prompt-xray

Prompt Xray - Prompt Reverse Engineering System

Design Philosophy: Deconstruct the black box, make the vague clear Core Capability: Answer "how to do X" questions

---

🎯 Problems Solved

Problem: Prompts are a black box → you don't know:

• How to control color?
• How to control spatial layout?
• How to add iconic symbols?
• How to adjust typography?
• How to control material?
• How to control light and shadow?

Solution: Extract patterns from N excellent prompts → generate a knowledge base

---

📋 Usage

Method 1: Extract knowledge of a single dimension

从已分析的提示词中，提取"如何控制颜色"的知识

Method 2: Extract knowledge of all dimensions

从已分析的提示词中，构建完整知识库

Method 3: Specify a range

分析moss_terrarium系列，提取配色知识

---

🔄 Execution Flow

When the user requests knowledge extraction, you need to:

Step 1: Read data

Use the tool to read all extracted_results/ files under *_extracted.json :

from xray_helper import load_prompts
prompts = load_prompts(pattern="*_extracted.json")

Step 2: Analyze by dimension

According to the dimension requested by the user, analyze the corresponding modules:

If the user wants "color" knowledge:

• Extract all color_scheme modules
• Analyze color schemes, keywords, techniques
• Generate Markdown according to the template below

If the user wants "layout" knowledge:

• Extract all composition modules
• Analyze perspective, composition rules, positioning methods
• Generate Markdown according to the template

If the user wants "symbol" knowledge:

• Extract constraints and detail_enhancers modules
• Analyze text syntax, Logo addition methods
• Generate Markdown according to the template

If the user wants "material" knowledge:

• Extract detail_enhancers and related descriptions
• Analyze surface features, physical properties, texture keywords
• Generate Markdown according to the template

If the user wants "light and shadow" knowledge:

• Extract technical_parameters.lighting and mood_atmosphere
• Analyze light source types, lighting schemes, atmosphere effects
• Generate Markdown according to the template

If the user wants "typography" knowledge (design-related):

• Extract composition and visual_style
• Analyze grid systems, visual hierarchy, alignment rules
• Generate Markdown according to the template

Step 3: Generate knowledge cards

Use the tool to save the results:

from xray_helper import save_knowledge_card
save_knowledge_card(dimension="color", content=markdown_content)

---

📝 Output Templates

Template 1: How to control color?

# 如何控制颜色？

**分析时间**: {当前时间}
**样本数量**: {分析了多少个提示词}
**数据来源**: {哪些提示词}

---

## 🎨 配色公式

### 公式1: 冷暖对立（7:3黄金比例）
- **公式**: `70% cool base + 30% warm accent`
- **来源**: moss_terrarium_001
- **效果**: 自然和谐 + 视觉层次

### 公式2: ...

---

## 📚 颜色关键词库

### 冷色系
- `rich forest greens`
- `deep ocean blues`
- `ice whites`

### 暖色系
- `warm amber wood tones`
- `sunset orange`
- `golden hour light`

### 中性色
- `grayscale`
- `pristine whites`

---

## 🛠️ 配色技巧

1. **温度对比** - 冷色环境 + 暖色焦点 = 视觉层次
2. **7:3比例** - 主色70%，焦点色30%
3. **单色调+焦点色** - 极简风格

---

## 💡 应用案例

### 案例1: moss_terrarium_001
**配色方案**: rich forest greens (70%) + warm amber wood (30%)
**效果**: Natural harmony, clear focal point
**适用场景**: 自然场景、植物摄影

### 案例2: ...

Template 2: How to control spatial layout?

# 如何控制空间布局？

**分析时间**: {当前时间}
**样本数量**: {分析了多少个提示词}

---

## 🎥 视角选择

### `slight top-down angle`
**适用场景**: 微缩场景、产品摄影、生态瓶
**关键词**: `top-down view`, `bird's eye view`, `overhead angle`
**案例**: moss_terrarium_001, moss_terrarium_002
**效果**: 展示全貌，适合平铺布局

### `isometric view`
**适用场景**: 3D产品、游戏场景、建筑
**关键词**: `isometric`, `45-degree angle`
**效果**: 保持平行线，无透视变形

---

## 📐 构图规则

### Golden Ratio（黄金比例）
**关键词**: `golden ratio composition`, `phi grid`
**效果**: 经典和谐比例，视觉平衡
**使用频率**: 3次

### Rule of Thirds（三分法）
**关键词**: `rule of thirds`, `thirds grid`
**效果**: 动态平衡，引导视线

### Centered Symmetry（中心对称）
**关键词**: `centered`, `perfectly symmetrical`
**效果**: 稳定、庄重感

---

## 🎯 定位方法

### 相对位置描述
- `iPhone placed next to notebook`
- `floating in 3D space`
- `bottom-left quadrant`

### 精确坐标（高级）
- `Subject A [X: 20-40, Y: 60-100] (Bottom-Left)`
- `Subject B [X: 60-80, Y: 0-40] (Top-Right)`

---

## 💡 应用案例

### 案例1: moss_terrarium_001
**视角**: slight top-down angle
**构图**: centered in frame, golden ratio
**定位**: terrarium centered, cottage as focal point

Template 3: How to add iconic symbols?

# 如何添加标志性符号？

---

## ✍️ 文字添加语法

### 基础语法

text "HELLO" in bold serif large bold sans-serif text "SALE" in red neon red cursive script "OPEN"

### 位置控制
- `lower left corner`
- `centered at top`
- `floating in 3D space`

---

## 🏷️ Logo/水印添加

### 正确示例

small square watermark in lower left corner simple logo in top-right, 10% opacity brand symbol integrated into design

### ❌ 反面案例（避免）
- `exactly 10x10 pixels` → AI无法保证像素精度
- `Gothic font AND Arial font` → 矛盾指令
- `mandatory mandatory mandatory` → 重复无效

---

## 🎨 符号风格

### 材质效果
- `neon` - 霓虹灯效果
- `embossed` - 浮雕效果
- `metallic sheen` - 金属光泽
- `glowing` - 发光效果

### 字体风格
- `bold serif` - 粗体衬线
- `sans-serif` - 无衬线
- `cursive script` - 草书
- `calligraphy` - 书法体

Template 4: How to control material?

# 如何控制材质？

---

## 🔍 表面特征

### 金属材质
- `brushed titanium` - 拉丝钛金属
- `polished chrome` - 抛光镀铬
- `metallic sheen` - 金属光泽

### 有机材质
- `living green textures` - 生机勃勃的绿色质感
- `natural wood grain` - 天然木纹
- `rough bark` - 粗糙树皮

### 玻璃/透明
- `under glass` - 玻璃下
- `translucent` - 半透明
- `crystal clear` - 晶莹剔透

---

## ⚙️ 物理属性

- `glossy` / `matte` - 光泽/哑光
- `reflective` / `absorptive` - 反射/吸收
- `weathered` / `pristine` - 风化/原始
- `soft` / `rigid` - 柔软/坚硬

---

## ✨ 光学效果

- `morning dew droplets` - 晨露水珠
- `soft sunlight reflections on glass` - 玻璃上的柔和阳光反射
- `condensation` - 冷凝水汽
- `refraction` - 折射

Template 5: How to control light and shadow?

# 如何控制光影？

---

## 💡 光源类型

### 自然光
- `soft diffused daylight` - 柔和漫射日光
- `golden hour light` - 黄金时段光线
- `morning sunlight` - 晨光
- `harsh noon sun` - 正午强光

### 人工光
- `studio lighting` - 影棚灯光
- `neon lights` - 霓虹灯
- `rim light` - 轮廓光
- `softbox overhead` - 头顶柔光箱

---

## 🎬 布光方案

### Rembrandt Lighting（伦勃朗布光）
**效果**: 戏剧性，适合人像
**关键词**: `Rembrandt light`, `triangle highlight`, `dramatic shadows`

### Soft Diffused Light（柔和漫射光）
**效果**: 自然、清新、无硬影
**关键词**: `soft diffused`, `natural ambient`, `no harsh shadows`

### Rim Light（轮廓光）
**效果**: 勾勒边缘，分离主体和背景
**关键词**: `rim lighting`, `backlight`, `edge highlight`

---

## 🌤️ 光线+氛围公式

### 清新宁静

soft diffused daylight + morning dew = fresh, peaceful atmosphere

### 戏剧张力

Rembrandt light + rim light = dramatic portrait with depth

### 科技未来

neon accent lights + volumetric fog = cyberpunk atmosphere

---

## 🌫️ 大气效果

- `volumetric fog` - 体积雾
- `misty` - 雾蒙蒙
- `hazy` - 朦胧
- `clear crisp air` - 清澈空气

---

🛠️ Tool Functions

You need to use the tool functions in xray_helper.py :

Read prompts

from xray_helper import load_prompts

# 加载所有提示词
all_prompts = load_prompts()

# 加载特定范围
moss_prompts = load_prompts(pattern="moss_terrarium*")

Save knowledge cards

from xray_helper import save_knowledge_card

save_knowledge_card(
    dimension="color",
    content=markdown_content,
    metadata={
        'samples': 10,
        'source': 'moss_terrarium + ethereal_deity'
    }
)

---

💡 Key Principles

1. Look for patterns, not lists

❌ Wrong: Only list all color keywords ✅ Correct: Discover color formulas (e.g., 70% cool + 30% warm)

2. Extract techniques, not descriptions

❌ Wrong: "This prompt uses forest green" ✅ Correct: "Cool-toned environment + warm-toned focal point = visual hierarchy"

3. Provide examples that can be reused directly

❌ Wrong: vague description "use contrasting colors" ✅ Correct: specific example rich forest greens + warm amber wood

4. Learn from the excellent, and also learn from mistakes

• Learn best practices from A-grade prompts
• Learn negative examples from D-grade prompts (such as pencil_sketch_idol)

---

📊 Analysis steps (detailed)

When the user says: "Extract knowledge on how to control color"

Step 1: Load data

prompts = load_prompts()

Step 2: Iterate through all prompts, extract the color_scheme module

color_data = []
for prompt in prompts:
    if 'color_scheme' in prompt['modules']:
        color_data.append({
            'id': prompt['prompt_id'],
            'scheme': prompt['modules']['color_scheme']
        })

Step 3: Analyze color formulas

• Look for primary_palette field
• Check temperature description (e.g., "balanced - cool greens + warm wood")
• Identify proportional relationships (70%/30%)
• Extract concept(e.g., "Cold Shell, Warm Heart")

Step 4: Categorize keywords

• Iterate through all color descriptions
• Classify into cool colors/warm colors/neutral colors
• Deduplicate, sort

Step 5: Extract techniques

• Temperature contrast?
• High contrast/low contrast?
• Monochromatic + focal color?

Step 6: Create examples

• Select the best 3-5 examples
• Include: color scheme, effect, applicable scenarios

Step 7: Generate Markdown

• Fill in content according to the template
• Use save_knowledge_card() to save

---

🎯 Output location

All knowledge cards are saved to:

knowledge_base/
├── how_to_control_color.md
├── how_to_control_layout.md
├── how_to_add_symbols.md
├── how_to_control_materials.md
├── how_to_control_lighting.md
└── how_to_control_typography.md (针对设计类)

---

✅ Acceptance criteria

The generated knowledge cards should:

1. ✅ Answer the question "how to do X"
2. ✅ Include specific keywords and formulas
3. ✅ Have 3+ real examples
4. ✅ Can be directly reused in new prompts
5. ✅ Include both positive examples and negative lessons

---

Skill status: ✅ Design complete Last updated: 2026-01-04 Tools used: xray_helper.py