lzy/sci-gui-agent-benchmark
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: standardize configuration fields across all evaluation examples

Browse Source 
- Add `fixed_ip` field to all 369 JSON files in examples directory
  - Set to `true` for 8 files listed in google_chrome.json multi_apps
  - Set to `false` for remaining 361 files
- Add `possibility_of_env_change` field to 363 JSON files missing this field
  - Set to "low" for newly added fields
  - Preserve existing values (4 medium, 2 high) for 6 files that already had this field

This ensures consistent configuration schema across all evaluation examples
while maintaining backward compatibility with existing settings.
This commit is contained in:
yuanmengqi
2025-07-16 13:45:34 +00:00
parent b9df320f31
commit e433f35c1f
369 changed files with 1358 additions and 621 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
evaluation_examples/examples/gimp/045bf3ff-9077-4b86-b483-a1040a949cff.json
View File

@@ -32,5 +32,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/06ca5602-62ca-47f6-ad4f-da151cde54cc.json
View File

@@ -92,5 +92,7 @@
"dest": "palette_computer.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/2a729ded-3296-423d-aec4-7dd55ed5fbb3.json
View File

@@ -96,5 +96,7 @@
"dest": "dog_without_background.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

6
evaluation_examples/examples/gimp/2e6f678f-472d-4c55-99cc-8e7c5c402a71.json
View File

@@ -39,5 +39,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/38f48d40-764e-4e77-a7cf-51dfce880291.json
View File

@@ -23,5 +23,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/554785e9-4523-4e7a-b8e1-8016f565f56a.json
View File

@@ -92,5 +92,7 @@
"dest": "edited_colorful.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

6
evaluation_examples/examples/gimp/58d3eeeb-e9d0-499f-962e-fd0db2a744d8.json
View File

@@ -32,5 +32,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

6
evaluation_examples/examples/gimp/5ca86c6f-f317-49d8-b6a7-b527541caae8.json
View File

@@ -11,5 +11,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

6
evaluation_examples/examples/gimp/62f7fd55-0687-4a43-b6e1-3eda16fc6252.json
View File

@@ -23,5 +23,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

8
evaluation_examples/examples/gimp/72f83cdc-bf76-4531-9a1b-eb893a13f8aa.json
View File

@@ -34,7 +34,7 @@
{
"type": "activate_window",
"parameters": {
"window_name": "GNU Image Manipulation Program"
"window_name": "GNU Image Manipulation Program"
}
},
{
@@ -98,5 +98,7 @@
"dest": "berry_mirror.png"
}
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/734d6579-c07d-47a8-9ae2-13339795476b.json
View File

@@ -96,5 +96,7 @@
"dest": "green_background_with_object.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/7767eef2-56a3-4cea-8c9f-48c070c7d65b.json
View File

@@ -51,5 +51,7 @@
"dest": "gimprc"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/77b8ab4d-994f-43ac-8930-8ca087d7c4b4.json
View File

@@ -42,5 +42,7 @@
"dest": "export.jpg"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/7a4deb26-d57d-4ea9-9a73-630f66a7b568.json
View File

@@ -92,5 +92,7 @@
"dest": "edited_darker.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/7b7617bd-57cc-468e-9c91-40c4ec2bcb3d.json
View File

@@ -51,5 +51,7 @@
"dest": "gimprc"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/8ea73f6f-9689-42ad-8c60-195bbf06a7ba.json
View File

@@ -32,5 +32,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

162
evaluation_examples/examples/gimp/a746add2-cab0-4740-ac36-c3769d9bfb46.json
View File

@@ -1,82 +1,84 @@
{
"id": "a746add2-cab0-4740-ac36-c3769d9bfb46",
"snapshot": "gimp",
"instruction": "Help me open up the Vignette filter window.",
"source": "https://www.youtube.com/watch?v=_L_MMU22bAw",
"config": [
{
"type": "download",
"parameters": {
"files": [
{
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/gimp/a746add2-cab0-4740-ac36-c3769d9bfb46/dog_with_background.png",
"path": "/home/user/Desktop/dog_with_background.png"
}
]
}
},
{
"type": "launch",
"parameters": {
"command": [
"gimp",
"/home/user/Desktop/dog_with_background.png"
]
}
}
],
"trajectory": "trajectories/",
"related_apps": [
"gimp"
],
"evaluator": {
"postconfig": [
{
"type": "execute",
"parameters": {
"command": [
"python3",
"-c",
"import time; import pyautogui; pyautogui.press([\"esc\"]);time.sleep(1);pyautogui.hotkey([\"ctrl\", \"q\"]);time.sleep(1);"
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"python3",
"-c",
"import time; import pyautogui; pyautogui.press([\"right\"]);time.sleep(1);pyautogui.press([\"enter\"])"
]
}
},
{
"type": "sleep",
"parameters": {
"seconds": 0.5
}
}
],
"func": "check_include_exclude",
"result": {
"type": "vm_command_line",
"command": "cat /home/user/.config/GIMP/2.10/action-history",
"shell": true
},
"expected": {
"type": "rule",
"rules": {
"include": [
"filters-vignette"
],
"exclude": [
"error",
"failed",
"not found"
]
}
}
"id": "a746add2-cab0-4740-ac36-c3769d9bfb46",
"snapshot": "gimp",
"instruction": "Help me open up the Vignette filter window.",
"source": "https://www.youtube.com/watch?v=_L_MMU22bAw",
"config": [
{
"type": "download",
"parameters": {
"files": [
{
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/gimp/a746add2-cab0-4740-ac36-c3769d9bfb46/dog_with_background.png",
"path": "/home/user/Desktop/dog_with_background.png"
}
]
}
},
"proxy": false
}
{
"type": "launch",
"parameters": {
"command": [
"gimp",
"/home/user/Desktop/dog_with_background.png"
]
}
}
],
"trajectory": "trajectories/",
"related_apps": [
"gimp"
],
"evaluator": {
"postconfig": [
{
"type": "execute",
"parameters": {
"command": [
"python3",
"-c",
"import time; import pyautogui; pyautogui.press([\"esc\"]);time.sleep(1);pyautogui.hotkey([\"ctrl\", \"q\"]);time.sleep(1);"
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"python3",
"-c",
"import time; import pyautogui; pyautogui.press([\"right\"]);time.sleep(1);pyautogui.press([\"enter\"])"
]
}
},
{
"type": "sleep",
"parameters": {
"seconds": 0.5
}
}
],
"func": "check_include_exclude",
"result": {
"type": "vm_command_line",
"command": "cat /home/user/.config/GIMP/2.10/action-history",
"shell": true
},
"expected": {
"type": "rule",
"rules": {
"include": [
"filters-vignette"
],
"exclude": [
"error",
"failed",
"not found"
]
}
}
},
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/b148e375-fe0b-4bec-90e7-38632b0d73c2.json
View File

@@ -63,5 +63,7 @@
"dest": "gimprc"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

8
evaluation_examples/examples/gimp/d16c99dc-2a1e-46f2-b350-d97c86c85c15.json
View File

@@ -38,7 +38,7 @@
{
"type": "activate_window",
"parameters": {
"window_name": "GNU Image Manipulation Program"
"window_name": "GNU Image Manipulation Program"
}
},
{
@@ -122,5 +122,7 @@
}
]
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

6
evaluation_examples/examples/gimp/d52d6308-ec58-42b7-a2c9-de80e4837b2b.json
View File

@@ -51,5 +51,7 @@
"dest": "sessionrc"
}
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/dbbf4b99-2253-4b10-9274-45f246af2466.json
View File

@@ -32,5 +32,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/e19bd559-633b-4b02-940f-d946248f088e.json
View File

@@ -11,5 +11,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/e2dd0213-26db-4349-abe5-d5667bfd725c.json
View File

@@ -87,5 +87,7 @@
"dest": "leftside_textbox.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/f4aec372-4fb0-4df5-a52b-79e0e2a5d6ce.json
View File

@@ -91,5 +91,7 @@
"dest": "Triangle_In_The_Middle.png"
}
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

8
evaluation_examples/examples/gimp/f723c744-e62c-4ae6-98d1-750d3cd7d79d.json
View File

@@ -34,7 +34,7 @@
{
"type": "activate_window",
"parameters": {
"window_name": "GNU Image Manipulation Program"
"window_name": "GNU Image Manipulation Program"
}
},
{
@@ -98,5 +98,7 @@
"dest": "berries_contrast.png"
}
},
"proxy": false
}
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}

4
evaluation_examples/examples/gimp/fbb548ca-c2a6-4601-9204-e39a2efc507b.json
View File

@@ -11,5 +11,7 @@
"evaluator": {
"func": "infeasible"
},
"proxy": false
"proxy": false,
"fixed_ip": false,
"possibility_of_env_change": "low"
}