♻️ standardize prompt vars with Jinja2 templating

# Conflicts:
#	backend/services/elasticsearch_service.py
#	backend/utils/attachment_utils.py

Author: Summer-Si

backend/agents/create_agent_info.py

@@ -13,9 +13,9 @@
 from services.tenant_config_service import get_selected_knowledge_list
 from utils.prompt_template_utils import get_agent_prompt_template
 from utils.config_utils import config_manager, tenant_config_manager, get_model_name_from_config
-from smolagents.agents import populate_template
 from smolagents.utils import BASE_BUILTIN_MODULES
 from services.memory_config_service import build_memory_context
+from jinja2 import Template, StrictUndefined
 
 from nexent.memory.memory_service import search_memory_in_levels
 
@@ -98,9 +98,8 @@ async def create_agent_config(agent_id, tenant_id, user_id, language: str = 'zh'
         # TODO: 前端展示"已抽取 xx 条回忆"
 
     # Assemble system_prompt
-    system_prompt = populate_template(
-        prompt_template["system_prompt"],
-        variables={
+    if (duty_prompt or constraint_prompt or few_shots_prompt):
+        system_prompt = Template(prompt_template["system_prompt"], undefined=StrictUndefined).render({
             "duty": duty_prompt,
             "constraint": constraint_prompt,
             "few_shots": few_shots_prompt,
@@ -110,8 +109,9 @@ async def create_agent_config(agent_id, tenant_id, user_id, language: str = 'zh'
             "APP_NAME": app_name,
             "APP_DESCRIPTION": app_description,
             "memory_list": memory_list
-        },
-    ) if (duty_prompt or constraint_prompt or few_shots_prompt) else agent_info.get("prompt", "")
+        })
+    else:
+        system_prompt = agent_info.get("prompt", "")
 
     # special logic
     try:

backend/prompts/analyze_file.yaml

@@ -3,7 +3,7 @@
 
 image_analysis:
   system_prompt: |-
-    用户提出了一个问题：{query}，请从回答这个问题的角度精简、仔细描述一下这个图片，200字以内。
+    用户提出了一个问题：{{ query }}，请从回答这个问题的角度精简、仔细描述一下这个图片，200字以内。
     
     **图片分析要求：**
     1. 重点关注与用户问题相关的图片内容
@@ -16,7 +16,7 @@ image_analysis:
 
 long_text_analysis:
   system_prompt: |-
-    用户提出了一个问题：{query}，请从回答这个问题的角度精简、仔细描述一下这段文本，200字以内。
+    用户提出了一个问题：{{ query }}，请从回答这个问题的角度精简、仔细描述一下这段文本，200字以内。
     
     **文本分析要求：**
     1. 重点提取与用户问题相关的文本内容
@@ -25,4 +25,4 @@ long_text_analysis:
     4. 避免冗余信息，专注于问题相关内容
     
   user_prompt: |
-    请仔细阅读并分析这段文本：{file_context} 
+    请仔细阅读并分析这段文本：{{ file_context }} 

backend/prompts/analyze_file_en.yaml

@@ -3,7 +3,7 @@
 
 image_analysis:
   system_prompt: |-
-    The user has asked a question: {query}. Please provide a concise and careful description of this image from the perspective of answering this question, within 200 words.
+    The user has asked a question: {{ query }}. Please provide a concise and careful description of this image from the perspective of answering this question, within 200 words.
     
     **Image Analysis Requirements:**
     1. Focus on image content relevant to the user's question
@@ -16,7 +16,7 @@ image_analysis:
 
 long_text_analysis:
   system_prompt: |-
-    The user has asked a question: {query}. Please provide a concise and careful description of this text from the perspective of answering this question, within 200 words.
+    The user has asked a question: {{ query }}. Please provide a concise and careful description of this text from the perspective of answering this question, within 200 words.
     
     **Text Analysis Requirements:**
     1. Focus on extracting text content relevant to the user's question
@@ -25,4 +25,4 @@ long_text_analysis:
     4. Avoid redundant information, focus on question-relevant content
     
   user_prompt: |
-    Please carefully read and analyze this text: {file_context} 
+    Please carefully read and analyze this text: {{ file_context }}

backend/prompts/knowledge_summary_agent.yaml

@@ -13,6 +13,4 @@ system_prompt: |-
   
 user_prompt: |
   请根据以下关键词内容生成一个简洁的知识库总结，不超过150个字：
-  关键词内容：{content}
- 
-    
+  关键词内容：{{ content }}

backend/prompts/knowledge_summary_agent_en.yaml

@@ -13,4 +13,4 @@ system_prompt: |-
   
 user_prompt: |
   Please generate a concise knowledge base summary based on the following keyword content, no more than 150 words:
-  Keyword content: {content} 
+  Keyword content: {{ content }} 

backend/services/conversation_management_service.py

@@ -250,8 +250,7 @@ def call_llm_for_title(content: str, tenant_id: str, language: str = 'zh') -> st
         api_key=model_config.get("api_key", ""), temperature=0.7, top_p=0.95)
 
     # Build messages
-    compiled_template = Template(prompt_template["USER_PROMPT"], undefined=StrictUndefined)
-    user_prompt = compiled_template.render({
+    user_prompt = Template(prompt_template["USER_PROMPT"], undefined=StrictUndefined).render({
         "content": content
     })
     messages = [{"role": "system",

backend/services/elasticsearch_service.py

@@ -27,6 +27,7 @@
 from utils.config_utils import tenant_config_manager, get_model_name_from_config
 from utils.file_management_utils import get_all_files_status, get_file_size
 from utils.prompt_template_utils import get_knowledge_summary_prompt_template
+from jinja2 import Template, StrictUndefined
 from database.knowledge_db import create_knowledge_record, get_knowledge_record, update_knowledge_record, delete_knowledge_record
 from database.attachment_db import delete_file
 
@@ -51,10 +52,15 @@ def generate_knowledge_summary_stream(keywords: str, language: str, tenant_id: s
 
     # Load prompt words based on language
     prompts = get_knowledge_summary_prompt_template(language)
+
+    # Render templates using Jinja2
+    system_prompt = Template(prompts['system_prompt'], undefined=StrictUndefined).render({})
+    user_prompt = Template(prompts['user_prompt'], undefined=StrictUndefined).render({'content': keywords})
+
     # Build messages
     messages: List[ChatCompletionMessageParam] = [
-        {"role": "system", "content": prompts['system_prompt']},
-        {"role": "user", "content": prompts['user_prompt'].format(content=keywords)}
+        {"role": "system", "content": system_prompt},
+        {"role": "user", "content": user_prompt}
     ]
 
     # Get model configuration from tenant config manager

backend/services/prompt_service.py

@@ -196,8 +196,7 @@ def join_info_for_generate_system_prompt(prompt_for_generate, sub_agent_info_lis
     agent_description = "\n".join(
         [f"- {sub_agent_info['name']}: {sub_agent_info['description']}" for sub_agent_info in sub_agent_info_list])
     # Generate content using template
-    compiled_template = Template(prompt_for_generate["USER_PROMPT"], undefined=StrictUndefined)
-    content = compiled_template.render({
+    content = Template(prompt_for_generate["USER_PROMPT"], undefined=StrictUndefined).render({
         "tool_description": tool_description,
         "agent_description": agent_description,
         "task_description": task_description,
@@ -237,8 +236,7 @@ def fine_tune_prompt(system_prompt: str, command: str, tenant_id: str, language:
         with open(fine_tune_config_path, "r", encoding="utf-8") as f:
             prompt_for_fine_tune = yaml.safe_load(f)
 
-        compiled_template = Template(prompt_for_fine_tune["FINE_TUNE_USER_PROMPT"], undefined=StrictUndefined)
-        content = compiled_template.render({
+        content = Template(prompt_for_fine_tune["FINE_TUNE_USER_PROMPT"], undefined=StrictUndefined).render({
             "prompt": system_prompt,
             "command": command
         })

backend/utils/attachment_utils.py

@@ -3,6 +3,7 @@
 
 from utils.config_utils import tenant_config_manager, get_model_name_from_config
 from utils.prompt_template_utils import get_analyze_file_prompt_template
+from jinja2 import Template, StrictUndefined
 
 from nexent.core.models.openai_vlm import OpenAIVLModel
 from nexent.core.models.openai_long_context_model import OpenAILongContextModel
@@ -37,8 +38,8 @@ def convert_image_to_text(query: str, image_input: Union[str, BinaryIO], tenant_
 
     # Load prompts from yaml file
     prompts = get_analyze_file_prompt_template(language)
-    system_prompt = prompts['image_analysis']['system_prompt'].format(query=query)
-    
+    system_prompt = Template(prompts['image_analysis']['system_prompt'], undefined=StrictUndefined).render({'query': query})
+
     return image_to_text_model.analyze_image(image_input=image_input, system_prompt=system_prompt).content
 
 
@@ -65,7 +66,7 @@ def convert_long_text_to_text(query: str, file_context: str, tenant_id: str, lan
 
     # Load prompts from yaml file
     prompts = get_analyze_file_prompt_template(language)
-    system_prompt = prompts['long_text_analysis']['system_prompt'].format(query=query)
-    user_prompt = prompts['long_text_analysis']['user_prompt'].format(file_context=file_context)
+    system_prompt = Template(prompts['long_text_analysis']['system_prompt'], undefined=StrictUndefined).render({'query': query})
+    user_prompt = Template(prompts['long_text_analysis']['user_prompt'], undefined=StrictUndefined).render({'file_context': file_context})
 
     return long_text_to_text_model.analyze_long_text(file_context, system_prompt, user_prompt)

sdk/nexent/core/agents/core_agent.py

@@ -7,7 +7,7 @@
 from rich.console import Group
 from rich.text import Text
 
-from smolagents.agents import CodeAgent, populate_template, handle_agent_output_types, AgentError, AgentType
+from smolagents.agents import CodeAgent, handle_agent_output_types, AgentError, AgentType
 from smolagents.local_python_executor import fix_final_answer_code
 from smolagents.memory import ActionStep, ToolCall, TaskStep, SystemPromptStep
 from smolagents.models import ChatMessage
@@ -16,6 +16,7 @@
     truncate_content
 
 from ..utils.observer import MessageObserver, ProcessType
+from jinja2 import Template, StrictUndefined
 
 
 def convert_code_format(text):
@@ -169,8 +170,9 @@ def __call__(self, task: str, **kwargs):
         """Adds additional prompting for the managed agent, runs it, and wraps the output.
         This method is called only by a managed agent.
         """
-        full_task = populate_template(self.prompt_templates["managed_agent"]["task"],
-            variables=dict(name=self.name, task=task), )
+        full_task = Template(self.prompt_templates["managed_agent"]["task"], undefined=StrictUndefined).render({
+            "name": self.name, "task": task, **self.state
+        })
         report = self.run(full_task, **kwargs)
 
         # When a sub-agent finishes running, return a marker
@@ -179,8 +181,9 @@ def __call__(self, task: str, **kwargs):
         except:
             self.observer.add_message(self.name, ProcessType.AGENT_FINISH, "")
 
-        answer = populate_template(self.prompt_templates["managed_agent"]["report"],
-            variables=dict(name=self.name, final_answer=report))
+        answer = Template(self.prompt_templates["managed_agent"]["report"], undefined=StrictUndefined).render({
+            "name": self.name, "final_answer": report
+        })
         if self.provide_run_summary:
             answer += "\n\nFor more detail, find below a summary of this agent's work:\n<summary_of_work>\n"
             for message in self.write_memory_to_messages(summary_mode=True):