Spaces:
Sleeping
Sleeping
Upload 3 files
Browse files
main.py
CHANGED
|
@@ -6,7 +6,7 @@ from solver import solve_quiz
|
|
| 6 |
import logging
|
| 7 |
|
| 8 |
# Configure logging
|
| 9 |
-
logging.basicConfig(level=logging.INFO)
|
| 10 |
logger = logging.getLogger(__name__)
|
| 11 |
|
| 12 |
load_dotenv()
|
|
@@ -20,81 +20,6 @@ class QuizRequest(BaseModel):
|
|
| 20 |
class Config:
|
| 21 |
extra = "allow"
|
| 22 |
|
| 23 |
-
from fastapi.responses import HTMLResponse
|
| 24 |
-
|
| 25 |
-
@app.get("/", response_class=HTMLResponse)
|
| 26 |
-
async def home():
|
| 27 |
-
return """
|
| 28 |
-
<!DOCTYPE html>
|
| 29 |
-
<html>
|
| 30 |
-
<head>
|
| 31 |
-
<title>Quiz Solver</title>
|
| 32 |
-
<style>
|
| 33 |
-
body { font-family: sans-serif; max-width: 600px; margin: 40px auto; padding: 20px; }
|
| 34 |
-
.form-group { margin-bottom: 15px; }
|
| 35 |
-
label { display: block; margin-bottom: 5px; }
|
| 36 |
-
input { width: 100%; padding: 8px; box-sizing: border-box; }
|
| 37 |
-
button { padding: 10px 20px; background: #007bff; color: white; border: none; cursor: pointer; }
|
| 38 |
-
button:hover { background: #0056b3; }
|
| 39 |
-
#status { margin-top: 20px; padding: 10px; display: none; }
|
| 40 |
-
.success { background: #d4edda; color: #155724; }
|
| 41 |
-
.error { background: #f8d7da; color: #721c24; }
|
| 42 |
-
</style>
|
| 43 |
-
</head>
|
| 44 |
-
<body>
|
| 45 |
-
<h1>Quiz Solver Agent</h1>
|
| 46 |
-
<div class="form-group">
|
| 47 |
-
<label>Email:</label>
|
| 48 |
-
<input type="email" id="email" value="22f3000730@ds.iitm.ac.in">
|
| 49 |
-
</div>
|
| 50 |
-
<div class="form-group">
|
| 51 |
-
<label>Secret:</label>
|
| 52 |
-
<input type="password" id="secret" placeholder="Enter your secret">
|
| 53 |
-
</div>
|
| 54 |
-
<div class="form-group">
|
| 55 |
-
<label>Start URL:</label>
|
| 56 |
-
<input type="text" id="url" value="https://tds-llm-analysis.s-anand.net/project2">
|
| 57 |
-
</div>
|
| 58 |
-
<button onclick="startSolver()">Start Solving</button>
|
| 59 |
-
<div id="status"></div>
|
| 60 |
-
|
| 61 |
-
<script>
|
| 62 |
-
async function startSolver() {
|
| 63 |
-
const email = document.getElementById('email').value;
|
| 64 |
-
const secret = document.getElementById('secret').value;
|
| 65 |
-
const url = document.getElementById('url').value;
|
| 66 |
-
const statusDiv = document.getElementById('status');
|
| 67 |
-
|
| 68 |
-
statusDiv.style.display = 'block';
|
| 69 |
-
statusDiv.className = '';
|
| 70 |
-
statusDiv.textContent = 'Sending request...';
|
| 71 |
-
|
| 72 |
-
try {
|
| 73 |
-
const response = await fetch('/', {
|
| 74 |
-
method: 'POST',
|
| 75 |
-
headers: { 'Content-Type': 'application/json' },
|
| 76 |
-
body: JSON.stringify({ email, secret, url })
|
| 77 |
-
});
|
| 78 |
-
|
| 79 |
-
const data = await response.json();
|
| 80 |
-
|
| 81 |
-
if (response.ok) {
|
| 82 |
-
statusDiv.className = 'success';
|
| 83 |
-
statusDiv.textContent = 'Success: ' + data.message;
|
| 84 |
-
} else {
|
| 85 |
-
statusDiv.className = 'error';
|
| 86 |
-
statusDiv.textContent = 'Error: ' + (data.detail || 'Unknown error');
|
| 87 |
-
}
|
| 88 |
-
} catch (e) {
|
| 89 |
-
statusDiv.className = 'error';
|
| 90 |
-
statusDiv.textContent = 'Network Error: ' + e.message;
|
| 91 |
-
}
|
| 92 |
-
}
|
| 93 |
-
</script>
|
| 94 |
-
</body>
|
| 95 |
-
</html>
|
| 96 |
-
"""
|
| 97 |
-
|
| 98 |
@app.post("/", status_code=200)
|
| 99 |
async def solve_quiz_endpoint(request: QuizRequest, background_tasks: BackgroundTasks):
|
| 100 |
logger.info(f"Received quiz request for URL: {request.url}")
|
|
|
|
| 6 |
import logging
|
| 7 |
|
| 8 |
# Configure logging
|
| 9 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
| 10 |
logger = logging.getLogger(__name__)
|
| 11 |
|
| 12 |
load_dotenv()
|
|
|
|
| 20 |
class Config:
|
| 21 |
extra = "allow"
|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
@app.post("/", status_code=200)
|
| 24 |
async def solve_quiz_endpoint(request: QuizRequest, background_tasks: BackgroundTasks):
|
| 25 |
logger.info(f"Received quiz request for URL: {request.url}")
|
solver.py
CHANGED
|
@@ -10,9 +10,14 @@ from agent import get_agent
|
|
| 10 |
logger = logging.getLogger(__name__)
|
| 11 |
|
| 12 |
async def solve_quiz(initial_url: str, email: str, secret: str):
|
| 13 |
-
|
|
|
|
|
|
|
| 14 |
|
| 15 |
current_url = initial_url
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
|
| 18 |
async with async_playwright() as p:
|
|
@@ -22,10 +27,24 @@ async def solve_quiz(initial_url: str, email: str, secret: str):
|
|
| 22 |
|
| 23 |
try:
|
| 24 |
while current_url:
|
|
|
|
|
|
|
| 25 |
# Generate a NEW session ID for each task/URL to keep memory clean
|
| 26 |
import uuid
|
| 27 |
session_id = str(uuid.uuid4())
|
| 28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
|
| 30 |
logger.info(f"Navigating to {current_url}")
|
| 31 |
await page.goto(current_url)
|
|
@@ -118,8 +137,20 @@ async def solve_quiz(initial_url: str, email: str, secret: str):
|
|
| 118 |
# Use agent to solve (initialize here if needed, but we use get_agent() outside if we wanted persistent agent object,
|
| 119 |
# but we want fresh memory per task, so we rely on session_id)
|
| 120 |
agent = get_agent()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 121 |
|
| 122 |
prompt = f"""
|
|
|
|
| 123 |
You are a highly capable Quiz Solver Agent.
|
| 124 |
Current Page URL: {current_url}
|
| 125 |
|
|
@@ -134,6 +165,7 @@ Solve the task on the current page.
|
|
| 134 |
**GUIDELINES**
|
| 135 |
- **Conciseness**: Plan and explain in **2-3 lines maximum**.
|
| 136 |
- **Action**: Respond **IMMEDIATELY** with a tool call or the final JSON. **DO NOT** output conversational text or plans like "I need to...". Just run the code.
|
|
|
|
| 137 |
|
| 138 |
**TOOL USAGE**
|
| 139 |
- **Secret Codes**: Return exactly as requested (no extra spaces).
|
|
@@ -263,6 +295,7 @@ Now return the final JSON for submission:
|
|
| 263 |
logger.info("Answer correct! Moving to next URL.")
|
| 264 |
|
| 265 |
current_url = next_url
|
|
|
|
| 266 |
break # Break retry loop to process new URL
|
| 267 |
|
| 268 |
# No new URL provided
|
|
@@ -272,8 +305,10 @@ Now return the final JSON for submission:
|
|
| 272 |
break # Break retry loop
|
| 273 |
else:
|
| 274 |
logger.warning(f"Answer incorrect: {submission_response.get('reason')}")
|
| 275 |
-
logger.info("No new URL provided. Retrying same URL in
|
| 276 |
-
|
|
|
|
|
|
|
| 277 |
# Break inner loop to refresh page and try again
|
| 278 |
break
|
| 279 |
|
|
|
|
| 10 |
logger = logging.getLogger(__name__)
|
| 11 |
|
| 12 |
async def solve_quiz(initial_url: str, email: str, secret: str):
|
| 13 |
+
import uuid
|
| 14 |
+
task_id = str(uuid.uuid4())[:8]
|
| 15 |
+
logger.info(f"[{task_id}] Starting quiz solver workflow for {email}")
|
| 16 |
|
| 17 |
current_url = initial_url
|
| 18 |
+
last_url = None
|
| 19 |
+
url_attempts = 0
|
| 20 |
+
failure_details = None
|
| 21 |
|
| 22 |
|
| 23 |
async with async_playwright() as p:
|
|
|
|
| 27 |
|
| 28 |
try:
|
| 29 |
while current_url:
|
| 30 |
+
# Generate a NEW session ID for each task/URL to keep memory clean
|
| 31 |
+
import uuid
|
| 32 |
# Generate a NEW session ID for each task/URL to keep memory clean
|
| 33 |
import uuid
|
| 34 |
session_id = str(uuid.uuid4())
|
| 35 |
+
|
| 36 |
+
# Update attempt counter
|
| 37 |
+
if current_url == last_url:
|
| 38 |
+
url_attempts += 1
|
| 39 |
+
else:
|
| 40 |
+
last_url = current_url
|
| 41 |
+
url_attempts = 1
|
| 42 |
+
|
| 43 |
+
if url_attempts > 60:
|
| 44 |
+
logger.error(f"Max attempts (60) reached for {current_url}. Stopping loop.")
|
| 45 |
+
break
|
| 46 |
+
|
| 47 |
+
logger.info(f"Started new agent session for {current_url} (Attempt {url_attempts}): {session_id}")
|
| 48 |
|
| 49 |
logger.info(f"Navigating to {current_url}")
|
| 50 |
await page.goto(current_url)
|
|
|
|
| 137 |
# Use agent to solve (initialize here if needed, but we use get_agent() outside if we wanted persistent agent object,
|
| 138 |
# but we want fresh memory per task, so we rely on session_id)
|
| 139 |
agent = get_agent()
|
| 140 |
+
|
| 141 |
+
failure_info = ""
|
| 142 |
+
if failure_details:
|
| 143 |
+
failure_info = f"""
|
| 144 |
+
***PREVIOUS FAILED SUBMISSION***
|
| 145 |
+
- The previous answer you submitted was INCORRECT.
|
| 146 |
+
- **Submitted Answer**: {json.dumps(failure_details['answer'].get('answer', 'UNKNOWN'))}
|
| 147 |
+
- **Error/Reason**: {failure_details['reason']}
|
| 148 |
+
- **GUIDANCE**: Analyze the error message carefully. Do NOT repeat the same answer. Try a different approach or format.
|
| 149 |
+
***
|
| 150 |
+
"""
|
| 151 |
|
| 152 |
prompt = f"""
|
| 153 |
+
{failure_info}
|
| 154 |
You are a highly capable Quiz Solver Agent.
|
| 155 |
Current Page URL: {current_url}
|
| 156 |
|
|
|
|
| 165 |
**GUIDELINES**
|
| 166 |
- **Conciseness**: Plan and explain in **2-3 lines maximum**.
|
| 167 |
- **Action**: Respond **IMMEDIATELY** with a tool call or the final JSON. **DO NOT** output conversational text or plans like "I need to...". Just run the code.
|
| 168 |
+
- Have a dummy answer for the intial request
|
| 169 |
|
| 170 |
**TOOL USAGE**
|
| 171 |
- **Secret Codes**: Return exactly as requested (no extra spaces).
|
|
|
|
| 295 |
logger.info("Answer correct! Moving to next URL.")
|
| 296 |
|
| 297 |
current_url = next_url
|
| 298 |
+
failure_details = None # Reset on success
|
| 299 |
break # Break retry loop to process new URL
|
| 300 |
|
| 301 |
# No new URL provided
|
|
|
|
| 305 |
break # Break retry loop
|
| 306 |
else:
|
| 307 |
logger.warning(f"Answer incorrect: {submission_response.get('reason')}")
|
| 308 |
+
logger.info("No new URL provided. Retrying same URL in 20 seconds...")
|
| 309 |
+
current_url = current_url # Just to be explicit, logic loops anyway
|
| 310 |
+
failure_details = {"answer": answer_payload, "reason": submission_response.get("reason")}
|
| 311 |
+
await asyncio.sleep(20)
|
| 312 |
# Break inner loop to refresh page and try again
|
| 313 |
break
|
| 314 |
|