Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,9 +1,8 @@
|
|
| 1 |
import ast
|
| 2 |
import torch
|
| 3 |
import torch.nn as nn
|
| 4 |
-
from fastapi import FastAPI, HTTPException
|
| 5 |
from pydantic import BaseModel
|
| 6 |
-
from typing import Optional
|
| 7 |
from transformers import (
|
| 8 |
T5ForConditionalGeneration,
|
| 9 |
RobertaTokenizer,
|
|
@@ -12,225 +11,163 @@ from transformers import (
|
|
| 12 |
)
|
| 13 |
import pandas as pd
|
| 14 |
import os
|
| 15 |
-
import threading
|
| 16 |
|
| 17 |
-
|
| 18 |
-
from train_engine import train_on_devign
|
| 19 |
-
|
| 20 |
-
app = FastAPI(title="Revcode AI ULTRA Orchestrator")
|
| 21 |
-
|
| 22 |
-
# Global training status
|
| 23 |
-
training_lock = threading.Lock()
|
| 24 |
-
is_training = False
|
| 25 |
-
|
| 26 |
-
# ---------------------------------------------------------
|
| 27 |
-
# 1. DATA MODELS
|
| 28 |
-
# ---------------------------------------------------------
|
| 29 |
-
class CodeInput(BaseModel):
|
| 30 |
-
code: str
|
| 31 |
-
filename: Optional[str] = "snippet.js"
|
| 32 |
|
| 33 |
# ---------------------------------------------------------
|
| 34 |
-
#
|
| 35 |
# ---------------------------------------------------------
|
| 36 |
class DeepVulnerabilityScanner:
|
| 37 |
def __init__(self):
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
self.
|
| 42 |
-
self.
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
self.
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
self.tokenizer = AutoTokenizer.from_pretrained(self.tokenizer_name)
|
| 50 |
-
self.model = AutoModelForSequenceClassification.from_pretrained(self.model_name)
|
| 51 |
-
self.model.eval()
|
| 52 |
-
|
| 53 |
def scan(self, code: str) -> dict:
|
| 54 |
-
|
|
|
|
|
|
|
|
|
|
| 55 |
with torch.no_grad():
|
| 56 |
logits = self.model(**inputs).logits
|
| 57 |
|
| 58 |
probs = torch.softmax(logits, dim=1)
|
| 59 |
vuln_prob = probs[0][1].item()
|
| 60 |
|
| 61 |
-
reasoning = "Analyzing code logic for Devign-pattern vulnerabilities."
|
| 62 |
-
if vuln_prob > 0.9:
|
| 63 |
-
reasoning = "CRITICAL: High-confidence fingerprint of a known vulnerability pattern (e.g., Buffer Overflow, Improper Sanitization)."
|
| 64 |
-
elif vuln_prob > 0.5:
|
| 65 |
-
reasoning = "WARNING: Code semantics mirror dangerous patterns found in the Devign security dataset."
|
| 66 |
-
elif vuln_prob < 0.1:
|
| 67 |
-
reasoning = "SAFE: Code logic is clean of any recognized vulnerability fingerprints."
|
| 68 |
-
|
| 69 |
return {
|
| 70 |
"is_vulnerable": vuln_prob > 0.5,
|
| 71 |
"risk_score": round(vuln_prob * 100, 2),
|
| 72 |
-
"
|
| 73 |
-
"reasoning": reasoning
|
| 74 |
}
|
| 75 |
|
| 76 |
# ---------------------------------------------------------
|
| 77 |
-
#
|
| 78 |
-
# ---------------------------------------------------------
|
| 79 |
-
class StructuralScanner:
|
| 80 |
-
@staticmethod
|
| 81 |
-
def scan_patterns(code: str, filename: str) -> list:
|
| 82 |
-
findings = []
|
| 83 |
-
if "os.system(" in code or "subprocess.Popen(..., shell=True)" in code:
|
| 84 |
-
findings.append({
|
| 85 |
-
"type": "Security",
|
| 86 |
-
"title": "Command Injection Risk",
|
| 87 |
-
"reasoning": "Detected use of shell=True or os.system which can lead to Remote Code Execution."
|
| 88 |
-
})
|
| 89 |
-
if "pickle.load" in code or "yaml.load(..., Loader=None)" in code:
|
| 90 |
-
findings.append({
|
| 91 |
-
"type": "Security",
|
| 92 |
-
"title": "Insecure Deserialization",
|
| 93 |
-
"reasoning": "Insecure loading of serialized data can lead to arbitrary code execution."
|
| 94 |
-
})
|
| 95 |
-
if "Password =" in code or "API_KEY =" in code:
|
| 96 |
-
findings.append({
|
| 97 |
-
"type": "Compliance",
|
| 98 |
-
"title": "Hardcoded Secret",
|
| 99 |
-
"reasoning": "Sensitive credentials found in source code. Use environment variables instead."
|
| 100 |
-
})
|
| 101 |
-
return findings
|
| 102 |
-
|
| 103 |
-
# ---------------------------------------------------------
|
| 104 |
-
# 4. AUTOMATED REPAIR ENGINE (The "Surgeon" + Context)
|
| 105 |
# ---------------------------------------------------------
|
| 106 |
class AutomatedRepairEngine:
|
| 107 |
def __init__(self):
|
| 108 |
-
print("Loading Repair Engine (CodeT5
|
|
|
|
| 109 |
self.model_name = "Salesforce/codet5p-220m"
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
with torch.no_grad():
|
| 118 |
outputs = self.model.generate(
|
| 119 |
**inputs,
|
| 120 |
-
max_length=
|
| 121 |
num_beams=5,
|
| 122 |
temperature=0.7,
|
| 123 |
early_stopping=True
|
| 124 |
)
|
|
|
|
| 125 |
return self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 126 |
|
| 127 |
# ---------------------------------------------------------
|
| 128 |
-
#
|
| 129 |
# ---------------------------------------------------------
|
| 130 |
-
class
|
| 131 |
@staticmethod
|
| 132 |
-
def
|
| 133 |
try:
|
| 134 |
ast.parse(code)
|
| 135 |
-
return True
|
| 136 |
-
except Exception
|
| 137 |
-
return False
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 138 |
|
| 139 |
# ---------------------------------------------------------
|
| 140 |
-
#
|
| 141 |
# ---------------------------------------------------------
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
struct_scanner = StructuralScanner()
|
| 145 |
-
guardrails = Guardrails()
|
| 146 |
|
| 147 |
-
def get_scanner(
|
| 148 |
-
global
|
| 149 |
-
if
|
| 150 |
-
|
| 151 |
-
return
|
| 152 |
|
| 153 |
def get_repairer():
|
| 154 |
-
global
|
| 155 |
-
if
|
| 156 |
-
|
| 157 |
-
return
|
| 158 |
-
|
| 159 |
-
# ---------------------------------------------------------
|
| 160 |
-
# 7. TRAINING WRAPPER
|
| 161 |
-
# ---------------------------------------------------------
|
| 162 |
-
def run_training():
|
| 163 |
-
global is_training
|
| 164 |
-
with training_lock:
|
| 165 |
-
is_training = True
|
| 166 |
-
try:
|
| 167 |
-
print("--- STARTING BACKGROUND TRAINING CYCLE ---")
|
| 168 |
-
train_on_devign(output_dir="./trained_model")
|
| 169 |
-
print("--- TRAINING CYCLE COMPLETED. RELOADING SCANNER ---")
|
| 170 |
-
get_scanner(force_reload=True)
|
| 171 |
-
finally:
|
| 172 |
-
with training_lock:
|
| 173 |
-
is_training = False
|
| 174 |
|
| 175 |
# ---------------------------------------------------------
|
| 176 |
-
#
|
| 177 |
# ---------------------------------------------------------
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
return {
|
| 181 |
-
"status": "Revcode AI ULTRA Orchestrator Operational",
|
| 182 |
-
"is_training": is_training,
|
| 183 |
-
"features": ["XAI", "Structural-Scan", "Context-Injection", "Auto-Train"]
|
| 184 |
-
}
|
| 185 |
-
|
| 186 |
-
@app.post("/train")
|
| 187 |
-
async def trigger_training(background_tasks: BackgroundTasks):
|
| 188 |
-
global is_training
|
| 189 |
-
if is_training:
|
| 190 |
-
return {"status": "error", "message": "Training already in progress."}
|
| 191 |
-
|
| 192 |
-
background_tasks.add_task(run_training)
|
| 193 |
-
return {"status": "success", "message": "Training started in background."}
|
| 194 |
|
| 195 |
@app.post("/analyze")
|
| 196 |
async def analyze_security(data: CodeInput):
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
if structural_findings:
|
| 201 |
-
res["is_vulnerable"] = True
|
| 202 |
-
res["reasoning"] += " | Structural rules flagged: " + ", ".join([f['title'] for f in structural_findings])
|
| 203 |
-
res["verdict"] = "CRITICAL_VULNERABILITY"
|
| 204 |
-
|
| 205 |
return {
|
| 206 |
-
"is_vulnerable":
|
| 207 |
-
"confidence":
|
| 208 |
-
"verdict":
|
| 209 |
-
"
|
| 210 |
-
"
|
| 211 |
-
"is_training": is_training,
|
| 212 |
-
"provider": "DeepScanner-ULTRA"
|
| 213 |
}
|
| 214 |
|
| 215 |
@app.post("/fix")
|
| 216 |
async def fix_code(data: CodeInput):
|
| 217 |
-
|
| 218 |
-
|
| 219 |
-
|
| 220 |
-
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
|
| 224 |
-
|
| 225 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
|
| 227 |
-
@app.post("/verify")
|
| 228 |
-
async def verify_fix(data: CodeInput):
|
| 229 |
-
is_valid, msg = guardrails.validate(data.code)
|
| 230 |
return {
|
| 231 |
-
"
|
| 232 |
-
"
|
| 233 |
-
"
|
| 234 |
}
|
| 235 |
|
| 236 |
@app.post("/feedback")
|
|
@@ -239,3 +176,11 @@ async def store_feedback(data: dict):
|
|
| 239 |
df = pd.DataFrame([data])
|
| 240 |
df.to_csv(feedback_file, mode='a', header=not os.path.exists(feedback_file), index=False)
|
| 241 |
return {"status": "Feedback stored for retraining"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import ast
|
| 2 |
import torch
|
| 3 |
import torch.nn as nn
|
| 4 |
+
from fastapi import FastAPI, HTTPException
|
| 5 |
from pydantic import BaseModel
|
|
|
|
| 6 |
from transformers import (
|
| 7 |
T5ForConditionalGeneration,
|
| 8 |
RobertaTokenizer,
|
|
|
|
| 11 |
)
|
| 12 |
import pandas as pd
|
| 13 |
import os
|
|
|
|
| 14 |
|
| 15 |
+
app = FastAPI(title="Revcode AI Strong Orchestrator")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
# ---------------------------------------------------------
|
| 18 |
+
# 1. ADVANCED SECURITY SCANNER (The "Brain")
|
| 19 |
# ---------------------------------------------------------
|
| 20 |
class DeepVulnerabilityScanner:
|
| 21 |
def __init__(self):
|
| 22 |
+
print("Loading Deep Security Scanner (DistilRoBERTa)...")
|
| 23 |
+
self.model_name = "distilroberta-base"
|
| 24 |
+
try:
|
| 25 |
+
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
|
| 26 |
+
self.model = AutoModelForSequenceClassification.from_pretrained(self.model_name, num_labels=2)
|
| 27 |
+
self.model.eval()
|
| 28 |
+
except Exception as e:
|
| 29 |
+
print(f"Failed to load Deep Scanner: {e}")
|
| 30 |
+
self.model = None
|
| 31 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
def scan(self, code: str) -> dict:
|
| 33 |
+
if not self.model:
|
| 34 |
+
return {"is_vulnerable": False, "risk_score": 0.0, "details": "Scanner unavailable"}
|
| 35 |
+
|
| 36 |
+
inputs = self.tokenizer(code, return_tensors="pt", truncation=True, max_length=512)
|
| 37 |
with torch.no_grad():
|
| 38 |
logits = self.model(**inputs).logits
|
| 39 |
|
| 40 |
probs = torch.softmax(logits, dim=1)
|
| 41 |
vuln_prob = probs[0][1].item()
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
return {
|
| 44 |
"is_vulnerable": vuln_prob > 0.5,
|
| 45 |
"risk_score": round(vuln_prob * 100, 2),
|
| 46 |
+
"details": "Potential vulnerability detected in code logic." if vuln_prob > 0.5 else "Clean code."
|
|
|
|
| 47 |
}
|
| 48 |
|
| 49 |
# ---------------------------------------------------------
|
| 50 |
+
# 2. AUTOMATED REPAIR ENGINE (The "Surgeon")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
# ---------------------------------------------------------
|
| 52 |
class AutomatedRepairEngine:
|
| 53 |
def __init__(self):
|
| 54 |
+
print("Loading Repair Engine (CodeT5)...")
|
| 55 |
+
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 56 |
self.model_name = "Salesforce/codet5p-220m"
|
| 57 |
+
try:
|
| 58 |
+
self.tokenizer = RobertaTokenizer.from_pretrained(self.model_name)
|
| 59 |
+
self.model = T5ForConditionalGeneration.from_pretrained(self.model_name).to(self.device)
|
| 60 |
+
self.model.eval()
|
| 61 |
+
except Exception as e:
|
| 62 |
+
print(f"Failed to load Repair Engine: {e}")
|
| 63 |
+
self.model = None
|
| 64 |
+
|
| 65 |
+
def repair(self, buggy_code: str) -> str:
|
| 66 |
+
if not self.model:
|
| 67 |
+
return buggy_code
|
| 68 |
|
| 69 |
+
prompt = f"Fix the security vulnerability: {buggy_code}"
|
| 70 |
+
inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=256).to(self.device)
|
| 71 |
+
|
| 72 |
with torch.no_grad():
|
| 73 |
outputs = self.model.generate(
|
| 74 |
**inputs,
|
| 75 |
+
max_length=256,
|
| 76 |
num_beams=5,
|
| 77 |
temperature=0.7,
|
| 78 |
early_stopping=True
|
| 79 |
)
|
| 80 |
+
|
| 81 |
return self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 82 |
|
| 83 |
# ---------------------------------------------------------
|
| 84 |
+
# 3. SYNTAX & LOGIC VALIDATOR (The "Quality Control")
|
| 85 |
# ---------------------------------------------------------
|
| 86 |
+
class CodeValidator:
|
| 87 |
@staticmethod
|
| 88 |
+
def is_syntax_valid(code: str) -> bool:
|
| 89 |
try:
|
| 90 |
ast.parse(code)
|
| 91 |
+
return True
|
| 92 |
+
except Exception:
|
| 93 |
+
return False
|
| 94 |
+
|
| 95 |
+
@staticmethod
|
| 96 |
+
def check_security_patterns(code: str) -> list:
|
| 97 |
+
issues = []
|
| 98 |
+
dangerous_calls = ["eval(", "exec(", "os.system(", "subprocess.call("]
|
| 99 |
+
for call in dangerous_calls:
|
| 100 |
+
if call in code:
|
| 101 |
+
issues.append(f"Dangerous call found: {call}")
|
| 102 |
+
return issues
|
| 103 |
|
| 104 |
# ---------------------------------------------------------
|
| 105 |
+
# 4. GLOBAL HANDLERS (Lazy Loading)
|
| 106 |
# ---------------------------------------------------------
|
| 107 |
+
_scanner = None
|
| 108 |
+
_repairer = None
|
|
|
|
|
|
|
| 109 |
|
| 110 |
+
def get_scanner():
|
| 111 |
+
global _scanner
|
| 112 |
+
if _scanner is None:
|
| 113 |
+
_scanner = DeepVulnerabilityScanner()
|
| 114 |
+
return _scanner
|
| 115 |
|
| 116 |
def get_repairer():
|
| 117 |
+
global _repairer
|
| 118 |
+
if _repairer is None:
|
| 119 |
+
_repairer = AutomatedRepairEngine()
|
| 120 |
+
return _repairer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 121 |
|
| 122 |
# ---------------------------------------------------------
|
| 123 |
+
# 5. API ENDPOINTS
|
| 124 |
# ---------------------------------------------------------
|
| 125 |
+
class CodeInput(BaseModel):
|
| 126 |
+
code: str
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
|
| 128 |
@app.post("/analyze")
|
| 129 |
async def analyze_security(data: CodeInput):
|
| 130 |
+
scanner = get_scanner()
|
| 131 |
+
result = scanner.scan(data.code)
|
| 132 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 133 |
return {
|
| 134 |
+
"is_vulnerable": result["is_vulnerable"],
|
| 135 |
+
"confidence": result["risk_score"],
|
| 136 |
+
"verdict": "VULNERABLE" if result["is_vulnerable"] else "SECURE",
|
| 137 |
+
"details": result["details"],
|
| 138 |
+
"provider": "DistilRoBERTa-Strong"
|
|
|
|
|
|
|
| 139 |
}
|
| 140 |
|
| 141 |
@app.post("/fix")
|
| 142 |
async def fix_code(data: CodeInput):
|
| 143 |
+
repairer = get_repairer()
|
| 144 |
+
validator = CodeValidator()
|
| 145 |
+
|
| 146 |
+
# ML Repair
|
| 147 |
+
suggestion = repairer.repair(data.code)
|
| 148 |
+
|
| 149 |
+
# Validation Loop
|
| 150 |
+
status = "PASSED"
|
| 151 |
+
msg = "Valid syntax"
|
| 152 |
+
|
| 153 |
+
if not validator.is_syntax_valid(suggestion):
|
| 154 |
+
status = "FAILED"
|
| 155 |
+
msg = "Repair generated invalid syntax"
|
| 156 |
+
# Heuristic fallback (from user's logic)
|
| 157 |
+
suggestion = data.code.replace("eval(", "safe_eval(")
|
| 158 |
+
|
| 159 |
+
# Final Security Pattern Check
|
| 160 |
+
final_issues = validator.check_security_patterns(suggestion)
|
| 161 |
+
if final_issues:
|
| 162 |
+
for issue in final_issues:
|
| 163 |
+
call_name = issue.split(": ")[1]
|
| 164 |
+
suggestion = suggestion.replace(call_name, f"# BLOCKED_{call_name.replace('(', '')}")
|
| 165 |
+
msg += f" | Blocked {len(final_issues)} dangerous calls"
|
| 166 |
|
|
|
|
|
|
|
|
|
|
| 167 |
return {
|
| 168 |
+
"suggestion": suggestion,
|
| 169 |
+
"guardrail_status": status,
|
| 170 |
+
"guardrail_msg": msg
|
| 171 |
}
|
| 172 |
|
| 173 |
@app.post("/feedback")
|
|
|
|
| 176 |
df = pd.DataFrame([data])
|
| 177 |
df.to_csv(feedback_file, mode='a', header=not os.path.exists(feedback_file), index=False)
|
| 178 |
return {"status": "Feedback stored for retraining"}
|
| 179 |
+
|
| 180 |
+
@app.get("/")
|
| 181 |
+
async def health():
|
| 182 |
+
return {"status": "Revcode AI Strong Engine is alive"}
|
| 183 |
+
|
| 184 |
+
if __name__ == "__main__":
|
| 185 |
+
import uvicorn
|
| 186 |
+
uvicorn.run(app, host="0.0.0.0", port=8000)
|