Dynamic builtin resolution bypasses detection(malicious skill example that bypasses this static detection model)

Summary

The scanner detects direct use of exec(), but misses the same execution behavior when the function is resolved dynamically through getattr().

Original:

import base64

decrypted_data = base64.b64decode('just a test')
exec(decrypted_data.decode('utf-8'))

Result:

HIGH: AST1 - exec() call detected
Confidence: 85%

Variant:

import base64
import builtins

decrypted_data = base64.b64decode('just a test')

_runner = getattr(builtins, ''.join(['e','x','e','c']))
_runner(decrypted_data.decode('utf-8'))

Result:

LOW: AST7 - Dynamic attribute access via getattr()
Confidence: 50%

Expected Behavior

The variant should still trigger the high-severity exec() finding because the dynamically constructed attribute resolves to builtins.exec.

Impact

Dynamic builtin resolution downgrades the finding from HIGH arbitrary code execution to a generic LOW getattr() warning.

Suggested Fix

Add constant folding and builtin resolution for getattr() calls, including statically constructed names such as ''.join(['e','x','e','c']).

Start by tracing the existing AST1 exec() detection and AST7 getattr() warning in the scanner. Compare how the direct and dynamic examples are classified, then investigate constant folding and builtin resolution for getattr() names such as ''.join(['e','x','e','c']). Done means the dynamic variant produces the same HIGH exec() finding as the direct form.