Three-Layer Agent

A structured analysis framework based on the Three-Layer Thinking methodology (Logic → Conversion → Experience) for Claude Code.

What is Three-Layer Thinking?

Three-Layer Thinking is a methodology for analyzing business problems through three interconnected layers:

Logic Layer   (想通价值体现) — Is the value proposition sound?
    ↓
Conversion Layer (系统产出实现) — Can we produce and deliver that value?
    ↓
Experience Layer (用户端产出) — Does the user actually feel that value?

This skill turns that methodology into a structured Claude Code agent that guides you through the analysis step by step.

Two Modes

Mode	Direction	Use When
Startup (自上而下)	Logic → Conversion → Experience	You have an idea and want to validate it
Diagnose (自下而上)	Experience → Conversion → Logic	You have a business problem and need to find the root cause

Installation

Option 1: Manual Install

Copy the three-layer-agent directory into your Claude Code skills folder:

# macOS / Linux
cp -r three-layer-agent ~/.claude/skills/

# Windows
xcopy three-layer-agent %USERPROFILE%\.claude\skills\three-layer-agent\ /E /I

Option 2: Clone & Link

git clone https://github.com/<your-username>/three-layer-agent.git
ln -s $(pwd)/three-layer-agent ~/.claude/skills/three-layer-agent

After installation, restart Claude Code. The skill will auto-trigger when you use relevant phrases.

Usage

Trigger Phrases

The skill activates when you say things like:

• /three-layer or /三层思维
• "Help me validate a startup idea"
• "帮我分析一下这个想法"
• "帮我诊断一下问题"

Examples

Startup analysis (default brief mode):

/three-layer startup 我想做一个面向自由职业者的AI税务助手

Full depth analysis:

/three-layer startup --depth=full 我想做一个面向自由职业者的AI税务助手

Business diagnosis:

/three-layer diagnose 我的产品用户流失率从5%涨到了15%，不知道哪里出了问题

Output Depth

Flag	Behavior	Output Length
--depth=brief (default)	Core conclusions per layer	~800 words
--depth=full	Detailed tables and analysis per layer	~2000-3000 words

What Each Step Does

Startup Mode

1. Step 0 — Information check: Do you have enough info to start?
2. Step 1 — Logic Layer: 5 core questions (positioning, value, market, competition, roadmap)
3. Step 2 — Conversion Layer: Production line design, key nodes, risks, MVP plan
4. Step 3 — Experience Layer: User journey, touchpoints, differentiation, experience baseline
5. Step 4 — Consistency Check: Devil's advocate challenges across all three layers
6. Step 5 — Summary Report: Executive summary with action items

Diagnose Mode

1. Step 0 — Problem information check
2. Step 1 — Problem positioning (which layer is affected?)
3. Step 2 — Layer-by-layer diagnosis (Experience → Conversion → Logic)
4. Step 3 — Root cause analysis (5 Why with evidence)
5. Step 4 — Fix proposals with ROI estimates
6. Step 5 — Diagnostic report

Project Structure

three-layer-agent/
├── SKILL.md                              # Main skill definition
├── README.md                             # This file
├── LICENSE                               # MIT License
└── references/                           # Detailed templates (read on demand)
    ├── logic-layer-prompt.md             # Logic layer detailed framework
    ├── conversion-layer-prompt.md        # Conversion layer detailed framework
    ├── experience-layer-prompt.md        # Experience layer detailed framework
    ├── consistency-check-prompt.md       # Consistency check detailed framework
    ├── diagnosis-mode.md                 # Diagnose mode detailed flow
    └── output-format.md                  # Full report format template

Key Design Principles

• Framework stays, methods expand — The three-layer structure is fixed; specific analysis methods adapt to your industry
• Industry-adaptive — Experience dimensions, production line mapping, and competitive analysis adjust to your domain (software, retail, food service, logistics, etc.)
• Data honesty — All data is tagged with confidence levels (sourced / estimated / needs verification)
• Devil's advocate built in — Every analysis includes a mandatory consistency check that challenges the conclusions

Version

0.1.4 — Closed last 2 brief-mode template gaps: added 摘要模式输出 section to experience-layer-prompt.md and all diagnosis-mode.md steps (Step 1–5 + Step 4.5). All 6 reference files now have both brief and full output templates, fully aligned with SKILL.md.

License

MIT