TravelContentCreator/utils/prompt_manager.py

585 lines
32 KiB
Python
Raw Normal View History

2025-04-22 14:19:21 +08:00
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Manages the construction of prompts for different AI generation tasks.
"""
import os
import traceback
2025-04-22 17:36:29 +08:00
import logging # Add logging
2025-04-22 14:19:21 +08:00
from .resource_loader import ResourceLoader # Use relative import within the same package
class PromptManager:
"""Handles the loading and construction of prompts."""
def __init__(self,
topic_system_prompt_path: str,
topic_user_prompt_path: str,
content_system_prompt_path: str,
2025-04-24 21:44:05 +08:00
prompts_dir: str = None, # 兼容旧配置
prompts_config: list = None, # 新的配置方式
resource_dir_config: list = None,
topic_gen_num: int = 1, # Default values if needed
topic_gen_date: str = ""
):
self.topic_system_prompt_path = topic_system_prompt_path
self.topic_user_prompt_path = topic_user_prompt_path
self.content_system_prompt_path = content_system_prompt_path
2025-04-24 21:44:05 +08:00
self.prompts_dir = prompts_dir # 保留兼容旧配置
self.prompts_config = prompts_config or [] # 新的配置方式
self.resource_dir_config = resource_dir_config or []
self.topic_gen_num = topic_gen_num
self.topic_gen_date = topic_gen_date
2025-04-24 21:44:05 +08:00
# 缓存加载的文件内容
self._style_cache = {}
self._demand_cache = {}
self._refer_cache = {}
self._system_prompt_cache = {} # 新增:系统提示词缓存
self._user_prompt_cache = {} # 新增:用户提示词缓存
self._dateline_cache = None # 新增:日期线缓存
# 初始化时预加载配置的文件
self._preload_prompt_files()
def _preload_prompt_files(self):
"""预加载配置中的提示文件到缓存"""
# 预加载系统提示词和用户提示词文件
if self.topic_system_prompt_path and os.path.exists(self.topic_system_prompt_path):
content = ResourceLoader.load_file_content(self.topic_system_prompt_path)
if content:
self._system_prompt_cache["topic"] = content
logging.info(f"预加载系统提示词: {self.topic_system_prompt_path}")
if self.topic_user_prompt_path and os.path.exists(self.topic_user_prompt_path):
content = ResourceLoader.load_file_content(self.topic_user_prompt_path)
if content:
self._user_prompt_cache["topic"] = content
logging.info(f"预加载用户提示词: {self.topic_user_prompt_path}")
if self.content_system_prompt_path and os.path.exists(self.content_system_prompt_path):
content = ResourceLoader.load_file_content(self.content_system_prompt_path)
if content:
self._system_prompt_cache["content"] = content
logging.info(f"预加载内容系统提示词: {self.content_system_prompt_path}")
# 预加载日期线文件
if self.topic_user_prompt_path:
user_prompt_dir = os.path.dirname(self.topic_user_prompt_path)
dateline_path = os.path.join(user_prompt_dir, "2025各月节日宣传节点时间表.md")
if os.path.exists(dateline_path):
self._dateline_cache = ResourceLoader.load_file_content(dateline_path)
logging.info(f"预加载日期线文件: {dateline_path}")
# 加载prompts_config配置的文件
if not self.prompts_config:
return
for config_item in self.prompts_config:
prompt_type = config_item.get("type", "").lower()
file_paths = config_item.get("file_path", [])
if prompt_type == "style":
for path in file_paths:
if os.path.exists(path):
filename = os.path.basename(path)
content = ResourceLoader.load_file_content(path)
if content:
self._style_cache[filename] = content
name_without_ext = os.path.splitext(filename)[0]
self._style_cache[name_without_ext] = content # 同时缓存不带扩展名的版本
elif prompt_type == "demand":
for path in file_paths:
if os.path.exists(path):
filename = os.path.basename(path)
content = ResourceLoader.load_file_content(path)
if content:
self._demand_cache[filename] = content
name_without_ext = os.path.splitext(filename)[0]
self._demand_cache[name_without_ext] = content # 同时缓存不带扩展名的版本
elif prompt_type == "refer":
for path in file_paths:
if os.path.exists(path):
filename = os.path.basename(path)
content = ResourceLoader.load_file_content(path)
if content:
self._refer_cache[filename] = content
def _get_style_content(self, style_name):
"""获取Style文件内容优先从缓存获取如果不存在则尝试从目录加载"""
# 首先检查缓存
if style_name in self._style_cache:
return self._style_cache[style_name]
# 确保有扩展名
if not style_name.lower().endswith('.txt'):
style_file = f"{style_name}.txt"
else:
style_file = style_name
style_name = os.path.splitext(style_name)[0] # 移除扩展名
# 尝试模糊匹配缓存中的文件名
for cache_key in self._style_cache.keys():
cache_key_lower = cache_key.lower()
style_name_lower = style_name.lower()
# 完全匹配
if cache_key_lower == style_name_lower:
return self._style_cache[cache_key]
# 部分匹配
# 攻略风格
if ("攻略" in style_name_lower or "干货" in style_name_lower) and "攻略" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Style文件: '{cache_key}' 匹配 '{style_name}'")
return self._style_cache[cache_key]
# 轻奢风格
if "轻奢" in style_name_lower and "轻奢" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Style文件: '{cache_key}' 匹配 '{style_name}'")
return self._style_cache[cache_key]
# 推荐风格
if ("推荐" in style_name_lower or "种草" in style_name_lower) and "推荐" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Style文件: '{cache_key}' 匹配 '{style_name}'")
return self._style_cache[cache_key]
# 美食风格
if "美食" in style_name_lower and "美食" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Style文件: '{cache_key}' 匹配 '{style_name}'")
return self._style_cache[cache_key]
# 如果没有在缓存中找到模糊匹配尝试从prompts_dir加载
if self.prompts_dir:
style_path = os.path.join(self.prompts_dir, "Style", style_file)
if os.path.exists(style_path):
content = ResourceLoader.load_file_content(style_path)
if content:
# 保存到缓存
self._style_cache[style_name] = content
self._style_cache[style_file] = content
return content
# 如果直接加载失败,尝试列出目录中的所有文件并进行模糊匹配
style_dir = os.path.join(self.prompts_dir, "Style")
if os.path.isdir(style_dir):
try:
files = os.listdir(style_dir)
style_name_lower = style_name.lower()
for file in files:
file_lower = file.lower()
# 检查关键词匹配
matched = False
if ("攻略" in style_name_lower or "干货" in style_name_lower) and "攻略" in file_lower:
matched = True
elif "轻奢" in style_name_lower and "轻奢" in file_lower:
matched = True
elif ("推荐" in style_name_lower or "种草" in style_name_lower) and "推荐" in file_lower:
matched = True
elif "美食" in style_name_lower and "美食" in file_lower:
matched = True
if matched:
matched_path = os.path.join(style_dir, file)
logging.info(f"模糊匹配 - 在目录中找到部分匹配的Style文件: '{file}' 匹配 '{style_name}'")
content = ResourceLoader.load_file_content(matched_path)
if content:
# 保存到缓存
self._style_cache[style_name] = content
self._style_cache[file] = content
return content
except Exception as e:
logging.warning(f"尝试列出Style目录内容时出错: {e}")
return None
def _get_demand_content(self, demand_name):
"""获取Demand文件内容优先从缓存获取如果不存在则尝试从目录加载"""
# 首先检查缓存
if demand_name in self._demand_cache:
return self._demand_cache[demand_name]
# 确保有扩展名
if not demand_name.lower().endswith('.txt'):
demand_file = f"{demand_name}.txt"
else:
demand_file = demand_name
demand_name = os.path.splitext(demand_name)[0] # 移除扩展名
# 尝试模糊匹配缓存中的文件名
for cache_key in self._demand_cache.keys():
cache_key_lower = cache_key.lower()
demand_name_lower = demand_name.lower()
# 完全匹配
if cache_key_lower == demand_name_lower:
return self._demand_cache[cache_key]
# 部分匹配:检查需求名称是否是缓存键的一部分,或者缓存键是否是需求名称的一部分
# 例如"亲子家庭文旅需求"能匹配到"亲子向文旅需求"
if "亲子" in demand_name_lower and "亲子" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
if "情侣" in demand_name_lower and "情侣" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
if "职场" in demand_name_lower and "职场" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
if "学生" in demand_name_lower and "学生" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
if "银发" in demand_name_lower and "夕阳红" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
if "夕阳红" in demand_name_lower and "银发" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
if "周边" in demand_name_lower and "周边" in cache_key_lower:
logging.info(f"模糊匹配 - 找到部分匹配的Demand文件: '{cache_key}' 匹配 '{demand_name}'")
return self._demand_cache[cache_key]
# 如果没有在缓存中找到模糊匹配尝试从prompts_dir加载向后兼容
if self.prompts_dir:
demand_path = os.path.join(self.prompts_dir, "Demand", demand_file)
if os.path.exists(demand_path):
content = ResourceLoader.load_file_content(demand_path)
if content:
# 保存到缓存
self._demand_cache[demand_name] = content
self._demand_cache[demand_file] = content
return content
# 如果直接加载失败,尝试列出目录中的所有文件并进行模糊匹配
demand_dir = os.path.join(self.prompts_dir, "Demand")
if os.path.isdir(demand_dir):
try:
files = os.listdir(demand_dir)
demand_name_lower = demand_name.lower()
for file in files:
file_lower = file.lower()
# 检查关键词匹配
matched = False
if "亲子" in demand_name_lower and "亲子" in file_lower:
matched = True
elif "情侣" in demand_name_lower and "情侣" in file_lower:
matched = True
elif "职场" in demand_name_lower and "职场" in file_lower:
matched = True
elif "学生" in demand_name_lower and "学生" in file_lower:
matched = True
elif ("银发" in demand_name_lower or "夕阳红" in demand_name_lower) and ("银发" in file_lower or "夕阳红" in file_lower):
matched = True
elif "周边" in demand_name_lower and "周边" in file_lower:
matched = True
if matched:
matched_path = os.path.join(demand_dir, file)
logging.info(f"模糊匹配 - 在目录中找到部分匹配的Demand文件: '{file}' 匹配 '{demand_name}'")
content = ResourceLoader.load_file_content(matched_path)
if content:
# 保存到缓存
self._demand_cache[demand_name] = content
self._demand_cache[file] = content
return content
except Exception as e:
logging.warning(f"尝试列出Demand目录内容时出错: {e}")
# 如果所有尝试都失败
logging.warning(f"未能找到Demand文件: '{demand_name}',尝试过以下位置: 缓存, {self.prompts_dir}/Demand/")
return None
def _get_all_refer_contents(self):
"""获取所有Refer文件内容"""
# 如果缓存中有内容,先使用缓存
if self._refer_cache:
refer_content_all = ""
for filename, content in self._refer_cache.items():
refer_content_all += f"--- Refer File: {filename} ---\n{content}\n\n"
return refer_content_all
# 如果缓存为空尝试从prompts_dir加载向后兼容
refer_content_all = ""
if self.prompts_dir:
refer_dir = os.path.join(self.prompts_dir, "Refer")
if os.path.isdir(refer_dir):
refer_files = [f for f in os.listdir(refer_dir) if os.path.isfile(os.path.join(refer_dir, f))]
for refer_file in refer_files:
refer_path = os.path.join(refer_dir, refer_file)
content = ResourceLoader.load_file_content(refer_path)
if content:
refer_content_all += f"--- Refer File: {refer_file} ---\n{content}\n\n"
# 保存到缓存
self._refer_cache[refer_file] = content
return refer_content_all
2025-04-22 14:19:21 +08:00
def get_topic_prompts(self):
"""Constructs the system and user prompts for topic generation."""
2025-04-22 17:36:29 +08:00
logging.info("Constructing prompts for topic generation...")
2025-04-22 14:19:21 +08:00
try:
# --- System Prompt ---
2025-04-24 21:44:05 +08:00
system_prompt = self._system_prompt_cache.get("topic")
2025-04-22 14:19:21 +08:00
if not system_prompt:
2025-04-24 21:44:05 +08:00
if not self.topic_system_prompt_path:
logging.error("Topic system prompt path not provided during PromptManager initialization.")
return None, None
system_prompt = ResourceLoader.load_file_content(self.topic_system_prompt_path)
if system_prompt:
self._system_prompt_cache["topic"] = system_prompt
else:
logging.error(f"Failed to load topic system prompt from '{self.topic_system_prompt_path}'.")
return None, None
2025-04-22 14:19:21 +08:00
# --- User Prompt ---
2025-04-24 21:44:05 +08:00
base_user_prompt = self._user_prompt_cache.get("topic")
if not base_user_prompt:
if not self.topic_user_prompt_path:
logging.error("Topic user prompt path not provided during PromptManager initialization.")
return None, None
base_user_prompt = ResourceLoader.load_file_content(self.topic_user_prompt_path)
if base_user_prompt:
self._user_prompt_cache["topic"] = base_user_prompt
else:
logging.error(f"Failed to load base topic user prompt from '{self.topic_user_prompt_path}'.")
return None, None
2025-04-22 14:19:21 +08:00
2025-04-24 21:44:05 +08:00
# --- Build the dynamic part of the user prompt ---
2025-04-22 14:19:21 +08:00
user_prompt_dynamic = "你拥有的创作资料如下:\n"
2025-04-24 21:44:05 +08:00
# 添加prompts_config配置的文件信息
if self.prompts_config:
for config_item in self.prompts_config:
prompt_type = config_item.get("type", "").lower()
file_paths = config_item.get("file_path", [])
if file_paths:
user_prompt_dynamic += f"{prompt_type.capitalize()}文件列表:\n"
for path in file_paths:
filename = os.path.basename(path)
user_prompt_dynamic += f"- {filename}\n"
user_prompt_dynamic += "\n"
# 兼容旧配置Add genPrompts directory structure
elif self.prompts_dir and os.path.isdir(self.prompts_dir):
2025-04-22 14:19:21 +08:00
try:
gen_prompts_list = os.listdir(self.prompts_dir)
2025-04-22 14:19:21 +08:00
for gen_prompt_folder in gen_prompts_list:
folder_path = os.path.join(self.prompts_dir, gen_prompt_folder)
2025-04-22 14:19:21 +08:00
if os.path.isdir(folder_path):
try:
# List files, filter out subdirs if needed
gen_prompts_files = [f for f in os.listdir(folder_path) if os.path.isfile(os.path.join(folder_path, f))]
user_prompt_dynamic += f"{gen_prompt_folder}\n{gen_prompts_files}\n"
except OSError as e:
2025-04-22 17:36:29 +08:00
logging.warning(f"Could not list directory {folder_path}: {e}")
2025-04-22 14:19:21 +08:00
except OSError as e:
logging.warning(f"Could not list base prompts directory {self.prompts_dir}: {e}")
2025-04-22 14:19:21 +08:00
else:
2025-04-24 21:44:05 +08:00
logging.warning(f"Neither prompts_config nor prompts_dir provided or valid.")
2025-04-22 14:19:21 +08:00
# Add resource directory contents
for dir_info in self.resource_dir_config:
2025-04-22 14:19:21 +08:00
source_type = dir_info.get("type", "UnknownType")
source_file_paths = dir_info.get("file_path", [])
for file_path in source_file_paths:
# Use ResourceLoader's static method
file_content = ResourceLoader.load_file_content(file_path)
if file_content:
user_prompt_dynamic += f"{source_type}信息:\n{os.path.basename(file_path)}\n{file_content}\n\n"
else:
2025-04-22 17:36:29 +08:00
logging.warning(f"Could not load resource file {file_path}")
2025-04-22 14:19:21 +08:00
# Add dateline information (optional)
2025-04-24 21:44:05 +08:00
if self._dateline_cache:
user_prompt_dynamic += f"\n{self._dateline_cache}"
else:
user_prompt_dir = os.path.dirname(self.topic_user_prompt_path)
dateline_path = os.path.join(user_prompt_dir, "2025各月节日宣传节点时间表.md") # Consider making this configurable
if os.path.exists(dateline_path):
dateline_content = ResourceLoader.load_file_content(dateline_path)
if dateline_content:
self._dateline_cache = dateline_content
user_prompt_dynamic += f"\n{dateline_content}"
2025-04-22 14:19:21 +08:00
# Combine dynamic part, base template, and final parameters
user_prompt = user_prompt_dynamic + base_user_prompt
user_prompt += f"\n选题数量:{self.topic_gen_num}\n选题日期:{self.topic_gen_date}\n"
2025-04-22 14:19:21 +08:00
2025-04-22 17:36:29 +08:00
logging.info(f"Topic prompts constructed. System: {len(system_prompt)} chars, User: {len(user_prompt)} chars.")
2025-04-22 14:19:21 +08:00
return system_prompt, user_prompt
except Exception as e:
2025-04-22 17:36:29 +08:00
logging.exception("Error constructing topic prompts:")
2025-04-22 14:19:21 +08:00
return None, None
def get_content_prompts(self, topic_item):
"""Constructs the system and user prompts for content generation based on a topic item."""
2025-04-22 17:36:29 +08:00
logging.info(f"Constructing content prompts for topic: {topic_item.get('object', 'N/A')}...")
2025-04-22 14:19:21 +08:00
try:
# --- System Prompt ---
2025-04-24 21:44:05 +08:00
system_prompt = self._system_prompt_cache.get("content")
2025-04-22 14:19:21 +08:00
if not system_prompt:
2025-04-24 21:44:05 +08:00
if not self.content_system_prompt_path:
logging.error("Content system prompt path not provided during PromptManager initialization.")
return None, None
system_prompt = ResourceLoader.load_file_content(self.content_system_prompt_path)
if system_prompt:
self._system_prompt_cache["content"] = system_prompt
else:
logging.error(f"Failed to load content system prompt from '{self.content_system_prompt_path}'.")
return None, None
2025-04-22 14:19:21 +08:00
2025-04-24 21:44:05 +08:00
# --- User Prompt ---
2025-04-22 14:19:21 +08:00
user_prompt = ""
2025-04-22 15:12:46 +08:00
# 1. 添加Demand部分 (直接使用 topic_item['logic'] 的描述性文本)
2025-04-22 14:19:21 +08:00
try:
2025-04-22 15:12:46 +08:00
demand_description = topic_item.get('logic')
if demand_description:
user_prompt += f"Demand Logic:\n{demand_description}\n"
else:
2025-04-22 17:36:29 +08:00
logging.warning("Warning: 'logic' key missing or empty in topic_item for Demand prompt.")
2025-04-22 14:19:21 +08:00
except Exception as e:
2025-04-22 17:36:29 +08:00
logging.exception("Error processing Demand description:")
2025-04-22 14:19:21 +08:00
2025-04-22 18:41:20 +08:00
# 2. Object Info - 先列出所有可用文件,再注入匹配文件的内容
2025-04-22 14:19:21 +08:00
try:
2025-04-22 18:41:20 +08:00
object_name_from_topic = topic_item.get('object') # e.g., "尚书第建筑群"
object_file_basenames = []
matched_object_file_path = None
matched_object_basename = None
# 遍历查找 Object 文件
2025-04-24 21:44:05 +08:00
for dir_info in self.resource_dir_config:
2025-04-22 14:19:21 +08:00
if dir_info.get("type") == "Object":
for file_path in dir_info.get("file_path", []):
2025-04-22 18:41:20 +08:00
basename = os.path.basename(file_path)
object_file_basenames.append(basename)
# 尝试匹配当前 topic 的 object (仅当尚未找到匹配时)
if object_name_from_topic and not matched_object_file_path:
cleaned_resource_name = basename
if cleaned_resource_name.startswith("景点信息-"):
cleaned_resource_name = cleaned_resource_name[len("景点信息-"):]
if cleaned_resource_name.endswith(".txt"):
cleaned_resource_name = cleaned_resource_name[:-len(".txt")]
if cleaned_resource_name and cleaned_resource_name in object_name_from_topic:
matched_object_file_path = file_path
matched_object_basename = basename
# 注意:这里不 break继续收集所有文件名
# 构建提示词 - Part 1: 文件列表
if object_file_basenames:
user_prompt += "Object信息:\n"
# user_prompt += f"{object_file_basenames}\n\n" # 直接打印列表可能不够清晰
for fname in object_file_basenames:
user_prompt += f"- {fname}\n"
user_prompt += "\n" # 加一个空行
logging.info(f"Listed {len(object_file_basenames)} available object files.")
2025-04-22 14:19:21 +08:00
else:
2025-04-22 18:41:20 +08:00
logging.warning("No resource directory entry found with type 'Object', or it has no file paths.")
# 构建提示词 - Part 2: 注入匹配文件内容
if matched_object_file_path:
logging.info(f"Attempting to load content for matched object file: {matched_object_basename}")
matched_object_content = ResourceLoader.load_file_content(matched_object_file_path)
if matched_object_content:
user_prompt += f"{matched_object_basename}\n{matched_object_content}\n\n"
logging.info(f"Successfully loaded and injected content for: {matched_object_basename}")
else:
logging.warning(f"Object file matched ({matched_object_basename}) but could not be loaded or is empty.")
elif object_name_from_topic: # 只有当 topic 中指定了 object 但没找到匹配文件时才警告
logging.warning(f"Could not find a matching Object resource file to inject content for '{object_name_from_topic}'. Only the list of files was provided.")
except KeyError:
logging.warning("Warning: 'object' key potentially missing in topic_item.")
2025-04-22 14:19:21 +08:00
except Exception as e:
2025-04-22 18:41:20 +08:00
logging.exception("Error processing Object prompt section:")
2025-04-22 14:19:21 +08:00
2025-04-22 15:12:46 +08:00
# 3. 添加Product信息 (if applicable)
2025-04-22 14:19:21 +08:00
try:
product_name = topic_item.get('product')
2025-04-22 15:12:46 +08:00
product_logic_description = topic_item.get('product_logic') # Directly use this description
if product_name:
# Add Product Logic description first (if available)
if product_logic_description:
user_prompt += f"Product Logic:\n{product_logic_description}\n"
else:
2025-04-22 17:36:29 +08:00
logging.warning(f"Warning: 'product_logic' key missing or empty for product '{product_name}'.")
2025-04-22 15:12:46 +08:00
# Then, load Product Info file
2025-04-22 14:19:21 +08:00
product_file_path = None
2025-04-24 21:44:05 +08:00
for dir_info in self.resource_dir_config:
2025-04-22 14:19:21 +08:00
if dir_info.get("type") == "Product":
for file_path in dir_info.get("file_path", []):
if product_name in os.path.basename(file_path):
2025-04-22 14:19:21 +08:00
product_file_path = file_path
break
if product_file_path: break
if product_file_path:
product_content = ResourceLoader.load_file_content(product_file_path)
if product_content:
2025-04-22 15:12:46 +08:00
user_prompt += f"Product Info:\n{product_content}\n"
2025-04-22 14:19:21 +08:00
else:
2025-04-22 17:36:29 +08:00
logging.warning(f"Product file could not be loaded: {product_file_path}")
2025-04-22 14:19:21 +08:00
else:
2025-04-22 17:36:29 +08:00
logging.warning(f"Product file path not found in config for: {product_name}")
except KeyError:
logging.warning("Warning: Missing 'product' key in topic_item for Product prompt.")
2025-04-22 14:19:21 +08:00
except Exception as e:
2025-04-22 17:36:29 +08:00
logging.exception("Error processing Product prompt:")
2025-04-22 14:19:21 +08:00
2025-04-24 21:44:05 +08:00
# 4. 添加Style信息 (加载文件 based on topic_item['style'])
2025-04-22 14:19:21 +08:00
try:
2025-04-24 21:44:05 +08:00
style_name = topic_item.get('style')
if style_name:
style_content = self._get_style_content(style_name)
2025-04-22 15:12:46 +08:00
if style_content:
2025-04-24 21:44:05 +08:00
user_prompt += f"Style Info:\n{style_content}\n"
2025-04-22 15:12:46 +08:00
else:
2025-04-24 21:44:05 +08:00
logging.warning(f"Style file not found or empty for: {style_name}")
2025-04-22 14:19:21 +08:00
else:
2025-04-22 17:36:29 +08:00
logging.warning("Warning: 'style' key missing or empty in topic_item.")
2025-04-22 14:19:21 +08:00
except Exception as e:
2025-04-22 17:36:29 +08:00
logging.exception("Error processing Style prompt:")
2025-04-22 14:19:21 +08:00
2025-04-24 21:44:05 +08:00
# 5. 添加Target Audience信息 (加载文件 based on topic_item['target_audience'])
2025-04-22 14:19:21 +08:00
try:
2025-04-24 21:44:05 +08:00
target_audience_name = topic_item.get('target_audience')
if target_audience_name:
target_audience_content = self._get_demand_content(target_audience_name)
2025-04-22 15:12:46 +08:00
if target_audience_content:
user_prompt += f"Target Audience Info:\n{target_audience_content}\n"
else:
2025-04-24 21:44:05 +08:00
logging.warning(f"Target Audience file not found or empty for: {target_audience_name}")
else:
2025-04-22 17:36:29 +08:00
logging.warning("Warning: 'target_audience' key missing or empty in topic_item.")
2025-04-22 15:12:46 +08:00
except Exception as e:
2025-04-22 17:36:29 +08:00
logging.exception("Error processing Target Audience prompt:")
2025-04-22 15:12:46 +08:00
2025-04-24 21:44:05 +08:00
# 6. 添加Refer信息 (加载所有Refer文件的内容)
2025-04-22 15:12:46 +08:00
try:
2025-04-24 21:44:05 +08:00
refer_content_all = self._get_all_refer_contents()
if refer_content_all:
user_prompt += f"Refer Info:\n{refer_content_all}"
2025-04-22 14:19:21 +08:00
else:
2025-04-24 21:44:05 +08:00
logging.warning("No content loaded from Refer files.")
2025-04-22 14:19:21 +08:00
except Exception as e:
2025-04-24 21:44:05 +08:00
logging.exception("Error processing Refer files:")
2025-04-22 15:12:46 +08:00
# --- End of prompt construction logic ---
2025-04-22 14:19:21 +08:00
2025-04-22 17:36:29 +08:00
logging.info(f"Content prompts constructed. System: {len(system_prompt)} chars, User: {len(user_prompt)} chars.")
2025-04-22 14:19:21 +08:00
return system_prompt, user_prompt
except KeyError as e:
# Catch potential KeyErrors from accessing topic_item if a required key is missing early on
2025-04-22 17:36:29 +08:00
logging.error(f"Error constructing content prompts: Missing essential key '{e}' in topic_item: {topic_item}")
2025-04-22 14:19:21 +08:00
return None, None
except Exception as e:
2025-04-22 17:36:29 +08:00
logging.exception("Error constructing content prompts:")
2025-04-22 14:19:21 +08:00
return None, None