refactor(agents): 抽取阶段常量与对话上下文;快档 LLM;图片 prompt 可禁止回退

访谈与阶段
- 新增 app/agents/stage_constants.py:集中 CHAT_STAGES、章节分类/顺序、阶段到默认 memoir 类别等,与 MemoirState 默认槽位顺序对齐;减少散落在 prompts 内的重复常量。
- 新增 app/agents/chat/prompt_context.py:以 ChatPromptContext 汇总 guided 系统提示所需字段(阶段、槽位、轮次、人设、记忆证据、回复长度模式、背景声线、职业等),统一走 get_guided_conversation_prompt。
- 大幅收敛 app/agents/chat/prompts_conversation.py;调整 prompts.py、stage_prompts.py、stage_detection.py;同步 interview_agent、profile_agent、helpers 与 state_schema,使对话侧构造提示的方式一致、可测。

回忆录流水线
- memoir/prompts.py 删除已迁至 stage_constants / 独立模板的大段常量与图片占位相关逻辑;classification / extraction / fidelity / narrative agents 与 orchest(全量历史仍可用于计数,注入模型时按轮次与字符上限截断)、image_prompt_fallback_disabled。
- dependencies 增加 get_llm_provider_fast(LRU 缓存,可与默认共用密钥与 base_url)。

任务与编排
- memoir_tasks:prepare_batches 注入 llm_fast;开启独立快档模型时打结构化日志。
- chapter_cover_tasks、story_image_tasks:与图片 prompt / JSON 工具路径或策略变更对齐(import 与行为一致)。
- story_pipeline_sync 等小处同步。

其它核心
- langchain_llm、text_normalize 随上述调用链微调。

开发者体验
- .cursor/settings.json:启用 redis-development、postman 插件。

测试
- 新增 test_image_prompt_policy:覆盖「禁止回退」等图片 prompt 策略。
- 更新 test_interview_prompts、test_interview_reply_length、test_experience_regressions、test_json_and_memory_utils,匹配新常量位置、json_utils 与对话/长度行为。
This commit is contained in:
Kevin
2026-04-02 12:00:00 +08:00
parent 43ef260ae2
commit bb16d3a5c9
42 changed files with 894 additions and 580 deletions

View File

@@ -13,6 +13,7 @@ from celery import shared_task
from PIL import Image
from sqlalchemy import and_, func, or_, select, update
from app.agents.image_prompt import get_image_prompt_orchestrator
from app.core.db import get_sync_db
from app.core.dependencies import get_image_generator
from app.core.logging import get_logger
@@ -87,22 +88,6 @@ def _normalize_image_bytes(image_bytes: bytes) -> bytes:
return output.getvalue()
def _build_story_image_prompt(
prompt_brief: str,
story_title: str = "",
story_stage: str | None = None,
style_profile: str | None = None,
) -> str:
"""从 intent.prompt_brief 构建出图 prompt。"""
from app.agents.memoir.prompts import IMAGE_PLACEHOLDER_TEMPLATE
base = IMAGE_PLACEHOLDER_TEMPLATE
if prompt_brief and prompt_brief.strip():
return f"{base}{prompt_brief.strip()}"
fallback = "".join(filter(None, [story_title, story_stage or ""])) or "人生故事"
return f"{base}{fallback}"
def _story_image_claimable_clause(now: datetime):
cutoff = now - timedelta(seconds=STORY_IMAGE_CLAIM_TTL_SECONDS)
return or_(
@@ -218,16 +203,24 @@ def generate_story_image(self, story_id: str):
storage = TencentCosStorageService.from_env()
settings = img_cfg
prompt_final = _build_story_image_prompt(
intent.prompt_brief or "",
orch = get_image_prompt_orchestrator()
intent_style = (getattr(intent, "style_profile", None) or "").strip() or None
prompt_out = orch.build_story_primary_prompt(
story_title=story.title or "",
story_stage=story.stage,
style_profile=intent.style_profile or settings.default_style,
prompt_brief=intent.prompt_brief or "",
style_profile=intent_style,
)
prompt_final = prompt_out["prompt"]
style_for_image = (
(prompt_out.get("style") or "").strip()
or intent_style
or settings.default_style
)
result = image_generator.generate(
prompt_final,
settings.default_size,
intent.style_profile or settings.default_style,
prompt_out.get("size") or settings.default_size,
style_for_image,
)
if result.status != TaskStatus.COMPLETED or not result.image_url:
raise RuntimeError(result.error or "Image generation failed")
@@ -262,7 +255,7 @@ def generate_story_image(self, story_id: str):
storage_key=cos_key,
url=url,
provider=settings.provider,
style_profile=intent.style_profile or settings.default_style,
style_profile=style_for_image,
prompt_final=prompt_final,
status="completed",
)