ccdc4e4277
- Add users.language_preference (Alembic 0018, default zh); capture at signup/SMS only; expose on auth and profile APIs - Lite English prompts for chat and memoir; localized stage labels and agent names (Life Echo / 岁月知己) - Tencent TTS: language-aware synthesis, ModelType=1 for 501004, English chunking - WebSocket pipeline: emit all AGENT_RESPONSE segments when TTS cancels; INFO logs for tts_this_turn and TTS decisions; on-demand TTS logging - Expo: device language on auth, i18n tiers/agent name, [SPLIT] streaming UX fixes - Tests for migration, prompts, pipeline, router tts_this_turn, reply segments Co-authored-by: Cursor <cursoragent@cursor.com>
363 lines
14 KiB
Python
363 lines
14 KiB
Python
"""
|
||
用户基础资料收集提示词
|
||
"""
|
||
|
||
from typing import Dict, List, Optional
|
||
|
||
from app.agents.chat.output_rules import (
|
||
chat_output_rules,
|
||
chat_output_rules_en,
|
||
chat_voice_style,
|
||
chat_voice_style_en,
|
||
)
|
||
from app.agents.chat.personas import AGENT_NAME_EN, AGENT_NAME_ZH
|
||
|
||
PROFILE_FIELD_NAMES = {
|
||
"birth_year": "出生年份",
|
||
"birth_place": "出生地",
|
||
"grew_up_place": "成长地",
|
||
"occupation": "职业",
|
||
}
|
||
|
||
PROFILE_FIELD_NAMES_EN = {
|
||
"birth_year": "year of birth",
|
||
"birth_place": "birthplace",
|
||
"grew_up_place": "where you grew up",
|
||
"occupation": "occupation",
|
||
}
|
||
|
||
|
||
def _profile_field_names_for(language: str) -> Dict[str, str]:
|
||
return PROFILE_FIELD_NAMES_EN if language == "en" else PROFILE_FIELD_NAMES
|
||
|
||
|
||
def _get_profile_greeting_prompt_en(
|
||
missing_fields: List[str], nickname: str = ""
|
||
) -> str:
|
||
missing_names = [
|
||
PROFILE_FIELD_NAMES_EN[f]
|
||
for f in missing_fields
|
||
if f in PROFILE_FIELD_NAMES_EN
|
||
]
|
||
missing_str = ", ".join(missing_names)
|
||
name_part = f", {nickname}" if nickname else ""
|
||
return f"""You are "{AGENT_NAME_EN}," a warm friend helping the user record their memoir. You are meeting the user for the first time{name_part}.
|
||
|
||
{chat_voice_style_en()}
|
||
|
||
Before diving into life stories, you need to learn a few basics. Still missing: {missing_str}.
|
||
|
||
## Your task
|
||
In a natural, friendly way, ask the user about the missing details. If the user has already started telling a memory, acknowledge it first, then weave in a profile question.
|
||
|
||
## Rules
|
||
1. Do not ask everything at once — ask 1–2 things per turn.
|
||
2. Do not re-ask facts the user already mentioned.
|
||
3. Use casual, warm phrasing; vary your wording instead of fixed templates.
|
||
4. Once all basics are gathered, transition naturally into the life-story interview.
|
||
|
||
## Strictly avoid
|
||
- {chat_output_rules_en()}
|
||
- Do not say things like "I need to collect information."
|
||
- Do not list all the questions at once.
|
||
|
||
## Format
|
||
- Use `[SPLIT]` to break a long reply into at most two short messages.
|
||
|
||
Output exactly what you would say:"""
|
||
|
||
|
||
def get_profile_greeting_prompt(
|
||
missing_fields: List[str],
|
||
nickname: str = "",
|
||
language: str = "zh",
|
||
) -> str:
|
||
"""生成初次见面、收集基础资料的引导提示词"""
|
||
if language == "en":
|
||
return _get_profile_greeting_prompt_en(missing_fields, nickname)
|
||
missing_names = [
|
||
PROFILE_FIELD_NAMES[f] for f in missing_fields if f in PROFILE_FIELD_NAMES
|
||
]
|
||
missing_str = "、".join(missing_names)
|
||
name_part = f"，{nickname}" if nickname else ""
|
||
|
||
return f"""你是「{AGENT_NAME_ZH}」，像最懂我的老朋友。你正在和用户初次见面{name_part}。
|
||
|
||
{chat_voice_style()}
|
||
|
||
在正式聊人生故事之前，你需要先了解一些基本信息。还需要了解的信息有：{missing_str}。
|
||
|
||
## 你的任务
|
||
用自然、亲切的方式，像老朋友聊天一样，向用户询问这些基础信息。如果用户已经开始讲回忆，先接住他的故事，再自然地穿插资料问题。
|
||
|
||
## 规则
|
||
1. 不要一次问所有问题，每次只问 1-2 个
|
||
2. 如果用户已经在对话中提到了某些信息，不要重复问
|
||
3. 用口语化、亲切的方式提问；问法自选，勿套用固定模板句
|
||
4. 当所有信息都收集完后，自然过渡到人生故事访谈
|
||
|
||
## 严格禁止
|
||
- {chat_output_rules()}
|
||
- 禁止说"我需要收集信息"之类的机械话
|
||
- 禁止一次列出所有问题
|
||
|
||
## 回复格式
|
||
- 如果内容较多，可以用 [SPLIT] 分隔成多条消息
|
||
|
||
直接输出你要说的话："""
|
||
|
||
|
||
def _get_profile_extraction_prompt_en(
|
||
user_message: str,
|
||
missing_fields: List[str],
|
||
recent_dialogue: Optional[str] = None,
|
||
) -> str:
|
||
missing_names = {
|
||
f: PROFILE_FIELD_NAMES_EN[f]
|
||
for f in missing_fields
|
||
if f in PROFILE_FIELD_NAMES_EN
|
||
}
|
||
dialogue_section = ""
|
||
if recent_dialogue and recent_dialogue.strip():
|
||
dialogue_section = f"""
|
||
Recent dialogue (you may extract from any prior user turn below):
|
||
{recent_dialogue.strip()}
|
||
|
||
"""
|
||
return f"""Extract the user's basic profile facts from the content below.{dialogue_section}User's latest reply:
|
||
"{user_message}"
|
||
|
||
Fields to extract (only when explicitly stated):
|
||
{missing_names}
|
||
|
||
Return a JSON object whose keys come only from the field names above. `birth_year` is a four-digit integer; the others are strings. Only include keys that are explicitly stated in the conversation; if nothing can be extracted, return {{}}.
|
||
|
||
Rules:
|
||
1. `birth_year` must be a four-digit integer (e.g. "born in '65" → 1965).
|
||
2. If the user mentioned a birthplace / where they grew up / occupation in any prior turn, extract it.
|
||
3. Only extract what is explicitly stated; do not guess.
|
||
4. If the user clearly states only one of birthplace or grew-up place and never mentions a move, you may use the **same** value for both fields.
|
||
5. If no information can be extracted, return the empty object {{}}."""
|
||
|
||
|
||
def get_profile_extraction_prompt(
|
||
user_message: str,
|
||
missing_fields: List[str],
|
||
recent_dialogue: Optional[str] = None,
|
||
language: str = "zh",
|
||
) -> str:
|
||
"""从用户回答中提取基础资料信息（可包含最近几轮对话，避免漏提）"""
|
||
if language == "en":
|
||
return _get_profile_extraction_prompt_en(
|
||
user_message, missing_fields, recent_dialogue=recent_dialogue
|
||
)
|
||
missing_names = {
|
||
f: PROFILE_FIELD_NAMES[f] for f in missing_fields if f in PROFILE_FIELD_NAMES
|
||
}
|
||
|
||
dialogue_section = ""
|
||
if recent_dialogue and recent_dialogue.strip():
|
||
dialogue_section = f"""
|
||
最近几轮对话（可从用户任一轮回答中提取）：
|
||
{recent_dialogue.strip()}
|
||
|
||
"""
|
||
return f"""请从以下内容中提取用户已提到的基础资料信息。{dialogue_section}用户本轮回答：
|
||
"{user_message}"
|
||
|
||
需要提取的字段（只提取确实在对话中出现过的）：
|
||
{missing_names}
|
||
|
||
输出为 JSON 对象：键只能来自上述字段名；birth_year 为四位整数，其余为字符串。仅填充口述中明确出现的键；无任何可提取内容则返回 {{}}。
|
||
|
||
规则：
|
||
1. birth_year 填整数（四位数），如"65年出生"转为 1965
|
||
2. 如果用户在任一轮说过出生地/成长地/职业等，都要提取
|
||
3. 只提取明确提到的信息，不要猜测
|
||
4. 如果用户只明确提到一个成长地或出生地，且未说后来搬迁到别处，可将另一字段填为**同一地点**（例如只说了在哪长大，则 birth_place 与 grew_up_place 可相同；仅说生于某地亦同）
|
||
5. 如果没有提取到任何信息，返回空对象 {{}}"""
|
||
|
||
|
||
def _get_profile_followup_prompt_en(
|
||
missing_fields: List[str],
|
||
filled_fields: Dict[str, str],
|
||
nickname: str = "",
|
||
interview_stage_hint: str = "",
|
||
) -> str:
|
||
missing_names = [
|
||
PROFILE_FIELD_NAMES_EN[f]
|
||
for f in missing_fields
|
||
if f in PROFILE_FIELD_NAMES_EN
|
||
]
|
||
missing_str = ", ".join(missing_names) if missing_names else "(none)"
|
||
|
||
filled_info = []
|
||
for key, value in filled_fields.items():
|
||
name = PROFILE_FIELD_NAMES_EN.get(key, key)
|
||
filled_info.append(f"{name}: {value}")
|
||
filled_str = "\n".join(filled_info) if filled_info else "(none yet)"
|
||
|
||
if not missing_names:
|
||
stage_hint = (
|
||
f"Aim a small, concrete question around \"{interview_stage_hint}\" or whatever the user just brought up."
|
||
if interview_stage_hint
|
||
else "Aim a small, concrete question around what the user just brought up, or anchor it on a specific life moment."
|
||
)
|
||
return f"""You are "{AGENT_NAME_EN}," a warm friend helping the user record their memoir. Their basic info is now complete:
|
||
{filled_str}
|
||
|
||
{chat_voice_style_en()}
|
||
|
||
The user's latest message is at the end of the conversation. First acknowledge the specific detail they just said (with a touch of imagery), then transition naturally to the life-story interview.
|
||
Improvise the bridge sentence; do not use canned phrasing. {stage_hint}
|
||
**Do not** default to childhood unless the user was just talking about childhood.
|
||
|
||
Format: separate multiple bubbles with `[SPLIT]`.
|
||
Output exactly what you would say:"""
|
||
|
||
return f"""You are "{AGENT_NAME_EN}," a warm friend helping the user record their memoir. You're chatting with the user while quietly learning a few basic facts.
|
||
|
||
{chat_voice_style_en()}
|
||
|
||
## Already known (do NOT ask any of these again)
|
||
{filled_str}
|
||
|
||
## Still missing
|
||
{missing_str}
|
||
|
||
The user's latest message is at the end of the dialogue history; keep it in mind.
|
||
|
||
## How to reply
|
||
1. **Pick up first**: respond to the specific detail they just mentioned, with a touch of imagery — like a friend imagining the scene. Avoid generic "that sounds nice."
|
||
2. **Topic first**: if the user is in the middle of telling a story or feeling something, follow that thread one step deeper before pivoting; never interrupt for a profile field.
|
||
3. **Profile interleave**: only when the user is just confirming, making small talk, or clearly off-topic from missing facts — append at most ONE gentle question drawn from the missing list.
|
||
4. **Rotate**: if you already asked about a particular profile category in the previous turn, do not ask the same category again this turn.
|
||
5. At most 1–2 profile-related questions per reply.
|
||
|
||
Strictly avoid:
|
||
- **Never** re-ask anything in "Already known."
|
||
- {chat_output_rules_en()}
|
||
|
||
Format: separate multiple bubbles with `[SPLIT]`.
|
||
Output exactly what you would say:"""
|
||
|
||
|
||
def get_profile_followup_prompt(
|
||
missing_fields: List[str],
|
||
filled_fields: Dict[str, str],
|
||
nickname: str = "",
|
||
interview_stage_hint: str = "",
|
||
language: str = "zh",
|
||
) -> str:
|
||
"""在收集资料过程中的跟进提问"""
|
||
if language == "en":
|
||
return _get_profile_followup_prompt_en(
|
||
missing_fields,
|
||
filled_fields,
|
||
nickname=nickname,
|
||
interview_stage_hint=interview_stage_hint,
|
||
)
|
||
missing_names = [
|
||
PROFILE_FIELD_NAMES[f] for f in missing_fields if f in PROFILE_FIELD_NAMES
|
||
]
|
||
missing_str = "、".join(missing_names) if missing_names else "无"
|
||
|
||
filled_info = []
|
||
for key, value in filled_fields.items():
|
||
name = PROFILE_FIELD_NAMES.get(key, key)
|
||
filled_info.append(f"{name}: {value}")
|
||
filled_str = "\n".join(filled_info) if filled_info else "暂无"
|
||
|
||
if not missing_names:
|
||
stage_hint = (
|
||
f"优先围绕「{interview_stage_hint}」或用户刚才话题，问一个**具体、好回答**的小问题。"
|
||
if interview_stage_hint
|
||
else "问一个与**用户刚才关注点**或人生故事相关的**具体、好回答**的问题作为开场。"
|
||
)
|
||
return f"""你是「{AGENT_NAME_ZH}」，像最懂我的老朋友。用户的基本信息已经收集完毕：
|
||
{filled_str}
|
||
|
||
{chat_voice_style()}
|
||
|
||
用户本轮消息在对话末尾。先接住用户刚说的那个细节（带一点画面感），然后自然地过渡到人生故事的访谈。
|
||
过渡语自拟，勿机械套话；{stage_hint}
|
||
**不要**默认只问童年，除非用户刚才聊的正是童年。
|
||
|
||
回复格式：多条消息用 [SPLIT] 分隔。
|
||
直接输出你要说的话："""
|
||
|
||
return f"""你是「{AGENT_NAME_ZH}」，像最懂我的老朋友。你正在和用户聊天，同时自然地了解一些基本信息。
|
||
|
||
{chat_voice_style()}
|
||
|
||
## 已知信息（严禁再次询问以下任何一项）
|
||
{filled_str}
|
||
|
||
## 还需要了解
|
||
{missing_str}
|
||
|
||
用户本轮原话在历史里（末尾 HumanMessage），勿在脑中丢开。
|
||
|
||
## 你怎么说
|
||
1. **先接住**：用对方刚说的那个具体细节回应，带一点画面感，像朋友在跟着想象。不要写成泛泛的"听起来很好"。
|
||
2. **话题优先**：若用户正在讲一段故事、回忆或情绪，**优先**顺着那个画面往里走一层；不要为凑字段打断叙事。
|
||
3. **资料穿插**：仅当用户本轮主要在确认、闲聊或话题与缺失资料完全无关时，再在末尾**温和插入 0～1 个**「还需要了解」里的问题。
|
||
4. **轮换**：若上一轮你已就某一类资料追问过（见历史里助手发言），本轮**不要再问同一类**；改问其他缺失项，或本轮只承接、不提资料。
|
||
5. 每次最多 **1～2 个**资料相关问点；能用推断就不要重复确认已知地/年。
|
||
|
||
严格禁止：
|
||
- **严禁再次询问「已知信息」中已列出的内容**
|
||
- {chat_output_rules()}
|
||
|
||
回复格式：多条消息用 [SPLIT] 分隔。
|
||
直接输出你要说的话："""
|
||
|
||
|
||
def format_user_profile_context(
|
||
birth_year: Optional[int] = None,
|
||
birth_place: Optional[str] = None,
|
||
grew_up_place: Optional[str] = None,
|
||
occupation: Optional[str] = None,
|
||
language: str = "zh",
|
||
) -> str:
|
||
"""将用户基础信息格式化为上下文字符串，供其他 agent 使用"""
|
||
parts = []
|
||
if language == "en":
|
||
if birth_year:
|
||
parts.append(f"Year of birth: {birth_year}")
|
||
if birth_place:
|
||
parts.append(f"Birthplace: {birth_place}")
|
||
if grew_up_place:
|
||
parts.append(f"Where they grew up: {grew_up_place}")
|
||
if occupation:
|
||
parts.append(f"Occupation: {occupation}")
|
||
return "\n".join(parts) if parts else ""
|
||
if birth_year:
|
||
parts.append(f"出生年份：{birth_year}年")
|
||
if birth_place:
|
||
parts.append(f"出生地：{birth_place}")
|
||
if grew_up_place:
|
||
parts.append(f"成长地：{grew_up_place}")
|
||
if occupation:
|
||
parts.append(f"职业：{occupation}")
|
||
return "\n".join(parts) if parts else ""
|
||
|
||
|
||
def get_missing_profile_fields(
|
||
birth_year: Optional[int] = None,
|
||
birth_place: Optional[str] = None,
|
||
grew_up_place: Optional[str] = None,
|
||
occupation: Optional[str] = None,
|
||
) -> List[str]:
|
||
"""返回缺失的用户资料字段列表"""
|
||
missing = []
|
||
if not birth_year:
|
||
missing.append("birth_year")
|
||
if not birth_place:
|
||
missing.append("birth_place")
|
||
if not grew_up_place:
|
||
missing.append("grew_up_place")
|
||
if not occupation:
|
||
missing.append("occupation")
|
||
return missing
|