Add scripts/validate.py
Browse files- scripts/validate.py +32 -0
scripts/validate.py
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import sys
|
| 2 |
+
from codereview_env.models import TaskId
|
| 3 |
+
from codereview_env.env import CodeReviewEnv
|
| 4 |
+
|
| 5 |
+
def validate_environment():
|
| 6 |
+
print("=== OpenEnv Validation Suite ===")
|
| 7 |
+
env = CodeReviewEnv()
|
| 8 |
+
|
| 9 |
+
tasks = [TaskId.BUG_DETECTION, TaskId.SECURITY_AUDIT, TaskId.ARCHITECTURAL_REVIEW]
|
| 10 |
+
total_scenarios = 30
|
| 11 |
+
passed = 0
|
| 12 |
+
|
| 13 |
+
for task in tasks:
|
| 14 |
+
print(f"\nValidating Task: {task}")
|
| 15 |
+
for seed in range(10): # Check first 10 scenarios per task
|
| 16 |
+
try:
|
| 17 |
+
env.reset(task, seed=seed)
|
| 18 |
+
passed += 1
|
| 19 |
+
print(f" [PASS] Scenario {seed}")
|
| 20 |
+
except Exception as e:
|
| 21 |
+
print(f" [FAIL] Scenario {seed}: {e}")
|
| 22 |
+
|
| 23 |
+
print(f"\nFinal Result: {passed}/{total_scenarios} scenarios reachable.")
|
| 24 |
+
if passed == 30:
|
| 25 |
+
print("ENVIRONMENT VALIDATED SUCCESSFULLY.")
|
| 26 |
+
sys.exit(0)
|
| 27 |
+
else:
|
| 28 |
+
print("VALIDATION FAILED.")
|
| 29 |
+
sys.exit(1)
|
| 30 |
+
|
| 31 |
+
if __name__ == "__main__":
|
| 32 |
+
validate_environment()
|