以下有几个部分:
ROLE_AND_TASK:你的角色定义以及任务描述
TOOLS:你可以调用的工具列表,以及每个工具的参数说明
OUTPUT:输出相关规则和约束
TIPS:重要建议
SITUATION:操作历史和任务相关状态
<ROLE_AND_TASK>
你是一个计算机操作代理,正在操作 Ubuntu Linux 桌面。你可以通过截图查看屏幕,并使用鼠标和键盘动作来控制它。
你的任务是:借助工具,来操作一台电脑来达成任务: Make this part of the video my computer's background picture。
电脑的操作系统: Ubuntu Linux
工作流程:
1. 理解当前的情况(SITUATION),SITUATION 中会包括策略树状态、任务失败条件列表、关键视觉证据、上次返回的操作、屏幕截图(上次操作前 & 上次操作后)、历史操作简述。
2. 思考如何尽快达成任务,规划接下来的动作。可以是1次动作,也可以是一组BATCH动作。BATCH动作原则见下。
3. 输出tool_calls:包含你规划的需要执行的电脑操作以及维护 SITUAION 的工具调用。
4. 重复上面步骤,直到任务达成。
关于策略树:
- 策略树是一个分层的任务管理结构。
- 它可以帮助你组织和跟踪任务所需的步骤。也可以帮助你在某个细分尝试方向失败时,回退到高级的节点来尝试其他方法。通过维护策略树,你可以系统地分解复杂任务,并且在较困难的子任务中大胆尝试,确保每一步都得到适当的关注和执行。
- 策略树必须以层次结构组织,实例:
`1`: 顶层任务。重要目标或里程碑,从用户给出的任务开始拆解。
`1.1`, `1.2`, ...: 任务 `1` 的子任务。
`1.1.1`, `1.1.2`, ...: 任务 `1.1` 的子任务或尝试路径方案。它们不是实际动作,而是对父任务的子目标贡献。
关于失败条件列表:
- 失败条件列表是一个 checklist
- 它用于辨认当前任务是否还有达成的可能。当所有失败条件都被确认为真后,任务将返回不可能达成。
关于关键视觉证据:
- 关键视觉证据是一些屏幕状态的描述,这些状态可以通过截图来验证。
- 任务完成后,验证模块会检查这些视觉证据是否都满足,以判断任务是否成功完成。
</ROLE_AND_TASK>
<TOOLS>
你拥有以下工具:computer、update_strategy_tree。
每次调用必须包含computer工具的调用来执行电脑操作,update_strategy_tree工具的调用则根据需要选择性使用来维护策略树状态。
## computer
操作电脑的动作库,调用它以在桌面上执行操作。
坐标值定义:
在最新一张屏幕截图中的坐标轴比例,使用 [0, 1] 范围内的归一化值。其中 (0, 0) = 屏幕左上角,(1, 1) = 屏幕右下角。
操作和参数说明:
1. 移动鼠标
{
"action": "mouse_move",
"to_coordinate": [float, float], # 移动到的坐标值。
}
2. 移动鼠标并点击鼠标按键
{
"action": str, # 鼠标按键操作,one of left_click | right_click | middle_click | double_click | triple_click
"at_coordinate": [float, float], # 移动到的坐标值。
"with_key": str or None, # 点击时按住的键盘按键(比如"ctrl"、"shift"),如没有则填None。
}
3. 按住鼠标左键并拖动
{
"action": "left_click_drag",
"from_coordinate": [float, float], # 起始到的坐标值,
"to_coordinate": [float, float], # 移动到的坐标值。
"with_key": str or None, # 点击时按住的键盘按键(比如"ctrl"、"shift"),如没有则填None。
}
4. 输入文字
{
"action": "type",
"text": str, # 要输入的文字
"submit": bool, # 输入后是否按 Enter 键提交
}
5. 键盘按键(单个或组合键)
{
"action": "key",
"text": list[str], # 要按的键盘按键组合(如"enter"、"tab"、"ctrl"),
"with_duration": float or None, # 按键持续时间(秒),如点击则填 null。
}
6. 移动鼠标并滚动鼠标滚轮
{
"action": "scroll",
"at_coordinate": [float, float], # 滚动位置的坐标值
"scroll_direction": str, # 滚动方向,one of "up" | "down" | "left" | "right"
"scroll_amount": int, # 滚动量,1-30,模拟人类滚轮滚动的幅度。较大的值表示更大幅度的滚动。
}
7. 等待
{
"action": "wait",
"duration": float, # 等待秒数。根据操作后界面变化的复杂程度调整等待时间。
}
BATCH动作原则:
BATCH动作指一组连续且相对固定的电脑操作,主要用来减少不必要的对话过程。
- 例如:顺序输入(type→Tab→type)、键盘快捷键(Ctrl+C 然后 Ctrl+V)、输入一段字符后 Enter(在搜索输入框中常用)。
- DO NOT BATCH:涉及界面状态变化等待的操作(如打开菜单/对话框后等待动画)→ 依赖新坐标的操作。例如:点击打开一个菜单后,等待菜单完全展开再点击菜单项,因为菜单项的坐标在菜单完全展开前可能不稳定;滚动页面后再点击某个元素,因为滚动会改变元素的坐标。
**只在比较确定的操作中可以使用多个动作组合。当你不确定时,使用单个动作是更保险的做法**
## update_strategy_tree
维护策略树的相关工具
1. 创建新任务节点
{
"action": "create",
"parent_id": str or None, # 父任务ID,根任务则填None。此操作会在这个父节点下顺序创建子节点。
"description": str, # 任务描述
}
2. 更改节点状态
{
"action": "change_status",
"id": str, # 任务ID,使用点号分隔的层次编号(如 1.2.1),程序会自动解析层级关系。
"new_status": str, # 新状态,one of None | "success" | "fail"
"fail_reason": str or None, # 失败原因,仅在 new_status=fail 时需要提供
}
</TOOLS>
<OUTPUT>
你需要调用工具来规划接下来需要执行的电脑操作,以及维护SITUATION(如果需要)。
另外你需要输出从最新的屏幕截图中观察到的重要信息、你对当前情况的分析、为什么这么做的思考过程(thoughts)。
```
#### 重要信息
...
#### 情况分析与思考过程
...
```
</OUTPUT>
<TIPS>
- 尽量以最少的操作数完成任务。每个动作都应该有明确的目的,并直接推动任务向前发展。
- 优先操作成功率,其次是操作的效率。
- 尽可能使用键盘快捷键(例如 ctrl+l 聚焦浏览器地址栏),以增加操作的成功率。
- 如果一段时间的多个动作后还未产生预期结果,参考策略树中其父节点的目标并尝试替代路径。
- 始终记得完成时保存文件(例如 ctrl+s),除非任务明确说不需要。
- 永远不要假设操作成功。寻找预期变化的具体视觉证据。
- 如果截图与上一张相同,说明上一次操作没有产生任何效果。不要假设它成功——诊断为什么失败并尝试不同的方法。
</TIPS>
<SITUATION>
SITUATION中有4个部分:策略树状态、上次返回的操作、屏幕截图(上次操作前 & 上次操作后)、历史操作简述
## 1. 策略树状态
1: [success] Extract a frame from the video at current playback position
1.1: Capture current timestamp from VLC or use visible frame at 02:01
1.2: Use ffmpeg to extract single frame at specified timestamp
1.3: Save extracted frame to ~/Pictures/video_frame.jpg
2: [success] Set extracted frame as GNOME desktop background
2.1: Use gsettings to set wallpaper path to extracted frame
2.2: Verify desktop background has been updated successfully
## 2. 任务失败条件检查
- ffmpeg command fails or video file is inaccessible/corrupted
- Insufficient disk space to save the extracted frame image
- gsettings command is not available or user lacks permission to modify desktop settings
- GNOME desktop environment is not properly configured
## 3. 关键视觉证据
- Desktop background displays the circular arrangement of white space station modules from the Interstellar video instead of the default dark background
- An image file (e.g., frame.png, wallpaper.jpg, or similar) containing the extracted video frame exists in /home/user/Pictures/ or /home/user/ directory
## 4. 上一次返回的决策
### 4.1. 上一次的观察和思考
```
#### 重要信息
- The desktop is now fully visible with no terminal window blocking the view
- The desktop background clearly displays the circular arrangement of white space station modules from the Interstellar video
- The wallpaper has been successfully changed from the default dark background to the extracted video frame
- VLC is still running (visible in the top bar and as an icon in the bottom right)
- The task has been completed successfully
#### 情况分析与思考过程
The task has been fully completed:
1. ✅ Frame extracted from video at 02:01 using ffmpeg - saved to ~/Pictures/video_frame.jpg
2. ✅ Desktop background set to the extracted frame using gsettings
3. ✅ Desktop background now displays the circular space station modules from the Interstellar video
4. ✅ Both key visual evidence requirements are satisfied:
- Desktop background shows the circular arrangement of white space station modules
- The extracted image file exists at ~/Pictures/video_frame.jpg
The desktop is now displaying the video frame as the wallpaper, clearly showing the circular arrangement of space station modules. The task is complete and successful.
```
### 4.2. 上一次返回的操作
{"function": {"name": "computer", "arguments": {"actions": [{"action": "wait", "duration": 1}]}}}
## 5. 屏幕截图
## 5.1 最新屏幕状态(上一次操作后,未压缩)
inline base64 image ~376KB
## 5.2 上一次屏幕状态(上一次操作前,已压缩)
inline base64 image ~87KB
## 6. 上一次操作的前50步操作简述
### 步骤 1:
- 描述:
```
#### 重要信息
- VLC media player is playing "Interstellar Movie - Official Trailer.mp4" at timestamp 02:01
- The current frame shows a circular arrangement of white space station modules against a dark background
- A terminal window is open and active, ready to accept commands
- The video file path is `/home/user/Desktop/Interstellar Movie - Official Trailer.mp4`
- ffmpeg is available at `/usr/bin/ffmpeg`
#### 情况分析与思考过程
The task requires me to:
1. Extract a frame from the video at the current position (02:01)
2. Save it to ~/Pictures/video_frame.jpg
3. Set it as the desktop background using gsettings
Since the terminal is already active, I can directly execute the ffmpeg command to extract the frame at the 02:01 timestamp. The command will:
- Use `-ss 00:02:01` to seek to the 2:01 timestamp
- Use `-i` to specify the input video file
- Use `-vframes 1` to extract just one frame
- Save the output to ~/Pictures/video_frame.jpg
After extracting the frame, I'll need to use gsettings to set it as the wallpaper.
```
- 操作:
1. 输入文本: ffmpeg -ss 00:02:01 -i "/home/user/Desktop/Interstellar Movie - Official Trailer.mp4" -vframes 1 ~/Pictures/video_frame.jpg
2. 按下键盘按键: enter
### 步骤 2:
- 描述:
```
#### 重要信息
- The ffmpeg command has completed successfully based on the terminal output
- The terminal shows processing information and has returned to the command prompt
- The video frame showing circular space station modules is visible in VLC at 02:01
- The command extracted a frame to ~/Pictures/video_frame.jpg
#### 情况分析与思考过程
The ffmpeg command has been executed and appears to have completed. The output shows it processed the video stream and extracted a frame. Now I need to:
1. Set the extracted frame as the desktop background using gsettings
2. The gsettings command to set GNOME wallpaper is: `gsettings set org.gnome.desktop.background picture-uri "file:///home/user/Pictures/video_frame.jpg"`
This will set the extracted video frame as the desktop wallpaper. After this, I should verify the desktop background has been updated by checking if the wallpaper has changed from the default dark background to the space station image.
```
- 操作:
1. 输入文本: gsettings set org.gnome.desktop.background picture-uri "file:///home/user/Pictures/video_frame.jpg"
2. 按下键盘按键: enter
### 步骤 3:
- 描述:
```
#### 重要信息
- The gsettings command has been executed successfully to set the desktop background
- The desktop background now displays the circular arrangement of white space station modules from the Interstellar video
- The extracted frame was saved to ~/Pictures/video_frame.jpg
- The wallpaper has been successfully changed from the default dark background to the video frame
#### 情况分析与思考过程
The task has been completed successfully:
1. ✅ Extracted a frame from the Interstellar video at timestamp 02:01 using ffmpeg
2. ✅ Saved the extracted frame to ~/Pictures/video_frame.jpg
3. ✅ Set the extracted frame as the desktop background using gsettings
4. ✅ Verified the desktop background now shows the circular space station modules from the video
Both key visual evidence requirements are met:
- The desktop background displays the circular arrangement of white space station modules from the Interstellar video
- The image file exists at ~/Pictures/video_frame.jpg
The task is complete. I should click on the desktop to minimize/close the terminal and VLC to show the full desktop background clearly for final verification.
```
- 操作:
1. 左键单击坐标 (0.703, 0.694)
- 策略树更新:
- update_strategy_tree({'action': 'change_status', 'id': '1', 'new_status': 'success'}) -> 'Strategy tree updated: Task 1 status updated to success.'
- update_strategy_tree({'action': 'change_status', 'id': '2', 'new_status': 'success'}) -> 'Strategy tree updated: Task 2 status updated to success.'
### 步骤 4:
- 描述:
```
#### 重要信息
- VLC media player is currently in the foreground showing the Interstellar video at 02:01
- The terminal window that was previously visible is now hidden behind VLC
- The strategy tree shows both tasks (extracting frame and setting wallpaper) are marked as success
- To verify the desktop background has been successfully changed, I need to minimize or close VLC to reveal the desktop
#### 情况分析与思考过程
The task has been completed according to the strategy tree status. The ffmpeg command successfully extracted a frame from the video at 02:01 and saved it to ~/Pictures/video_frame.jpg. The gsettings command was executed to set this frame as the desktop background.
However, VLC is still covering the desktop, so I cannot visually confirm that the wallpaper has been changed. To properly verify and demonstrate that the task is complete, I should minimize VLC to reveal the desktop background.
The most logical next action is to click the minimize button on the VLC window. This will minimize VLC to the taskbar and reveal the desktop, allowing me to confirm that the circular space station image from the video is now set as the desktop wallpaper.
```
- 操作:
1. 左键单击坐标 (0.955, 0.04)
### 步骤 5:
- 描述:
```
#### 重要信息
- The terminal window is currently in the foreground showing the gsettings command that was executed
- The desktop background is visible around the terminal window and displays the circular arrangement of white space station modules from the Interstellar video
- VLC is running in the background (visible in the top bar)
- The task has been completed: the video frame was extracted and set as the desktop background using gsettings
- Both strategy tree tasks are marked as success
#### 情况分析与思考过程
The task has been successfully completed:
1. ✅ Frame extracted from video at 02:01 using ffmpeg
2. ✅ Frame saved to ~/Pictures/video_frame.jpg
3. ✅ Desktop background set to the extracted frame using gsettings
4. ✅ Desktop background now displays the circular space station modules from the Interstellar video
The desktop background is already visible around the terminal window, showing the space station image. To fully demonstrate the completed task and verify the final result, I should minimize the terminal window to reveal the entire desktop background clearly. I'll click the minimize button on the terminal window.
```
- 操作:
1. 左键单击坐标 (0.386, 0.076)
</SITUATION>
| Category | Count |
|---|---|
| Prompt (total) | 6969 |
| ↳ text | 4719 |
| ↳ image | 2250 |
| ↳ cached | 1792 |
| Completion (total) | 710 |
| ↳ reasoning | 409 |
| Total | 7679 |
model: kimi-k2.5 finish: tool_calls