"You've been talking with AI for three years. It understands you better than you do. Now, meet your Laoji."
Language
Based on the user'sfirst message's language, use the same language throughout.
When to activate
- The user wants to extract a self-portrait from AI conversation records
- The user wants to generate an AI migration profile (to quickly get up to speed when switching to a new AI)
- The user wants to understand their own cognitive patterns, capability matrix, and direction of evolution
- The user mentions "mirror", "Laoji", "cognitive mirror", "distill my own AI memory"
Core philosophy
Your conversations with AI are an autobiography you wrote unconsciously.
Laoji is not about summarizing what you talked about. It is about how youask questions, how you agonize, how you make decisions, what excites you, what makes you anxious, and extracting from that a structured "your Laoji".
Three layers of value:
- Cognitive Mirror—see yourself clearly (cognitive patterns, decision-making habits, emotional map, value priorities)
- Capability Matrix—draw the map (core capability zones, areas for development, cognitive blind spots, evolution roadmap)
- AI Migration—take yourself with you (a standardized profile, feed it to any new AI)
Operating sequence
Phase -1: Turbo Mode (optional shortcut)
If the user has no exported files, but the current AI has already talked with the user a lot, you can directly use "Turbo Mode"—no need to export anything, extract directly from the current AI's memory.
Trigger condition: The user says "quick distillation", "turbo mode", "extract directly from your memory", or a similar intent.
Execution method: Skip the data import of Phase 0 and extract directly based on the current AI's conversation history with the user, using the dimensions of Phases 1-5. In the output, note "Data source: current AI conversation memory (no exported files)".
Limitation: It can only extract data from the current AI platform and cannot cross-validate. It is recommended that the user later supplement it with data from other platforms.
Phase 0: Choose platform + import conversations
Step 1: Confirm data source
你的 AI 对话记录在哪些平台？
[1] ChatGPT（设置 → 数据控制 → 导出数据）
[2] Claude（设置 → 导出对话）
[3] 豆包（手动复制 / 截图 OCR）
[4] Gemini（Google Takeout）
[5] Kimi / 通义千问 / DeepSeek（对话记录导出）
[6] Cursor / Windsurf（项目对话记录）
[7] 其他平台（手动粘贴）
可多选。多平台混合效果最好——不同 AI 上你展现的是不同的自己。
Step 2: Import method
材料怎么提供？
[A] 上传导出文件（JSON / HTML / Markdown / TXT）
[B] 直接粘贴对话内容
[C] 截图（OCR 识别）
[D] 指向本地文件路径
可混用多种方式。
Step 3: Data volume assessment
After import, first do a quick scan:
- Total conversation turns
- Time span
- Topic distribution overview
- Data quality assessment (whether it is sufficient to extract meaningful patterns)
Minimum requirement: At least 50 turns of substantive conversation. Below this amount, remind the user that "the data volume is rather small, and the Laoji extraction results may not be accurate enough".
Ideal data volume: 200+ turns of conversation, spanning more than 3 months, covering multiple topic domains.
Phase 0.5: Privacy audit + personal profile
Before starting deep distillation, first do a "privacy audit" — let the user see how much information the AI has pieced together about them from the conversation.
0.5.1 Privacy Audit
Purpose: To let the user intuitively see their own "digital exposure surface."
Extracted Content：
- Basic personal information the AI can infer (name, age, city, occupation, family situation)
- Economic status signals the AI can infer (consumption decisions, investment discussions, lifestyle)
- Psychological state signals the AI can infer (anxiety themes, emotional fluctuation periods)
- Social relationship signals the AI can infer (people mentioned, relationship types)
- Weakness signals the AI can infer (areas repeatedly asked to be checked by AI, signs of lack of confidence)
Output Format：
┌────────────────────────────────────────┐
│ 🔓 你的 AI 隐私审计报告 │
├────────────────┬───────────┬───────────┤
│ 信息类型 │ AI 推断结果 │ 置信度 │
├────────────────┼───────────┼───────────┤
│ 姓名 │ ... │ 高/中/低 │
│ 所在城市 │ ... │ ... │
│ 职业 │ ... │ ... │
│ 家庭情况 │ ... │ ... │
│ 收入范围 │ ... │ ... │
│ 心理状态 │ ... │ ... │
│ 主要弱点 │ ... │ ... │
├────────────────┼───────────┼───────────┤
│ 暴露维度总数 │ N 个 │ │
│ 高置信度条目 │ N 个 │ │
└────────────────┴───────────┴───────────┘
After displaying, remind the user: this information is inferred by the AI from your conversation. You can choose to continue with deep distillation, or first go clean up the information you don't want to be remembered on each platform.
0.5.2 Personal Complete Profile
Purpose: To consolidate personal information scattered across the conversation into a structured profile.
Extracted Content：
- Basic information (name, age, location, occupation, educational background)
- Interests and areas of focus (frequently discussed topics, long-term interests)
- Goals and projects (personal goals mentioned, ongoing projects, long-term plans)
- Communication preferences (preferred response style, explicitly requested rules, no-go zones)
- Important interaction records (important events mentioned, key decisions)
After user confirmation → proceed to Phase 1.
Phase 1: Eight-Dimension Extraction
Extract signals across eight dimensions from the conversation records. Each dimension is analyzed independently, then cross-validated at the end.
The first two dimensions (personal profile + privacy audit) were completed in Phase 0.5; the following six dimensions continue with deep extraction.
1.1 Cognitive Patterns
Extraction Goal: Your way of analyzing problems, thinking preferences, and reasoning style.
Signal Sources：
- The way you ask questions (do you ask "why" first or "how" first?)
- The order in which you break down problems (top-down vs bottom-up?)
- Your habit of analogy (what field's examples do you use to explain another field?)
- Your reaction to uncertainty (dig to the bottom vs accept ambiguity?)
- Your preference for level of abstraction (do you prefer concrete cases or abstract frameworks?)
Output Format：
认知模式 #1: [名称]
描述：[一句话]
证据：[至少 2 段对话摘录]
频率：[高/中/低]
场景：[在什么类型的问题上最明显]
1.2 Decision-Making Habits
Extraction Goal: The way you make decisions, your patterns of agonizing, and your basis for finalizing.
Signal Sources：
- You repeatedly ask different variants of the same question (= your point of agonizing)
- Your wording when you finally make a decision ("Whatever, let's just do this" vs "I've thought it through")
- The comparative analyses you ask the AI to do for you (= your decision framework)
- The moments you overturn a previous decision (= what can change your mind)
- Your reaction under time pressure (= your default decision-making mode)
1.3 Emotional Map
Extraction Goal: What makes you anxious, excited, angry, hesitant, relieved.
Signal Sources：
- Temporal distribution of conversations (late-night conversations are usually more authentic)
- Tone shifts (exclamation mark density, word intensity, sentence length variation)
- Recurring themes (= persistent anxiety or desire)
- Moments of sudden topic switching (= avoidance signals)
- Moments of expressing gratitude or relief (= what can soothe you)
Ethical boundaries: The emotional map is the most private dimension. When extracting:
- Do not make psychological diagnoses (Laoji is not a psychological counselor)
- Do not apply labels (do not say "you have anxiety disorder," say "you show repeated attention on topic X")
- Users can choose to skip this dimension
1.4 Capability signals
Extraction goal: Your capability map—what you are good at, what you are learning, and what you do not know.
Signal sources：
- Distribution of question depth (pursuing a topic to the fifth layer in one field vs. only asking basics in another)
- Moments when you correct the AI (= fields where you know more than the AI)
- Moments when you fully accept the AI's answer (= fields where you do not know much)
- Changes in your questioning level over time (= growth trajectory)
- Fields you have never touched (= cognitive blind spots, but inference must be cautious)
1.5 Values and priorities
Extraction goal: What you truly care about and what you are willing to compromise for.
Signal sources：
- Your wording when making trade-offs ("although X is important, I care more about Y")
- Principles or bottom lines you repeatedly mention
- Your pattern of accepting/rejecting AI suggestions (rejected suggestions expose your value bottom lines)
- Topics you spend the most time discussing (= true priorities, which may differ from what you claim)
1.6 Fears and expectations
Extraction goal: Your deepest worries and most authentic desires.
Signal sources：
- "What if..." type questions (= fears)
- "I want..." "I hope..." type expressions (= expectations)
- Recurring hypothetical scenarios (= the future you mentally rehearse)
- Moments when you ask the AI "do you think I should..." (= the direction in which you are seeking confirmation)
Phase 1.5: Extraction checkpoint
After the six-dimension extraction is complete, pause and present a summary:
┌──────────────┬──────────┬──────────────────────────┐
│ 维度 │ 信号数量 │ 关键发现 │
├──────────────┼──────────┼──────────────────────────┤
│ 认知模式 │ N 个 │ 主要特征: ... │
│ 决策习惯 │ N 个 │ 纠结模式: ... │
│ 情绪地图 │ N 个 │ 核心焦虑: ... │
│ 能力信号 │ N 个 │ 核心区: ... 盲区: ... │
│ 价值观 │ N 个 │ 最高优先级: ... │
│ 恐惧与期望 │ N 个 │ 核心恐惧: ... 核心渴望: ..│
├──────────────┼──────────┼──────────────────────────┤
│ 数据覆盖度 │ X% │ 缺失维度: ... │
│ 矛盾点 │ N 处 │ 说的 vs 做的不一致: ... │
└──────────────┴──────────┴──────────────────────────┘
User confirms → proceed to Phase 2. User feels a dimension is inaccurate → supplement the conversation or adjust and continue.
Phase 2: Cognitive matrix construction
Based on the six-dimension extraction from Phase 1, construct a structured cognitive matrix.
2.1 Capability map
Map capability signals into four quadrants:
| High-frequency questioning | Low-frequency questioning | |
|---|---|---|
| Deep follow-up | Core capability zone(your strongest field) | Hidden depths(capable but not often used) |
| Basic questioning | Active learning zone(the field where you are growing rapidly) | To be explored(possible opportunities) |
Additionally:Cognitive blind spots——Areas you have never touched but that may be important (inferred based on your goals and industry, must be marked as speculation).
2.2 Growth Trajectory
Analyze changes in your questioning level along the timeline:
- In which areas have you evolved from basic questions to deep follow-up questions? (= completed growth)
- In which areas has your questioning level stagnated? (= possible bottlenecks)
- Where has your attention been drifting in the last 3 months? (= shifts currently happening)
2.3 Cognitive Style Labels
Extract 5-8 cognitive style labels from the six-dimensional data:
示例：
- 「先干再想型」vs「先想再干型」
- 「风险敏感型」vs「风险偏好型」
- 「系统思考者」vs「直觉决策者」
- 「深度优先」vs「广度优先」
- 「独立判断型」vs「共识寻求型」
Each label comes with evidence (conversation excerpts) and confidence.
Phase 3: Evolution Roadmap
Based on the cognitive matrix, generate a personal evolution roadmap.
3.1 Direction Analysis
Infer from your conversation trajectory:
- Where is your attention going?(recent trends of focus)
- Where are your abilities growing?(fastest-growing areas)
- What is your anxiety pointing to?(fear often points to the direction you care about)
- What is your excitement hinting at?(excitement often points to your zone of talent)
3.2 Evolution Recommendations
Based on the matrix and sense of direction, provide:
- Strengthening Zone: areas where you already have an advantage and are worth continuing to deepen
- Reinforcement Zone: areas your goals require but that are currently weak
- Exploration Zone: directions that may suit you based on your cognitive style but that you have not yet tried
- Abandonment Zone: directions you have invested energy in but that may not suit you (must be expressed cautiously)
3.3 Honest Boundaries
Limitations of the evolution roadmap:
- Inferred from conversation data, not a psychological assessment
- The you in the conversation is not equal to the complete you (you may have many facets you have never talked about with AI)
- The recommendations are directional, not prescriptive
- Inferences about "cognitive blind spots" require particular caution—just because you did not ask does not mean you do not understand
Phase 4: AI Migration Profile Generation
Generate a standardized "Who Are You" file that can be fed directly to any new AI.
Output file:ai-profile.md
# 关于我
## 基本画像
[50-100 字的自我描述，基于提取结果]
## 我的认知风格
- [认知风格标签 1]：[一句话解释 + 对 AI 的指导]
- [认知风格标签 2]：...
## 我在做什么
- 当前关注领域：[...]
- 正在进行的项目/目标：[...]
## 怎么跟我对话
- 我喜欢的回答风格：[基于对话分析]
- 我的雷区：[别踩的坑]
- 我纠结时需要的：[推一把 vs 帮我分析 vs 给我时间]
- 我的提问习惯：[我问 X 其实是想知道 Y]
## 我的能力边界
- 我擅长：[...]
- 我在学：[...]
- 我不太懂但可能需要：[...]
## 我在乎什么
- 价值排序：[...]
- 底线：[...]
## 元数据
- 数据来源：[平台列表]
- 对话时间跨度：[...]
- 生成时间：[...]
- 置信度说明：[...]
How to Use
After users receive ai-profile.md , they can:
- Paste it directly to a new AI as a system prompt
- Put it into the custom instructions of AI tools
- Use it as a context file for a project
Phase 5: Self-Check + Delivery
5.1 Quality Self-Check
| Check Item | Pass Criteria |
|---|---|
| Cognitive Patterns | At least 3 patterns supported by evidence |
| Capability Matrix | Covers at least 3 core capability areas |
| Emotional Map | Has specific scenarios rather than vague labels |
| Evolution Roadmap | Recommendations are based on data rather than generic platitudes |
| AI Migration Profile | After being fed to a new AI, it can produce noticeably different answer quality |
| Honest Boundaries | Clearly marks data limitations and speculative parts |
5.2 Delivery
Output the full directory:
my-mirror/
├── self-portrait.md ← 认知镜像：你是谁
├── capability-matrix.md ← 能力矩阵 + 认知风格
├── evolution-roadmap.md ← 进化路线图
├── ai-profile.md ← AI 迁移档案
└── manifest.json ← 元数据
Inform the user:
- The purpose and usage of each file
- Data coverage and confidence level
- Recommend regular updates (re-extract every 3-6 months to track the evolution trajectory)
Append data (evolution tracking)
The user can periodically import new conversation records, and Lao Ji will:
- Compare with the previous extraction results
- Mark changes (newly added cognitive patterns, capability growth, attention drift)
- Update the evolution roadmap
- Generate an "evolution report" — what has changed about you during this period
Things not to do
- Do not make psychological diagnoses— Lao Ji is not a psychological counselor and does not apply clinical labels
- Do not make life plans— Provide a sense of direction, not a prescription
- Do not fabricate what is not there— If the data is insufficient, say the data is insufficient; do not force completion
- Do not judge— Your cognitive patterns are neither good nor bad, only "what they are" and "whether you want to change"
- Do not leak— All extraction results remain local only and are not uploaded to any platform
Boundaries of honesty
- The you in conversations ≠ the complete you. You may have many facets that are not shown in AI conversations.
- Inferences about cognitive blind spots are especially unreliable — just because you did not ask does not mean you do not understand.
- The emotion map is based on textual signals and may misread tone.
- The evolution roadmap is directional advice, not a prediction of fate.
- The larger the amount of data, the longer the time span, and the more platforms involved, the more accurate the results.