feat(api): 访谈人格/回复长度策略、口述归一、背景语气与输入净稿全链路

Chat 访谈
- 新增 persona 系统(default / warm_listener / curious_guide)与 background_voice 语气层
- 回复长度由 compute_reply_plan 统一决策(brief / standard / expanded),融合信息密度启发式
- 输入净稿(input_normalize):编排层可选 rules/llm 归一用户口语后再喂模型与记忆检索
- 记忆证据注入:按用户话检索 memory evidence 并注入 prompt

Memoir 回忆录
- 口述归一(oral_normalize):segment 原文保留,story 管线取派生净稿作叙事输入
- segment 入队批次门闸:累计字数 + 最长等待秒数,减少零碎提交
- fidelity_check / prompts / narrative_agent 微调
- Alembic 0005:清理跨章节 story 外键

Infra
- Dockerfile 加入 ffmpeg
- pyproject.toml 新增依赖并同步 uv.lock
- .env.example / .env.production 补全新配置项

Tests
- 新增 test_background_voice、test_chat_input_normalize、test_experience_regressions
- 扩展 test_interview_prompts、test_interview_reply_length、test_story_route_oral_invariant

Made-with: Cursor
This commit is contained in:
Kevin
2026-03-31 23:55:26 +08:00
parent 42ae2a5e91
commit 69a673e6c6
44 changed files with 2998 additions and 259 deletions

View File

@@ -10,6 +10,7 @@ from __future__ import annotations
import json
import re
from dataclasses import dataclass
from typing import Any
from app.agents.memoir.prompts import (
@@ -95,6 +96,14 @@ def _normalize_llm_category(raw: str) -> str:
return s
@dataclass(frozen=True)
class ChapterClassifyResult:
"""章节分类结果;``llm_said_none`` 仅当走 LLM 且解析为 none 时为 Truefragment 启发式不为 True"""
category: str
llm_said_none: bool = False
def _parse_category_from_llm_response(raw: str) -> str:
"""优先解析 JSON ``{"category": "..."}``,失败则按纯文本 key 处理。"""
s = (raw or "").strip()
@@ -119,10 +128,11 @@ class ClassificationAgent:
llm: Any,
*,
segment_id: str | None = None,
) -> str:
) -> ChapterClassifyResult:
"""
分类到 8 个章节类别之一。
LLM 返回 none 或启发式为零散档案时,返回 ``summary``(仍走回忆录流水线)。
LLM 返回 none 或启发式为零散档案时,``category`` 为 ``summary``(仍走回忆录流水线
``llm_said_none`` 仅在 LLM 明确返回 none 时为 True供空转抑制判断
llm 需支持 .invoke(prompt) 同步调用。
"""
if _looks_like_fragment_only(text):
@@ -133,7 +143,10 @@ class ClassificationAgent:
len(text or ""),
_SUMMARY_FALLBACK_CATEGORY,
)
return _SUMMARY_FALLBACK_CATEGORY
return ChapterClassifyResult(
category=_SUMMARY_FALLBACK_CATEGORY,
llm_said_none=False,
)
if llm:
try:
@@ -153,14 +166,18 @@ class ClassificationAgent:
len(text or ""),
_SUMMARY_FALLBACK_CATEGORY,
)
return _SUMMARY_FALLBACK_CATEGORY
return ChapterClassifyResult(
category=_SUMMARY_FALLBACK_CATEGORY,
llm_said_none=True,
)
if category in CHAPTER_CATEGORIES:
return category
return ChapterClassifyResult(category=category, llm_said_none=False)
except Exception as e:
logger.warning("ClassificationAgent LLM 章节分类失败: {}", e)
stage = _detect_stage(text, fallback_stage)
return _STAGE_TO_DEFAULT_CATEGORY.get(
cat = _STAGE_TO_DEFAULT_CATEGORY.get(
stage,
_STAGE_TO_DEFAULT_CATEGORY.get(fallback_stage, "childhood"),
)
return ChapterClassifyResult(category=cat, llm_said_none=False)