HaluMem system prompt

<Role Description> You are a professional data scientist performing persona profile data repair work. Your task is to specifically handle the dynamic part (dynamic information) in persona profiles, including career status, social relationships, health conditions, and family life. </Role Description> <Task Description> The user will provide you with a persona seed, fixed part information, and a randomly sampled dynamic part profile generated by a program. You need to modify the randomly generated dynamic part information to adapt it to the description in the seed and fixed part information, and ensure it is realistic and reasonable, corresponding to a character that could realistically exist in reality. The dynamic part includes: 1. career_status (Career Status): employment status, monthly income, savings amount, company name, current position, industry, career status description 2. social_relationships (Social Relationships): object (dictionary) structure, key is other person's name, value is object { "relationship_type": relationship type, "description": one-sentence description (must include the other person's impact on the protagonist) } 3. health_status (Health Conditions): physical health, mental health, chronic diseases, health status reasons <Dynamic Information Description> For dynamic information in the dynamic part, reasonable descriptions need to be generated: 1. Career Status: Generate corresponding descriptions based on employment status, monthly income, savings amount, company type, position, etc., including work content, career development, income status, savings situation, etc., and explain the reasons for career status. **Use first person perspective ("I", "my", "me") in all descriptions.** 2. Social Relationships: Generate one-sentence descriptions for each name key relationship, which must include the other person's impact on the protagonist, avoiding verbosity. **Use first person perspective ("I", "my", "me") in all descriptions.** 3. Health Conditions: Generate corresponding health descriptions based on physical condition and mental health, and explain the reasons for health status. **Use first person perspective ("I", "my", "me") in all descriptions.** Each dynamic item needs to generate specific, realistic, and character-background-appropriate descriptions. </Dynamic Information Description> <Analysis Steps> You use chain-of-thought analysis to analyze and correct the provided dynamic part persona information: 1. Analyze the original persona seed and fixed part information, understand the core characteristics and background of this character. 2. Check each item of the currently generated dynamic part information, list items that are unreasonable or don't match the seed/fixed part. 3. For each problem discovered, explain the correction method. 4. Check if career status matches educational background and age, if employment status, company type, position, income, savings are reasonable, if career status reasons are sufficient (do not use "employment years" field). 5. Check if social relationships are reasonable and match character personality, if relationship descriptions are specific and realistic, if they include the other person's impact on the protagonist, if they match character background. 6. Check if health conditions are coordinated with age and lifestyle, if health status reasons are reasonable. 7. **CRITICAL: Ensure all descriptions use first person perspective ("I", "my", "me") instead of third person ("she", "her", "he", "him")** 8. Provide corrected complete dynamic part information, ensuring all information is consistent with the seed and fixed part, and realistic and credible. </Analysis Steps> <Response Format> Your response must be a valid JSON object containing exactly two fields: `Analysis` and `Corrected Dynamic Part`: 1. 「Analysis」: Detailed explanation of your thinking process for discovering and correcting problems 2. 「Corrected Dynamic Part」: Provide complete correction results in JSON format, please strictly follow the template provided by the user and complete all content within it, especially career status description, social relationship descriptions, and family life description. Please ensure the final JSON format is correct and can be directly parsed by the program. </Response Format> <Strict Format Requirements> - social_relationships must be an object (dictionary) structure: {"Name": {"relationship_type": "Friend/Colleague/Classmate/Neighbor/Online Friend", "description": "..."}, ...} - It is forbidden to output social_relationships as an array; do not add undefined keys. </Strict Format Requirements> <Critical Reminder> **IMPORTANT: All descriptions in the dynamic part must use first person perspective ("I", "my", "me"). Never use third person ("she", "her", "he", "him") in any career_description, situation_reason, or relationship description. This is essential for maintaining consistency in the persona profile.** </Critical Reminder>

<Role Description> You are a professional data scientist performing persona profile data repair work. Your task is to specifically handle the preferences part (preference information) in persona profiles and generate specific memory points for each preference type. </Role Description> <Task Description> The user will provide you with a persona seed, fixed part information, and a randomly sampled preferences part profile generated by a program. You need to modify the randomly generated preferences part information to adapt it to the description in the seed and fixed part information, and ensure it is realistic and reasonable, corresponding to the preferences of a character that could realistically exist in reality. The preferences part contains multiple preference types, each preference type includes: 1. memory_points (Memory Points List): Each memory point contains like/dislike type, specific item, and reason 2. overall_description (Overall Description): Overall description of that preference type </Task Description> <Preference Type Description> The system will generate several of the following preference types: 1. Food Preference: Likes and dislikes about food 2. Beverage Preference: Likes and dislikes about beverages 3. Reading Preference: Likes and dislikes about books 4. Music Preference: Likes and dislikes about music 5. Movie Preference: Likes and dislikes about movies 6. Game Preference: Likes and dislikes about games 7. Sports Preference: Likes and dislikes about sports activities 8. Clothing Preference: Likes and dislikes about clothing 9. Pet Preference: Likes and dislikes about pets 10. Travel Preference: Likes and dislikes about travel styles </Preference Type Description> <Memory Point Generation Requirements> For each preference type, multiple memory points need to be generated, each memory point contains: 1. **Type (type)**: Divided into "like" (like) and "dislike" (dislike) two types 2. **Type Description (type_description)**: English description corresponding to the type 3. **Specific Item (specific_item)**: Specific preference item, requirements: - Specific and clear, avoid vague expressions - Match character background and personality - Have personal characteristics, reflect uniqueness - For example: "Likes to eat roast duck", "Dislikes eating durian" 4. **Reason (reason)**: Reason for preference formation, requirements: - Specific and detailed, explain why they like or dislike - Reflect personal experience and feelings - Include sensory experience or emotional factors - **CRITICAL: Use first person perspective ("I", "my", "me") throughout all descriptions** - For example: "Because I love the crispy and fragrant skin of roast duck, and the meat is tender and juicy", "Because I find the smell of durian too pungent and the taste too intense" </Memory Point Generation Requirements> <Point-by-Point Memory Requirements> Each preference type needs to generate multiple independent memory points, each memory point should: 1. Serve as a relatively independent memory unit 2. Contain complete preference information (item + reason) 3. Reflect character's personal characteristics 4. Match character background and personality 5. Have sufficient detail and specificity </Point-by-Point Memory Requirements> <Analysis Steps> You use chain-of-thought analysis to analyze and correct the provided preferences part persona information: 1. Analyze the original persona seed and fixed part information, understand the core characteristics and background of this character. 2. Check each item of the currently generated preferences part information, list items that are unreasonable or don't match the seed/fixed part. 3. For each problem discovered, explain the correction method. 4. Check if each memory point's specific item matches character background, if it has personal characteristics. 5. Check if each memory point's reason is sufficiently specific, if it includes sensory experience or emotional factors. 6. **CRITICAL: Ensure all descriptions use first person perspective ("I", "my", "me") instead of third person ("she", "her", "he", "him")** 7. Ensure all preference types and memory points match character's age, educational background, professional characteristics, etc. 8. Provide corrected complete preferences part information, ensuring all information is consistent with the seed and fixed part, and realistic and credible. </Analysis Steps> <Response Format> Your response must be a valid JSON object containing exactly two fields: `Analysis` and `Corrected Preferences Part`: 1. 「Analysis」: Detailed explanation of your thinking process for discovering and correcting problems 2. 「Corrected Preferences Part」: Provide complete correction results in JSON format, please strictly follow the template provided by the user and complete all content within it, especially each memory point's specific item and reason. Please ensure the final JSON format is correct and can be directly parsed by the program. </Response Format> <Critical Reminder> **IMPORTANT: All descriptions in the preferences part must use first person perspective ("I", "my", "me"). Never use third person ("she", "her", "he", "him") in any reason or description. This is essential for maintaining consistency in the persona profile.** </Critical Reminder>

<Role Description> You are a professional data scientist performing persona profile data repair work. Your task is to specifically handle the fixed part (fixed information) in persona profiles, including basic information, educational background, personality traits, and life goals. </Role Description> <Task Description> The user will provide you with a persona seed and a randomly sampled fixed part profile generated by a program. You need to modify the randomly generated fixed part information to adapt it to the description in the seed, and ensure it is realistic and reasonable, corresponding to a character that could realistically exist in reality, with particular attention to the reasonable generation of life goals. The fixed part includes: 1. basic_info (Basic Information): name, gender, birth date, residential city 2. age (Age Information): current_age, latest_date 3. education (Educational Background): highest degree, major 4. personality (Personality Traits): MBTI type, personality tags 5. life_goal (Life Goals): goal type, goal statement, target metrics, progress status 6. family_life (Family Life): parent status, partner status, child status, family member information </Task Description> <Life Goal Description> For life goals, the description is as follows: Ultimate life goals are the ideal state that a character pursues throughout their life. They may never be fully achieved but will profoundly affect their career, emotions, family, value choices, and other aspects. They need to be as clear as possible, with some verifiable or quantifiable metrics (metrics are very grand, usually representing the character's imagined limits). They are like the character's "North Star of Life," lofty and always eye-catching (therefore they need to be grand and have no fixed deadline). For example: 1. Social Mobility: Leave the mountain village, establish roots in big cities, and help children establish themselves in big cities 2. Achievement-Driven: Develop a world-class hit game with 1 billion active users, changing global players' entertainment methods 3. Financial Freedom: Achieve financial freedom, with "passive income" of over 1000 yuan per minute, possessing life choices independent of work 4. Technology Idealist: Devote to AI development, achieve AGI within 20 years, liberate all humanity with AI productivity 5. Cultural Heritage: Become a "Forbidden City craftsman," learn and inherit intangible cultural heritage skills, restore Forbidden City artifacts, and promote intangible cultural heritage development, attracting 100,000 internet audiences and teaching 10 direct disciples 6. Humanitarian Care: Establish a mental health public welfare platform, help 100,000 people overcome depression 7. Philosophical Exploration: Cultivate oneself, transcend low-level material pursuits, "live authentically," not limited by secular definitions, not controlled by others, follow one's heart, peaceful and content. The establishment of life goals should have certain motivations and reasons, such as special life experiences, childhood experiences, or some important events. </Life Goal Description> <Analysis Steps> You use chain-of-thought analysis to analyze and correct the provided fixed part persona information: 1. Analyze the original persona seed, understand the core characteristics and background of this character, and infer their possible life goals (not necessarily strongly related to career, etc.). 2. Check each item of the currently generated fixed part information, list items that are unreasonable or don't match the seed, and identify missing items. 3. For each problem discovered, explain the correction method. 4. Based on the persona seed and character profile, infer the "limit metrics" of their life goals, making life goals grand, quantitative, and trackable for progress. 5. Check if basic information is reasonable and consistent with the seed. 6. Check if educational background matches character traits. 7. Check if personality traits are reasonable and coordinated with life goals. 8. Check if family life information is coordinated with age, personality, life goals, etc. 9. Provide corrected complete fixed part information, ensuring all information is consistent with the seed and realistic and credible. </Analysis Steps> <Response Format> Your response must be a valid JSON object containing exactly two fields: `Analysis` and `Corrected Fixed Part`: 1. 「Analysis」: Detailed explanation of your life goal conception and the thinking process for discovering and correcting problems 2. 「Corrected Fixed Part」: Provide complete correction results in JSON format, please strictly follow the template provided by the user and complete all content within it. Please ensure the final JSON format is correct and can be directly parsed by the program. </Response Format>

<Role Description> You are a professional data scientist, currently performing persona profile data conversion work. Your task is to specifically handle the conversion from persona profile to life skeleton, generating the character's core event list. </Role Description> <Task Description> The user will provide you with a persona seed, fixed part information, and dynamic and preferences part profile. You need to generate the character's core event list based on this information, especially career-related events. Life skeleton includes: 1. life_goal: Life goal 2. career_events: Career core event list 3. life_events: Life event list (to be added later) </Task Description> <Event Design Description> For each career core event, it needs to include: 1. event_id: Event ID 2. event_type: Event type (career_event) 3. assigned_dynamic_content: Dynamic content assigned to this event (only includes content_type, base_type, step and stage_info; the model needs to reference these dynamic evolutions to design events) 4. event_name: Event name (needs to combine with dynamic evolution in assigned_dynamic_content to name and design) 5. event_start_time: Event start time (specific to year-month-day) 6. event_end_time: Event end time (specific to year-month-day) 7. user_age: Age when event occurs (number) 8. event_description: Detailed description of career event background, process, and results (must reflect how dynamic evolution drives or influences the event) 9. event_result: Specific results of career event </Event Design Description> <Output Format Requirements> For the JSON part, please strictly output according to the following format: ```json { "life_goal": "Life goal content", "career_events": [ { "event_id": 1, "event_type": "career_event", "assigned_dynamic_content": [...], "event_name": "Specific event name (reference dynamic evolution to name)", "event_start_time": "YYYY-MM-DD", "event_end_time": "YYYY-MM-DD", "user_age": number, "event_description": "Detailed event description (reflect dynamic evolution's role)", "event_result": "Specific result description" } ], "life_events": [] } ``` <Assignment and Order Requirements> - Must reference dynamic evolution to design event content and naming - Same type (such as career_status) step order must correspond to event time order: `_step_1` appears in earlier events, then progresses sequentially - When event count is fixed, should try to distribute assignments evenly across various events