@
feat(圆桌): Step3 三产物并行 + 大屏多页模式 + 后台挂起查看恢复流程 后台挂起 Step3: - engine 总结报告/结果绘制/大屏数据 asyncio.gather 三流并行,故障隔离 - request.state 回归修复(进程内虚拟 request 补 state),恢复后台报告 - ROUNDTABLE_JOB_GATEWAY 默认 inprocess;产物自愈重试(截断/缺失纠偏) - 大屏后端只产 report-json(dashboardJson),前端按当前主题渲染 结果绘制大屏多页模式(模式二): - report-template 模板渲染(report-json→大屏HTML),零截断、主题跟随 - 与模式一同款交互(流式+沙箱+可修改);沙箱新窗口打开按钮 最终方案展示:左侧席位栏+总控头像、交付/文件合并、收展动画、内联沙箱 后台挂起查看=恢复实时流程: - handleLoadDraft 返回 Promise,await hydrate 后订阅 streamJob SSE 跟随 - saveDraftNow 强制存(发起前+查看前各一次)防丢 第二步优化:总控判断串/并行(并行单列可写)、席位上下文+交付模板、 peerDeliveries、失败重试、共识标记、流程图小窗(缩放/拖动/并行批) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> @
This commit is contained in:
1 parent
a7596001c5
commit
d5f319d9cb
51 files changed
+5016
-1186
No files matched your search
@@ -14,6 +14,7 @@
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import logging
|
||||
import uuid
|
||||
from datetime import UTC, datetime
|
||||
@@ -26,6 +27,7 @@ from app.gateway.roundtable_run_policy import roundtable_run_policy
|
||||
from app.gateway.routers._roundtable_seed import (
|
||||
COORDINATOR_AGENT_ID,
|
||||
REPORT_AGENT_ID,
|
||||
SUMMARY_AGENT_ID,
|
||||
ensure_roundtable_functional_agents,
|
||||
)
|
||||
from deerflow.agents.roundtable_orchestrator.types import (
|
||||
@@ -33,6 +35,7 @@ from deerflow.agents.roundtable_orchestrator.types import (
|
||||
ReportResult,
|
||||
SeatRef,
|
||||
SeatTurn,
|
||||
SummaryResult,
|
||||
)
|
||||
from deerflow.config.paths import get_paths
|
||||
from deerflow.runtime.serialization import serialize_channel_values
|
||||
@@ -41,6 +44,7 @@ logger = logging.getLogger(__name__)
|
||||
|
||||
_ROUNDTABLE_THREAD_METADATA: dict[str, Any] = {"thread_type": "roundtable", "system": True}
|
||||
_REPORT_FILE = "方案总览.html"
|
||||
_SUMMARY_FILE = "方案总结报告.md"
|
||||
|
||||
# 各角色的工具裁剪 / thinking / reasoning 全部取自单一数据源 roundtable_run_policy
|
||||
# (与 multi_agent 的 _leader_run / _special_run 同源),不再在本文件硬编码——历史上
|
||||
@@ -227,6 +231,92 @@ def _read_report_html(app, thread_id: str, user_id: str) -> str | None:
|
||||
return None
|
||||
|
||||
|
||||
def _read_summary_md(app, thread_id: str, user_id: str) -> str | None:
|
||||
"""从 summary thread 的 outputs 目录读出总结报告 Markdown。
|
||||
|
||||
优先 ``方案总结报告.md``;找不到就扫目录里**任意 .md**(取体积最大的那个),
|
||||
容忍总结 agent 用了别的文件名。镜像 ``_read_report_html``。
|
||||
"""
|
||||
try:
|
||||
outputs_dir = get_paths().sandbox_outputs_dir(thread_id, user_id=user_id)
|
||||
preferred = outputs_dir / _SUMMARY_FILE
|
||||
if preferred.exists():
|
||||
return preferred.read_text(encoding="utf-8")
|
||||
if outputs_dir.exists():
|
||||
mds = sorted(
|
||||
outputs_dir.glob("*.md"),
|
||||
key=lambda p: p.stat().st_size,
|
||||
reverse=True,
|
||||
)
|
||||
if mds:
|
||||
logger.info("summary md: 方案总结报告.md not found, using %s", mds[0].name)
|
||||
return mds[0].read_text(encoding="utf-8")
|
||||
logger.warning("summary md: no .md found in %s", outputs_dir)
|
||||
except Exception: # noqa: BLE001
|
||||
logger.warning("read summary md failed for thread %s", thread_id, exc_info=True)
|
||||
return None
|
||||
|
||||
|
||||
# Step3 产物自愈重试次数(首轮之外的纠偏轮数)。
|
||||
# 实测痛点(2026-06-12,h:\xiugai\后台日志.txt):deepseek 报告 agent 说「let me write
|
||||
# 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
|
||||
|
||||
_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"
|
||||
"(最常见原因:试图一次写入整份长 HTML,输出超长被截断,工具调用没有真正落盘)。\n"
|
||||
"请立即重新写入,并**严格分段**:\n"
|
||||
f"1. 先用 write_file 写入 /mnt/user-data/outputs/{_REPORT_FILE} 的开头部分(<!doctype html> 起,约 200-300 行);\n"
|
||||
"2. 再用 write_file 的 append=true 参数分多次续写后续片段,每段同样控制在 300 行以内;\n"
|
||||
"3. 直到文档以 </html> 收尾。\n"
|
||||
"不要在对话里粘贴 HTML 源码,写完用一两句话总结即可。"
|
||||
)
|
||||
|
||||
_REPORT_RETRY_TRUNCATED = (
|
||||
f"检测到 /mnt/user-data/outputs/{_REPORT_FILE} 已写入但**不完整**(没有 </html> 收尾,疑似上一轮输出被截断)。\n"
|
||||
"请用 write_file 的 append=true 参数从中断处**继续补全剩余内容**(不要重写已有部分),"
|
||||
"每段控制在 300 行以内,分多次续写直到文档以 </html> 收尾。不要在对话里粘贴源码。"
|
||||
)
|
||||
|
||||
_SUMMARY_RETRY_MISSING = (
|
||||
f"检测到 /mnt/user-data/outputs/{_SUMMARY_FILE} 还没有生成——你上一轮的 write_file 没有执行成功"
|
||||
"(可能是单次输出过长被截断)。请立即用 write_file 重新写入完整报告;若内容很长,"
|
||||
"先写前半部分、再用 write_file 的 append=true 参数分段续写。写完后把报告正文也输出到对话回复里。"
|
||||
)
|
||||
|
||||
_DASHBOARD_FENCE = "```report-json"
|
||||
_DASHBOARD_RETRY = (
|
||||
"你刚才没有输出有效的 ```report-json 代码块(或其中的 JSON 无法解析)。"
|
||||
"请重新在回复**最末尾**输出一个完整且**严格合法**的 ```report-json 代码块"
|
||||
"(标准 JSON:无注释、无尾逗号、字符串用双引号),代码块之外只允许一两句说明;"
|
||||
"不要调用任何工具、不要写文件、不要输出 HTML。"
|
||||
)
|
||||
|
||||
|
||||
def _extract_report_json(text: str) -> str | None:
|
||||
"""从模型回复里提取最后一个 ```report-json 代码块的 JSON 字符串(须可解析)。"""
|
||||
start = text.rfind(_DASHBOARD_FENCE)
|
||||
if start < 0:
|
||||
return None
|
||||
after = text[start + len(_DASHBOARD_FENCE) :]
|
||||
close = after.find("```")
|
||||
body = (after[:close] if close >= 0 else after).strip()
|
||||
if not body:
|
||||
return None
|
||||
try:
|
||||
json.loads(body)
|
||||
except Exception: # noqa: BLE001
|
||||
return None
|
||||
return body
|
||||
|
||||
|
||||
def _html_complete(html: str | None) -> bool:
|
||||
"""报告 HTML 是否完整可用(非空 + 有 </html> 收尾,截断检测)。"""
|
||||
return bool(html and html.strip()) and "</html>" in html.lower()
|
||||
|
||||
|
||||
def _build_seat_intro(agents: list[SeatRef]) -> str:
|
||||
"""镜像 init 注入 coordinator 的「可调度席位清单」。"""
|
||||
lines = [f"- agent_name: {a.agent_id}({a.name})" for a in agents]
|
||||
@@ -328,18 +418,111 @@ class InProcessRoundtableGateway:
|
||||
return await _create_thread(self._app)
|
||||
|
||||
async def run_report(self, *, report_thread_id, prompt, model) -> ReportResult:
|
||||
"""跑报告 agent,**带产物自愈**:首轮后 HTML 缺失/截断(无 </html>)→ 同线程追加
|
||||
纠偏消息(分段 + append 续写)重试,至多 ``_ARTIFACT_RETRY_LIMIT`` 轮。"""
|
||||
policy = roundtable_run_policy("report")
|
||||
messages = await _run_agent_turn(
|
||||
self._app,
|
||||
thread_id=report_thread_id,
|
||||
agent_name=REPORT_AGENT_ID,
|
||||
message=prompt,
|
||||
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,
|
||||
)
|
||||
summary = _latest_ai_text(messages)
|
||||
html = _read_report_html(self._app, report_thread_id, self._user_id) or summary
|
||||
return ReportResult(html=html, summary=summary, model=model or self._model)
|
||||
summary = ""
|
||||
html: str | None = None
|
||||
message = prompt
|
||||
for attempt in range(1 + _ARTIFACT_RETRY_LIMIT):
|
||||
messages = await _run_agent_turn(
|
||||
self._app,
|
||||
thread_id=report_thread_id,
|
||||
agent_name=REPORT_AGENT_ID,
|
||||
message=message,
|
||||
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,
|
||||
)
|
||||
summary = _latest_ai_text(messages) or summary
|
||||
html = _read_report_html(self._app, report_thread_id, self._user_id)
|
||||
if _html_complete(html):
|
||||
break
|
||||
if attempt >= _ARTIFACT_RETRY_LIMIT:
|
||||
break
|
||||
message = _REPORT_RETRY_TRUNCATED if (html and html.strip()) else _REPORT_RETRY_MISSING
|
||||
logger.warning(
|
||||
"report html %s after attempt %d for thread %s; retrying with corrective prompt",
|
||||
"truncated" if (html and html.strip()) else "missing",
|
||||
attempt + 1,
|
||||
report_thread_id,
|
||||
)
|
||||
# 截断的 HTML 也比纯文字强(浏览器容忍未闭合标签),仅在完全没有文件时才回退正文。
|
||||
final_html = html if (html and html.strip()) else summary
|
||||
return ReportResult(html=final_html, summary=summary, model=model or self._model)
|
||||
|
||||
async def create_summary_thread(self) -> str:
|
||||
ensure_roundtable_functional_agents()
|
||||
return await _create_thread(self._app)
|
||||
|
||||
async def run_summary(self, *, summary_thread_id, prompt, model) -> SummaryResult:
|
||||
"""进程内驱动 roundtable-summary 写「方案总结报告.md」+ 对话正文(与前端
|
||||
useStep3Summary 同一智能体 / 同一 run policy / 同款产物),返回与
|
||||
DraftStep3SummarySnapshot 对齐的结果。线程是真实持久线程——加载草稿后可续问。
|
||||
|
||||
带产物自愈:md 文件缺失时同线程追加纠偏消息重试(同 ``run_report``)。"""
|
||||
policy = roundtable_run_policy("summary")
|
||||
closing = ""
|
||||
md: str | None = None
|
||||
message = prompt
|
||||
for attempt in range(1 + _ARTIFACT_RETRY_LIMIT):
|
||||
messages = await _run_agent_turn(
|
||||
self._app,
|
||||
thread_id=summary_thread_id,
|
||||
agent_name=SUMMARY_AGENT_ID,
|
||||
message=message,
|
||||
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,
|
||||
)
|
||||
closing = _latest_ai_text(messages) or closing
|
||||
md = _read_summary_md(self._app, summary_thread_id, self._user_id)
|
||||
if md and md.strip():
|
||||
break
|
||||
if attempt >= _ARTIFACT_RETRY_LIMIT:
|
||||
break
|
||||
message = _SUMMARY_RETRY_MISSING
|
||||
logger.warning(
|
||||
"summary md missing after attempt %d for thread %s; retrying with corrective prompt",
|
||||
attempt + 1,
|
||||
summary_thread_id,
|
||||
)
|
||||
final_md = md if (md and md.strip()) else closing
|
||||
return SummaryResult(md=final_md, closing=closing, model=model or self._model, thread_id=summary_thread_id)
|
||||
|
||||
async def run_dashboard_data(self, *, prompt, model) -> str | None:
|
||||
"""大屏多页(模式二)的结构化数据:进程内驱动 roundtable-report 输出
|
||||
```report-json(prompt 已声明禁工具/禁文件),返回可解析的 JSON 字符串。
|
||||
缺失 / 坏 JSON → 同线程纠偏重试(同 run_report/run_summary 的自愈语义)。"""
|
||||
ensure_roundtable_functional_agents()
|
||||
thread_id = await _create_thread(self._app)
|
||||
policy = roundtable_run_policy("report")
|
||||
message = prompt
|
||||
for attempt in range(1 + _ARTIFACT_RETRY_LIMIT):
|
||||
messages = await _run_agent_turn(
|
||||
self._app,
|
||||
thread_id=thread_id,
|
||||
agent_name=REPORT_AGENT_ID,
|
||||
message=message,
|
||||
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,
|
||||
)
|
||||
body = _extract_report_json(_latest_ai_text(messages))
|
||||
if body:
|
||||
return body
|
||||
if attempt >= _ARTIFACT_RETRY_LIMIT:
|
||||
break
|
||||
message = _DASHBOARD_RETRY
|
||||
logger.warning(
|
||||
"dashboard report-json missing/invalid after attempt %d for thread %s; retrying",
|
||||
attempt + 1,
|
||||
thread_id,
|
||||
)
|
||||
return None
|
||||
@@ -16,9 +16,10 @@ from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
import logging
|
||||
from dataclasses import dataclass, field
|
||||
from collections.abc import Callable
|
||||
from dataclasses import dataclass
|
||||
from datetime import UTC, datetime
|
||||
from typing import Any, Callable
|
||||
from typing import Any
|
||||
|
||||
from deerflow.agents.roundtable_orchestrator import (
|
||||
JobProgressWriter,
|
||||
@@ -29,14 +30,13 @@ from deerflow.runtime.user_context import reset_current_user, set_current_user
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# ⚠️ 临时禁用「后台挂起」的 Step3 结果绘制(报告 HTML 生成)。
|
||||
# 进程内驱动报告 agent 在真模型环境下仍无法稳定写出 HTML(前端手动调
|
||||
# /api/multi-agent/report/init + run/stream 可以,但后台 start_run 路径不行,
|
||||
# 根因尚未定位)。临时方案:后台只跑完 Step2 研讨即收尾,不自动画报告——用户需要
|
||||
# 报告时可在前端手动跑 Step3(那条路是通的)。
|
||||
# 待修复后把本开关改回 True 即恢复后台自动报告。
|
||||
# 详见 frontend-web/docs/multi-agent-background-run-遗留问题与下次继续.md「后台报告 HTML」。
|
||||
ROUNDTABLE_BACKGROUND_REPORT_ENABLED = False
|
||||
# 「后台挂起」的 Step3 自动产出(方案总结报告 md + 结果绘制 HTML)总开关。
|
||||
# 历史:2026-06-05 因后台 start_run 路径报告不稳定临时禁用;2026-06-12 定位到根因是
|
||||
# ``_enforce_agent_access`` 对进程内虚拟 request 的 ``request.state`` 裸访问(AttributeError
|
||||
# 杀死编排,见 tests/test_roundtable_inprocess_request.py),修复后恢复开启。
|
||||
# 现 Step2 收口后自动跑:① roundtable-summary 写「方案总结报告.md」② roundtable-report
|
||||
# 画「方案总览.html」,两段故障隔离(engine._run_step3),产物按手动同款形状写回草稿。
|
||||
ROUNDTABLE_BACKGROUND_REPORT_ENABLED = True
|
||||
|
||||
|
||||
class _JobUser:
|
||||
@@ -176,8 +176,8 @@ class StartParams:
|
||||
# dag 模式分层编排计划(前端 OrchestrationPlan dict):驱动引擎 _run_dag + 写进草稿 run
|
||||
# 供前端流程图渲染。None = 非 dag(引擎按 mode 回退 chain/recommend)。
|
||||
orchestration_plan: dict[str, Any] | None = None
|
||||
# 是否生成 Step3 报告。默认沿用全局开关(当前临时禁用,见 ROUNDTABLE_BACKGROUND_REPORT_ENABLED)。
|
||||
# 测试可显式传 True 覆盖以验证报告链路。
|
||||
# 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
|
||||
|
||||
|
||||
|
||||
@@ -58,6 +58,11 @@ _FUNCTIONAL_AGENT_IDS: tuple[str, ...] = (
|
||||
# seeder 会复制资源目录里的这些文件(存在即复制,缺失则跳过)。
|
||||
_SEED_FILES: tuple[str, ...] = ("config.yaml", "SOUL.md")
|
||||
|
||||
# 内置「知识库检索」技能:素材收集专家通用检索的默认技能。skills/ 目录被
|
||||
# .gitignore 排除,所以与 agent 目录一样需要 seeder 在缺失时从资源补建。
|
||||
KNOWLEDGE_SEARCH_SKILL_NAME = "knowledge-base-search"
|
||||
_SKILL_ASSETS_DIR = _ASSETS_DIR / "skills"
|
||||
|
||||
|
||||
def _ensure_one(agent_id: str) -> bool:
|
||||
"""Copy the verified SOUL.md + config.yaml for one functional agent into
|
||||
@@ -91,10 +96,69 @@ def _ensure_one(agent_id: str) -> bool:
|
||||
return True
|
||||
|
||||
|
||||
def _skills_public_dir() -> Path:
|
||||
"""skills/public 物理目录(config.yaml skills.path 解析);独立成函数便于测试替身。"""
|
||||
from deerflow.skills.storage import get_or_new_skill_storage
|
||||
|
||||
return Path(get_or_new_skill_storage().get_skills_root_path()) / "public"
|
||||
|
||||
|
||||
def _ensure_default_search_skill() -> bool:
|
||||
"""Seed the built-in knowledge-base-search skill into ``skills/public/``
|
||||
when missing. Never overwrites an existing skill directory.
|
||||
|
||||
Returns True if the skill was just created.
|
||||
"""
|
||||
src = _SKILL_ASSETS_DIR / KNOWLEDGE_SEARCH_SKILL_NAME / "SKILL.md"
|
||||
if not src.exists():
|
||||
logger.error("Missing seed asset for skill '%s' (expected at %s)", KNOWLEDGE_SEARCH_SKILL_NAME, src)
|
||||
return False
|
||||
|
||||
skill_dir = _skills_public_dir() / KNOWLEDGE_SEARCH_SKILL_NAME
|
||||
if skill_dir.exists():
|
||||
return False
|
||||
skill_dir.mkdir(parents=True, exist_ok=True)
|
||||
(skill_dir / "SKILL.md").write_bytes(src.read_bytes())
|
||||
return True
|
||||
|
||||
|
||||
def _ensure_researcher_default_skills() -> bool:
|
||||
"""给**已存在**的 researcher agent config.yaml 补默认技能配置(就地升级)。
|
||||
|
||||
seeder 对已存在目录绝不整体覆盖,但旧版种子没有 ``skills`` 字段——通用
|
||||
检索改为技能驱动后,老部署的智能体管理页会显示"未配置技能"(运行时虽有
|
||||
knowledge-base-search 兜底,但用户看不见、也没法在它基础上增删)。只在
|
||||
``skills`` 键**完全缺失**时补默认值;用户显式改过(包括清空成 ``[]``)
|
||||
的配置不动。Returns True if the file was patched.
|
||||
"""
|
||||
import yaml
|
||||
|
||||
config_file = get_paths().agent_dir(RESEARCHER_AGENT_ID) / "config.yaml"
|
||||
if not config_file.exists():
|
||||
return False
|
||||
try:
|
||||
data = yaml.safe_load(config_file.read_text(encoding="utf-8")) or {}
|
||||
except yaml.YAMLError:
|
||||
logger.warning("researcher config.yaml unparsable, skip default-skills patch")
|
||||
return False
|
||||
if not isinstance(data, dict) or "skills" in data:
|
||||
return False
|
||||
data["skills"] = [KNOWLEDGE_SEARCH_SKILL_NAME]
|
||||
config_file.write_text(
|
||||
yaml.safe_dump(data, allow_unicode=True, sort_keys=False),
|
||||
encoding="utf-8",
|
||||
)
|
||||
return True
|
||||
|
||||
|
||||
def ensure_ai_writing_functional_agents() -> list[str]:
|
||||
"""Seed the four ai-writing functional agents if their on-disk directories
|
||||
are missing. Idempotent — fast no-op when all exist.
|
||||
|
||||
Also seeds the built-in knowledge-base-search skill (the researcher's
|
||||
default retrieval skill) and patches a pre-existing researcher config.yaml
|
||||
that predates the ``skills`` field.
|
||||
|
||||
Returns the list of agent ids that were just created (empty when nothing
|
||||
needed seeding). Errors are logged but never raised — a partial failure
|
||||
must not block the request; the graph node falls back to the legacy
|
||||
@@ -109,6 +173,16 @@ def ensure_ai_writing_functional_agents() -> list[str]:
|
||||
logger.exception("Failed to seed ai-writing functional agent '%s'", agent_id)
|
||||
if created:
|
||||
logger.info("Seeded missing ai-writing functional agents: %s", ", ".join(created))
|
||||
try:
|
||||
if _ensure_default_search_skill():
|
||||
logger.info("Seeded built-in skill '%s'", KNOWLEDGE_SEARCH_SKILL_NAME)
|
||||
except Exception:
|
||||
logger.exception("Failed to seed built-in skill '%s'", KNOWLEDGE_SEARCH_SKILL_NAME)
|
||||
try:
|
||||
if _ensure_researcher_default_skills():
|
||||
logger.info("Patched researcher config.yaml with default skills")
|
||||
except Exception:
|
||||
logger.exception("Failed to patch researcher default skills")
|
||||
return created
|
||||
|
||||
|
||||
|
||||
+5
-1
@@ -1,3 +1,7 @@
|
||||
description: AI 写作流程的素材收集专家,负责为写作任务生成检索关键词、并对检索回来的素材按相关度排序筛选
|
||||
description: AI 写作流程的素材收集专家,负责为写作任务生成检索关键词、用配置的技能编排检索、并对检索回来的素材按相关度排序筛选
|
||||
id: ai-writing-researcher
|
||||
name: 素材收集专家
|
||||
# 通用检索使用的技能(按顺序编排)。默认配置内置「知识库检索」;
|
||||
# admin/用户可在智能体管理页替换为其它检索技能或追加多个。
|
||||
skills:
|
||||
- knowledge-base-search
|
||||
+16
-7
@@ -25,9 +25,10 @@
|
||||
## 产出要求(最重要)
|
||||
|
||||
### ⓪ 先判断模式:用户是否要求了结构化输出 / 流程图
|
||||
当用户**明确要求**「结构化输出 / 结构化数据 / 流程图 / 格式化输出 / 可视化流程」时(要求可能写在业务链总控编排提示里,也可能是对话中的新消息),本轮进入**流程图模式**:
|
||||
当用户**要求**「结构化输出 / 结构化数据 / 流程图 / 格式化输出 / 可视化流程」之类时(要求可能写在业务链总控编排提示里,也可能是对话中的新消息;**按语义理解,不要求逐字命中这些词**),本轮进入**流程图模式**:
|
||||
- **跳过下面的 A**——不要调用 `write_file`、不要生成任何 .md 文件;
|
||||
- 把**精炼的文字版总结报告**直接写在对话回复里(文字展示给用户读,结构遵循用户要求);
|
||||
- 文字版总结的**默认结构**(用户只要求了结构化/格式化、**没指定**具体结构时):**对各席位交付内容逐席位格式化输出**——每个席位一节:`席位名` → 核心结论(一句话)→ 关键交付要点(编号列表,逐条提炼)→ 风险/建议(如有),最后加一节「总控共识」。用户指定了结构就完全按用户的来;
|
||||
- 在回复**最末尾必须**输出一个完整的 ```flow-json 代码块(契约见「结构化流程数据」节)——这是本轮的**主产物**,绝不能省略。
|
||||
|
||||
未进入流程图模式时,每一轮都要做 A + B **两件事**,缺一不可:
|
||||
@@ -77,27 +78,33 @@
|
||||
- 用户没有要求结构化 / 流程图时,**不要**输出该块。
|
||||
|
||||
## 多轮交互(用户会在输入框里继续和你对话)
|
||||
用户读完报告 / 流程图后,可能在输入框里继续给你发消息。你要**自己判断**意图——报告和流程图都是可修改的,分四种情况处理:
|
||||
用户读完报告 / 流程图后,可能在输入框里继续给你发消息。你要**按语义自己判断**意图——**绝不要求用户使用特定关键词或固定句式**。用户不会说「请重新结构化输出」这种话,他只会说「把验收环节加上」「改一下」「太长了精简点」「不对,重来」——这些都是有效的修改指令,你必须直接理解并动手,不要反问用户「您是想要结构化输出吗」。
|
||||
|
||||
1. **修改报告**(如「把执行路线写得更细」「补一节预算分析」「风险部分太长,精简」):
|
||||
**意图判断的两条核心规则**:
|
||||
- **修改类消息默认作用于上一轮的主产物**:用户没说明改哪个产物时——上一轮你输出的是流程图(flow-json),那「加一个环节」「调整一下」「精简点」就是改流程图;上一轮主产物是 .md 报告,就是改报告。不要因为用户没提「流程图 / 结构化」字眼就把改流程图的意图当成改报告(反之亦然)。
|
||||
- **意图实在太模糊、无法判断要改什么**(如只发「不行」「再改改」「不是这个意思」,且上下文也推不出指向)→ **不要修改任何产物、不要乱猜**,在对话正文里向用户提出一个**简短的澄清问题**并给出可选方向(如「您是想调整流程图的节点结构,还是修改文字报告的某个章节?可以具体说说想改哪里」),等用户回复后再动手。注意:澄清就写在正文里(你没有专门的澄清工具)。
|
||||
|
||||
报告和流程图都是可修改的,分四种情况处理:
|
||||
|
||||
1. **修改报告**(如「把执行路线写得更细」「补一节预算分析」「风险部分太长,精简」,或上一轮主产物是报告时的笼统修改意见):
|
||||
- 在**原报告基础上做增量修改**——只改用户提到的地方,其余内容保持不变;
|
||||
- 用 `write_file` **整文件重写**到同一路径 `/mnt/user-data/outputs/方案总结报告.md`(`str_replace` 已被禁用,整文件重写才能让右侧预览实时刷新);
|
||||
- 然后在对话回复里给出修改后的完整报告正文;
|
||||
- 若此前已经输出过流程图、且这次修改影响了流程结构,可在回复末尾顺带输出更新后的完整 ```flow-json 块。
|
||||
|
||||
2. **修改流程图**(如「流程图里加上验收环节」「把风险节点拆成两个」「换成按时间阶段排」):
|
||||
2. **修改流程图**(如「流程图里加上验收环节」「把风险节点拆成两个」「换成按时间阶段排」;**只要上一轮主产物是流程图,笼统的修改意见也默认按本条处理**,无需用户提到「流程图 / 结构化」字眼):
|
||||
- **不要**改写 .md 报告文件;
|
||||
- 在对话里简短说明改了什么,并在回复**最末尾**输出**完整修订版** ```flow-json 块——必须是**全量数据**(所有节点 + 连线),不是只给改动部分,前端会整体重绘。
|
||||
|
||||
3. **要求结构化输出 / 流程图**(如「总结成格式化输出」「画个流程图」「给我结构化数据」):
|
||||
- 进入流程图模式:**不写 / 不改 .md 文件**,正文给一段简短的文字说明,末尾输出完整 ```flow-json 块;
|
||||
3. **要求结构化输出 / 流程图**(如「总结成格式化输出」「画个流程图」「给我结构化数据」,或语义等价的说法):
|
||||
- 进入流程图模式:**不写 / 不改 .md 文件**,正文按第 ⓪ 条的默认结构(逐席位格式化各席位交付内容;用户指定结构则按用户的)给出文字总结,末尾输出完整 ```flow-json 块;
|
||||
- 已有的 .md 报告**不要动**——用户随时可以在前端切回去看它。
|
||||
|
||||
4. **就报告 / 研讨内容提问**(如「为什么没采用方案 B?」「风险最高的是哪一项?」):
|
||||
- **只在对话里用文字回答**,依据已有的研讨原料和报告内容作答;
|
||||
- **不要**改写报告文件、**不要**重新 `write_file`、**不要**输出 flow-json。
|
||||
|
||||
判断不准时,倾向于「先简短回答,再问用户是否需要据此更新报告 / 流程图」。
|
||||
修改意图明确但细节有歧义时(能判断改哪个产物、只是改法有多种)→ 按最合理的理解直接改,并在回复里说明你的理解,邀请用户继续纠正;完全判断不了才走上面的「澄清提问」。
|
||||
|
||||
## 禁止事项
|
||||
- ❌ 生成 HTML / 绘制可视化页面 / 画图表(只写 Markdown 文本)
|
||||
@@ -107,6 +114,8 @@
|
||||
- ❌ 写完文件后只回一句「已写入文件」而不给出报告正文(用户要在对话里读到报告)
|
||||
- ❌ 用户只是提问时还去改写报告文件
|
||||
- ❌ 用户要求了结构化输出 / 流程图,却只写 .md 报告而不输出 flow-json 块
|
||||
- ❌ 要求用户必须说出「结构化输出 / 流程图」等关键词才肯修改流程图(上一轮主产物是流程图时,笼统的修改意见就是改流程图)
|
||||
- ❌ 用户意图模糊时不提澄清问题、自行乱猜乱改一通
|
||||
- ❌ 流程图模式下仍调用 `write_file` 生成 / 改写 .md 文件
|
||||
- ❌ 修订流程图时只输出改动的节点(必须输出全量 flow-json,前端整体重绘)
|
||||
- ❌ 在报告任何位置添加「AI 自动生成」「AI generated」之类的标签、徽章或水印
|
||||
|
||||
@@ -29,11 +29,10 @@ from fastapi import APIRouter, HTTPException, Request
|
||||
from fastapi.responses import StreamingResponse
|
||||
from pydantic import BaseModel, Field
|
||||
|
||||
from app.gateway.roundtable_job_executor import StartParams
|
||||
from deerflow.agents.roundtable_orchestrator import SeatRef
|
||||
from deerflow.runtime.user_context import get_effective_user_id
|
||||
|
||||
from app.gateway.roundtable_job_executor import StartParams
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
router = APIRouter(prefix="/api/roundtable-jobs", tags=["roundtable-jobs"])
|
||||
|
||||
@@ -182,6 +181,15 @@ async def start_job(request: Request, body: StartJobRequest) -> RoundtableJobRes
|
||||
if not body.agents:
|
||||
raise HTTPException(status_code=400, detail="agents is required")
|
||||
|
||||
# 防重复挂起(服务端幂等闸门):同一草稿已有未终态作业(queued/running/awaiting_input)
|
||||
# 时直接返回它、不再新建。前端的「手动挂起按钮 / 离开页面自动挂起 / beforeunload 刷新
|
||||
# 兜底」可能对同一研讨多次触发本端点 —— 客户端去重只能挡住同页内的重复,跨触发源 /
|
||||
# 跨标签页的并发只能靠这里兜住,否则同一研讨会被双跑。
|
||||
if body.draftId:
|
||||
existing = await store.get_by_draft(body.draftId, user_id=user_id)
|
||||
if existing is not None and (existing.get("status") or "") not in _TERMINAL:
|
||||
return _to_response(existing)
|
||||
|
||||
agents = [SeatRef(a.agentId, a.name or a.agentId) for a in body.agents]
|
||||
job_id = uuid4().hex
|
||||
row = await store.create(
|
||||
|
||||
Reference in new issue
Block a user