geolocation-skill

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Geolocation Reasoning

地理定位推理

Method: Iterative reasoning loop (max 5 iterations) Inspiration: GeoVista research framework

方法： 迭代推理循环（最多5次迭代） 灵感来源： GeoVista研究框架

Core Workflow

核心工作流程

1. Visual Clue Extraction

1. 视觉线索提取

Category	Look for
Text/Signage	Languages, scripts, street signs, license plates, phone formats
Infrastructure	Road markings, traffic lights, poles, utilities, bollards
Architecture	Materials, styles, roof types, density patterns
Nature	Vegetation, terrain, geology, sun position, weather
Culture	Vehicles, driving side, clothing, shops, brands

类别	关注要点
文字/标识	语言、文字脚本、街道标识、车牌、电话号码格式
基础设施	道路标线、交通信号灯、电线杆、公共设施、护柱
建筑	材料、风格、屋顶类型、密度模式
自然环境	植被、地形、地质、太阳位置、天气
文化特征	车辆、驾驶方向、服饰、商铺、品牌

2. Hypothesis Formation

2. 假设形成

Form hypotheses at multiple scales:

Scale	Example
Continent/Region	Europe, Southeast Asia
Country	Specific nation
Region/State	Sub-national
City/Town	Locality
Precise	Street-level

For each: Note supporting evidence, confidence (H/M/L), conflicts.

在多个尺度上形成假设：

尺度	示例
大陆/区域	欧洲、东南亚
国家	特定国家
地区/州	国家级以下行政区
城镇	地方区域
精确位置	街道级

针对每个假设：记录支持证据、置信度（高/中/低）、矛盾点。

3. Image Examination

3. 图像检查

When details unclear:

Crop regions of interest
Enhance text/signage
Isolate architectural details

当细节不清晰时：

裁剪感兴趣区域
增强文字/标识清晰度
分离建筑细节

4. Web Search Verification

4. 网络搜索验证

Strategy	Query pattern
Sign text	`"[exact text]" location`
Features	`[feature] [suspected region]`
Business	`"[name]" address`
Landmark	`[description] landmark [region]`
Road	`[designation] [country]`

Cross-reference multiple clues, look for consistent attribution.

策略	查询模式
标识文字	`"[准确文字]" location`
特征	`[特征] [疑似区域]`
商业场所	`"[名称]" address`
地标	`[描述] landmark [区域]`
道路	`[道路名称] [国家]`

交叉参考多个线索，寻找一致的归属信息。

5. Iterative Refinement

5. 迭代优化

After each cycle:

Update hypothesis confidence
Identify remaining uncertainties
Determine if more examination helps
Continue until confident or limit reached

每个循环后：

更新假设置信度
识别剩余不确定性
判断是否需要进一步检查
持续迭代直至获得足够置信度或达到次数上限

Output Format

输出格式

**Location Analysis**

Observed Clues:
- [Key observations with confidence]

Reasoning Process:
- [Hypothesis formation and verification]

**Conclusion**
- Most likely location: [As specific as evidence allows]
- Confidence: High/Medium/Low
- Key evidence: [Primary determinants]
- Uncertainties: [What remains unclear]

**Location Analysis**

Observed Clues:
- [Key observations with confidence]

Reasoning Process:
- [Hypothesis formation and verification]

**Conclusion**
- Most likely location: [As specific as evidence allows]
- Confidence: High/Medium/Low
- Key evidence: [Primary determinants]
- Uncertainties: [What remains unclear]

Important Considerations

重要注意事项

Aspect	Guideline
Confidence	High requires multiple independent confirmations
Single clues	Can mislead (similar signage in multiple countries)
Privacy	Don't identify individuals or residential addresses
Iteration	Max 5 cycles, stop when marginal value low

方面	指导原则
置信度	高置信度需要多个独立证据确认
单一线索	可能产生误导（多个国家存在相似标识）
隐私	不得识别个人或住宅地址
迭代次数	最多5次循环，当边际价值较低时停止

References

参考资料

```
references/visual-clues.md
```
— Clue categories
```
references/regional-signatures.md
```
— Region-specific indicators
```
references/verification-strategies.md
```
— Search patterns

🎯 COMPLETED: [SKILL:geolocation-skill] [location identified from image] 🗣️ CUSTOM COMPLETED: [SKILL:geolocation-skill] [Location found]

```
references/visual-clues.md
```
— 线索类别
```
references/regional-signatures.md
```
— 区域特定指标
```
references/verification-strategies.md
```
— 搜索模式

🎯 COMPLETED: [SKILL:geolocation-skill] [location identified from image] 🗣️ CUSTOM COMPLETED: [SKILL:geolocation-skill] [Location found]