Identify the real-world shooting location of the user's current desktop wallpaper and
provide a rich introduction to that location.
Follow these steps in order:
Run the bundled script to obtain the absolute path of the current wallpaper:
python scripts/get_wallpaper_path.py
HKCU\Control Panel\Desktop\WallPaper;
falls back to the TranscodedWallpaper cache file for theme-based wallpapers.
osascript to query System Events.
gsettings (GNOME) then xfconf-query (XFCE).
If the script outputs an ERROR: line, report the problem to the user and stop.
Use the Read tool with the path returned by the script to load the image into
context. WorkBuddy is multimodal and can directly interpret the visual content.
If the path points to a file without a standard image extension (e.g. the Windows
TranscodedWallpaper cache has no extension), still pass it to Read—it is a
valid JPEG/PNG image and will be processed correctly.
Carefully examine the image and identify:
national park, etc.
vegetation, water bodies, sky color/weather, and any other identifiable features.
reasoning for each.
Select the single most probable shooting location. State clearly:
supports the identification.
If the wallpaper is a digital illustration, abstract art, or a clearly synthetic
image with no identifiable real location, inform the user and stop here.
Write a compelling, informative introduction to the identified location covering:
| Section | Content |
|---------|---------|
| 概况 / Overview | Location, country, administrative region, coordinates (approximate) |
| 自然地理 / Geography | Terrain, climate, notable natural features |
| 历史文化 / History & Culture | Historical background, cultural significance, famous events |
| 旅游亮点 / Travel Highlights | Best viewpoints, must-see attractions, nearby places of interest |
| 最佳旅行时间 / Best Time to Visit | Season recommendation and reason |
| 实用信息 / Practical Info | How to get there, tips for visitors |
Write in the same language the user used (default: Chinese). Use headings and
bullet points for readability. Aim for 400–800 words total.
End with a short prompt asking if the user wants to know more, e.g.:
> "如需了解更多关于该地点的信息(如签证要求、住宿推荐、摄影技巧等),欢迎继续提问!"
Only the image content (pixels) is analysed by the model within this session.
candidate locations and explain the ambiguity rather than guessing wrongly.
model's visual understanding—no reverse image search or external API is required.
If the model is uncertain, say so honestly.
generated art, skip geolocation and instead describe the artwork and its style.
共 2 个版本