diff --git a/docs/source/changes.md b/docs/source/changes.md
index 5b9038fd..20541a4e 100644
--- a/docs/source/changes.md
+++ b/docs/source/changes.md
@@ -28,6 +28,7 @@ releases are available on [PyPI](https://pypi.org/project/pytask) and
 - {pull}`540` changes the CLI entry-point and allow `pytask.build(tasks=task_func)` as
   the signatures suggested.
 - {pull}`542` refactors the plugin manager.
+- {pull}`543` fixes imports in tests and related issues.
 
 ## 0.4.4 - 2023-12-04
 
diff --git a/src/pytask/__init__.py b/src/pytask/__init__.py
index 612dc537..8c530cc2 100644
--- a/src/pytask/__init__.py
+++ b/src/pytask/__init__.py
@@ -6,7 +6,6 @@
 from _pytask.build import build
 from _pytask.capture_utils import CaptureMethod
 from _pytask.capture_utils import ShowCapture
-from _pytask.cli import cli
 from _pytask.click import ColoredCommand
 from _pytask.click import ColoredGroup
 from _pytask.click import EnumChoice
@@ -76,6 +75,10 @@
 from _pytask.warnings_utils import warning_record_to_str
 from _pytask.warnings_utils import WarningReport
 
+# _pytask.cli needs to be imported last because it triggers extending the cli and
+# therefore loading plugins which will attempt to import modules that might only be
+# partially initialized. Maybe not here, but definitely for plugins.
+from _pytask.cli import cli  # noreorder
 
 __all__ = [
     "BaseTable",
diff --git a/tests/conftest.py b/tests/conftest.py
index ee77081b..6b0ac497 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import os
 import re
 import sys
 from contextlib import contextmanager
@@ -7,10 +8,11 @@
 from typing import Any
 
 import pytest
-from _pytask.pluginmanager import storage
 from click.testing import CliRunner
+from nbmake.pytest_items import NotebookItem
 from packaging import version
 from pytask import console
+from pytask import storage
 
 
 @pytest.fixture(autouse=True)
@@ -106,3 +108,11 @@ def invoke(self, *args, **kwargs):
 @pytest.fixture()
 def runner():
     return CustomCliRunner()
+
+
+def pytest_collection_modifyitems(session, config, items) -> None:  # noqa: ARG001
+    """Add markers to Jupyter notebook tests."""
+    if sys.platform == "debian" and "CI" in os.environ:  # pragma: no cover
+        for item in items:
+            if isinstance(item, NotebookItem):
+                item.add_marker(pytest.mark.xfail(reason="Fails regularly on MacOS"))
diff --git a/tests/test_collect.py b/tests/test_collect.py
index b6cbf122..6d01eec6 100644
--- a/tests/test_collect.py
+++ b/tests/test_collect.py
@@ -8,12 +8,12 @@
 import pytest
 from _pytask.collect import _find_shortest_uniquely_identifiable_name_for_tasks
 from _pytask.collect import pytask_collect_node
-from _pytask.exceptions import NodeNotCollectedError
-from _pytask.models import NodeInfo
 from pytask import build
 from pytask import cli
 from pytask import CollectionOutcome
 from pytask import ExitCode
+from pytask import NodeInfo
+from pytask import NodeNotCollectedError
 from pytask import Session
 from pytask import Task
 
diff --git a/tests/test_collect_command.py b/tests/test_collect_command.py
index 0af0c229..94788da8 100644
--- a/tests/test_collect_command.py
+++ b/tests/test_collect_command.py
@@ -9,10 +9,10 @@
 import pytest
 from _pytask.collect_command import _find_common_ancestor_of_all_nodes
 from _pytask.collect_command import _print_collected_tasks
-from _pytask.nodes import PathNode
 from attrs import define
 from pytask import cli
 from pytask import ExitCode
+from pytask import PathNode
 from pytask import Task
 
 
diff --git a/tests/test_collect_utils.py b/tests/test_collect_utils.py
index 4969097f..55ee009d 100644
--- a/tests/test_collect_utils.py
+++ b/tests/test_collect_utils.py
@@ -12,9 +12,9 @@
 from _pytask.collect_utils import _find_args_with_product_annotation
 from _pytask.collect_utils import _merge_dictionaries
 from _pytask.collect_utils import _Placeholder
-from _pytask.typing import Product  # noqa: TCH002
 from pytask import depends_on
 from pytask import produces
+from pytask import Product
 from typing_extensions import Annotated
 
 
diff --git a/tests/test_database.py b/tests/test_database.py
index 5c4fd5ca..1639360c 100644
--- a/tests/test_database.py
+++ b/tests/test_database.py
@@ -3,12 +3,12 @@
 import textwrap
 
 import pytest
-from _pytask.database_utils import create_database
-from _pytask.database_utils import DatabaseSession
-from _pytask.database_utils import State
 from pytask import build
 from pytask import cli
+from pytask import create_database
+from pytask import DatabaseSession
 from pytask import ExitCode
+from pytask import State
 from pytask.path import hash_path
 from sqlalchemy.engine import make_url
 
diff --git a/tests/test_execute.py b/tests/test_execute.py
index d9774dfb..ea8f0819 100644
--- a/tests/test_execute.py
+++ b/tests/test_execute.py
@@ -683,7 +683,7 @@ def task_example() -> Annotated[Dict[str, str], nodes]:
 @pytest.mark.end_to_end()
 def test_execute_tasks_and_pass_values_only_by_python_nodes(runner, tmp_path):
     source = """
-    from _pytask.nodes import PathNode
+    from pytask import PathNode
     from pytask import PythonNode
     from typing_extensions import Annotated
     from pathlib import Path
diff --git a/tests/test_hook_module.py b/tests/test_hook_module.py
index 24784c1d..40a712d9 100644
--- a/tests/test_hook_module.py
+++ b/tests/test_hook_module.py
@@ -9,6 +9,7 @@
 from pytask import ExitCode
 
 
+@pytest.mark.end_to_end()
 @pytest.mark.parametrize(
     "module_name",
     [
@@ -55,6 +56,7 @@ def pytask_extend_command_line_interface(cli):
     assert "--new-option" in result.stdout.decode()
 
 
+@pytest.mark.end_to_end()
 @pytest.mark.parametrize(
     "module_name",
     [
@@ -100,6 +102,7 @@ def pytask_extend_command_line_interface(cli):
     assert "--new-option" in result.stdout.decode()
 
 
+@pytest.mark.end_to_end()
 def test_error_when_hook_module_path_does_not_exist(tmp_path):
     result = subprocess.run(  # noqa: PLW1510
         ("pytask", "build", "--hook-module", "hooks.py", "--help"),
@@ -110,6 +113,7 @@ def test_error_when_hook_module_path_does_not_exist(tmp_path):
     assert b"Error: Invalid value for '--hook-module'" in result.stderr
 
 
+@pytest.mark.end_to_end()
 def test_error_when_hook_module_module_does_not_exist(tmp_path):
     result = subprocess.run(  # noqa: PLW1510
         ("pytask", "build", "--hook-module", "hooks", "--help"),
@@ -120,6 +124,7 @@ def test_error_when_hook_module_module_does_not_exist(tmp_path):
     assert b"Error: Invalid value for '--hook-module':" in result.stderr
 
 
+@pytest.mark.end_to_end()
 def test_error_when_hook_module_is_no_iterable(tmp_path):
     tmp_path.joinpath("pyproject.toml").write_text(
         "[tool.pytask.ini_options]\nhook_module = 'hooks'"
diff --git a/tests/test_live.py b/tests/test_live.py
index 0058691a..468ce36b 100644
--- a/tests/test_live.py
+++ b/tests/test_live.py
@@ -6,8 +6,8 @@
 import pytest
 from _pytask.live import LiveExecution
 from _pytask.live import LiveManager
-from _pytask.reports import ExecutionReport
 from pytask import cli
+from pytask import ExecutionReport
 from pytask import ExitCode
 from pytask import Task
 from pytask import TaskOutcome
diff --git a/tests/test_logging.py b/tests/test_logging.py
index e14e90f4..972716ff 100644
--- a/tests/test_logging.py
+++ b/tests/test_logging.py
@@ -8,9 +8,9 @@
 from _pytask.logging import _format_plugin_names_and_versions
 from _pytask.logging import _humanize_time
 from _pytask.logging import pytask_log_session_footer
-from _pytask.outcomes import ExitCode
-from _pytask.outcomes import TaskOutcome
 from pytask import cli
+from pytask import ExitCode
+from pytask import TaskOutcome
 
 
 class DummyDist(NamedTuple):
diff --git a/tests/test_mark_utils.py b/tests/test_mark_utils.py
index 0050acea..fdcbc2a2 100644
--- a/tests/test_mark_utils.py
+++ b/tests/test_mark_utils.py
@@ -4,7 +4,7 @@
 
 import pytask
 import pytest
-from _pytask.models import CollectionMetadata
+from pytask import CollectionMetadata
 from pytask import get_all_marks
 from pytask import get_marks
 from pytask import has_mark
diff --git a/tests/test_outcomes.py b/tests/test_outcomes.py
index cda4325e..5253ca21 100644
--- a/tests/test_outcomes.py
+++ b/tests/test_outcomes.py
@@ -1,10 +1,10 @@
 from __future__ import annotations
 
 import pytest
-from _pytask.reports import CollectionReport
-from _pytask.reports import ExecutionReport
 from pytask import CollectionOutcome
+from pytask import CollectionReport
 from pytask import count_outcomes
+from pytask import ExecutionReport
 from pytask import TaskOutcome
 
 
diff --git a/tests/test_path.py b/tests/test_path.py
index 265f92b1..e7df4389 100644
--- a/tests/test_path.py
+++ b/tests/test_path.py
@@ -15,8 +15,8 @@
 from _pytask.path import find_case_sensitive_path
 from _pytask.path import find_closest_ancestor
 from _pytask.path import find_common_ancestor
-from _pytask.path import import_path
 from _pytask.path import relative_to
+from pytask.path import import_path
 
 
 @pytest.mark.unit()
diff --git a/tests/test_persist.py b/tests/test_persist.py
index 047b69ff..db57e2c8 100644
--- a/tests/test_persist.py
+++ b/tests/test_persist.py
@@ -3,8 +3,6 @@
 import textwrap
 
 import pytest
-from _pytask.database_utils import State
-from _pytask.path import hash_path
 from _pytask.persist import pytask_execute_task_process_report
 from pytask import build
 from pytask import create_database
@@ -12,7 +10,9 @@
 from pytask import ExitCode
 from pytask import Persisted
 from pytask import SkippedUnchanged
+from pytask import State
 from pytask import TaskOutcome
+from pytask.path import hash_path
 
 from tests.conftest import restore_sys_path_and_module_after_test_execution
 
diff --git a/tests/test_profile.py b/tests/test_profile.py
index 8dbafdd7..6c3bfcda 100644
--- a/tests/test_profile.py
+++ b/tests/test_profile.py
@@ -3,13 +3,13 @@
 import textwrap
 
 import pytest
-from _pytask.cli import cli
 from _pytask.profile import _to_human_readable_size
-from _pytask.profile import Runtime
 from pytask import build
+from pytask import cli
 from pytask import create_database
 from pytask import DatabaseSession
 from pytask import ExitCode
+from pytask import Runtime
 
 
 @pytest.mark.end_to_end()
diff --git a/tests/test_shared.py b/tests/test_shared.py
index 0547f742..22cd5cb2 100644
--- a/tests/test_shared.py
+++ b/tests/test_shared.py
@@ -4,10 +4,10 @@
 from contextlib import ExitStack as does_not_raise  # noqa: N813
 
 import pytest
-from _pytask.outcomes import ExitCode
 from _pytask.shared import convert_to_enum
 from _pytask.shared import find_duplicates
 from pytask import build
+from pytask import ExitCode
 from pytask import ShowCapture
 
 
diff --git a/tests/test_task.py b/tests/test_task.py
index b1086198..e4b00014 100644
--- a/tests/test_task.py
+++ b/tests/test_task.py
@@ -708,7 +708,7 @@ def func():
     source = """
     from pytask import task
     from pathlib import Path
-    from _pytask.path import import_path
+    from pytask.path import import_path
     import inspect
 
     _ROOT_PATH = Path(__file__).parent
diff --git a/tests/test_tree_util.py b/tests/test_tree_util.py
index 99d010a5..8c93cb15 100644
--- a/tests/test_tree_util.py
+++ b/tests/test_tree_util.py
@@ -4,11 +4,11 @@
 import textwrap
 
 import pytest
-from _pytask.outcomes import ExitCode
-from _pytask.tree_util import tree_map
-from _pytask.tree_util import tree_structure
 from pytask import build
 from pytask import cli
+from pytask import ExitCode
+from pytask.tree_util import tree_map
+from pytask.tree_util import tree_structure
 
 
 @pytest.mark.end_to_end()
@@ -52,7 +52,7 @@ def test_profile_with_pytree(tmp_path, runner):
     source = """
     import time
     import pytask
-    from _pytask.tree_util import tree_leaves
+    from pytask.tree_util import tree_leaves
 
     @pytask.mark.produces([{"out_1": "out_1.txt"}, {"out_2": "out_2.txt"}])
     def task_example(produces):
diff --git a/tests/test_typing.py b/tests/test_typing.py
index eb60d461..1a084c7f 100644
--- a/tests/test_typing.py
+++ b/tests/test_typing.py
@@ -3,7 +3,7 @@
 import functools
 
 import pytest
-from _pytask.typing import is_task_function
+from pytask import is_task_function
 
 
 @pytest.mark.unit()