Idiomatic package tidy-ups: empty grouping __init__, _-prefixed test helpers

- Empty the four layer/grouping __init__.py (core, ui, app, app/transcribe) to
  match the agent/tts/streaming precedent — a no-op `from __future__` import in
  an otherwise-empty package init isn't meaningful.
- Rename tests/setup_helpers.py -> _setup_helpers.py and
  tests/replay_fixtures.py -> _replay_fixtures.py so every non-test support
  module in tests/ shares the underscore-prefix convention (pytest won't collect
  them as test modules); update the three importers and the two doc references.

No behavior change.

Author: claude

aai_cli/app/__init__.py

@@ -1 +0,0 @@
-from __future__ import annotations

aai_cli/app/transcribe/__init__.py

@@ -1 +0,0 @@
-from __future__ import annotations

aai_cli/core/__init__.py

@@ -1 +0,0 @@
-from __future__ import annotations

aai_cli/ui/__init__.py

@@ -1 +0,0 @@
-from __future__ import annotations

tests/AGENTS.md

@@ -102,7 +102,7 @@ these live in the default suite. Three moving parts:
   Refresh after an API shape change: `ASSEMBLYAI_API_KEY=… uv run python scripts/record_fixtures.py`.
   The key comes from the env; the AMS session JWT + `account_id` from the keyring/`config.toml`
   of whoever ran `assembly login` (profile `default`) — neither is ever written to a fixture.
-- **`tests/replay_fixtures.py`** — rebuilds the boundary objects from JSON. A transcript is a
+- **`tests/_replay_fixtures.py`** — rebuilds the boundary objects from JSON. A transcript is a
   real `aai.Transcript` via `Transcript.from_response`; an LLM response is rebuilt with
   `ChatCompletion.model_construct` (**not** `model_validate`) because the gateway returns
   Anthropic-flavored fields — `finish_reason="end_turn"`, token counts under

tests/replay_fixtures.py tests/_replay_fixtures.pytests/replay_fixtures.py renamed to tests/_replay_fixtures.py

@@ -1,7 +1,7 @@
 """End-to-end replay tests: drive real CLI commands against recorded API responses.
 
 Each test patches the command's network boundary (``client.* / llm.* / ams.*``) to
-return an object rebuilt from a real, scrubbed fixture (see ``tests/replay_fixtures.py``
+return an object rebuilt from a real, scrubbed fixture (see ``tests/_replay_fixtures.py``
 and ``scripts/record_fixtures.py``), then invokes the command through Typer and asserts
 on the rendered output. The transport stays offline — pytest-socket is untouched — but
 the command's own parsing, formatting, and rendering all run against a real payload.
@@ -13,7 +13,7 @@
 
 from aai_cli.core import config
 from aai_cli.main import app
-from tests import replay_fixtures as rf
+from tests import _replay_fixtures as rf
 
 runner = CliRunner()
 

tests/setup_helpers.py tests/_setup_helpers.pytests/setup_helpers.py renamed to tests/_setup_helpers.py

@@ -5,7 +5,13 @@
 from typer.testing import CliRunner
 
 from aai_cli.main import app
-from tests.setup_helpers import FakeRun, _all_tools_present, _cli_skill_path, _skill_path, _statuses
+from tests._setup_helpers import (
+    FakeRun,
+    _all_tools_present,
+    _cli_skill_path,
+    _skill_path,
+    _statuses,
+)
 
 runner = CliRunner()
 

tests/test_replay_e2e.py

@@ -5,7 +5,7 @@
 from typer.testing import CliRunner
 
 from aai_cli.main import app
-from tests.setup_helpers import (
+from tests._setup_helpers import (
     FakeRun,
     _all_tools_present,
     _cli_skill_path,