||||
generated_files=[],
|
||||
)
|
||||
|
||||
thread_result = await _result_for_thread_id(record["thread_id"], request)
|
||||
return EmbedSessionResultResponse(
|
||||
session_id=session_id,
|
||||
registered=True,
|
||||
thread_id=thread_result.thread_id,
|
||||
finished=thread_result.finished,
|
||||
thread_status=thread_result.thread_status,
|
||||
has_active_run=thread_result.has_active_run,
|
||||
latest_run_status=thread_result.latest_run_status,
|
||||
messages=thread_result.messages,
|
||||
generated_text=thread_result.generated_text,
|
||||
generated_files=thread_result.generated_files,
|
||||
)
|
||||
|
||||
|
||||
@router.get(
|
||||
"/threads/{thread_id}/result",
|
||||
response_model=EmbedThreadResultResponse,
|
||||
summary="Fetch embed Q&A messages, generated text, and files by thread_id",
|
||||
)
|
||||
async def get_embed_thread_result(thread_id: str, request: Request) -> EmbedThreadResultResponse:
|
||||
"""Direct thread lookup when the parent already knows thread_id."""
|
||||
return await _result_for_thread_id(thread_id, request)
|
||||
Reference in new issue
Block a user