life-echo/api/app/agents/chat/background_voice.py

"""
从用户档案「职业」等文本推断访谈/叙事语气维度（干部形、军队形）。
与 chat_interview_persona（温柔倾听等）正交，可叠加。
"""

from __future__ import annotations

from typing import Final, Literal

BackgroundVoice = Literal["default", "cadre", "military"]

# 军队系优先：含「军、部队」等则走军队形，避免与泛「干部」冲突。
_MILITARY_NEEDLES: Final[tuple[str, ...]] = (
    "军人",
    "军官",
    "士兵",
    "部队",
    "入伍",
    "服役",
    "退伍",
    "转业",
    "武警",
    "解放军",
    "陆军",
    "海军",
    "空军",
    "火箭军",
    "军区",
    "军营",
    "军校",
    "文职干部",
    "军队文职",
    "现役",
    "预备役",
)

# 干部/机关系（避免过短词误判：如「机关」→机关枪、「主任」→班主任）
_CADRE_NEEDLES: Final[tuple[str, ...]] = (
    "公务员",
    "党政机关",
    "党政",
    "组织部",
    "党委书记",
    "党组书记",
    "书记",
    "处长",
    "科长",
    "局长",
    "厅长",
    "部长",
    "国企",
    "事业单位",
    "干部",
    "科级",
    "处级",
    "厅级",
)


def infer_background_voice(occupation: str | None) -> BackgroundVoice:
    """
    据职业自由文本推断背景语气。军队关键词优先于干部关键词。
    无匹配或未填 → default。
    """
    if not occupation or not str(occupation).strip():
        return "default"
    t = str(occupation).strip().casefold()
    for n in _MILITARY_NEEDLES:
        if n.casefold() in t:
            return "military"
    for n in _CADRE_NEEDLES:
        if n.casefold() in t:
            return "cadre"
    return "default"


def normalize_background_voice(voice: str | None) -> BackgroundVoice:
    """调用方传入已归一化枚举或原始职业文本均可。"""
    if not voice:
        return "default"
    s = voice.strip()
    if s in ("default", "cadre", "military"):
        return s  # type: ignore[return-value]
    return infer_background_voice(s)


def get_background_voice_tone_hint(voice: str | None) -> str:
    """一句背景语气提示，融入主 system prompt；default 返回空串。"""
    v = normalize_background_voice(voice)
    if v == "default":
        return ""
    if v == "military":
        return (
            "语气简洁利落、得体；称呼自然；不写命令式、不堆砌军事辞藻；"
            "仅当用户口述已出现相关事实时才呼应军旅语境，不编造经历。"
        )
    return (
        "语气稳重有分寸、敬语适度；避免官样排比与公文套话；"
        "不得编造用户未提及的职级、单位与荣誉。"
    )


def get_background_voice_narrative_block(voice: str | None) -> str:
    """附在叙事系统提示后的文体补充；default 返回空串。"""
    v = normalize_background_voice(voice)
    if v == "default":
        return ""
    if v == "military":
        return (
            "## 背景文体（军队，须遵守上文事实边界）\n"
            "叙事紧凑、层次清楚；若口述已出现纪律、集体、任务等语境，可适度用书面语呼应，**禁止**堆砌口号式军事辞藻或虚构军旅细节。\n"
            "不新增军衔、单位番号、表彰等口述未出现的信息。\n"
            "用户已退役/转业，以回忆军旅岁月为基调，不要预设其仍在服役。"
        )
    return (
        "## 背景文体（干部/机关，须遵守上文事实边界）\n"
        "段落层次清晰，用语庄重自然，避免口语碎词与段子感；**不得编造**职务、荣誉、单位名称与组织细节。\n"
        "文采服务于真实内容，不写成公文或汇报腔。\n"
        "用户已退休，以回顾和怀念工作岁月为基调，不要预设其仍在岗。"
    )