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

ver Jan15th

Browse Source 
merged zdy into main
This commit is contained in:
David Chang
2024-01-15 12:13:44 +08:00
parent 7ffb5de551 fc289a3427
commit a9e9ebbc80
6 changed files with 185 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
desktop_env/controllers/setup.py
View File

@@ -205,15 +205,15 @@ class SetupController:
except requests.exceptions.RequestException as e:
logger.error("An error occurred while trying to send the request: %s", e)
def _launch_setup(self, command: Union[str, List[str]]):
def _launch_setup(self, command: Union[str, List[str]], shell: bool = False):
if not command:
raise Exception("Empty command to launch.")
if isinstance(command, str) and len(command.split()) > 1:
if not shell and isinstance(command, str) and len(command.split()) > 1:
logger.warning("Command should be a list of strings. Now it is a string. Will split it by space.")
command = command.split()
payload = json.dumps({"command": command})
payload = json.dumps({"command": command, "shell": shell})
headers = {"Content-Type": "application/json"}
try:
@@ -245,7 +245,7 @@ class SetupController:
while not terminates:
try:
response = requests.post(self.http_server_setup_root + "/execute", headers=headers, data=payload)
response = requests.post(self.http_server + "/setup" + "/execute", headers=headers, data=payload)
if response.status_code == 200:
results: Dict[str, str] = response.json()
if stdout:

7
desktop_env/server/main.py
View File

@@ -58,11 +58,12 @@ def execute_command():
@app.route('/setup/launch', methods=["POST"])
def launch_app():
data = request.json
command: List[str] = data.get("command", [])
shell = data.get("shell", False)
command: List[str] = data.get("command", "" if shell else [])
try:
subprocess.Popen(command)
return "{:} launched successfully".format(" ".join(command))
subprocess.Popen(command, shell=shell)
return "{:} launched successfully".format(command if shell else " ".join(command))
except Exception as e:
return jsonify({"status": "error", "message": str(e)}), 500

1
desktop_env/server/requirements.txt
View File

@@ -5,3 +5,4 @@ git+https://github.com/moses-palmer/pynput.git@refs/pull/541/head # to make sure
requests
flask
numpy
lxml

2
evaluation_examples/examples/thunderbird/06fe7178-4491-4589-810f-2e2bc9502122.json
View File

@@ -1,7 +1,7 @@
{
"id": "06fe7178-4491-4589-810f-2e2bc9502122",
"snapshot": "thunderbird",
"instruction": "Could you help me back up all the email files in my profile to ~/email.bak? Please save them separately in eml format.",
"instruction": "Could you help me back up all the email files in my profile to ~/emails.bak? Please save them separately in eml format.",
"source": "https://www.quora.com/How-do-I-backup-email-files-in-Mozilla-Thunderbird",
"config": [
{

98
evaluation_examples/examples/thunderbird/d088f539-cab4-4f9a-ac92-9999fc3a656e.json Normal file
View File

@@ -0,0 +1,98 @@
{
"id": "d088f539-cab4-4f9a-ac92-9999fc3a656e",
"snapshot": "thunderbird",
"instruction": "Attach the my AWS bill to the email. The bill is stored at ~/aws-bill.pdf. Don't close it or send it. I haven't finish all the contents.",
"source": "https://support.mozilla.org/en-US/kb/how-use-attachments",
"config": [
{
"type": "download",
"parameters": {
"files": [
{
"url": "https://drive.usercontent.google.com/download?id=1JDUxe8Cs0D9La9ECoYbHJAIQ95DZfkDT&export=download&authuser=0&confirm=t&uuid=1e0f94d8-4fb1-4f9b-878a-516cf4a7ebc4&at=APZUnTXHBZKTuW_vQUzi-1XKuAYg:1705242554801",
"path": "/home/user/aws-bill.pdf"
},
{
"url": "https://drive.usercontent.google.com/download?id=1EHLRWzBCOsyERkSMUnTF2pnsR0n6ZvtR&export=download&authuser=0&confirm=t&uuid=de09bd5e-bef8-499a-b599-c642af190e10&at=APZUnTXqOsQkxl0zMSX6R1Sgp_v3:1704362491712",
"path": "/home/user/thunderbird-profile.tar.gz"
},
{
"url": "https://drive.usercontent.google.com/download?id=17LutYTkEXhM__sxvrdldar7p-eQba4Pm&export=download&authuser=0&confirm=t&uuid=34730e1c-0fac-401a-ab91-102a47f02224&at=APZUnTX4eFTnG9vnYlQEJX6N0jqP:1705227498314",
"path": "/home/user/.aws-bill-mail-body.html"
}
]
}
},
{
"type": "execute",
"parameters": {
"command": [
"tar",
"-xzv",
"--recursive-unlink",
"-f",
"/home/user/thunderbird-profile.tar.gz",
"-C",
"/home/user/"
]
}
},
{
"type": "launch",
"parameters": {
"command": "/usr/bin/thunderbird -compose \"from='Anonym Tester <anonym-x2024@outlook.com>',to=assistant@outlook.com,subject='New-month AWS Bill',body='$(cat /home/user/.aws-bill-mail-body.html)'\"",
"shell": true
}
}
],
"trajectory": "trajectories/d088f539-cab4-4f9a-ac92-9999fc3a656e",
"related_apps": [
"thunderbird"
],
"evaluator": {
"postconfig": [
{
"type": "download",
"parameters": {
"files": [
{
"url": "https://drive.usercontent.google.com/download?id=1abb7PPPRGGy3elpq2vTXe2aKTMOrbm7c&export=download&authuser=0&confirm=t&uuid=652a3d4c-278e-499c-bb26-58d0f70aa317&at=APZUnTXs_ZTZzOF3QrBa_rMjMr08:1705287009269",
"path": "/home/user/show-thunderbird-attachments.py"
}
]
}
},
{
"type": "execute",
"parameters": {
"command": ["pip", "install", "cssselect"]
}
},
{
"type": "execute",
"parameters": {
"command": [
"python",
"/home/user/show-thunderbird-attachments.py",
"New-month AWS Bill",
"aws-bill.pdf"
],
"stdout": "check-attachments.output"
}
}
],
"func": "check_list",
"result": {
"type": "cache_file",
"path": "check-attachments.output"
},
"expected": {
"type": "rule",
"rules": {
"expect": [
"Attachment added!"
]
}
}
}
}

77
evaluation_examples/examples/thunderbird/d088f539-cab4-4f9a-ac92-9999fc3a656e.json.nosetup Normal file
View File

@@ -0,0 +1,77 @@
{
"id": "d088f539-cab4-4f9a-ac92-9999fc3a656e",
"snapshot": "thunderbird",
"instruction": "Attach the my AWS bill to the email. The bill is stored at ~/aws-bill.pdf. Don't close it or send it. I haven't finish all the contents.",
"source": "https://support.mozilla.org/en-US/kb/how-use-attachments",
"config": [
{
"type": "download",
"parameters": {
"files": [
{
"url": "https://drive.usercontent.google.com/download?id=1JDUxe8Cs0D9La9ECoYbHJAIQ95DZfkDT&export=download&authuser=0&confirm=t&uuid=1e0f94d8-4fb1-4f9b-878a-516cf4a7ebc4&at=APZUnTXHBZKTuW_vQUzi-1XKuAYg:1705242554801",
"path": "/home/user/aws-bill.pdf"
},
{
"url": "https://drive.usercontent.google.com/download?id=1EHLRWzBCOsyERkSMUnTF2pnsR0n6ZvtR&export=download&authuser=0&confirm=t&uuid=de09bd5e-bef8-499a-b599-c642af190e10&at=APZUnTXqOsQkxl0zMSX6R1Sgp_v3:1704362491712",
"path": "/home/user/thunderbird-profile.tar.gz"
},
{
"url": "https://drive.usercontent.google.com/download?id=17LutYTkEXhM__sxvrdldar7p-eQba4Pm&export=download&authuser=0&confirm=t&uuid=34730e1c-0fac-401a-ab91-102a47f02224&at=APZUnTX4eFTnG9vnYlQEJX6N0jqP:1705227498314",
"path": "/home/user/.aws-bill-mail-body.html"
}
]
}
}
],
"trajectory": "trajectories/d088f539-cab4-4f9a-ac92-9999fc3a656e",
"related_apps": [
"thunderbird"
],
"evaluator": {
"postconfig": [
{
"type": "download",
"parameters": {
"files": [
{
"url": "https://drive.usercontent.google.com/download?id=1abb7PPPRGGy3elpq2vTXe2aKTMOrbm7c&export=download&authuser=0&confirm=t&uuid=652a3d4c-278e-499c-bb26-58d0f70aa317&at=APZUnTXs_ZTZzOF3QrBa_rMjMr08:1705287009269",
"path": "/home/user/show-thunderbird-attachments.py"
}
]
}
},
{
"type": "execute",
"parameters": {
"command": ["pip", "install", "cssselect"]
}
},
{
"type": "execute",
"parameters": {
"command": [
"python",
"/home/user/show-thunderbird-attachments.py",
"New-month AWS Bill",
"aws-bill.pdf"
],
"stdout": "check-attachments.output"
}
}
],
"func": "check_list",
"result": {
"type": "cache_file",
"path": "check-attachments.output"
},
"expected": {
"type": "rule",
"rules": {
"expect": [
"Attachment added!"
]
}
}
}
}