isort sorts ALL_CAPS objects before other objects, while autoflake sorts purely lexically after removing an unused import object from a from ... import ... line - which causes discrepancies and means isort has to be run after autoflake in CI's.

Example:

initial file

from typing import TYPE_CHECKING, Any, NamedTuple, TypeVar

if TYPE_CHECKING:
    T = TypeVar('T', bound=NamedTuple)

first line after running autoflake

from typing import NamedTuple, TYPE_CHECKING, TypeVar

which then gets modified if we run isort afterwards

from typing import TYPE_CHECKING, NamedTuple, TypeVar

I suggest this be solved either by adhering to isort's sorting order when removing an import - or not resorting the import objects at all when removing and entirely leaving that up to isort.