Features:

• Maximum Security: Complete isolation in Docker container
• Resource Limits: Memory (32-512MB), CPU (0.5 cores)
• Network Isolation: No network access during execution
• Filesystem: Read-only with limited /tmp access (10MB)
• User Security: Non-root execution (nexus:1000)
• Cleanup: Automatic container and image removal

Features:

• Local Execution: Runs on the host system when Docker is unavailable
• Restricted Environment: Limited builtins and import restrictions
• Resource Limits: Memory and CPU limits via system calls
• Security: Code validation and import filtering
• Isolation: Process-level isolation only

Scenario: Convert proprietary date format

Task: Convert "20250907-143022" format to ISO format

# LLM generates this code dynamically
python_code = """
from datetime import datetime

custom_date = "20250907-143022"
date_part, time_part = custom_date.split("-")

dt = datetime(
    int(date_part[:4]),    # year
    int(date_part[4:6]),   # month  
    int(date_part[6:8]),   # day
    int(time_part[:2]),    # hour
    int(time_part[2:4]),   # minute
    int(time_part[4:6])    # second
)

print(f"ISO: {dt.isoformat()}")
print(f"Human: {dt.strftime('%Y-%m-%d %H:%M:%S')}")
"""

# Claude calls this automatically
result = create_and_run_tool(python_code, timeout=30, memory_limit_mb=64)

Output:

ISO: 2025-09-07T14:30:22
Human: 2025-09-07 14:30:22

Scenario: Convert legacy pipe-separated data to JSON

Task: Parse "JOHN|25|NYC|50000" format into structured JSON

# Dynamic tool for legacy data conversion
python_code = """
import json

legacy_data = "JOHN|25|NYC|50000\\nJANE|30|LA|75000"
records = []

for line in legacy_data.strip().split("\\n"):
    if line.strip():
        name, age, city, salary = line.split("|")
        records.append({
            "name": name,
            "age": int(age),
            "city": city,
            "salary": int(salary)
        })

print(json.dumps(records, indent=2))
"""

result = create_and_run_tool(python_code, timeout=45)

Output:

[
  {
    "name": "JOHN",
    "age": 25,
    "city": "NYC",
    "salary": 50000
  },
  {
    "name": "JANE",
    "age": 30,
    "city": "LA",
    "salary": 75000
  }
]

Scenario: Custom mathematical function

Task: Calculate the nth pentagonal number with verification

# Specialized mathematical calculation
python_code = """
def pentagonal_number(n):
    '''Calculate the nth pentagonal number: n(3n-1)/2'''
    return n * (3 * n - 1) // 2

def verify_pentagonal(p):
    '''Verify if a number is pentagonal'''
    # Using quadratic formula: n = (1 + sqrt(1 + 24p)) / 6
    import math
    discriminant = 1 + 24 * p
    if discriminant < 0:
        return False
    
    sqrt_d = math.sqrt(discriminant)
    n = (1 + sqrt_d) / 6
    
    return n == int(n) and n > 0

# Calculate pentagonal numbers
for i in range(1, 11):
    pent = pentagonal_number(i)
    is_valid = verify_pentagonal(pent)
    print(f"P({i}) = {pent}, Valid: {is_valid}")

# Test specific number
test_num = 145
print(f"\\nIs {test_num} pentagonal? {verify_pentagonal(test_num)}")
"""

result = create_and_run_tool(python_code, timeout=30, memory_limit_mb=64)