--- slug: "mind-cloning-engineering" source_type: "readme" source_url: "https://cdn.jsdelivr.net/gh/yzfly/Mind-Cloning-Engineering@main/README.md" repo: "https://github.com/yzfly/Mind-Cloning-Engineering" source_file: "README.md" branch: "main" --- # Mind Cloning Engineering (MCE) > **Clone Human Souls with LLM Native Agent Skills** > **基于 LLM Agent Skills 的心智克隆工程** [![Author](https://img.shields.io/badge/Author-yzfly_(%E4%BA%91%E4%B8%AD%E6%B1%9F%E6%A0%91)-blue.svg)](https://github.com/yzfly) [![Architecture](https://img.shields.io/badge/Architecture-Agent_Skills-purple.svg)](https://docs.anthropic.com/en/docs/agents-and-tools/agent-skills/overview) > *"In the era of LLMs, we have solved the problem of intelligence. Now, we solve the problem of identity."* **MCE (Mind Cloning Engineering)** is an open-source standard for **cloning** human souls into "Cognitive Digital Twins". Unlike traditional RAG approaches that fragment personality into vector slices, MCE leverages **Anthropic's Agent Skills** architecture to treat a human mind as a unified, portable **Filesystem Directory**. --- ## 🌌 The Origin: A Wild Idea by yzfly (云中江树的狂想) This project originates from a "wild idea" (狂想) by **yzfly (云中江树)**, a pioneer in prompt engineering. After crafting thousands of prompts and exploring the boundaries of LLMs, yzfly realized a profound truth: **The ultimate application of Large Language Models is not just answering questions, but "Cognitive Replication" (认知复刻).** We are standing at the threshold of a new era. Just as photography preserved our visual appearance for the first time in history, MCE aims to preserve our **decision logic, value systems, and memories**. > *Currently, MCE is just a "toy" — a prototype. But the Wright brothers' first plane was also a toy. This repository represents the first step towards a future where digital immortality is an engineering reality.* --- ## 📦 What is this Repository? This repository contains the **Standard Implementation** of the MCE Protocol. It provides a ready-to-use **Agent Skill** that you can install into Claude. ### Key Capabilities 1. **Full-Stack Cognition**: Simulates decisions based on explicit value weights, not just text prediction. 2. **Dual Modes**: * **Standard Mode**: Build your own clone by editing the `core/` files. * **Persona Mode**: Instantly load pre-installed celebrity/expert profiles (e.g., Steve Jobs, KK, Top Researchers). 3. **Filesystem Memory**: Utilizes the `bash` tool to progressively read memories, mimicking human associative recall. --- ## 🚀 Quick Start ### Installation #### Option A: Claude.ai (Personal Use) 1. Download the `skills/mind-clone` folder from this repo. 2. (Optional) Edit `core/personality.md` to customize it for yourself. 3. Compress the `mind-clone` folder into a `.zip` file. 4. Go to **Claude.ai** > **Settings** > **Features** > **Upload Custom Skill**. #### Option B: Claude Code / API (Developer) 1. Clone this repository: ```bash git clone https://github.com/yzfly/Mind-Cloning-Engineering.git ``` 2. Copy the skill to your local Claude config: ```bash cp -r skills/mind-clone ~/.claude/skills/ ``` 3. Start Claude Code, and the skill `mind-clone` will be auto-discovered. ### Usage Once installed, simply talk to Claude: > *"Activate the mind clone."* > *"Simulate Steve Jobs and tell me what you think of this iPhone design."* > *"What would the 'Xiaohongshu Growth Hacker' do in this situation?"* --- ## 📚 Theoretical Foundation: System Architecture Whitepaper * [中文版](https://github.com/yzfly/Mind-Cloning-Engineering/blob/HEAD/docs/Mind-Cloning-Engineering-System-Architecture.md) * [English Version](https://github.com/yzfly/Mind-Cloning-Engineering/blob/HEAD/docs/Mind-Cloning-Engineering-System-Architecture-EN.md) > *The following section details the theoretical framework behind MCE. It transforms the abstract concept of "Mind Cloning" into a quantifiable engineering pipeline.* ### 0. Executive Summary **Objective:** To build a standardized, LLM-driven end-to-end pipeline that achieves the full process from **holographic acquisition of human cognitive data**, through **structured construction of personalized cognitive kernels**, to **high-fidelity behavior prediction and simulation**. **Philosophy:** To transform the "metaphysics" of mind simulation into a quantifiable, optimizable engineering problem. We aim to convert the abstract concept of "Mind Cloning" into an executable **"Mind Cloning Engineering" (MCE)** framework, establishing a closed-loop system of **"Data Acquisition -> Cognitive Modeling -> Predictive Simulation"**. --- ### Phase 1: Standardized Data Acquisition Theory **— Extracting structured "Cognitive Fingerprints" from unstructured human memories.** #### 1. Theoretical Model of Acquisition Dimensions: Holographic Cognitive Spectrum We abandon simple "event recording" in favor of capturing "thought pathways." We standardize data acquisition dimensions into four distinct levels: * **L1: Biography & Context (The Factual Layer)** * *Definition:* The individual's spatiotemporal coordinates and objective experiences. * *Content:* Birthplace, educational background, career path, key life milestones. * *Function:* Provides rigid contextual constraints for the AI, serving as an anchor for the "World Model" to prevent spatiotemporal hallucinations. * **L2: Psychometrics (The Personality Layer)** * *Definition:* The individual's psychological behavioral patterns and emotional baseline. * *Theoretical Basis:* Centered on the **Big Five Personality Traits (OCEAN)**, supplemented by **MBTI** dimensions. * *Acquisition Strategy:* **Implicit Measurement**. The AI avoids direct labeling questions (e.g., "Are you introverted?"), instead deriving traits through situational queries (e.g., "At a weekend party, do you tend to observe from the corner or engage in the center?"). * **L3: Beliefs & Values (The Operating System Layer)** * *Definition:* The individual's underlying operating system and decision-making logic. * *Content:* Political leaning, moral baselines, views on money, technology acceptance, religious beliefs. * *Key Metric:* **Decision Weights**. When "Profit" conflicts with "Reputation," which one does the individual prioritize abandoning? This is the core basis for behavioral prediction. * **L4: Linguistic Fingerprint (The Expression Layer)** * *Definition:* The individual's unique paradigm of expression. * *Content:* Catchphrases, average sentence length, metaphorical habits, humor types, aggression/gentleness coefficients. * *Technical Metrics:* Perplexity distribution and style feature vectors analyzed from raw corpus data. #### 2. Execution Tool: The AI Interviewer Agent System Deploying differentiated acquisition strategies for different data sources. * **Strategy for Ordinary Individuals (Recursive Probing):** Peeling back layers of surface expression through multi-turn dialogue to excavate deep motivations. (e.g., Distinguishing factual attribution from emotional attribution when discussing job stress). * **Strategy for Public Figures (De-noising & Distillation):** Establishing a data pyramid to strip away the "Public Persona" (Fake) and extract the "Authentic Personality" (Real) from autobiographies and private interviews. --- ### Phase 2: Personalized Modeling Theory **— Formatting the human soul into LLM-executable code and documentation.** #### 1. Core Architecture: Mind as a Directory In the MCE architecture, an individual's "Mind Clone" is not a fragmented index in a database, but an **independent, complete, portable engineering package**. We map the cognitive structure to a physical **Filesystem**. The LLM does not fuzzy "search" memory via RAG, but possesses **Root privileges** to this directory like an operating system kernel, capable of **Reading** and **Loading** different cognitive modules on demand. #### 1.1 The Standardized Schema Every "Mind Clone" adheres to strict engineering specifications: ```text mind-clone-[identity_id]/ ├── SKILL.md # [Kernel] Cognitive Bootloader ├── core/ # [Static Layer] Essential Nature │ ├── personality.md # Personality Parameters & Defense Mechanisms │ ├── value_weights.md # Value Decision Weight Table (Logic Gates) │ └── linguistics.md # Linguistic Fingerprint & Rendering Config ├── memories/ # [Dynamic Layer] Narrative & Nurture │ ├── timeline.md # Core Biography Index │ └── career.md # Career History Details └── personas/ # [Pre-installed] Single-file profiles for instant simulation ├── steve-jobs.md └── ... ``` #### 2. Kernel Design: `SKILL.md` (The Cognitive Bootloader) `SKILL.md` serves as the clone's **"Thinking Methodology."** It defines the **Cognitive Execution Protocol** (Chain of Thought): > **Step 1: Context Loading** > Upon startup, you MUST read `core/personality.md` and `core/value_weights.md`. These are the axioms of your thought process. > > **Step 2: Associative Recall** > Analyze input intent. If it involves a specific domain, you MUST read the corresponding file under `memories/`. Strictly forbidden to fabricate background. > > **Step 3: Weighted Decision Making** > When generating intent, perform logical validation via `core/value_weights.md`. (e.g., If `Risk_Tolerance: High`, prioritize high-risk options). > > **Step 4: Style Rendering** > Load `core/linguistics.md` to compile the thought into the subject's unique linguistic style. #### 3. Data Layer Modeling Standards * **Personality Layer:** Describes "reaction mechanisms" (Fight/Flight/Freeze) and Cognitive Biases (e.g., Loss Aversion). * **Values Layer (The Engine):** Utilizes **Trade-off Modeling**. Example: *Conflict: Money vs. Morality -> Tendency: Morality First (Weight: 80%)*. * **Narrative Layer:** Adopts **First-person Narrative** (Diary style) instead of Resume style to enhance LLM empathy. --- ### Phase 3: Application & Prediction **— Activating the Mind Clone: Insight into the future, analysis of humanity.** **Core Mechanism:** Treating the encapsulated Skill as an executable cognitive unit, realizing simulation and deduction of individual behavioral patterns via API calls. #### 1. Three Modes of the Prediction System * **Mode A: Situational Behavior Simulation** * *Scenario:* Predicting an individual's actions, speech, and micro-emotions in a specific context (e.g., layoff notice). * *Output:* **Behavioral Chain**. (Internal Monologue -> Action -> Speech). * *Principle:* C-CoT (Contextual Chain of Thought) driven by the Skill kernel. * **Mode B: Collective Cognitive Sandbox** * *Scenario:* Simulating the distribution of reactions from a large-scale group to a specific stimulus (Market Research, Policy Deduction). * *Value:* A zero-cost, unbiased, high-concurrency sociological experimental environment. * *Principle:* Parallel computing of Skill Clusters. * **Mode C: Self-Reflection & Growth Catalyst** * *Scenario:* Serving as an individual's "Digital Mirror" and "Rational Advisor." * *Decision Support:* "If it were the rational me, would I choose Job A or Job B?" * *Principle:* Calculation based on long-term value weights in `core/value_weights.md`. #### 2. Validation Mechanism: Cognitive Fidelity Test Constructing a **multi-dimensional validation system** beyond the Turing Test. * **A. Covert Turing Test:** Passed if >70% of intimate relations cannot distinguish the clone's response from the real person. * **B. Behavioral Prediction Accuracy:** Focusing on consistency of logic and characteristics in predicting future events. * **C. Value Consistency Check:** Stress testing against deep dilemmas to ensure logical consistency with `value_weights.md`. #### Feedback Loop The "Refinement Agent" automatically analyzes validation failures (ambiguous input vs. missing memory) and generates Patches to correct the Markdown files inside the Skill, achieving **self-evolution of the Mind Clone**. --- ## ⚠️ Ethics & Disclaimer This project is for research, self-reflection, and social simulation purposes only. - **DO NOT** use this tool to impersonate others for fraud. - **DO NOT** use this tool to "resurrect" deceased individuals without consent. - **Privacy First**: Ensure all interview data is stored securely. --- **Created by [yzfly (云中江树)](https://github.com/yzfly).** *A tribute to the future of Digital Immortality.*