deerflow-code/offline-backend-20260512/backend/app/gateway/routers/roundtable_chains.py

"""CRUD API for roundtable-planning business chains (业务链条).

A business chain is a named, reusable, *ordered* list of agents the user
assembles in the chain editor. In Step 2 the coordinator dispatches to the seats
following this order. Bound to the requesting user (not the browser), so it
follows the account across devices/sessions — same model as roundtable drafts.

Routes (prefix ``/api/roundtable-chains``):
    GET    ""               list current user's chains (full records, incl. seats)
    POST   ""               create a chain
    GET    "/{chain_id}"    one chain
    PUT    "/{chain_id}"    partial update (title / description / seats)
    DELETE "/{chain_id}"    delete
"""

from __future__ import annotations

import logging
from datetime import datetime
from uuid import uuid4

from fastapi import APIRouter, Depends, HTTPException, Request
from langchain_core.messages import HumanMessage, SystemMessage
from pydantic import BaseModel, Field

from app.gateway.deps import get_config, get_local_provider
from deerflow.config.app_config import AppConfig
from deerflow.models import create_chat_model
from deerflow.runtime.user_context import get_effective_user_id

logger = logging.getLogger(__name__)
router = APIRouter(prefix="/api/roundtable-chains", tags=["roundtable-chains"])

# Seat-count bounds mirror the roundtable selection bounds (2 ≤ N ≤ 8).
MIN_SEATS = 2
MAX_SEATS = 8


# ── schemas ────────────────────────────────────────────────────────────────


class ChainSeatSchema(BaseModel):
    agent_id: str = Field(min_length=1, max_length=128)
    name: str = ""
    description: str = ""


class ChainResponse(BaseModel):
    id: str
    title: str = ""
    description: str | None = None
    seats: list[ChainSeatSchema] = []
    # Optional layered orchestration: list of stages, each a list of agent_ids
    # run in parallel (stages run serially). None/absent = linear chain.
    stages: list[list[str]] | None = None
    # Optional per-stage goals aligned to ``stages`` (dag-only). None = no goals.
    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）。
    is_mine: bool = True
    # 配置者（owner）的 user_id 及解析后的可读用户名（email）。owner_name 由路由层
    # 用 LocalAuthProvider 解析；解析失败时为 None，前端回退到短 id 展示。
    owner_id: str | None = None
    owner_name: str | None = None
    created_at: datetime | str | None = None
    updated_at: datetime | str | None = None


class ChainListResponse(BaseModel):
    chains: list[ChainResponse]


class ChainCreateRequest(BaseModel):
    # Optional client-provided id so the browser can keep one stable id across
    # the create + subsequent updates without an extra round-trip.
    id: str | None = Field(default=None, max_length=64)
    title: str = Field(default="", max_length=512)
    description: str | None = Field(default=None, max_length=2048)
    seats: list[ChainSeatSchema] = []
    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):
    # All optional → callers send only what changed.
    title: str | None = Field(default=None, max_length=512)
    description: str | None = Field(default=None, max_length=2048)
    seats: list[ChainSeatSchema] | None = None
    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


# ── helpers ──────────────────────────────────────────────────────────────


def _current_user_id(request: Request) -> str:
    user = getattr(request.state, "user", None)
    if user is not None:
        return str(user.id)
    return get_effective_user_id()


async def _enrich_owner_names(rows: list[dict]) -> None:
    """把每行的 ``owner_id`` 解析成可读用户名（email），写入 ``owner_name``。

    与 knowledge 路由的 ``_enrich_actor_names`` 同思路：批量解析、按 id 缓存、best-effort
    （解析失败 / provider 不可用时 ``owner_name`` 留空，前端回退到短 id）。
    """
    try:
        provider = get_local_provider()
    except Exception:
        return
    cache: dict[str, str | None] = {}
    for row in rows:
        uid = row.get("owner_id")
        if not uid:
            row["owner_name"] = None
            continue
        if uid not in cache:
            try:
                user = await provider.get_user(uid)
                cache[uid] = user.email if user else None
            except Exception:
                cache[uid] = None
        row["owner_name"] = cache[uid]


def _get_store(request: Request):
    store = getattr(request.app.state, "roundtable_chain_store", None)
    if store is None:
        raise HTTPException(status_code=503, detail="Roundtable chain store not available")
    return store


def _validate_seats(seats: list[ChainSeatSchema]) -> None:
    """Enforce 2 ≤ N ≤ 8 and reject duplicate agent ids within a chain."""
    if len(seats) < MIN_SEATS or len(seats) > MAX_SEATS:
        raise HTTPException(
            status_code=400,
            detail=f"业务链条席位数须在 {MIN_SEATS}–{MAX_SEATS} 之间（当前 {len(seats)}）",
        )
    seen: set[str] = set()
    for s in seats:
        if s.agent_id in seen:
            raise HTTPException(status_code=400, detail=f"业务链条存在重复智能体: {s.agent_id}")
        seen.add(s.agent_id)


def _validate_stages(seats: list[ChainSeatSchema], stages: list[list[str]] | None) -> None:
    """Validate the optional layering. ``stages`` must partition the seats.

    Rules (None = linear chain, always valid):
      - at least one stage, no empty stage;
      - every agent_id in a stage must be one of the chain's seats;
      - no agent_id appears twice (within or across stages);
      - every seat must appear in exactly one stage (full coverage), so the
        stored layering and the seat list never drift apart.
    """
    if stages is None:
        return
    if len(stages) == 0:
        raise HTTPException(status_code=400, detail="编排分层不能为空（至少一个阶段）")
    seat_ids = {s.agent_id for s in seats}
    seen: set[str] = set()
    for idx, stage in enumerate(stages):
        if not stage:
            raise HTTPException(status_code=400, detail=f"第 {idx + 1} 阶段为空，请删除或放入席位")
        for agent_id in stage:
            if agent_id not in seat_ids:
                raise HTTPException(
                    status_code=400,
                    detail=f"编排分层包含不在席位列表中的智能体: {agent_id}",
                )
            if agent_id in seen:
                raise HTTPException(
                    status_code=400,
                    detail=f"智能体 {agent_id} 在编排分层中重复出现，每个席位只能跑一次",
                )
            seen.add(agent_id)
    if seen != seat_ids:
        missing = seat_ids - seen
        raise HTTPException(
            status_code=400,
            detail=f"编排分层未覆盖全部席位，缺少: {', '.join(sorted(missing))}",
        )


# Per-goal length cap to keep injected prompts bounded.
MAX_STAGE_GOAL_CHARS = 1000


def _validate_stage_goals(
    stages: list[list[str]] | None, stage_goals: list[str | None] | None
) -> None:
    """Validate the optional per-stage goals (None = no goals, always valid).

    Rules:
      - goals only make sense alongside a non-null layering (dag-only): reject
        goals sent without ``stages``;
      - length must match ``stages`` exactly (positional alignment);
      - each goal (when present) must be ≤ ``MAX_STAGE_GOAL_CHARS`` chars.
    """
    if stage_goals is None:
        return
    if stages is None:
        raise HTTPException(status_code=400, detail="阶段目标需与编排分层一起提交（线性链不支持阶段目标）")
    if len(stage_goals) != len(stages):
        raise HTTPException(
            status_code=400,
            detail=f"阶段目标数量（{len(stage_goals)}）与编排分层阶段数（{len(stages)}）不一致",
        )
    for idx, goal in enumerate(stage_goals):
        if goal is not None and len(goal) > MAX_STAGE_GOAL_CHARS:
            raise HTTPException(
                status_code=400,
                detail=f"第 {idx + 1} 阶段目标过长（上限 {MAX_STAGE_GOAL_CHARS} 字）",
            )


_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 ───────────────────────────────────────────────────────────────


@router.get("", response_model=ChainListResponse)
async def list_chains(request: Request, scope: str = "mine") -> ChainListResponse:
    """列出业务链条。

    ``scope`` 查询参数(筛选):
      - ``mine``（默认）：仅当前用户自己的链条；
      - ``public``：所有**已发布**的公共链条（任何账号均可查看 / 选入研讨）。
    其余值按 ``mine`` 处理。
    """
    normalized_scope = scope if scope in ("mine", "public") else "mine"
    store = _get_store(request)
    user_id = _current_user_id(request)
    rows = await store.list_chains(user_id, scope=normalized_scope)
    await _enrich_owner_names(rows)
    return ChainListResponse(chains=[ChainResponse(**r) for r in rows])


@router.post("", response_model=ChainResponse, status_code=201)
async def create_chain(request: Request, body: ChainCreateRequest) -> ChainResponse:
    _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(
        user_id,
        {
            "id": (body.id or uuid4().hex),
            "title": body.title.strip(),
            "description": body.description,
            "seats": [s.model_dump() for s in body.seats],
            "stages": body.stages,
            "stage_goals": body.stage_goals,
            "coordinator_prompt": body.coordinator_prompt,
            "seat_mode": body.seat_mode,
        },
    )
    await _enrich_owner_names([row])
    return ChainResponse(**row)


@router.get("/{chain_id}", response_model=ChainResponse)
async def get_chain(request: Request, chain_id: str) -> ChainResponse:
    store = _get_store(request)
    user_id = _current_user_id(request)
    row = await store.get_chain(chain_id, user_id)
    if row is None:
        raise HTTPException(status_code=404, detail="Chain not found")
    await _enrich_owner_names([row])
    return ChainResponse(**row)


@router.put("/{chain_id}", response_model=ChainResponse)
async def update_chain(request: Request, chain_id: str, body: ChainUpdateRequest) -> ChainResponse:
    if body.seats is not None:
        _validate_seats(body.seats)
    # A non-null stages layering can only be validated against seats; the editor
    # always saves both together. Reject a layering update that omits seats.
    if body.stages is not None:
        if body.seats is None:
            raise HTTPException(status_code=400, detail="更新编排分层时必须同时提交席位列表")
        _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.
    fields = body.model_dump(exclude_unset=True)
    row = await store.update_chain(chain_id, user_id, **fields)
    if row is None:
        raise HTTPException(status_code=404, detail="Chain not found")
    await _enrich_owner_names([row])
    return ChainResponse(**row)


@router.delete("/{chain_id}", status_code=204)
async def delete_chain(request: Request, chain_id: str) -> None:
    store = _get_store(request)
    user_id = _current_user_id(request)
    deleted = await store.delete_chain(chain_id, user_id)
    if not deleted:
        raise HTTPException(status_code=404, detail="Chain not found")


# ── 提示词优化（总控提示词 / 阶段目标）──────────────────────────────────────
# 链条编辑器的「AI 优化」按钮：把用户当前写的提示词 + 链条结构发来，单次直连
# LLM 重写后返回全文（同 suggestions 路由的直调模式，不走 agent / loopback）。


class OptimizePromptRequest(BaseModel):
    prompt_text: str = Field(min_length=1, max_length=8000, description="用户当前写的提示词原文")
    kind: str = Field(default="coordinator", description="coordinator=总控提示词 | stage_goal=阶段目标")
    context: str = Field(default="", max_length=8000, description="链条结构等上下文（标题 / 各 Stage 席位 / 目标）")
    model_name: str | None = Field(default=None, description="可选模型覆盖，缺省用 config models[0]")


class OptimizePromptResponse(BaseModel):
    optimized: str


_OPTIMIZE_SYSTEM_COORDINATOR = (
    "你是「多智能体业务链条」的提示词优化专家。用户在业务链条编辑器里给总控（协调）智能体写了一段编排提示词——"
    "总控会按它逐阶段给各席位智能体派活；第 3 步的方案总结报告也会把它当作最高优先级口径（决定总结的结构与重点）。\n\n"
    "请在保留用户原意的前提下重写这段提示词，让它更明确、更可执行：\n"
    "1. 明确派活方式：各阶段先后顺序、每阶段的产出目标、对上一阶段结论的承接要求；\n"
    "2. 明确交付口径：各席位交付聚焦本职、避免重复劳动，要求显式给出结论而非泛泛而谈；\n"
    "3. 若原文流露了对最终总结的期望（输出结构、重点席位、风格详略），把它整理成明确条目；若没有，不要凭空替用户添加；\n"
    "4. 语言精炼、命令式、分点表述。\n\n"
    "硬性要求：只输出优化后的提示词全文（纯文本，可用换行与编号），不要任何前后缀说明、不要 markdown 代码块；"
    "全程简体中文；总长不超过 800 字。"
)

_OPTIMIZE_SYSTEM_STAGE_GOAL = (
    "你是「多智能体业务链条」的提示词优化专家。用户在链条编辑器里为某一个 Stage（阶段）写了「本阶段目标」——"
    "总控按它给本阶段的席位派活，并要求本阶段交付承接上一阶段产出。\n\n"
    "请在保留用户原意的前提下重写这段阶段目标，让它更明确、更可执行：点明本阶段要产出什么（交付物 / 结论形态）、"
    "怎么承接上一阶段产出、本阶段各席位分别聚焦什么（如上下文给了席位名单）。语言精炼，两到四句话即可。\n\n"
    "硬性要求：只输出优化后的阶段目标全文（纯文本），不要任何前后缀说明、不要 markdown 代码块；"
    "全程简体中文；总长不超过 200 字。"
)


def _extract_llm_text(content: object) -> str:
    if isinstance(content, str):
        return content
    if isinstance(content, list):
        parts: list[str] = []
        for block in content:
            if isinstance(block, str):
                parts.append(block)
            elif isinstance(block, dict) and block.get("type") in {"text", "output_text"}:
                text = block.get("text")
                if isinstance(text, str):
                    parts.append(text)
        return "\n".join(parts) if parts else ""
    if content is None:
        return ""
    return str(content)


def _strip_code_fence(text: str) -> str:
    stripped = text.strip()
    if not stripped.startswith("```"):
        return stripped
    lines = stripped.splitlines()
    if len(lines) >= 3 and lines[0].startswith("```") and lines[-1].startswith("```"):
        return "\n".join(lines[1:-1]).strip()
    return stripped


@router.post("/optimize-prompt", response_model=OptimizePromptResponse)
async def optimize_chain_prompt(
    body: OptimizePromptRequest,
    config: AppConfig = Depends(get_config),
) -> OptimizePromptResponse:
    system = _OPTIMIZE_SYSTEM_STAGE_GOAL if body.kind == "stage_goal" else _OPTIMIZE_SYSTEM_COORDINATOR
    user_parts: list[str] = []
    if body.context.strip():
        user_parts.append(f"【业务链条结构】\n{body.context.strip()}")
    user_parts.append(f"【待优化的提示词原文】\n{body.prompt_text.strip()}")
    user_parts.append("请输出优化后的提示词全文。")
    try:
        model = create_chat_model(name=body.model_name, thinking_enabled=False, app_config=config)
        response = await model.ainvoke(
            [SystemMessage(content=system), HumanMessage(content="\n\n".join(user_parts))],
            config={"run_name": "chain_prompt_optimize"},
        )
        text = _strip_code_fence(_extract_llm_text(response.content))
    except Exception as exc:  # 模型 / 网络故障 → 502，前端 toast 原样展示
        logger.exception("chain prompt optimize failed: %s", exc)
        raise HTTPException(status_code=502, detail=f"大模型调用失败：{exc}") from exc
    if not text:
        raise HTTPException(status_code=502, detail="大模型未返回内容")
    return OptimizePromptResponse(optimized=text)