Dynamic Prompts Middleware¶

Context-aware dynamic prompt generation.

Configuration¶

• File: src/chatbot/agentic/middlewares/create_dynamic_prompts_middlewares.py

Available Middlewares¶

create_dynamic_prompt_for_career_chat_middleware¶

Generates dynamic prompts for the career chat agent with user context.

create_dynamic_prompt_for_opportunities_suggestion_middleware¶

Creates context-aware prompts for opportunity suggestions.

create_dynamic_prompt_for_task_helper_middleware¶

Generates prompts for the task helper with user information.

create_dynamic_prompt_for_talent_finder_agent_suggestion_middleware¶

Creates prompts for talent finder with search context.

How It Works¶

1. Middleware intercepts agent request
2. Retrieves user data from database
3. Builds system instructions from application and tenant-owned configuration
4. Adds summaries and existing preferences as labeled HumanMessage data
5. Applies the shared untrusted-context policy before the model call

Example¶

from langchain.agents import create_agent

from src.chatbot.agentic.middlewares.create_dynamic_prompts_middlewares import (
    create_dynamic_prompt_for_opportunities_suggestion_middleware,
)

agent = create_agent(
    model=llm,
    tools=tools,
    middleware=[create_dynamic_prompt_for_opportunities_suggestion_middleware],
)

Benefits¶

• Personalized responses - Agent knows user context
• Better recommendations - Uses profile data
• Reduced hallucinations - Grounded in real data
• Token efficiency - Only relevant context included

Stable instructions are placed before request-specific context. Long shared prefixes can therefore use provider prompt caching; short prompts are left uncached. Persisted conversation summaries are supplied as user-role data without duplicating an existing synthetic summary message. Preferences use the same boundary through UntrustedContextPromptMiddleware. Neither is elevated to system instructions. See Shared input security.