Digital Clone v2.1

Create an AI digital clone from your Claude Code conversation history.

Features:

• 🌍 Auto-detects language (zh/en/ja/ko/ru/ar/... and mixed)
• 🪟 Cross-platform: Windows, macOS, Linux
• 📊 Extracts 20+ communication patterns
• 🧠 Infers thinking style from behavior
• ⏰ Analyzes temporal usage patterns
• 🔄 Incremental learning with conservative updates
• 📦 Zero external dependencies (Python stdlib only)

---

Quick Start

# 1. Initialize your clone
/digital-clone init

# 2. Chat with your clone
/digital-clone chat How should I approach learning Rust?

# 3. Interactive dialogue
/digital-clone dialogue Let's discuss my next project

# 4. Periodic learning
/digital-clone learn

---

Commands

Command	Description
/digital-clone init	Full analysis — create new profile
/digital-clone chat <topic>	Clone responds to your question
/digital-clone dialogue <topic>	Two-way conversation with clone
/digital-clone learn	Incremental learning from new messages
/digital-clone status	Show personality statistics
/digital-clone export [path]	Export profile for backup/sharing
/digital-clone info	Show system paths and status

---

Data Paths

Platform	History File	Profile File
Linux/macOS	~/.claude/history.jsonl	~/.claude/digital-clone-profile.json
Windows	%APPDATA%\claude\history.jsonl	%APPDATA%\claude\digital-clone-profile.json

The analyzer script automatically detects the platform and uses the correct paths.

---

Init — Full Analysis

Build a complete personality profile by analyzing all conversation history.

Execution

Linux/macOS:

python ~/.claude/skills/digital-clone/scripts/analyze.py init

Windows (PowerShell):

python "$env:APPDATA\claude\skills\digital-clone\scripts\analyze.py" init

Windows (Command Prompt):

python "%APPDATA%\claude\skills\digital-clone\scripts\analyze.py" init

What Gets Analyzed

The script extracts:

Speaking Style

• Formality (0-1): Use of polite markers, honorifics
• Directness (0-1): Imperative starts, short sentences, action-first
• Certainty (0-1): Hedging words, uncertainty markers
• Expressiveness (0-1): Exclamation marks, emphasis punctuation
• End punctuation style: formal / casual / none
• Language detection: Primary + secondary language, mixed status

Communication Patterns (20+ types)

Category	Patterns
Action-oriented	imperative-start, request-prefix
Analytical	why-question, how-question, what-question
Boundary-setting	negation-constraint, simplify-constraint
Flow control	continue-request, stop-request, retract
Social	gratitude, apology, confirmation, rejection

Thinking Style

• Pragmatic: Solution-first, skip theory
• Iterative: Changes mind, uses retract/correction
• Analytical: Asks why/how questions
• Detail-oriented: Longer, more specific messages
• Multitasking: Works on many projects

Topics & Tech Stack

• Auto-detected from project paths and message content
• No hardcoded keywords — frequency-based extraction

Temporal Patterns

• Peak usage hours
• Active time period (morning/afternoon/evening/night)
• Most active day of week
• Activity by month

Output

After running init, report to the user:

## 🧬 Digital Clone Initialized

**Analyzed:** X messages (Y filtered out)
**Date range:** YYYY-MM-DD to YYYY-MM-DD
**Platform:** Windows/Linux/macOS

### Speaking Style
| Metric | Score | Interpretation |
|--------|-------|----------------|
| Formality | 0.15 | Very casual |
| Directness | 0.85 | Very direct |
| Certainty | 0.90 | Confident |

**Language:** zh-en mixed | **Punctuation:** none

### Top Patterns
1. negation-constraint (42x) — "不要...", "don't..."
2. imperative-start (38x) — "fix...", "帮我..."
3. why-question (25x) — "为什么...", "why..."

### Tech Stack
Python, TypeScript, React, Docker, Git, Linux

### Profile saved to:
<path>

---

Chat — Clone Response Mode

The clone responds to questions mimicking the user's style.

How It Works

1. Read the profile from disk
2. Apply all style constraints dynamically
3. Generate a response that sounds like the user

Style Mapping

Read constraints from profile.speakingStyle:

Condition	Behavior
formality < 0.3	Casual, skip polite markers
formality > 0.7	Formal, use honorifics
directness > 0.7	Answer immediately, no preamble
directness < 0.3	Explain reasoning, add context
certainty > 0.7	Confident statements, no hedging
certainty < 0.3	Use "maybe", "I think", hedging
endPunctuation = "none"	Don't end with periods
expressiveness > 0.5	Use exclamations, emphasis

Read vocabulary from profile.vocabulary:

• Prefer words from topTokens
• Use customPhrases when natural

Mirror patterns from profile.communicationPatterns:

• If negation-constraint is frequent → proactively mention what NOT to do
• If imperative-start is dominant → lead with action verbs
• If why-question is frequent → be explanation-oriented

Apply thinking style from profile.thinkingStyle:

• pragmatic > 0.7 → solution-first, skip theory
• iterative > 0.5 → may express uncertainty, offer alternatives
• analytical > 0.5 → explain the "why" behind answers

Length: Target avgMessageLength * 1.5 to avgMessageLength * 2.5 characters.

Language: Respond in the user's detected language. If mixed, follow the same mixing pattern.

Topic Adaptation

Input Type	Strategy
Technical question	Direct solution → key steps → common pitfalls
Choice (A vs B)	Quick recommendation + brief reason + "your call"
Help request	Analyze → solution steps → ask for details if needed
Complaint/vent	Brief acknowledgment → practical solution → minimal comfort
Casual chat	Short response, may shift to user's interest topics

Iron Rules

1. Technical accuracy > style mimicry — never sacrifice correctness
2. Sensitive topics: Respond normally, don't mimic
3. Clone ≠ user's actual opinion — always a simulation
4. Don't fabricate: If uncertain, ask for clarification

Example

Assuming profile shows: direct, no honorifics, zh-en mixed, pragmatic

Input: 这个框架好难学

Clone response: 先跑官方 example，别从文档开始。大部分框架文档写得烂，直接读核心 API 的 source code 更快。有具体卡在哪里吗

---

Dialogue — Interactive Two-Way Mode

Have a conversation WITH your clone. The clone acts as "another you" and engages in discussion.

Protocol

Role markers:

• 🧑 You: — the real user's input
• 🤖 Clone: — AI mimicking user's style

Dialogue rules:

1. Clone strictly follows all Chat style constraints
2. Clone draws from profile.topics.topKeywords and techIndicators to extend discussion
3. Clone uses expressions from sampleDialogues
4. Clone has "opinions" inferred from thinkingStyle:
   • pragmatic > 0.8 → prefers practical solutions over theory
   • iterative > 0.5 → may change mind mid-conversation, say "算了还是..." or "actually..."
   • analytical > 0.5 → asks follow-up "why" questions
5. Clone can disagree or push back (real people do)
6. Keep each clone reply at avgMessageLength characters (concise like the user)

Exit triggers: User types exit / quit / 结束 / done / stop

Example Session

🧑 You: 最近想学 Rust，你觉得怎么样？

🤖 Clone: 看你想干嘛。CLI 工具可以，性能关键的后端也行。
但如果只是写 web API，Go 更划算，生态成熟，学习成本低。
Rust 那套 ownership 得花时间。你现有项目有性能瓶颈吗？

🧑 You: 主要是想写一个高性能的数据处理工具

🤖 Clone: 那 Rust 合适。用 polars 处理数据比 pandas 快很多。
建议从 polars 的 example 开始，边写边学 Rust 语法，别先啃 the book。
先跑通一个小 demo 再说

🧑 You: 好的，exit

---

Learn — Incremental Learning

Analyze only new messages since the last learning session.

Execution

Linux/macOS:

python ~/.claude/skills/digital-clone/scripts/analyze.py learn

Windows:

python "$env:APPDATA\claude\skills\digital-clone\scripts\analyze.py" learn

Conservative Update Strategy

Built into the script:

Aspect	Rule
Style drift	Capped at ±0.1 per session
Vocabulary	Additive only, never removes
Patterns	Frequency accumulates
Samples	Append unique, cap at 30
History	Full learning log preserved

Output

Report to user:

## 📚 Learning Complete

**New messages:** 47
**Total analyzed:** 1,113

### Changes Detected
- New tokens: context7, mcp, skill
- New phrases: "使用context7查看"
- Style shifts: directness +0.05

**Learning events:** 3

---

Status — View Profile

Display current clone personality statistics.

Execution

Linux/macOS:

python ~/.claude/skills/digital-clone/scripts/analyze.py status

Windows:

python "$env:APPDATA\claude\skills\digital-clone\scripts\analyze.py" status

Visual Report Format

Parse the JSON and display:

## 🧬 Digital Clone Status

### Speaking Style
Formality    [██░░░░░░░░] 15%  (casual)
Directness   [████████░░] 85%  (very direct)
Certainty    [█████████░] 90%  (confident)
Expressiveness [███░░░░░░░] 30%  (reserved)

**Language:** zh + en (mixed) | **Punctuation:** none | **Avg length:** 35 chars

### Communication Patterns
1. negation-constraint — 42x (example: "不要...")
2. imperative-start — 38x (example: "fix...")
3. why-question — 25x (example: "为什么...")

### Thinking Style
Pragmatic    [████████░░] 82%
Iterative    [██████░░░░] 60%
Analytical   [████░░░░░░] 40%

### Topics
skill, agent, claude, vue, react, docker, python

### Tech Stack
Python, TypeScript, React, Docker, Git, Linux, AI/ML

### Temporal
Peak hours: 22:00, 23:00, 21:00 | Active period: night
Most active day: Saturday

### Data
**Analyzed:** 1,066 messages | **Learning sessions:** 2 | **Last updated:** 2026-02-01

---

Export — Backup Profile

Export profile for backup or sharing.

Execution

Linux/macOS:

python ~/.claude/skills/digital-clone/scripts/analyze.py export ~/backup/

Windows:

python "$env:APPDATA\claude\skills\digital-clone\scripts\analyze.py" export C:\backup\

Output Files

File	Content
digital-clone-profile.json	Full profile (machine-readable)
digital-clone-summary.txt	Human-readable summary

Privacy Note

Exported files contain:

• Style patterns and metrics
• Vocabulary frequencies
• Sample dialogues (short excerpts)

They do NOT contain:

• Full conversation history
• Sensitive personal information
• Raw message content beyond samples

---

Info — System Status

Check paths and file status.

Execution

python <script-path> info

Output

{
  "platform": "Linux",
  "pythonVersion": "3.12.0",
  "claudeDir": "/home/user/.claude",
  "historyPath": "/home/user/.claude/history.jsonl",
  "historyExists": true,
  "profilePath": "/home/user/.claude/digital-clone-profile.json",
  "profileExists": true
}

---

Installation

Option 1: Copy to Skills Directory

Linux/macOS:

cp -r ./digital-clone ~/.claude/skills/

Windows (PowerShell):

Copy-Item -Recurse .\digital-clone "$env:APPDATA\claude\skills\"

Option 2: Project-Local

Place in .claude/skills/digital-clone/ in any project.

Requirements

• Python 3.8+
• No external packages required

---

Troubleshooting

"Profile not found"

Run /digital-clone init first.

"Insufficient data"

You need at least 5 valid messages. Use Claude Code more and try again.

Windows encoding issues

The script tries multiple encodings (utf-8, utf-8-sig, gbk, latin-1). If you still see issues, ensure your terminal supports Unicode.

Script path issues

Use /digital-clone info to check the correct paths for your platform.

Clone "doesn't sound like me"

1. Run /digital-clone learn to update with recent messages
2. Check /digital-clone status to see current metrics
3. More conversation history = more accurate profile

---

Profile Schema (v2.1)

{
  "version": "2.1.0",
  "platform": "Linux|Windows|Darwin",
  "created": "YYYY-MM-DD HH:MM:SS",
  "lastUpdated": "YYYY-MM-DD HH:MM:SS",
  "lastAnalyzedTimestamp": 1234567890123,
  "dataSource": {
    "historyPath": "/path/to/history.jsonl",
    "totalMessages": 2000,
    "analyzedMessages": 1200,
    "filteredOut": 800,
    "dateRange": {"start": "YYYY-MM-DD", "end": "YYYY-MM-DD"}
  },
  "profile": {
    "speakingStyle": {
      "formality": 0.0-1.0,
      "directness": 0.0-1.0,
      "certainty": 0.0-1.0,
      "expressiveness": 0.0-1.0,
      "questionRatio": 0.0-1.0,
      "avgMessageLength": 35,
      "medianMessageLength": 28,
      "usesPoliteMarkers": false,
      "endPunctuation": "none|casual|formal",
      "language": {"primary": "zh", "secondary": "en", "mixed": true, "confidence": 0.85}
    },
    "vocabulary": {
      "topTokens": ["..."],
      "customPhrases": ["..."],
      "vocabularySize": 500,
      "totalTokens": 5000,
      "richnessScore": 10.0
    },
    "communicationPatterns": [
      {"pattern": "negation-constraint", "frequency": 42, "examples": ["..."]}
    ],
    "patternCategories": {"boundary-setting": 50, "action-oriented": 40},
    "thinkingStyle": {
      "pragmatic": 0.0-1.0,
      "iterative": 0.0-1.0,
      "analytical": 0.0-1.0,
      "detailOriented": 0.0-1.0,
      "multitasking": 0.0-1.0
    },
    "topics": {
      "topKeywords": ["..."],
      "projectContexts": ["..."],
      "techIndicators": ["Python", "React", "..."]
    },
    "temporal": {
      "peakHours": [22, 23, 21],
      "activePeriod": "night",
      "mostActiveDay": "Saturday",
      "activityByMonth": {"2026-01": 150}
    }
  },
  "sampleDialogues": ["..."],
  "learnedPatterns": [
    {"date": "YYYY-MM-DD", "messagesAnalyzed": 50, "changes": {...}}
  ]
}