feat(agent): align prompts and runtime with workflow@20261020

- archive workflow/20261020 export (diff vs 20261016: collection prompt only - state-decides-body rule, route when info complete) - prompts: collection.md updated; 6 photo/confirm prompts archived with source nodeIds; render_prompt replaces {{$vars}} (Asia/Shanghai clock, person fields, explicit 'not provided' for missing prefills) - llm: message history passed as user/assistant roles (chatNode history=50); optional LLM_TEMPERATURE; FakeLLMService supports history-aware responders - collect: context via system supplement, raw user input as user message (Text Editor#2 parity), shared validated_reply (one call, one repair) - welcome: Assigned Reply#3 original text, stored in dialogue history - transitions: 2004 plate-confirm entries (2003->2004/2005, 2004->2005) - verified: prompt files byte-identical to workflow JSON (parametrized test)
2026-10-21 15:10:51 +08:00
parent 1c6b1a513d
commit a5407e03b2
14 changed files with 7881 additions and 98 deletions
@@ -15,6 +15,8 @@
 调用方（API 层）将其转换为可识别错误响应。
 """

+from collections.abc import Callable
+import json
 import re

 from .fields import (
@@ -23,7 +25,7 @@ from .fields import (
    validate_candidate,
 )
 from .llm import LLMService
-from .prompts import COLLECTION, load_prompt
+from .prompts import COLLECTION, render_prompt
 from .state import AgentState
 from .transitions import COLLECT_STAGE, is_legal_transition

@@ -73,51 +75,60 @@ def _try_parse(text: str) -> tuple[str, str] | None:
        return None


-# 注入提示词的对话历史轮数上限（每轮 = user + assistant 两条）
-_HISTORY_TURNS = 10
+async def validated_reply(
+    llm: LLMService,
+    system_prompt: str,
+    user_message: str,
+    current: str,
+    *,
+    validator: Callable[[str, str], bool] = is_legal_transition,
+    history: list[dict[str, str]] | None = None,
+) -> tuple[str, str]:
+    """模型同时产出下一阶段和正文；格式与转换共用一次修复预算。
+
+    只校验模型提议，不按输入长度或关键字替模型决定。非法输出不写回
+    图状态；修复仍失败则抛出不含模型原文的错误。
+    """
+    for attempt in range(2):
+        prompt = system_prompt
+        if attempt:
+            prompt += (
+                _REPAIR_SUFFIX
+                + f"\n当前阶段为 {current}。请重新检查该阶段允许的转换"
+                "和当前 clientMode，不得跳步或切换到另一模式的出口。"
+            )
+        raw = await llm.complete(prompt, user_message, history=history)
+        parsed = _try_parse(raw)
+        if parsed is not None and validator(current, parsed[0]):
+            return raw, parsed[0]
+    raise CollectReplyError("模型回复格式或阶段转换不合规（一次修复后仍失败）")


-def _format_history(state: AgentState) -> str:
-    """把 dialogue 格式化为提示词可用的历史文本；为空返回“（无）”。"""
-    dialogue = state.get("dialogue") or []
-    if not dialogue:
-        return "（首轮对话）"
-    recent = dialogue[-_HISTORY_TURNS * 2:]
-    lines = []
-    for msg in recent:
-        role = "用户" if msg.get("role") == "user" else "助手"
-        content = str(msg.get("content", ""))
-        # 助手历史带 <state> 标签，去掉标签避免干扰模型本轮输出格式
-        content = STATE_PREFIX_RE.sub("", content).strip()
-        lines.append(f"{role}：{content}")
-    return "\n".join(lines)
+# 工作流 chatNode 的 history 配置为 50；保留独立消息及其角色。
+_HISTORY_MESSAGES = 50


-def _format_known_fields(state: AgentState) -> str:
-    """把已确认字段格式化为提示词摘要；为空返回“（暂无）”。"""
-    fields = state.get("accident_fields") or {}
-    if not fields:
-        return "（暂无）"
-    return "；".join(f"{key}={value}" for key, value in fields.items())
+def conversation_history(state: AgentState) -> list[dict[str, str]]:
+    """历史不拼入 user 消息，不剥除历史 assistant 的状态前缀。"""
+    return [dict(message) for message in (state.get("dialogue") or [])[-_HISTORY_MESSAGES:]]
+
+
+def context_prompt(state: AgentState) -> str:
+    """运行时上下文随 system 提示提供；业务判断仍由模型负责。"""
+    return "\n\n【当前会话上下文】\n" + json.dumps({
+        "stage_code": state.get("stage_code") or COLLECT_STAGE,
+        "client_mode": state.get("client_mode") or "direct",
+        "person_index": state.get("person_index") or 1,
+        "accident_fields": state.get("accident_fields") or {},
+        "person_fields": state.get("person_fields") or {},
+        "no_reply_count_before_this_turn": state.get("no_reply_count") or 0,
+    }, ensure_ascii=False)


 def _build_user_message(state: AgentState) -> str:
-    """把图状态翻译成本轮用户消息（代替工作流的变量注入）。
+    """对齐工作流 Text Editor#2：只传用户本轮原始文本。"""
+    return state.get("input_text", "")

-    注入三部分上下文（计划作业 4-5“跳过已知项”）：
-    - 对话历史：模型看不到 checkpoint，历史必须显式注入，
-      否则每轮都以为首次提问、重复询问已知信息；
-    - 已确认字段：信息齐全时模型按提示词直接分流（2000/2010），
-      不追加核对问题；
-    - 当前阶段与计数：供模型做转人工等判断。
-    """
-    return (
-        f"【对话历史】\n{_format_history(state)}\n"
-        f"【已确认信息】\n{_format_known_fields(state)}\n"
-        f"【当前对话阶段】{state.get('stage_code') or COLLECT_STAGE}\n"
-        f"【连续无回复次数】{state.get('no_reply_count') or 0}\n"
-        f"【用户最新输入】{state.get('input_text', '')}"
-    )


 def _extract_candidates(state: AgentState, body: str) -> list[FieldCandidate]:
@@ -159,9 +170,8 @@ def _bookkeeping(state: AgentState) -> tuple[int, int]:
    返回 (no_reply_count, clarify_count)。注意这只是给 LLM 的
    判断输入，是否转人工由模型按提示词决定。

-    clarify_count 的“计入”需要语义完整性判断（短促词/ASR 乱码
-    不计入，见 prompts/collection.md）；参考版用“普通输入”做
-    简化近似，接入真实 LLM 后按语义判定细化。
+    澄清失败次数由模型结合语义与历史判断；这里透传已有值，不把
+    每次普通输入都当作一次失败。
    """
    event = state.get("client_event") or ""
    no_reply_count = state.get("no_reply_count") or 0
@@ -170,10 +180,7 @@ def _bookkeeping(state: AgentState) -> tuple[int, int]:
    elif (state.get("input_text") or "").strip():
        no_reply_count = 0

-    clarify_count = state.get("clarify_count") or 0
-    if not event and (state.get("input_text") or "").strip():
-        clarify_count += 1
-    return no_reply_count, clarify_count
+    return no_reply_count, state.get("clarify_count") or 0


 async def run_collect(state: AgentState, llm: LLMService) -> dict:
@@ -184,41 +191,13 @@ async def run_collect(state: AgentState, llm: LLMService) -> dict:
    任何一步修复后仍不合规都抛 CollectReplyError：不提交业务状态、
    错误信息不携带原始模型内容。
    """
-    system_prompt = load_prompt(COLLECTION)
+    system_prompt = render_prompt(COLLECTION, state) + context_prompt(state)
    user_message = _build_user_message(state)
    current = state.get("stage_code") or COLLECT_STAGE
+    raw, stage = await validated_reply(
+        llm, system_prompt, user_message, current, history=conversation_history(state)
+    )

-    # 1) 第一次模型调用（计划作业 4-3：一次调用同时产出前缀与正文）
-    raw = await llm.complete(system_prompt, user_message)
-    parsed = _try_parse(raw)
-
-    # 2) 前缀不合规：受控修复一次（最多多一次调用）
-    if parsed is None:
-        raw = await llm.complete(system_prompt + _REPAIR_SUFFIX, user_message)
-        parsed = _try_parse(raw)
-        if parsed is None:
-            raise CollectReplyError(
-                "模型回复缺少合法 <state> 前缀（修复一次后仍不合规）"
-            )
-    stage, _body = parsed
-
-    # 3) 白名单守卫：LLM 提议的转换必须合法（计划作业 3 校验语义）
-    if not is_legal_transition(current, stage):
-        repair_hint = (
-            "\n另外：你上一条提议的状态码不合规。"
-            f"当前阶段是 {current}，只能选择合法的下一步状态码，"
-            "不得跳步或使用未定义的状态码。"
-        )
-        raw = await llm.complete(
-            system_prompt + _REPAIR_SUFFIX + repair_hint, user_message
-        )
-        parsed = _try_parse(raw)
-        if parsed is None or not is_legal_transition(current, parsed[0]):
-            raise CollectReplyError(
-                f"模型提议了非法的阶段转换 {current}→{stage}"
-                "（修复一次后仍不合规）"
-            )
-        stage, _body = parsed

    # 4) 字段候选提取与三态落库（提取本身见 _extract_candidates）
    accident_fields, confirm_status, patch = _apply_candidates(