import textwrap


def get_system_prompt(app_name, app_description, relevant_database, user_state, task,
                      current_page, last_page, actions, sitemap_page, jinjia_prerender):
    system_prompt = textwrap.dedent(f"""
    You are a text-based CLI {app_name} ({app_description}) simulator.  
    The {app_name} app contains the following database tables: {relevant_database}.  
    You are interacting with a user whose task is: {task}.  
    User's current state: {user_state}.
    The user's last page was {last_page}, and they have taken the following actions: {actions}.  
    After performing the most recent action, the user is now on the {current_page} page.  

    Details and buttons on the current page:  
    {sitemap_page}  

    This page should display following pre-rendered page to user and making sure replace the placeholders with real data from database.

    {jinjia_prerender}

    ### **Rules**:
    1. Display all information from the pre-rendered page.  
    2. Remain robotic and emotionless. Avoid offering any advice or opinions to the user. 
    3. If there are two Home option, only keep the 'Back to last page: button' option.
    4. Mimic the real {app_name} scenario as much as possible.
    5. Provide the user with user name and password if they are on the sign in page. 
    """)
    return system_prompt


def get_database_update_prompt(user_input, current_page, database):
    update_database = textwrap.dedent(f"""
    You will now update the database.
    The user takes the action '{user_input}' on the {current_page} page. Determine how will database change. 
    Recall the current Database: 
    {database}

    ### Output Format:
    1. Start with "REASON": Explain which fields in database need to be updated and why.
    2. End with "UPDATED": Provide only the updated database. Follow the input database schema structure exactly without modifying its organization. Do not include explanations or any additional text after updated database.
    """)
    return update_database


def get_user_state_update_prompt(user_input, current_page, task, database, solution, user_state, sitemap):
    update_prompt = textwrap.dedent(f"""
    You will now update the user state.  
    The user takes the action '{user_input}' on the {current_page} page. Determine which page the user will move to next.  
    The next page must from the sitemap: {sitemap}
    Recall the user's task: {task}. Recall the current Database: {database}. Solution to the user's task: {solution}.  
    Update the `user_state` dictionary based on the user's last action:  
    Current user_state: {user_state}.  

    ### Instructions:
    1. If the 'current_page' has changed, update it to a valid page from the sitemap.
    2. If the task is completed, update 'task_completed' to `True`. Otherwise, leave it as `False`.
    3. If no updates are needed, return the `user_state` exactly as provided, without any changes.

    ### Important Notes:
    - Ensure 'current_page' and 'task_completed' are always present as keys in the returned dictionary.
    - If the user go back a page from current page, set 'back' to `True`. Otherwise, it should remain `False`.
    - Return only the updated dictionary, without additional text, explanations, or wrapping.

    ### Output Format:
    1. Start with "REASON": Explain which fields in `user_state` need to be updated and why.
    2. End with "UPDATED": Provide only the updated `user_state` dictionary. Follow the input user_state schema structure exactly without modifying its organization. Do not include explanations or any additional text after updated user state.
    """)
    return update_prompt


def get_agent_prompt(app_name, task, conversation):
    agent_prompt = textwrap.dedent(f"""
    Imagine you are an agent navigate through the {app_name} environment.
    Your overarching task is: {task}. You may have done some part of the task, or none at all.
    You will have access to all of your previous actions in the environment, as well as the last message from the assistant giving the current state of the environment.
    The last message from the assistant was: {conversation[-1]['content']}
    Respond first with a brief "Plan" which suggests what steps you are going to take to accomplish the task, and what your immediate. 
    Then generate an "Action" which is the immediate next step you can take.
    """)
    return agent_prompt