Files
sci-gui-agent-benchmark/evaluation_examples/examples/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6.json

165 lines
5.0 KiB
JSON

{
"id": "48c46dc7-fe04-4505-ade7-723cba1aa6f6",
"snapshot": "chrome",
"instruction": "Help me to automatically set up my work space. To be specific, open project directory of OSWorld in terminal and file manager, then open github homepage and the online document of Python in chrome browser.",
"source": "authors",
"config": [
{
"type": "execute",
"parameters": {
"command": [
"mkdir",
"-p",
"/home/user/Documents/Finance/receipts",
"/home/user/Documents/Projects/OSWorld",
"/home/user/.local/share/applications"
]
}
},
{
"type": "launch",
"parameters": {
"command": [
"socat",
"tcp-listen:9222,fork",
"tcp:localhost:1337"
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"cp",
"/usr/share/applications/google-chrome.desktop",
"/home/user/.local/share/applications"
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"sed",
"-i.bak",
"-e",
"s#Exec=/usr/bin/google-chrome-stable#& --remote-debugging-port=1337#g",
"/home/user/.local/share/applications/google-chrome.desktop"
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"update-desktop-database"
]
}
},
{
"type": "download",
"parameters": {
"files": [
{
"path": "/home/user/Documents/Finance/receipts/aws-invoice-2308.pdf",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/invoice0123456789-2308.pdf"
},
{
"path": "/home/user/Documents/Finance/receipts/aws-invoice-2309.pdf",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/invoice0123456789-2309.pdf"
},
{
"path": "/home/user/Documents/Finance/receipts/aws-invoice-2310.pdf",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/invoice0123456789-2310.pdf"
},
{
"path": "/home/user/Documents/Finance/receipts/aws-invoice-2311.pdf",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/invoice0123456789-2311.pdf"
},
{
"path": "/home/user/Documents/Finance/receipts/aws-invoice-2312.pdf",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/invoice0123456789-2312.pdf"
},
{
"path": "/home/user/Documents/Finance/receipts/X-receipt-2312.pdf",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/Receipt-0123-4567.pdf"
},
{
"path": "/home/user/Documents/Finance/tally_book.xlsx",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/tally_book.xlsx"
},
{
"path": "/home/user/.projects.tar.xz",
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/multi_apps/48c46dc7-fe04-4505-ade7-723cba1aa6f6/projects.tar.xz"
}
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"tar",
"-xJvf",
"/home/user/.projects.tar.xz",
"-C",
"/home/user/Documents/Projects/OSWorld"
]
}
}
],
"trajectory": "trajectories/48c46dc7-fe04-4505-ade7-723cba1aa6f6",
"related_apps": [
"chrome",
"os"
],
"evaluator": {
"postconfig": [
{
"type": "execute",
"parameters": {
"command": [
"wmctrl",
"-lx"
],
"stdout": "wmctrl.out"
}
}
],
"func": [
"check_list",
"is_expected_tabs"
],
"result": [
{
"type": "cache_file",
"path": "wmctrl.out"
},
{
"type": "open_tabs_info"
}
],
"expected": [
{
"type": "rule",
"rules": {
"expect": [
"\\borg\\.gnome\\.Nautilus\\.Org\\.gnome\\.Nautilus\\b.*\\bOSWorld\\b",
"gnome-terminal-server\\.Gnome-terminal\\b.*~/Documents/Projects/OSWorld\\b"
]
}
},
{
"type": "rule",
"rules": {
"type": "url",
"urls": [
"https://github.com",
"https://docs.python.org/3/"
]
}
}
]
},
"proxy": false
}