v2.10.0 (#697)

Co-authored-by: Clément Drouin <clement.drouin@mistral.ai>
Co-authored-by: Corentin André <corentin.andre@mistral.ai>
Co-authored-by: Guillaume LE GOFF <guillaume.lgf@gmail.com>
Co-authored-by: Kim-Adeline Miguel <51720070+kimadeline@users.noreply.github.com>
Co-authored-by: Maxime Dolores <maxime.dolores@ext.mistral.ai>
Co-authored-by: Nelson PROIA <144663685+Nelson-PROIA@users.noreply.github.com>
Co-authored-by: Peter Evers <pevers90@gmail.com>
Co-authored-by: Pierre Rossinès <pierre.rossines@mistral.ai>
Co-authored-by: Quentin <quentin.torroba@mistral.ai>
Co-authored-by: Vincent G <10739306+VinceOPS@users.noreply.github.com>
Co-authored-by: MichisGitIsKing <MichisGitIsKing@users.noreply.github.com>
Co-authored-by: Mistral Vibe <vibe@mistral.ai>

tests/acp/test_initialize.py

@@ -34,7 +34,7 @@ class TestACPInitialize:
             ),
         )
         assert response.agent_info == Implementation(
-            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.9.6"
+            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.10.0"
         )
 
         assert response.auth_methods == []
@@ -62,7 +62,7 @@ class TestACPInitialize:
             ),
         )
         assert response.agent_info == Implementation(
-            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.9.6"
+            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.10.0"
         )
 
         assert response.auth_methods is not None

tests/acp/test_new_session.py

@@ -47,12 +47,6 @@ class TestACPNewSession:
             cwd=str(Path.cwd()), mcp_servers=[]
         )
 
-        new_session_events = [
-            e for e in telemetry_events if e.get("event_name") == "vibe.new_session"
-        ]
-        assert len(new_session_events) == 1
-        assert new_session_events[0]["properties"]["entrypoint"] == "acp"
-
         assert session_response.session_id is not None
         acp_session = next(
             (
@@ -63,6 +57,17 @@ class TestACPNewSession:
             None,
         )
         assert acp_session is not None
+
+        # Telemetry now fires from the background warm-up worker once
+        # `wait_until_ready` joins both MCP and experiments. Awaiting it here
+        # forces emission before assertions.
+        await acp_session.agent_loop.wait_until_ready()
+
+        new_session_events = [
+            e for e in telemetry_events if e.get("event_name") == "vibe.new_session"
+        ]
+        assert len(new_session_events) == 1
+        assert new_session_events[0]["properties"]["entrypoint"] == "acp"
         assert (
             acp_session.agent_loop.session_logger.session_id
             == session_response.session_id

tests/acp/test_telemetry_notification.py

@@ -74,6 +74,31 @@ class TestTelemetryNotification:
         assert props["file_extensions"] == {".py": 1}
         assert props["message_id"] == "msg-abc"
 
+    @pytest.mark.asyncio
+    async def test_user_rating_feedback_dispatches_telemetry(
+        self, acp_agent_loop: VibeAcpAgentLoop, telemetry_events: list[dict[str, Any]]
+    ) -> None:
+        session = await acp_agent_loop.new_session(cwd=str(Path.cwd()), mcp_servers=[])
+        telemetry_events.clear()
+
+        await acp_agent_loop.ext_notification(
+            "telemetry/send",
+            {
+                "event": "vibe.user_rating_feedback",
+                "session_id": session.session_id,
+                "properties": {"rating": 1},
+            },
+        )
+
+        rating_events = [
+            e
+            for e in telemetry_events
+            if e["event_name"] == "vibe.user_rating_feedback"
+        ]
+        assert len(rating_events) == 1
+        props = rating_events[0]["properties"]
+        assert props["rating"] == 1
+
     @pytest.mark.asyncio
     async def test_raises_on_invalid_params(
         self, acp_agent_loop: VibeAcpAgentLoop

tests/backend/test_openai_responses_adapter.py

@@ -139,7 +139,7 @@ class TestPrepareRequest:
             {"role": "user", "content": "Hi"},
         ]
 
-    def test_consecutive_user_messages_are_merged(self, adapter, provider):
+    def test_consecutive_user_messages_are_preserved(self, adapter, provider):
         payload = _prepare(
             adapter,
             provider,
@@ -148,7 +148,10 @@ class TestPrepareRequest:
                 LLMMessage(role=Role.user, content="Again"),
             ],
         )
-        assert payload["input"] == [{"role": "user", "content": "Hi\n\nAgain"}]
+        assert payload["input"] == [
+            {"role": "user", "content": "Hi"},
+            {"role": "user", "content": "Again"},
+        ]
 
     def test_multiple_system_messages_are_preserved(self, adapter, provider):
         payload = _prepare(

tests/browser_sign_in/stubs.py

@@ -1,7 +1,8 @@
 from __future__ import annotations
 
+from collections.abc import Awaitable, Callable
 from dataclasses import dataclass
-from datetime import datetime, timedelta
+from datetime import UTC, datetime, timedelta
 from urllib.parse import urlencode
 
 from vibe.setup.auth import (
@@ -11,6 +12,7 @@ from vibe.setup.auth import (
     BrowserSignInPollResult,
     BrowserSignInProcess,
 )
+from vibe.setup.auth.browser_sign_in import BrowserSignInService
 
 
 @dataclass
@@ -42,6 +44,7 @@ class StubBrowserSignInGateway(BrowserSignInGateway):
         self.polled_urls: list[str] = []
         self.exchange_requests: list[ExchangeRequestPayload] = []
         self.closed = False
+        self.poll_calls = 0
         self.process_number = 0
 
     async def create_process(self, code_challenge: str) -> BrowserSignInProcess:
@@ -55,10 +58,10 @@ class StubBrowserSignInGateway(BrowserSignInGateway):
 
     async def poll(self, poll_url: str) -> BrowserSignInPollResult:
         self.polled_urls.append(poll_url)
+        self.poll_calls += 1
         if not self._poll_results:
             msg = "StubBrowserSignInGateway requires scripted poll results."
             raise AssertionError(msg)
-
         result = self._poll_results.pop(0)
         if isinstance(result, BrowserSignInError):
             raise result
@@ -109,5 +112,80 @@ def build_poll_failed_error() -> BrowserSignInError:
     )
 
 
+def build_poll_results_from_outcomes(
+    outcomes: list[str],
+) -> tuple[
+    list[BrowserSignInProcess], list[BrowserSignInPollResult | BrowserSignInError]
+]:
+    processes: list[BrowserSignInProcess] = []
+    poll_results: list[BrowserSignInPollResult | BrowserSignInError] = []
+    now = datetime(2026, 3, 16, tzinfo=UTC)
+
+    for process_index, outcome in enumerate(outcomes, start=1):
+        process_id = f"process-{process_index}"
+        processes.append(build_sign_in_process(now, process_id=process_id))
+
+        match outcome:
+            case "completed":
+                poll_results.extend([
+                    BrowserSignInPollResult(status="pending"),
+                    BrowserSignInPollResult(
+                        status="completed", exchange_token=f"exchange-{process_id}"
+                    ),
+                ])
+            case "expired":
+                poll_results.append(BrowserSignInPollResult(status="expired"))
+            case "poll_failed":
+                poll_results.extend([
+                    BrowserSignInPollResult(status="pending"),
+                    build_poll_failed_error(),
+                    build_poll_failed_error(),
+                    build_poll_failed_error(),
+                ])
+            case _:
+                msg = f"Unsupported browser sign-in outcome: {outcome}"
+                raise AssertionError(msg)
+
+    return processes, poll_results
+
+
 async def noop_sleep(_: float) -> None:
     return None
+
+
+def build_browser_sign_in_service_factory(
+    outcomes: list[str],
+    *,
+    exchange_result: str = "sk-browser-onboarding-test-key",
+    open_browser: Callable[[str], bool] | None = None,
+    sleep: Callable[[float], Awaitable[None]] = noop_sleep,
+    now: Callable[[], datetime] | None = None,
+) -> tuple[
+    StubBrowserSignInGateway,
+    Callable[[], BrowserSignInService],
+    list[BrowserSignInService],
+]:
+    processes, poll_results = build_poll_results_from_outcomes(outcomes)
+    gateway = StubBrowserSignInGateway(
+        processes=processes, poll_results=poll_results, exchange_result=exchange_result
+    )
+    created_services: list[BrowserSignInService] = []
+
+    def build_service() -> BrowserSignInService:
+        service = BrowserSignInService(
+            gateway,
+            open_browser=open_browser or (lambda _: True),
+            sleep=sleep,
+            now=now
+            or (
+                lambda: (
+                    datetime(2026, 3, 16, tzinfo=UTC)
+                    + timedelta(seconds=gateway.poll_calls)
+                )
+            ),
+            poll_interval=0,
+        )
+        created_services.append(service)
+        return service
+
+    return gateway, build_service, created_services

tests/browser_sign_in/test_browser_sign_in.py

@@ -23,6 +23,7 @@ from vibe.setup.auth import (
     BrowserSignInPollResult,
     BrowserSignInProcess,
     BrowserSignInService,
+    BrowserSignInStatus,
 )
 
 TEST_NOW = datetime(2026, 3, 16, tzinfo=UTC)
@@ -68,7 +69,7 @@ def build_test_service(
 @pytest.mark.asyncio
 async def test_authenticate_returns_api_key_after_pending_poll() -> None:
     opened_urls: list[str] = []
-    statuses: list[str] = []
+    statuses: list[BrowserSignInStatus] = []
     gateway, service = build_test_service(
         poll_results=[
             BrowserSignInPollResult(status="pending"),
@@ -84,10 +85,10 @@ async def test_authenticate_returns_api_key_after_pending_poll() -> None:
     assert api_key == "sk-browser-key"
     assert opened_urls == [TEST_SIGN_IN_URL]
     assert statuses == [
-        "opening_browser",
-        "waiting_for_browser_sign_in",
-        "exchanging",
-        "completed",
+        BrowserSignInStatus.OPENING_BROWSER,
+        BrowserSignInStatus.WAITING_FOR_BROWSER_SIGN_IN,
+        BrowserSignInStatus.EXCHANGING,
+        BrowserSignInStatus.COMPLETED,
     ]
     assert gateway.polled_urls == [TEST_POLL_URL, TEST_POLL_URL]
     assert gateway.exchange_requests[0].exchange_token == "exchange-1"

tests/cli/plan_offer/test_http_whoami_gateway.py

@@ -4,6 +4,7 @@ import httpx
 import pytest
 import respx
 
+from tests.conftest import build_test_vibe_config
 from vibe.cli.plan_offer.adapters.http_whoami_gateway import HttpWhoAmIGateway
 from vibe.cli.plan_offer.ports.whoami_gateway import (
     WhoAmIGatewayError,
@@ -11,6 +12,7 @@ from vibe.cli.plan_offer.ports.whoami_gateway import (
     WhoAmIPlanType,
     WhoAmIResponse,
 )
+from vibe.core.config import DEFAULT_CONSOLE_BASE_URL
 
 
 @pytest.mark.asyncio
@@ -170,3 +172,53 @@ async def test_return_unknown_plan_on_unsupported_plan_type(
         plan_name="INDIVIDUAL",
         prompt_switching_to_pro_plan=False,
     )
+
+
+@pytest.mark.asyncio
+async def test_gateway_calls_custom_console_base_url_from_config(
+    respx_mock: respx.MockRouter,
+) -> None:
+    custom_url = "https://custom-console.example.com"
+    config = build_test_vibe_config(console_base_url=custom_url)
+
+    route = respx_mock.get(f"{custom_url}/api/vibe/whoami").mock(
+        return_value=httpx.Response(
+            200,
+            json={
+                "plan_type": "CHAT",
+                "plan_name": "INDIVIDUAL",
+                "prompt_switching_to_pro_plan": False,
+            },
+        )
+    )
+
+    gateway = HttpWhoAmIGateway(base_url=config.console_base_url)
+    response = await gateway.whoami("api-key")
+
+    assert route.called
+    assert response.plan_type == "CHAT"
+
+
+@pytest.mark.asyncio
+async def test_gateway_uses_default_console_url_when_not_configured(
+    respx_mock: respx.MockRouter,
+) -> None:
+    config = build_test_vibe_config()
+
+    assert config.console_base_url == DEFAULT_CONSOLE_BASE_URL
+
+    route = respx_mock.get(f"{DEFAULT_CONSOLE_BASE_URL}/api/vibe/whoami").mock(
+        return_value=httpx.Response(
+            200,
+            json={
+                "plan_type": "CHAT",
+                "plan_name": "INDIVIDUAL",
+                "prompt_switching_to_pro_plan": False,
+            },
+        )
+    )
+
+    gateway = HttpWhoAmIGateway(base_url=config.console_base_url)
+    await gateway.whoami("api-key")
+
+    assert route.called

tests/cli/test_approval_app_grace_period.py

@@ -0,0 +1,78 @@
+from __future__ import annotations
+
+from unittest.mock import patch
+
+from pydantic import BaseModel
+import pytest
+
+from vibe.cli.textual_ui.widgets.approval_app import ApprovalApp
+from vibe.core.config import VibeConfig
+
+_TEST_GRACE_PERIOD_S = 0.5
+
+
+class FakeArgs(BaseModel):
+    command: str = "echo hello"
+
+
+@pytest.fixture
+def approval_app(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(
+        "vibe.cli.textual_ui.widgets.approval_app._INPUT_GRACE_PERIOD_S",
+        _TEST_GRACE_PERIOD_S,
+    )
+    config = VibeConfig()
+    app = ApprovalApp(tool_name="bash", tool_args=FakeArgs(), config=config)
+    app._mount_time = 100.0
+    return app
+
+
+class TestGracePeriod:
+    def test_actions_ignored_within_grace_period(self, approval_app: ApprovalApp):
+        with (
+            patch("vibe.cli.textual_ui.widgets.approval_app.time") as mock_time,
+            patch.object(approval_app, "post_message") as posted,
+        ):
+            mock_time.monotonic.return_value = 100.0 + _TEST_GRACE_PERIOD_S - 0.01
+            assert approval_app.is_within_grace_period()
+
+            approval_app.action_select()
+            approval_app.action_select_1()
+            approval_app.action_select_2()
+            approval_app.action_select_3()
+            approval_app.action_reject()
+
+            posted.assert_not_called()
+
+    def test_actions_post_messages_after_grace_period(self, approval_app: ApprovalApp):
+        with (
+            patch("vibe.cli.textual_ui.widgets.approval_app.time") as mock_time,
+            patch.object(approval_app, "post_message") as posted,
+        ):
+            mock_time.monotonic.return_value = 100.0 + _TEST_GRACE_PERIOD_S + 0.01
+            assert not approval_app.is_within_grace_period()
+
+            approval_app.action_select_1()
+            approval_app.action_reject()
+
+            assert posted.call_count == 2
+            assert isinstance(
+                posted.call_args_list[0].args[0], ApprovalApp.ApprovalGranted
+            )
+            assert isinstance(
+                posted.call_args_list[1].args[0], ApprovalApp.ApprovalRejected
+            )
+
+    def test_arrow_keys_work_during_grace_period(self, approval_app: ApprovalApp):
+        with (
+            patch("vibe.cli.textual_ui.widgets.approval_app.time") as mock_time,
+            patch.object(approval_app, "_update_options"),
+        ):
+            mock_time.monotonic.return_value = 100.0 + 0.01
+            assert approval_app.is_within_grace_period()
+
+            assert approval_app.selected_option == 0
+            approval_app.action_move_down()
+            assert approval_app.selected_option == 1
+            approval_app.action_move_up()
+            assert approval_app.selected_option == 0

tests/cli/test_programmatic_setup.py

@@ -21,6 +21,7 @@ def _make_args(**overrides: object) -> argparse.Namespace:
         "agent": "default",
         "setup": False,
         "workdir": None,
+        "add_dir": [],
         "trust": False,
         "teleport": False,
         "continue_session": False,

tests/cli/test_question_app_grace_period.py

@@ -0,0 +1,111 @@
+from __future__ import annotations
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+from textual import events
+
+from vibe.cli.textual_ui.widgets.question_app import QuestionApp
+from vibe.core.tools.builtins.ask_user_question import (
+    AskUserQuestionArgs,
+    Choice,
+    Question,
+)
+
+_TEST_GRACE_PERIOD_S = 0.5
+
+
+@pytest.fixture
+def question_app(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(
+        "vibe.cli.textual_ui.widgets.question_app._INPUT_GRACE_PERIOD_S",
+        _TEST_GRACE_PERIOD_S,
+    )
+    args = AskUserQuestionArgs(
+        questions=[
+            Question(
+                question="Pick one",
+                header="Pick",
+                options=[Choice(label="A"), Choice(label="B")],
+            )
+        ]
+    )
+    app = QuestionApp(args)
+    app._mount_time = 100.0
+    return app
+
+
+class TestQuestionAppGracePeriod:
+    def test_select_and_cancel_ignored_within_grace_period(
+        self, question_app: QuestionApp
+    ):
+        with (
+            patch("vibe.cli.textual_ui.widgets.question_app.time") as mock_time,
+            patch.object(question_app, "post_message") as posted,
+        ):
+            mock_time.monotonic.return_value = 100.0 + _TEST_GRACE_PERIOD_S - 0.01
+            assert question_app.is_within_grace_period()
+
+            question_app.action_select()
+            question_app.action_cancel()
+
+            posted.assert_not_called()
+
+    def test_cancel_posts_message_after_grace_period(self, question_app: QuestionApp):
+        with (
+            patch("vibe.cli.textual_ui.widgets.question_app.time") as mock_time,
+            patch.object(question_app, "post_message") as posted,
+        ):
+            mock_time.monotonic.return_value = 100.0 + _TEST_GRACE_PERIOD_S + 0.01
+
+            question_app.action_cancel()
+
+            posted.assert_called_once()
+            assert isinstance(posted.call_args.args[0], QuestionApp.Cancelled)
+
+    def test_navigation_works_during_grace_period(self, question_app: QuestionApp):
+        with patch("vibe.cli.textual_ui.widgets.question_app.time") as mock_time:
+            mock_time.monotonic.return_value = 100.0 + 0.01
+            assert question_app.is_within_grace_period()
+
+            assert question_app.selected_option == 0
+            question_app.action_move_down()
+            assert question_app.selected_option == 1
+            question_app.action_move_up()
+            assert question_app.selected_option == 0
+
+    def test_number_key_consumed_but_not_acted_within_grace_period(
+        self, question_app: QuestionApp
+    ):
+        with (
+            patch("vibe.cli.textual_ui.widgets.question_app.time") as mock_time,
+            patch.object(question_app, "post_message") as posted,
+        ):
+            mock_time.monotonic.return_value = 100.0 + _TEST_GRACE_PERIOD_S - 0.01
+            event = MagicMock(spec=events.Key)
+            event.character = "1"
+
+            handled = question_app._handle_number_key(event)
+
+            assert handled is True
+            event.stop.assert_called_once()
+            event.prevent_default.assert_called_once()
+            posted.assert_not_called()
+
+    def test_number_key_selects_option_after_grace_period(
+        self, question_app: QuestionApp
+    ):
+        with (
+            patch("vibe.cli.textual_ui.widgets.question_app.time") as mock_time,
+            patch.object(question_app, "post_message") as posted,
+        ):
+            mock_time.monotonic.return_value = 100.0 + _TEST_GRACE_PERIOD_S + 0.01
+            event = MagicMock(spec=events.Key)
+            event.character = "1"
+
+            handled = question_app._handle_number_key(event)
+
+            assert handled is True
+            assert question_app.selected_option == 0
+            posted.assert_called_once()
+            assert isinstance(posted.call_args.args[0], QuestionApp.Answered)

tests/cli/test_typing_debounce.py

@@ -0,0 +1,35 @@
+from __future__ import annotations
+
+import pytest
+
+from vibe.cli.textual_ui.app import _TYPING_DEBOUNCE_ENV_VAR, _resolve_typing_debounce_s
+
+_TEST_DEFAULT_DEBOUNCE_MS = 1000
+
+
+@pytest.fixture(autouse=True)
+def _restore_default_debounce(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(
+        "vibe.cli.textual_ui.app._DEFAULT_TYPING_DEBOUNCE_MS", _TEST_DEFAULT_DEBOUNCE_MS
+    )
+
+
+class TestTypingDebounceEnvVar:
+    @pytest.mark.parametrize(
+        ("env_value", "expected_s"), [("500", 0.5), ("2000", 2.0), ("0", 0.0)]
+    )
+    def test_env_var_override(
+        self, env_value: str, expected_s: float, monkeypatch: pytest.MonkeyPatch
+    ):
+        monkeypatch.setenv(_TYPING_DEBOUNCE_ENV_VAR, env_value)
+        assert _resolve_typing_debounce_s() == expected_s
+
+    @pytest.mark.parametrize("env_value", [None, "not-a-number", "-100"])
+    def test_falls_back_to_default(
+        self, env_value: str | None, monkeypatch: pytest.MonkeyPatch
+    ):
+        if env_value is None:
+            monkeypatch.delenv(_TYPING_DEBOUNCE_ENV_VAR, raising=False)
+        else:
+            monkeypatch.setenv(_TYPING_DEBOUNCE_ENV_VAR, env_value)
+        assert _resolve_typing_debounce_s() == _TEST_DEFAULT_DEBOUNCE_MS / 1000

tests/cli/test_ui_session_exit.py

@@ -3,13 +3,14 @@ from __future__ import annotations
 import pytest
 
 from vibe.cli.textual_ui.session_exit import print_session_resume_message
+from vibe.core.config import SessionLoggingConfig
 from vibe.core.types import AgentStats
 
 
 def test_print_session_resume_message_skips_output_without_session_id(
     capsys: pytest.CaptureFixture[str],
 ) -> None:
-    print_session_resume_message(None, AgentStats())
+    print_session_resume_message(None, AgentStats(), SessionLoggingConfig())
 
     assert capsys.readouterr().out == ""
 
@@ -20,6 +21,7 @@ def test_print_session_resume_message_prints_resume_commands_and_usage(
     print_session_resume_message(
         "12345678-1234-1234-1234-123456789abc",
         AgentStats(session_prompt_tokens=14_867, session_completion_tokens=6),
+        SessionLoggingConfig(),
     )
 
     assert capsys.readouterr().out == (
@@ -34,7 +36,7 @@ def test_print_session_resume_message_prints_resume_commands_and_usage(
 def test_print_session_resume_message_prints_zero_usage_for_resumed_run_without_llm_activity(
     capsys: pytest.CaptureFixture[str],
 ) -> None:
-    print_session_resume_message("12345678", AgentStats())
+    print_session_resume_message("12345678", AgentStats(), SessionLoggingConfig())
 
     assert capsys.readouterr().out == (
         "\n"

tests/conftest.py

@@ -164,6 +164,18 @@ def _disable_feedback_bar(monkeypatch: pytest.MonkeyPatch) -> None:
     )
 
 
+@pytest.fixture(autouse=True)
+def _disable_input_grace_periods(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(
+        "vibe.cli.textual_ui.widgets.approval_app._INPUT_GRACE_PERIOD_S", 0
+    )
+    monkeypatch.setattr(
+        "vibe.cli.textual_ui.widgets.question_app._INPUT_GRACE_PERIOD_S", 0
+    )
+    monkeypatch.setattr("vibe.cli.textual_ui.app._DEFAULT_TYPING_DEBOUNCE_MS", 0)
+    monkeypatch.delenv("VIBE_TYPING_GRACE_PERIOD_MS", raising=False)
+
+
 @pytest.fixture(autouse=True)
 def telemetry_events(monkeypatch: pytest.MonkeyPatch) -> list[dict[str, Any]]:
     events: list[dict[str, Any]] = []

tests/core/experiments/test_client.py

@@ -0,0 +1,116 @@
+from __future__ import annotations
+
+import logging
+
+import httpx
+import pytest
+import respx
+
+from vibe.core.experiments._constants import build_eval_url
+from vibe.core.experiments.active import ExperimentName
+from vibe.core.experiments.client import RemoteEvalClient
+from vibe.core.experiments.models import ExperimentAttributes
+
+_TEST_API_HOST = "https://growthbook.test"
+_TEST_CLIENT_KEY = "sdk-test"
+_TEST_EVAL_URL = build_eval_url(_TEST_API_HOST, _TEST_CLIENT_KEY)
+assert _TEST_EVAL_URL is not None
+
+
+def _attrs() -> ExperimentAttributes:
+    return ExperimentAttributes(
+        userId="hashed",
+        entrypoint="cli",
+        agent_version="1.2.3",
+        client_name="vibe",
+        client_version="1.2.3",
+        os="darwin",
+    )
+
+
+def _make_client() -> RemoteEvalClient:
+    return RemoteEvalClient.from_settings(_TEST_API_HOST, _TEST_CLIENT_KEY)
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_evaluate_happy_path() -> None:
+    payload = {
+        "features": {
+            ExperimentName.SYSTEM_PROMPT.value: {
+                "defaultValue": "cli",
+                "rules": [{"force": "cli_v2", "tracks": []}],
+            }
+        }
+    }
+    route = respx.post(_TEST_EVAL_URL).mock(
+        return_value=httpx.Response(200, json=payload)
+    )
+    client = _make_client()
+    response = await client.evaluate(_attrs())
+    await client.aclose()
+
+    assert route.called
+    request_body = route.calls.last.request.read()
+    assert b'"userId":"hashed"' in request_body
+    assert b'"forcedVariations":{}' in request_body
+    assert b'"forcedFeatures":[]' in request_body
+    assert response is not None
+    assert (
+        response.features[ExperimentName.SYSTEM_PROMPT.value].resolved_value()
+        == "cli_v2"
+    )
+
+
+@pytest.fixture
+def _silence_vibe_logger(caplog: pytest.LogCaptureFixture) -> None:
+    # Failure-path tests legitimately emit WARNING logs. Silence the vibe
+    # logger so they don't leak into ~/.vibe/logs/vibe.log (the file handler
+    # is bound to the real path at module import, before VIBE_HOME is
+    # monkeypatched in the conftest fixture).
+    caplog.set_level(logging.CRITICAL, logger="vibe")
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_evaluate_returns_none_on_5xx(_silence_vibe_logger: None) -> None:
+    respx.post(_TEST_EVAL_URL).mock(return_value=httpx.Response(500, text="oops"))
+    client = _make_client()
+    assert await client.evaluate(_attrs()) is None
+    await client.aclose()
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_evaluate_returns_none_on_network_error(
+    _silence_vibe_logger: None,
+) -> None:
+    respx.post(_TEST_EVAL_URL).mock(side_effect=httpx.ConnectError("nope"))
+    client = _make_client()
+    assert await client.evaluate(_attrs()) is None
+    await client.aclose()
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_evaluate_returns_none_on_invalid_json(
+    _silence_vibe_logger: None,
+) -> None:
+    respx.post(_TEST_EVAL_URL).mock(return_value=httpx.Response(200, text="not json"))
+    client = _make_client()
+    assert await client.evaluate(_attrs()) is None
+    await client.aclose()
+
+
+@pytest.mark.asyncio
+async def test_evaluate_skips_request_when_url_unset() -> None:
+    client = RemoteEvalClient.from_settings(api_host="", client_key="sdk-x")
+    assert await client.evaluate(_attrs()) is None
+    await client.aclose()
+
+
+@pytest.mark.asyncio
+async def test_aclose_is_idempotent() -> None:
+    client = _make_client()
+    await client.aclose()
+    await client.aclose()

tests/core/experiments/test_manager.py

@@ -0,0 +1,427 @@
+from __future__ import annotations
+
+from typing import Any
+
+import pytest
+
+from vibe.core.experiments.active import DEFAULT_VARIANTS, ExperimentName
+from vibe.core.experiments.client import RemoteEvalClient
+from vibe.core.experiments.manager import ExperimentManager, hash_api_key
+from vibe.core.experiments.models import EvalResponse, ExperimentAttributes
+
+
+class _StubClient(RemoteEvalClient):
+    def __init__(self, response: EvalResponse | None) -> None:
+        self._response = response
+        self.calls: list[ExperimentAttributes] = []
+
+    async def evaluate(self, attributes: ExperimentAttributes) -> EvalResponse | None:
+        self.calls.append(attributes)
+        return self._response
+
+    async def aclose(self) -> None:
+        pass
+
+
+def _attrs() -> ExperimentAttributes:
+    return ExperimentAttributes(
+        userId="x", entrypoint="cli", agent_version="0", os="darwin"
+    )
+
+
+def _response(features: dict[str, Any]) -> EvalResponse:
+    return EvalResponse.model_validate({"features": features})
+
+
+def test_hash_api_key_is_stable_and_anonymous() -> None:
+    a = hash_api_key("sk-abc")
+    b = hash_api_key("sk-abc")
+    assert a == b
+    assert "sk-" not in a
+    assert len(a) == 32
+
+
+def test_hash_api_key_differs_per_key() -> None:
+    assert hash_api_key("sk-abc") != hash_api_key("sk-def")
+
+
+@pytest.mark.asyncio
+async def test_get_variant_returns_default_when_uninitialized() -> None:
+    manager = ExperimentManager(client=_StubClient(None))
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "cli"
+
+
+@pytest.mark.asyncio
+async def test_get_variant_or_none_returns_none_when_unassigned() -> None:
+    manager = ExperimentManager(client=_StubClient(None))
+    assert manager.get_variant_or_none(ExperimentName.SYSTEM_PROMPT) is None
+
+
+@pytest.mark.asyncio
+async def test_get_variant_or_none_returns_override() -> None:
+    manager = ExperimentManager(
+        client=_StubClient(None), overrides={ExperimentName.SYSTEM_PROMPT.value: "lean"}
+    )
+    assert manager.get_variant_or_none(ExperimentName.SYSTEM_PROMPT) == "lean"
+
+
+@pytest.mark.asyncio
+async def test_get_variant_or_none_returns_resolved_value() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "explore", "tracks": []}],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assert manager.get_variant_or_none(ExperimentName.SYSTEM_PROMPT) == "explore"
+
+
+@pytest.mark.asyncio
+async def test_get_variant_returns_resolved_value() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "cli_v2", "tracks": []}],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "cli_v2"
+
+
+@pytest.mark.asyncio
+async def test_overrides_take_precedence_over_remote() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "cli_v2", "tracks": []}],
+        }
+    })
+    manager = ExperimentManager(
+        client=_StubClient(response),
+        overrides={ExperimentName.SYSTEM_PROMPT.value: "forced"},
+    )
+    await manager.initialize(_attrs())
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "forced"
+
+
+@pytest.mark.asyncio
+async def test_get_variant_falls_back_when_remote_returns_no_match() -> None:
+    response = _response({"some_other_feature": {"defaultValue": True}})
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "cli"
+
+
+@pytest.mark.asyncio
+async def test_assignments_uses_resolved_variant_value() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [
+                {
+                    "force": "cli_v2",
+                    "tracks": [
+                        {
+                            "experiment": {"key": ExperimentName.SYSTEM_PROMPT.value},
+                            "result": {
+                                "key": "1",
+                                "variationId": 1,
+                                "inExperiment": True,
+                            },
+                        }
+                    ],
+                }
+            ],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assignments = manager.assignments()
+    assert assignments == {ExperimentName.SYSTEM_PROMPT.value: "cli_v2"}
+
+
+@pytest.mark.asyncio
+async def test_assignments_prefers_track_result_value() -> None:
+    # When the GrowthBook payload carries an explicit per-arm value, it wins
+    # over the rule's force value — it's the most precise label for the user.
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [
+                {
+                    "force": "cli_v2",
+                    "tracks": [
+                        {
+                            "experiment": {"key": ExperimentName.SYSTEM_PROMPT.value},
+                            "result": {
+                                "key": "1",
+                                "variationId": 1,
+                                "value": "cli_v3",
+                                "inExperiment": True,
+                            },
+                        }
+                    ],
+                }
+            ],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assert manager.assignments() == {ExperimentName.SYSTEM_PROMPT.value: "cli_v3"}
+
+
+@pytest.mark.asyncio
+async def test_assignments_keys_on_feature_id_when_experiment_key_differs() -> None:
+    # GrowthBook feature ID and experiment key can diverge. Telemetry must
+    # key on the feature ID (matches ExperimentName) so overrides, lookups
+    # and downstream analysis stay consistent.
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [
+                {
+                    "force": "cli_v2",
+                    "tracks": [
+                        {
+                            "experiment": {"key": "vibe-code-cli-system-prompt"},
+                            "result": {
+                                "key": "1",
+                                "variationId": 1,
+                                "inExperiment": True,
+                                "featureId": ExperimentName.SYSTEM_PROMPT.value,
+                            },
+                        }
+                    ],
+                }
+            ],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assignments = manager.assignments()
+    assert assignments == {ExperimentName.SYSTEM_PROMPT.value: "cli_v2"}
+    assert "vibe-code-cli-system-prompt" not in assignments
+
+
+@pytest.mark.asyncio
+async def test_initialize_does_nothing_on_failed_eval() -> None:
+    manager = ExperimentManager(client=_StubClient(None))
+    await manager.initialize(_attrs())
+    assert manager.assignments() == {}
+    for name in ExperimentName:
+        assert manager.get_variant(name) == DEFAULT_VARIANTS[name]
+
+
+def test_experiment_attributes_default_custom_system_prompt_to_false() -> None:
+    attrs = ExperimentAttributes(
+        userId="x", entrypoint="cli", agent_version="0", os="darwin"
+    )
+    assert attrs.custom_system_prompt is False
+    assert attrs.model_dump(exclude_none=True)["custom_system_prompt"] is False
+
+
+def test_experiment_attributes_serializes_custom_system_prompt() -> None:
+    attrs = ExperimentAttributes(
+        userId="x",
+        entrypoint="cli",
+        agent_version="0",
+        os="darwin",
+        custom_system_prompt=True,
+    )
+    assert attrs.model_dump(exclude_none=True)["custom_system_prompt"] is True
+
+
+def test_export_state_returns_none_before_initialize() -> None:
+    manager = ExperimentManager(client=_StubClient(None))
+    assert manager.export_state() is None
+
+
+@pytest.mark.asyncio
+async def test_export_state_returns_response_after_initialize() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "cli_v2", "tracks": []}],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+    assert manager.export_state() == response
+
+
+def test_hydrate_does_not_call_client() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "cli_v2", "tracks": []}],
+        }
+    })
+    stub = _StubClient(response)
+    manager = ExperimentManager(client=stub)
+    manager.hydrate(response)
+    assert stub.calls == []
+
+
+def test_hydrate_makes_get_variant_match_initialized_manager() -> None:
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [
+                {
+                    "force": "cli_v2",
+                    "tracks": [
+                        {
+                            "experiment": {"key": ExperimentName.SYSTEM_PROMPT.value},
+                            "result": {
+                                "key": "1",
+                                "variationId": 1,
+                                "inExperiment": True,
+                            },
+                        }
+                    ],
+                }
+            ],
+        }
+    })
+    hydrated = ExperimentManager(client=_StubClient(None))
+    hydrated.hydrate(response)
+    assert hydrated.get_variant(ExperimentName.SYSTEM_PROMPT) == "cli_v2"
+    assert hydrated.assignments() == {ExperimentName.SYSTEM_PROMPT.value: "cli_v2"}
+
+
+@pytest.mark.asyncio
+async def test_initialize_twice_replaces_response() -> None:
+    # Session reset re-runs initialize_experiments, which calls manager.initialize
+    # again. The second response must replace the first so the new session sees
+    # the up-to-date variant assignment.
+    first = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "explore", "tracks": []}],
+        }
+    })
+    second = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [{"force": "lean", "tracks": []}],
+        }
+    })
+
+    class _ScriptedClient(RemoteEvalClient):
+        def __init__(self, responses: list[EvalResponse]) -> None:
+            self._responses = responses
+            self.call_count = 0
+
+        async def evaluate(
+            self, attributes: ExperimentAttributes
+        ) -> EvalResponse | None:
+            response = self._responses[self.call_count]
+            self.call_count += 1
+            return response
+
+        async def aclose(self) -> None:
+            pass
+
+    client = _ScriptedClient([first, second])
+    manager = ExperimentManager(client=client)
+    await manager.initialize(_attrs())
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "explore"
+    await manager.initialize(_attrs())
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "lean"
+    assert client.call_count == 2
+
+
+@pytest.mark.asyncio
+async def test_assignments_excludes_tracks_not_in_experiment() -> None:
+    # Holdouts and forced overrides come back with inExperiment=False.
+    # We must NOT report them as active experiment participants — that would
+    # pollute downstream A/B analysis since the assignment map IS the
+    # exposure record (no separate exposure event).
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [
+                {
+                    "force": "cli",
+                    "tracks": [
+                        {
+                            "experiment": {"key": ExperimentName.SYSTEM_PROMPT.value},
+                            "result": {
+                                "key": "0",
+                                "variationId": 0,
+                                "inExperiment": False,
+                            },
+                        }
+                    ],
+                }
+            ],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+
+    assert manager.assignments() == {}
+
+
+@pytest.mark.asyncio
+async def test_assignments_excludes_tracks_with_missing_in_experiment() -> None:
+    # Defensive: if the proxy returns a track without inExperiment, treat it
+    # as "not confirmed in the experiment" rather than silently counting it.
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {
+            "defaultValue": "cli",
+            "rules": [
+                {
+                    "force": "cli_v2",
+                    "tracks": [
+                        {
+                            "experiment": {"key": ExperimentName.SYSTEM_PROMPT.value},
+                            "result": {"key": "1", "variationId": 1},
+                        }
+                    ],
+                }
+            ],
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+
+    assert manager.assignments() == {}
+
+
+@pytest.mark.asyncio
+async def test_initialize_drops_unknown_features() -> None:
+    # GrowthBook returns every feature defined in the org. Only the ones
+    # listed in ExperimentName should survive into manager state and the
+    # persisted snapshot.
+    response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {"defaultValue": "cli"},
+        "unrelated_org_flag": {"defaultValue": True},
+        "neko": {"defaultValue": False},
+        "gbdemo-checkout-layout": {"defaultValue": "dev"},
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(_attrs())
+
+    snapshot = manager.export_state()
+    assert snapshot is not None
+    assert set(snapshot.features.keys()) == {ExperimentName.SYSTEM_PROMPT.value}
+
+
+def test_hydrate_drops_unknown_features() -> None:
+    # When resuming a session saved by an older vibe version, the snapshot
+    # may contain experiments that have since been retired. Filter them on
+    # hydrate so the in-memory state matches the current ExperimentName.
+    legacy_response = _response({
+        ExperimentName.SYSTEM_PROMPT.value: {"defaultValue": "cli"},
+        "retired_experiment": {"defaultValue": "control"},
+    })
+    manager = ExperimentManager(client=_StubClient(None))
+    manager.hydrate(legacy_response)
+
+    snapshot = manager.export_state()
+    assert snapshot is not None
+    assert set(snapshot.features.keys()) == {ExperimentName.SYSTEM_PROMPT.value}

tests/core/experiments/test_models.py

@@ -0,0 +1,68 @@
+from __future__ import annotations
+
+from vibe.core.experiments.active import ExperimentName
+from vibe.core.experiments.models import EvalResponse, FeatureDefinition, FeatureRule
+
+
+class TestFeatureDefinition:
+    def test_resolved_value_returns_force_when_present(self) -> None:
+        feature = FeatureDefinition(
+            defaultValue="cli", rules=[FeatureRule(force="cli_v2")]
+        )
+        assert feature.resolved_value() == "cli_v2"
+
+    def test_resolved_value_falls_back_to_default(self) -> None:
+        feature = FeatureDefinition(defaultValue="cli")
+        assert feature.resolved_value() == "cli"
+
+    def test_resolved_value_picks_first_rule_with_force(self) -> None:
+        feature = FeatureDefinition(
+            defaultValue="cli",
+            rules=[FeatureRule(force=None), FeatureRule(force="cli_v2")],
+        )
+        assert feature.resolved_value() == "cli_v2"
+
+
+class TestEvalResponse:
+    def test_parses_real_proxy_payload(self) -> None:
+        # A reduced version of the live response captured during dev probing.
+        sp_key = ExperimentName.SYSTEM_PROMPT.value
+        raw = {
+            "features": {
+                "rbac_enabled": {"defaultValue": True},
+                sp_key: {
+                    "defaultValue": "cli",
+                    "rules": [
+                        {
+                            "force": "cli_v2",
+                            "tracks": [
+                                {
+                                    "experiment": {
+                                        "key": sp_key,
+                                        "variations": [{}, "cli_v2"],
+                                    },
+                                    "result": {
+                                        "key": "1",
+                                        "variationId": 1,
+                                        "value": "cli_v2",
+                                        "inExperiment": True,
+                                        "hashAttribute": "id",
+                                        "hashValue": "abc",
+                                    },
+                                }
+                            ],
+                        }
+                    ],
+                },
+            },
+            "experiments": [],
+            "dateUpdated": "2026-05-11T13:29:48.394Z",
+        }
+        response = EvalResponse.model_validate(raw)
+        assert response.features["rbac_enabled"].defaultValue is True
+        sp = response.features[sp_key]
+        assert sp.resolved_value() == "cli_v2"
+        track = sp.rules[0].tracks[0]
+        assert track.experiment.key == sp_key
+        assert track.result.variationId == 1
+        assert track.result.inExperiment is True

tests/core/experiments/test_resume.py

@@ -0,0 +1,168 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+
+from tests.conftest import build_test_vibe_config
+from vibe.core.agents import AgentManager
+from vibe.core.experiments.active import ExperimentName
+from vibe.core.experiments.client import RemoteEvalClient
+from vibe.core.experiments.manager import ExperimentManager
+from vibe.core.experiments.models import EvalResponse, ExperimentAttributes
+from vibe.core.session.session_loader import SessionLoader
+from vibe.core.skills.manager import SkillManager
+from vibe.core.system_prompt import get_universal_system_prompt
+from vibe.core.tools.manager import ToolManager
+from vibe.core.types import SessionMetadata
+
+
+class _StubClient(RemoteEvalClient):
+    def __init__(self, response: EvalResponse | None) -> None:
+        self._response = response
+        self.calls: list[ExperimentAttributes] = []
+
+    async def evaluate(self, attributes: ExperimentAttributes) -> EvalResponse | None:
+        self.calls.append(attributes)
+        return self._response
+
+    async def aclose(self) -> None:
+        pass
+
+
+def _attrs() -> ExperimentAttributes:
+    return ExperimentAttributes(
+        userId="x", entrypoint="cli", agent_version="0", os="darwin"
+    )
+
+
+def _response_forcing(variant: str) -> EvalResponse:
+    return EvalResponse.model_validate({
+        "features": {
+            ExperimentName.SYSTEM_PROMPT.value: {
+                "defaultValue": "cli",
+                "rules": [
+                    {
+                        "force": variant,
+                        "tracks": [
+                            {
+                                "experiment": {
+                                    "key": ExperimentName.SYSTEM_PROMPT.value
+                                },
+                                "result": {
+                                    "key": "1",
+                                    "variationId": 1,
+                                    "inExperiment": True,
+                                },
+                            }
+                        ],
+                    }
+                ],
+            }
+        }
+    })
+
+
+@pytest.mark.asyncio
+async def test_resume_round_trip_preserves_variant() -> None:
+    response = _response_forcing("explore")
+    fresh = ExperimentManager(client=_StubClient(response))
+    await fresh.initialize(_attrs())
+
+    persisted_state = fresh.export_state()
+    assert persisted_state is not None
+
+    serialized = persisted_state.model_dump_json()
+    rehydrated_response = EvalResponse.model_validate_json(serialized)
+
+    stub = _StubClient(None)
+    resumed = ExperimentManager(client=stub)
+    resumed.hydrate(rehydrated_response)
+
+    assert resumed.get_variant(ExperimentName.SYSTEM_PROMPT) == "explore"
+    assert resumed.assignments() == fresh.assignments()
+    assert stub.calls == []
+
+
+def test_resume_old_session_without_experiments_field_falls_back_to_defaults(
+    tmp_path: Path,
+) -> None:
+    legacy_metadata = {
+        "session_id": "legacy",
+        "start_time": "2025-01-01T00:00:00+00:00",
+        "end_time": None,
+        "git_commit": None,
+        "git_branch": None,
+        "environment": {"working_directory": str(tmp_path)},
+        "username": "tester",
+        "loops": [],
+    }
+    meta_path = tmp_path / "meta.json"
+    meta_path.write_text(json.dumps(legacy_metadata))
+
+    metadata = SessionLoader.load_metadata(tmp_path)
+
+    assert metadata.experiments is None
+
+
+def test_resume_overrides_still_win_over_hydrated_state() -> None:
+    response = _response_forcing("explore")
+    manager = ExperimentManager(
+        client=_StubClient(None), overrides={ExperimentName.SYSTEM_PROMPT.value: "lean"}
+    )
+    manager.hydrate(response)
+
+    assert manager.get_variant(ExperimentName.SYSTEM_PROMPT) == "lean"
+
+
+def test_session_metadata_round_trips_experiments_field() -> None:
+    response = _response_forcing("explore")
+    metadata = SessionMetadata(
+        session_id="s",
+        start_time="2025-01-01T00:00:00+00:00",
+        end_time=None,
+        git_commit=None,
+        git_branch=None,
+        environment={"working_directory": "/tmp"},
+        username="tester",
+        experiments=response,
+    )
+
+    serialized = metadata.model_dump_json()
+    restored = SessionMetadata.model_validate_json(serialized)
+
+    assert restored.experiments is not None
+    assert restored.experiments == response
+
+
+def _build_managers(config):
+    return (
+        ToolManager(lambda: config),
+        SkillManager(lambda: config),
+        AgentManager(lambda: config),
+    )
+
+
+@pytest.mark.asyncio
+async def test_graduated_experiment_with_deleted_variant_file_falls_back() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="cli",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    response = _response_forcing("removed_after_graduation_2025_07")
+    manager = ExperimentManager(client=_StubClient(None))
+    manager.hydrate(response)
+
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=manager
+    )
+
+    default_prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=None
+    )
+    assert prompt == default_prompt

tests/core/experiments/test_session_helpers.py

@@ -0,0 +1,208 @@
+from __future__ import annotations
+
+from typing import Any
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from vibe.core.experiments.client import RemoteEvalClient
+from vibe.core.experiments.manager import ExperimentManager
+from vibe.core.experiments.models import EvalResponse, ExperimentAttributes
+from vibe.core.experiments.session import (
+    hydrate_experiments_from_session,
+    initialize_experiments,
+)
+
+
+class _StubClient(RemoteEvalClient):
+    def __init__(self, response: EvalResponse | None) -> None:
+        self._response = response
+
+    async def evaluate(self, attributes: ExperimentAttributes) -> EvalResponse | None:
+        return self._response
+
+    async def aclose(self) -> None:
+        pass
+
+
+def _make_config(
+    *, enable_telemetry: bool = True, enable_experiments: bool = True
+) -> Any:
+    config = MagicMock()
+    config.enable_telemetry = enable_telemetry
+    config.experiments.enable = enable_experiments
+    return config
+
+
+@pytest.mark.asyncio
+async def test_initialize_returns_false_when_telemetry_disabled(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    persist = AsyncMock()
+    session_logger = MagicMock()
+    session_logger.persist_experiments = persist
+    manager = ExperimentManager(client=_StubClient(None))
+
+    result = await initialize_experiments(
+        config=_make_config(enable_telemetry=False),
+        manager=manager,
+        session_logger=session_logger,
+        entrypoint_metadata=None,
+    )
+
+    assert result is False
+    persist.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_initialize_returns_false_when_experiments_disabled(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    # `experiments.enable=False` must short-circuit even when telemetry is on
+    # — it is the user's opt-out for being assigned to A/B tests without
+    # disabling all telemetry.
+    persist = AsyncMock()
+    session_logger = MagicMock()
+    session_logger.persist_experiments = persist
+    manager = ExperimentManager(client=_StubClient(None))
+
+    result = await initialize_experiments(
+        config=_make_config(enable_experiments=False),
+        manager=manager,
+        session_logger=session_logger,
+        entrypoint_metadata=None,
+    )
+
+    assert result is False
+    persist.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_initialize_returns_false_when_no_mistral_provider(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        "vibe.core.experiments.session.get_mistral_provider_and_api_key",
+        lambda _config: None,
+    )
+    persist = AsyncMock()
+    session_logger = MagicMock()
+    session_logger.persist_experiments = persist
+    manager = ExperimentManager(client=_StubClient(None))
+
+    result = await initialize_experiments(
+        config=_make_config(),
+        manager=manager,
+        session_logger=session_logger,
+        entrypoint_metadata=None,
+    )
+
+    assert result is False
+    persist.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_initialize_returns_false_when_remote_eval_fails(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    # Regression: even if telemetry is enabled and a Mistral key is set,
+    # a failed remote eval (returns None) leaves manager state empty —
+    # the helper must NOT report success, so the caller skips the
+    # unnecessary system prompt refresh.
+    monkeypatch.setattr(
+        "vibe.core.experiments.session.get_mistral_provider_and_api_key",
+        lambda _config: (MagicMock(), "fake-key"),
+    )
+    monkeypatch.setattr(
+        "vibe.core.experiments.session._build_attributes",
+        lambda *_args, **_kwargs: ExperimentAttributes(
+            userId="x", entrypoint="cli", agent_version="0", os="darwin"
+        ),
+    )
+    persist = AsyncMock()
+    session_logger = MagicMock()
+    session_logger.persist_experiments = persist
+    manager = ExperimentManager(client=_StubClient(None))
+
+    result = await initialize_experiments(
+        config=_make_config(),
+        manager=manager,
+        session_logger=session_logger,
+        entrypoint_metadata=None,
+    )
+
+    assert result is False
+    persist.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_initialize_returns_true_and_persists_when_remote_eval_succeeds(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        "vibe.core.experiments.session.get_mistral_provider_and_api_key",
+        lambda _config: (MagicMock(), "fake-key"),
+    )
+    monkeypatch.setattr(
+        "vibe.core.experiments.session._build_attributes",
+        lambda *_args, **_kwargs: ExperimentAttributes(
+            userId="x", entrypoint="cli", agent_version="0", os="darwin"
+        ),
+    )
+    persist = AsyncMock()
+    session_logger = MagicMock()
+    session_logger.persist_experiments = persist
+    response = EvalResponse.model_validate({
+        "features": {"vibe_cli_system_prompt": {"defaultValue": "cli"}}
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+
+    result = await initialize_experiments(
+        config=_make_config(),
+        manager=manager,
+        session_logger=session_logger,
+        entrypoint_metadata=None,
+    )
+
+    assert result is True
+    persist.assert_awaited_once()
+
+
+@pytest.mark.asyncio
+async def test_hydrate_returns_false_when_telemetry_disabled() -> None:
+    session_logger = MagicMock()
+    response = EvalResponse.model_validate({
+        "features": {"vibe_cli_system_prompt": {"defaultValue": "cli"}}
+    })
+    session_logger.session_metadata.experiments = response
+    manager = ExperimentManager(client=_StubClient(None))
+
+    result = await hydrate_experiments_from_session(
+        config=_make_config(enable_telemetry=False),
+        manager=manager,
+        session_logger=session_logger,
+    )
+
+    assert result is False
+    assert manager.export_state() is None
+
+
+@pytest.mark.asyncio
+async def test_hydrate_returns_false_when_experiments_disabled() -> None:
+    # Regression: without this gate, a user who flipped experiments.enable
+    # to False between sessions would still resume into a hydrated variant.
+    session_logger = MagicMock()
+    response = EvalResponse.model_validate({
+        "features": {"vibe_cli_system_prompt": {"defaultValue": "cli"}}
+    })
+    session_logger.session_metadata.experiments = response
+    manager = ExperimentManager(client=_StubClient(None))
+
+    result = await hydrate_experiments_from_session(
+        config=_make_config(enable_experiments=False),
+        manager=manager,
+        session_logger=session_logger,
+    )
+
+    assert result is False
+    assert manager.export_state() is None

tests/core/experiments/test_system_prompt_variant.py

@@ -0,0 +1,186 @@
+from __future__ import annotations
+
+import pytest
+
+from tests.conftest import build_test_vibe_config
+from vibe.core.agents import AgentManager
+from vibe.core.experiments.active import ExperimentName
+from vibe.core.experiments.client import RemoteEvalClient
+from vibe.core.experiments.manager import ExperimentManager
+from vibe.core.experiments.models import EvalResponse, ExperimentAttributes
+from vibe.core.prompts import load_system_prompt
+from vibe.core.skills.manager import SkillManager
+from vibe.core.system_prompt import get_universal_system_prompt
+from vibe.core.tools.manager import ToolManager
+
+
+class _StubClient(RemoteEvalClient):
+    def __init__(self, response: EvalResponse | None) -> None:
+        self._response = response
+
+    async def evaluate(self, attributes: ExperimentAttributes) -> EvalResponse | None:
+        return self._response
+
+    async def aclose(self) -> None:
+        pass
+
+
+def _build_managers(config):
+    return (
+        ToolManager(lambda: config),
+        SkillManager(lambda: config),
+        AgentManager(lambda: config),
+    )
+
+
+@pytest.mark.asyncio
+async def test_system_prompt_uses_assigned_variant() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="cli",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    response = EvalResponse.model_validate({
+        "features": {
+            ExperimentName.SYSTEM_PROMPT.value: {
+                "defaultValue": "cli",
+                "rules": [{"force": "tests", "tracks": []}],
+            }
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(
+        ExperimentAttributes(
+            userId="x", entrypoint="cli", agent_version="0", os="darwin"
+        )
+    )
+
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=manager
+    )
+    assert prompt.startswith("You are Vibe, a super useful programming assistant.")
+
+
+@pytest.mark.asyncio
+async def test_system_prompt_falls_back_to_default_when_variant_unknown() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="cli",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    response = EvalResponse.model_validate({
+        "features": {
+            ExperimentName.SYSTEM_PROMPT.value: {
+                "defaultValue": "cli",
+                "rules": [{"force": "nonexistent_prompt", "tracks": []}],
+            }
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(
+        ExperimentAttributes(
+            userId="x", entrypoint="cli", agent_version="0", os="darwin"
+        )
+    )
+
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=manager
+    )
+    # Falls back to the default `cli` system prompt.
+    assert "You are Mistral Vibe" in prompt or "You are Vibe" in prompt
+
+
+def test_system_prompt_uses_default_when_no_manager() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="cli",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager
+    )
+    assert "You are Mistral Vibe" in prompt or "You are Vibe" in prompt
+
+
+def test_system_prompt_honors_user_config_when_manager_uninitialized() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="lean",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    manager = ExperimentManager(client=_StubClient(None))
+
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=manager
+    )
+    assert prompt == load_system_prompt("lean")
+
+
+@pytest.mark.asyncio
+async def test_system_prompt_honors_user_config_when_no_remote_assignment() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="lean",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    response = EvalResponse.model_validate({
+        "features": {"some_other_feature": {"defaultValue": True}}
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(
+        ExperimentAttributes(
+            userId="x", entrypoint="cli", agent_version="0", os="darwin"
+        )
+    )
+
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=manager
+    )
+    assert prompt == load_system_prompt("lean")
+
+
+@pytest.mark.asyncio
+async def test_user_config_overrides_assigned_experiment_variant() -> None:
+    config = build_test_vibe_config(
+        system_prompt_id="lean",
+        include_project_context=False,
+        include_prompt_detail=False,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    response = EvalResponse.model_validate({
+        "features": {
+            ExperimentName.SYSTEM_PROMPT.value: {
+                "defaultValue": "cli",
+                "rules": [{"force": "explore", "tracks": []}],
+            }
+        }
+    })
+    manager = ExperimentManager(client=_StubClient(response))
+    await manager.initialize(
+        ExperimentAttributes(
+            userId="x", entrypoint="cli", agent_version="0", os="darwin"
+        )
+    )
+
+    tool_manager, skill_manager, agent_manager = _build_managers(config)
+    prompt = get_universal_system_prompt(
+        tool_manager, config, skill_manager, agent_manager, experiment_manager=manager
+    )
+    assert prompt == load_system_prompt("lean")
+    assert prompt != load_system_prompt("explore")

tests/core/experiments/test_telemetry_integration.py

@@ -0,0 +1,31 @@
+from __future__ import annotations
+
+from tests.conftest import build_test_vibe_config
+from vibe.core.experiments.active import ExperimentName
+from vibe.core.telemetry.send import TelemetryClient
+
+
+def test_build_client_event_metadata_omits_experiments_when_empty() -> None:
+    config = build_test_vibe_config(enable_telemetry=True)
+    client = TelemetryClient(
+        config_getter=lambda: config, experiments_getter=lambda: {}
+    )
+    metadata = client.build_client_event_metadata()
+    assert "experiments" not in metadata
+
+
+def test_build_client_event_metadata_includes_experiments_when_present() -> None:
+    config = build_test_vibe_config(enable_telemetry=True)
+    sp_key = ExperimentName.SYSTEM_PROMPT.value
+    client = TelemetryClient(
+        config_getter=lambda: config, experiments_getter=lambda: {sp_key: "1"}
+    )
+    metadata = client.build_client_event_metadata()
+    assert metadata["experiments"] == {sp_key: "1"}
+
+
+def test_build_client_event_metadata_works_without_getter() -> None:
+    config = build_test_vibe_config(enable_telemetry=True)
+    client = TelemetryClient(config_getter=lambda: config)
+    metadata = client.build_client_event_metadata()
+    assert "experiments" not in metadata

tests/core/session/test_last_session_pointer.py

@@ -0,0 +1,123 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from vibe.core.config import SessionLoggingConfig
+from vibe.core.session import last_session_pointer
+
+
+@pytest.fixture
+def session_logging(tmp_path: Path) -> SessionLoggingConfig:
+    return SessionLoggingConfig(save_dir=str(tmp_path))
+
+
+def _set_tty(monkeypatch: pytest.MonkeyPatch, key: str | None) -> None:
+    monkeypatch.setattr(last_session_pointer, "current_tty_key", lambda: key)
+
+
+def test_load_returns_none_when_no_tty(
+    session_logging: SessionLoggingConfig, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    _set_tty(monkeypatch, None)
+    assert last_session_pointer.load(session_logging) is None
+
+
+def test_load_returns_none_when_no_pointer_written(
+    session_logging: SessionLoggingConfig, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    _set_tty(monkeypatch, "ttys001")
+    assert last_session_pointer.load(session_logging) is None
+
+
+def test_record_then_load_round_trip(
+    session_logging: SessionLoggingConfig, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    _set_tty(monkeypatch, "ttys001")
+    last_session_pointer.record(session_logging, "abc-123")
+    assert last_session_pointer.load(session_logging) == "abc-123"
+
+
+def test_record_skips_when_no_tty(
+    session_logging: SessionLoggingConfig, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    _set_tty(monkeypatch, None)
+    last_session_pointer.record(session_logging, "abc-123")
+    pointer_dir = Path(session_logging.save_dir) / last_session_pointer.POINTER_DIR_NAME
+    assert not pointer_dir.exists()
+
+
+def test_record_skips_when_logging_disabled(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    disabled = SessionLoggingConfig(save_dir=str(tmp_path), enabled=False)
+    _set_tty(monkeypatch, "ttys001")
+    last_session_pointer.record(disabled, "abc-123")
+    assert not (tmp_path / last_session_pointer.POINTER_DIR_NAME).exists()
+
+
+def test_pointers_are_per_tty(
+    session_logging: SessionLoggingConfig, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    _set_tty(monkeypatch, "ttys001")
+    last_session_pointer.record(session_logging, "session-a")
+
+    _set_tty(monkeypatch, "ttys002")
+    last_session_pointer.record(session_logging, "session-b")
+    assert last_session_pointer.load(session_logging) == "session-b"
+
+    _set_tty(monkeypatch, "ttys001")
+    assert last_session_pointer.load(session_logging) == "session-a"
+
+
+def test_record_ignores_empty_session_id(
+    session_logging: SessionLoggingConfig, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    _set_tty(monkeypatch, "ttys001")
+    last_session_pointer.record(session_logging, None)
+    last_session_pointer.record(session_logging, "")
+    assert last_session_pointer.load(session_logging) is None
+
+
+def test_current_tty_key_returns_none_when_ttyname_is_unavailable(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.delattr(last_session_pointer.os, "ttyname", raising=False)
+
+    assert last_session_pointer.current_tty_key() is None
+
+
+def _patch_windows(monkeypatch: pytest.MonkeyPatch, hwnd: int) -> None:
+    import ctypes
+    from types import SimpleNamespace
+
+    fake_windll = SimpleNamespace(
+        kernel32=SimpleNamespace(GetConsoleWindow=lambda: hwnd)
+    )
+    monkeypatch.setattr(last_session_pointer.sys, "platform", "win32")
+    monkeypatch.setattr(ctypes, "windll", fake_windll, raising=False)
+
+
+def test_current_tty_key_uses_console_hwnd_on_windows(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_windows(monkeypatch, hwnd=12345)
+    assert last_session_pointer.current_tty_key() == "conhost-12345"
+
+
+def test_current_tty_key_falls_back_to_wt_session(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_windows(monkeypatch, hwnd=0)
+    monkeypatch.setenv("WT_SESSION", "abcd-1234")
+    assert last_session_pointer.current_tty_key() == "wt-abcd-1234"
+
+
+def test_current_tty_key_falls_back_to_ppid_on_windows(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_windows(monkeypatch, hwnd=0)
+    monkeypatch.delenv("WT_SESSION", raising=False)
+    monkeypatch.setattr(last_session_pointer.os, "getppid", lambda: 4242)
+    assert last_session_pointer.current_tty_key() == "ppid-4242"

tests/core/test_add_dir.py

@@ -0,0 +1,389 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from vibe.core.config.harness_files import (
+    HarnessFilesManager,
+    get_harness_files_manager,
+    init_harness_files_manager,
+    reset_harness_files_manager,
+)
+from vibe.core.paths import AGENTS_MD_FILENAME
+from vibe.core.tools.utils import is_path_within_workdir
+from vibe.core.trusted_folders import trusted_folders_manager
+
+
+class TestHarnessFilesManagerAdditionalDirs:
+    def test_default_additional_dirs_empty(self) -> None:
+        mgr = HarnessFilesManager(sources=("user",))
+        assert mgr._additional_dirs == ()
+
+    def test_init_with_additional_dirs(self, tmp_path: Path) -> None:
+        d1 = tmp_path / "extra1"
+        d1.mkdir()
+        d2 = tmp_path / "extra2"
+        d2.mkdir()
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project", additional_dirs=[d1, d2])
+        mgr = get_harness_files_manager()
+        # init normalizes to resolved paths so the symlink/relative-vs-absolute
+        # spellings of the same dir compare equal on reinit.
+        assert mgr._additional_dirs == (d1.resolve(), d2.resolve())
+
+    def test_reinit_idempotent_when_paths_resolve_equal(self, tmp_path: Path) -> None:
+        target = tmp_path / "target"
+        target.mkdir()
+        link = tmp_path / "link"
+        link.symlink_to(target, target_is_directory=True)
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project", additional_dirs=[target])
+        # Second call with the symlink spelling must NOT raise — both resolve
+        # to the same dir.
+        init_harness_files_manager("user", "project", additional_dirs=[link])
+
+    def test_init_without_additional_dirs(self) -> None:
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project")
+        mgr = get_harness_files_manager()
+        assert mgr._additional_dirs == ()
+
+    def test_reinit_with_same_additional_dirs_is_idempotent(
+        self, tmp_path: Path
+    ) -> None:
+        d = tmp_path / "extra"
+        d.mkdir()
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project", additional_dirs=[d])
+        # Same sources + same additional_dirs: no-op, no raise.
+        init_harness_files_manager("user", "project", additional_dirs=[d])
+
+    def test_reinit_with_different_additional_dirs_raises(self, tmp_path: Path) -> None:
+        d1 = tmp_path / "extra1"
+        d1.mkdir()
+        d2 = tmp_path / "extra2"
+        d2.mkdir()
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project", additional_dirs=[d1])
+        with pytest.raises(RuntimeError, match="different configuration"):
+            init_harness_files_manager("user", "project", additional_dirs=[d2])
+
+
+class TestAdditionalDirsDiscovery:
+    def test_discovers_tools_from_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra_project"
+        tools_dir = extra / ".vibe" / "tools"
+        tools_dir.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert tools_dir in mgr.project_tools_dirs
+
+    def test_discovers_skills_from_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra_project"
+        skills_dir = extra / ".vibe" / "skills"
+        skills_dir.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert skills_dir in mgr.project_skills_dirs
+
+    def test_discovers_agents_from_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra_project"
+        agents_dir = extra / ".vibe" / "agents"
+        agents_dir.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert agents_dir in mgr.project_agents_dirs
+
+    def test_discovers_agents_skills_from_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra_project"
+        skills_dir = extra / ".agents" / "skills"
+        skills_dir.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert skills_dir in mgr.project_skills_dirs
+
+    def test_discovers_prompts_from_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra_project"
+        prompts_dir = extra / ".vibe" / "prompts"
+        prompts_dir.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert prompts_dir in mgr.project_prompts_dirs
+
+    def test_no_dirs_when_additional_dir_has_no_vibe(self, tmp_path: Path) -> None:
+        extra = tmp_path / "bare_project"
+        extra.mkdir()
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert mgr.project_tools_dirs == []
+        assert mgr.project_skills_dirs == []
+        assert mgr.project_agents_dirs == []
+
+
+class TestAdditionalDirsAgentsMd:
+    def test_load_project_docs_includes_additional_dir_agents_md(
+        self, tmp_path: Path
+    ) -> None:
+        extra = tmp_path / "extra_project"
+        extra.mkdir()
+        agents_md = extra / AGENTS_MD_FILENAME
+        agents_md.write_text("Extra project instructions", encoding="utf-8")
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        docs = mgr.load_project_docs()
+        assert any(d == extra and "Extra project" in content for d, content in docs)
+
+    def test_load_project_docs_skips_missing_agents_md(self, tmp_path: Path) -> None:
+        extra = tmp_path / "no_agents_md"
+        extra.mkdir()
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        docs = mgr.load_project_docs()
+        assert not any(d == extra for d, _ in docs)
+
+    def test_load_project_docs_skips_empty_agents_md(self, tmp_path: Path) -> None:
+        extra = tmp_path / "empty_agents_md"
+        extra.mkdir()
+        (extra / AGENTS_MD_FILENAME).write_text("   \n", encoding="utf-8")
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        docs = mgr.load_project_docs()
+        assert not any(d == extra for d, _ in docs)
+
+    def test_find_subdirectory_agents_md_in_additional_dir(
+        self, tmp_path: Path
+    ) -> None:
+        extra = tmp_path / "extra_project"
+        sub = extra / "sub"
+        sub.mkdir(parents=True)
+        agents_md = sub / AGENTS_MD_FILENAME
+        agents_md.write_text("Sub instructions", encoding="utf-8")
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        file_in_sub = sub / "main.py"
+        docs = mgr.find_subdirectory_agents_md(file_in_sub)
+        assert any("Sub instructions" in content for _, content in docs)
+
+    def test_find_subdirectory_agents_md_resolves_additional_dir(
+        self, tmp_path: Path
+    ) -> None:
+        extra = tmp_path / "extra_project"
+        sub = extra / "sub"
+        sub.mkdir(parents=True)
+        link = tmp_path / "extra_link"
+        link.symlink_to(extra, target_is_directory=True)
+        agents_md = sub / AGENTS_MD_FILENAME
+        agents_md.write_text("Sub instructions", encoding="utf-8")
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(link,))
+        docs = mgr.find_subdirectory_agents_md(sub / "main.py")
+        assert any("Sub instructions" in content for _, content in docs)
+
+    def test_find_subdirectory_agents_md_outside_all_dirs(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra"
+        extra.mkdir()
+        outside = tmp_path / "outside"
+        outside.mkdir()
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        docs = mgr.find_subdirectory_agents_md(outside / "file.py")
+        assert docs == []
+
+
+class TestFilePermissionsAdditionalDirs:
+    def test_is_within_workdir_for_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra"
+        extra.mkdir()
+        file_in_extra = extra / "some_file.py"
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project", additional_dirs=[extra])
+
+        assert is_path_within_workdir(str(file_in_extra))
+
+    def test_is_within_workdir_resolves_additional_dir(self, tmp_path: Path) -> None:
+        extra = tmp_path / "extra"
+        extra.mkdir()
+        link = tmp_path / "extra_link"
+        link.symlink_to(extra, target_is_directory=True)
+        file_in_extra = extra / "some_file.py"
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project", additional_dirs=[link])
+
+        assert is_path_within_workdir(str(file_in_extra))
+
+    def test_is_within_workdir_for_cwd(self, tmp_working_directory: Path) -> None:
+        file_in_cwd = tmp_working_directory / "file.py"
+        assert is_path_within_workdir(str(file_in_cwd))
+
+    def test_is_not_within_workdir_for_outside_path(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        outside = tmp_path / "outside"
+        outside.mkdir()
+        cwd = tmp_path / "cwd"
+        cwd.mkdir()
+        monkeypatch.chdir(cwd)
+
+        reset_harness_files_manager()
+        init_harness_files_manager("user", "project")
+
+        assert not is_path_within_workdir(str(outside / "file.py"))
+
+    def test_is_path_within_workdir_when_manager_uninitialized(
+        self, tmp_working_directory: Path
+    ) -> None:
+        reset_harness_files_manager()
+        # Without a manager, only cwd counts.
+        assert is_path_within_workdir(str(tmp_working_directory / "file.py"))
+
+
+class TestAdditionalDirsAreImplicitlyTrusted:
+    def test_load_project_docs_does_not_require_trust_for_additional_dir(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        # Untrusted cwd: primary workdir contributes nothing.
+        monkeypatch.setattr(trusted_folders_manager, "is_trusted", lambda _: False)
+
+        extra = tmp_path / "extra_project"
+        extra.mkdir()
+        (extra / AGENTS_MD_FILENAME).write_text(
+            "Trusted via --add-dir", encoding="utf-8"
+        )
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(extra,)
+        )
+        docs = mgr.load_project_docs()
+        assert any("Trusted via --add-dir" in content for _, content in docs)
+
+    def test_project_tools_dirs_skip_trust_check_for_additional_dir(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        monkeypatch.setattr(trusted_folders_manager, "is_trusted", lambda _: False)
+
+        extra = tmp_path / "extra_project"
+        tools_dir = extra / ".vibe" / "tools"
+        tools_dir.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(extra,)
+        )
+        assert tools_dir in mgr.project_tools_dirs
+
+
+class TestLoadProjectDocsDedupe:
+    def test_dedupes_when_cwd_under_additional_dir(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        extra = (tmp_path / "extra_project").resolve()
+        extra.mkdir()
+        sub = extra / "sub"
+        sub.mkdir()
+        (extra / AGENTS_MD_FILENAME).write_text("Root instructions", encoding="utf-8")
+        (sub / AGENTS_MD_FILENAME).write_text("Sub instructions", encoding="utf-8")
+        monkeypatch.chdir(sub)
+
+        trusted_folders_manager.trust_for_session(extra)
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(extra,)
+        )
+        docs = mgr.load_project_docs()
+
+        emitted_dirs = [d for d, _ in docs]
+        assert emitted_dirs.count(extra) == 1
+        assert sub.resolve() in emitted_dirs
+
+
+class TestHookFilesAdditionalDirs:
+    def test_hook_files_includes_additional_dir_hooks_toml(
+        self, tmp_path: Path
+    ) -> None:
+        extra = tmp_path / "extra"
+        extra.mkdir()
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(extra,))
+        assert extra / ".vibe" / "hooks.toml" in mgr.hook_files
+
+    def test_hook_files_deduplicates_workdir_and_additional_dir(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        workdir = tmp_path / "project"
+        workdir.mkdir()
+        monkeypatch.chdir(workdir)
+        trusted_folders_manager.trust_for_session(workdir)
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(workdir.resolve(),)
+        )
+        hook_file = workdir / ".vibe" / "hooks.toml"
+
+        assert mgr.hook_files.count(hook_file.resolve()) == 1
+
+
+class TestProjectPromptsDirsAdditionalDirs:
+    def test_project_prompts_dirs_deduplicates_workdir_and_additional_dir(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        workdir = tmp_path / "project"
+        (workdir / ".vibe" / "prompts").mkdir(parents=True)
+        monkeypatch.chdir(workdir)
+        trusted_folders_manager.trust_for_session(workdir)
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(workdir.resolve(),)
+        )
+
+        assert len(mgr.project_prompts_dirs) == 1
+
+
+class TestProjectRootsNestedDedup:
+    def test_nested_add_dirs_collapse(self, tmp_path: Path) -> None:
+        outer = (tmp_path / "outer").resolve()
+        inner = outer / "inner"
+        inner.mkdir(parents=True)
+
+        mgr = HarnessFilesManager(sources=("user",), _additional_dirs=(outer, inner))
+        assert mgr.project_roots == [outer]
+
+    def test_add_dir_containing_cwd_keeps_both(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        outer = (tmp_path / "outer").resolve()
+        inner = outer / "inner"
+        inner.mkdir(parents=True)
+        (outer / AGENTS_MD_FILENAME).write_text("Outer instructions", encoding="utf-8")
+        (inner / AGENTS_MD_FILENAME).write_text("Inner instructions", encoding="utf-8")
+        monkeypatch.chdir(inner)
+        trusted_folders_manager.trust_for_session(inner)
+        trusted_folders_manager.trust_for_session(outer)
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(outer,)
+        )
+        # cwd survives so its walk-up semantics still work; the add-dir keeps
+        # its own root-level discovery.
+        assert mgr.project_roots == [inner, outer]
+        docs = mgr.load_project_docs()
+        assert any("Outer" in c for _, c in docs)
+        assert any("Inner" in c for _, c in docs)
+
+    def test_add_dir_nested_under_trusted_workdir(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        outer = (tmp_path / "outer").resolve()
+        inner = outer / "inner"
+        inner.mkdir(parents=True)
+        monkeypatch.chdir(outer)
+        trusted_folders_manager.trust_for_session(outer)
+
+        mgr = HarnessFilesManager(
+            sources=("user", "project"), _additional_dirs=(inner,)
+        )
+        assert mgr.project_roots == [outer]

tests/core/test_config_paths.py

@@ -14,7 +14,7 @@ class TestTrustedWorkdir:
     ) -> None:
         monkeypatch.chdir(tmp_path)
         mgr = HarnessFilesManager(sources=("user",))
-        assert mgr.trusted_workdir is None
+        assert mgr._trusted_workdir is None
 
     def test_returns_none_when_not_trusted(
         self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
@@ -22,7 +22,7 @@ class TestTrustedWorkdir:
         monkeypatch.chdir(tmp_path)
         monkeypatch.setattr(trusted_folders_manager, "is_trusted", lambda _: False)
         mgr = HarnessFilesManager(sources=("user", "project"))
-        assert mgr.trusted_workdir is None
+        assert mgr._trusted_workdir is None
 
     def test_returns_cwd_when_project_in_sources_and_trusted(
         self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
@@ -30,7 +30,7 @@ class TestTrustedWorkdir:
         monkeypatch.chdir(tmp_path)
         monkeypatch.setattr(trusted_folders_manager, "is_trusted", lambda _: True)
         mgr = HarnessFilesManager(sources=("user", "project"))
-        assert mgr.trusted_workdir == tmp_path
+        assert mgr._trusted_workdir == tmp_path
 
 
 class TestProjectToolsDirs:

tests/core/test_local_config_walk.py

@@ -5,6 +5,7 @@ from pathlib import Path
 from vibe.core.paths._local_config_walk import (
     _MAX_DIRS,
     WALK_MAX_DEPTH,
+    ConfigWalkResult,
     walk_local_config_dirs,
 )
 
@@ -166,3 +167,29 @@ class TestWalkConfigDirs:
         assert resolved / ".vibe" in result.config_dirs
         assert resolved / ".agents" in result.config_dirs
         assert resolved / "sub" / ".vibe" in result.config_dirs
+
+
+class TestConfigWalkResultOr:
+    def test_or_concatenates_each_field(self) -> None:
+        a = ConfigWalkResult(
+            config_dirs=(Path("/a/.vibe"),),
+            tools=(Path("/a/.vibe/tools"),),
+            skills=(Path("/a/.vibe/skills"),),
+            agents=(Path("/a/.vibe/agents"),),
+        )
+        b = ConfigWalkResult(
+            config_dirs=(Path("/b/.vibe"),),
+            tools=(Path("/b/.vibe/tools"),),
+            skills=(Path("/b/.vibe/skills"),),
+            agents=(Path("/b/.vibe/agents"),),
+        )
+        merged = a | b
+        assert merged.config_dirs == (Path("/a/.vibe"), Path("/b/.vibe"))
+        assert merged.tools == (Path("/a/.vibe/tools"), Path("/b/.vibe/tools"))
+        assert merged.skills == (Path("/a/.vibe/skills"), Path("/b/.vibe/skills"))
+        assert merged.agents == (Path("/a/.vibe/agents"), Path("/b/.vibe/agents"))
+
+    def test_or_with_empty_is_identity(self) -> None:
+        a = ConfigWalkResult(tools=(Path("/a/.vibe/tools"),))
+        assert (a | ConfigWalkResult()) == a
+        assert (ConfigWalkResult() | a) == a

tests/core/test_rewind_manager.py

@@ -37,7 +37,7 @@ def _make_manager(
     async def save_messages() -> None:
         save_calls.append(True)
 
-    def reset_session() -> None:
+    async def reset_session() -> None:
         reset_calls.append(True)
 
     mgr = RewindManager(

tests/core/test_utils.py

@@ -102,6 +102,79 @@ class TestReadSafe:
             read_safe(tmp_path / "nope.txt")
 
 
+class TestReadSafeNewlines:
+    def test_lf(self, tmp_path: Path) -> None:
+        f = tmp_path / "lf.txt"
+        f.write_bytes(b"a\nb\nc\n")
+        got = read_safe(f)
+        assert got.text == "a\nb\nc\n"
+        assert got.newline == "\n"
+
+    def test_crlf(self, tmp_path: Path) -> None:
+        f = tmp_path / "crlf.txt"
+        f.write_bytes(b"a\r\nb\r\nc\r\n")
+        got = read_safe(f)
+        assert got.text == "a\nb\nc\n"
+        assert got.newline == "\r\n"
+
+    def test_cr(self, tmp_path: Path) -> None:
+        f = tmp_path / "cr.txt"
+        f.write_bytes(b"a\rb\rc\r")
+        got = read_safe(f)
+        assert got.text == "a\nb\nc\n"
+        assert got.newline == "\r"
+
+    def test_mixed_picks_most_frequent(self, tmp_path: Path) -> None:
+        f = tmp_path / "mixed.txt"
+        f.write_bytes(b"a\r\nb\r\nc\rd\n")
+        got = read_safe(f)
+        assert got.text == "a\nb\nc\nd\n"
+        assert got.newline == "\r\n"
+
+    @pytest.mark.parametrize(("linesep", "expected"), [("\n", "\n"), ("\r\n", "\r\n")])
+    def test_no_newline_defaults_to_os_linesep(
+        self,
+        tmp_path: Path,
+        monkeypatch: pytest.MonkeyPatch,
+        linesep: str,
+        expected: str,
+    ) -> None:
+        monkeypatch.setattr(io_utils.os, "linesep", linesep)
+        f = tmp_path / "single.txt"
+        f.write_bytes(b"hello")
+        got = read_safe(f)
+        assert got.text == "hello"
+        assert got.newline == expected
+
+    @pytest.mark.parametrize(("linesep", "expected"), [("\n", "\n"), ("\r\n", "\r\n")])
+    def test_empty_defaults_to_os_linesep(
+        self,
+        tmp_path: Path,
+        monkeypatch: pytest.MonkeyPatch,
+        linesep: str,
+        expected: str,
+    ) -> None:
+        monkeypatch.setattr(io_utils.os, "linesep", linesep)
+        f = tmp_path / "empty.txt"
+        f.write_bytes(b"")
+        got = read_safe(f)
+        assert got.text == ""
+        assert got.newline == expected
+
+    def test_decode_safe_reports_newline(self) -> None:
+        got = decode_safe(b"a\r\nb\r\n")
+        assert got.text == "a\nb\n"
+        assert got.newline == "\r\n"
+
+    @pytest.mark.asyncio
+    async def test_async_reports_newline(self, tmp_path: Path) -> None:
+        f = tmp_path / "crlf.txt"
+        f.write_bytes(b"a\r\nb\r\n")
+        got = await read_safe_async(f)
+        assert got.text == "a\nb\n"
+        assert got.newline == "\r\n"
+
+
 class TestReadSafeResultEncoding:
     def test_reports_utf8_for_plain_utf8_file(self, tmp_path: Path) -> None:
         f = tmp_path / "x.txt"

tests/e2e/common.py

@@ -29,6 +29,7 @@ def write_e2e_config(vibe_home: Path, api_base: str) -> None:
             'active_model = "mock-model"',
             "enable_update_checks = false",
             "enable_auto_update = false",
+            "disable_welcome_banner_animation = true",
             "",
             "[[providers]]",
             'name = "mock-provider"',

tests/onboarding/test_ui_onboarding.py

@@ -1,19 +1,49 @@
 from __future__ import annotations
 
+import asyncio
 from collections.abc import Callable
+from pathlib import Path
 from types import SimpleNamespace
+from typing import cast
 
 import pytest
 from textual.pilot import Pilot
 from textual.widgets import Input
 
-from vibe.core.config import ProviderConfig
+from tests.browser_sign_in.stubs import build_browser_sign_in_service_factory
+from tests.conftest import build_test_vibe_config
+from vibe.core.config import ModelConfig, ProviderConfig, VibeConfig
+from vibe.core.config._settings import (
+    DEFAULT_MISTRAL_BROWSER_AUTH_API_BASE_URL,
+    DEFAULT_MISTRAL_BROWSER_AUTH_BASE_URL,
+)
+from vibe.core.config.harness_files import (
+    init_harness_files_manager,
+    reset_harness_files_manager,
+)
 from vibe.core.paths import GLOBAL_ENV_FILE
 from vibe.core.telemetry.build_metadata import build_entrypoint_metadata
 from vibe.core.telemetry.send import TelemetryClient
 from vibe.core.types import Backend
+from vibe.setup.auth import (
+    BrowserSignInError,
+    BrowserSignInErrorCode,
+    BrowserSignInService,
+    BrowserSignInStatus,
+)
+import vibe.setup.onboarding as onboarding_module
 from vibe.setup.onboarding import OnboardingApp
 from vibe.setup.onboarding.screens.api_key import ApiKeyScreen, persist_api_key
+from vibe.setup.onboarding.screens.auth_method import AuthMethodScreen
+from vibe.setup.onboarding.screens.browser_sign_in import (
+    ERROR_HINT,
+    PENDING_HINT,
+    UNEXPECTED_ERROR_MESSAGE,
+    BrowserSignInScreen,
+)
+
+CONSOLE_URL = "https://console.mistral.ai"
+API_URL = "https://api.mistral.ai"
 
 
 async def _wait_for(
@@ -26,40 +56,653 @@ async def _wait_for(
     while not condition():
         await pilot.pause(interval)
         if (elapsed := elapsed + interval) >= timeout:
-            msg = "Timed out waiting for condition."
+            raise AssertionError("Timed out waiting for condition.")
+
+
+def _build_onboarding_config(
+    *,
+    provider_name: str = "mistral",
+    model_provider: str | None = None,
+    backend: Backend = Backend.MISTRAL,
+    api_key_env_var: str = "MISTRAL_API_KEY",
+    browser_auth_base_url: str | None = None,
+    browser_auth_api_base_url: str | None = None,
+    enable_experimental_browser_sign_in: bool = False,
+) -> VibeConfig:
+    provider = ProviderConfig(
+        name=provider_name,
+        api_base="https://api.mistral.ai/v1",
+        api_key_env_var=api_key_env_var,
+        browser_auth_base_url=browser_auth_base_url,
+        browser_auth_api_base_url=browser_auth_api_base_url,
+        backend=backend,
+    )
+    model = ModelConfig(
+        name="mistral-vibe-cli-latest",
+        provider=model_provider or provider_name,
+        alias="devstral-2",
+    )
+    return build_test_vibe_config(
+        providers=[provider],
+        models=[model],
+        enable_experimental_browser_sign_in=enable_experimental_browser_sign_in,
+    )
+
+
+def _build_browser_onboarding_app(
+    *, browser_sign_in_service_factory: Callable[[], BrowserSignInService] | None = None
+) -> OnboardingApp:
+    return OnboardingApp(
+        config=_build_onboarding_config(
+            browser_auth_base_url=CONSOLE_URL,
+            browser_auth_api_base_url=API_URL,
+            enable_experimental_browser_sign_in=True,
+        ),
+        browser_sign_in_service_factory=browser_sign_in_service_factory,
+    )
+
+
+def _patch_failing_browser_sign_in_service(
+    monkeypatch: pytest.MonkeyPatch, captured_base_urls: list[tuple[str, str]]
+) -> None:
+    class FakeGateway:
+        def __init__(self, browser_base_url: str, api_base_url: str) -> None:
+            captured_base_urls.append((browser_base_url, api_base_url))
+
+    class FakeService:
+        def __init__(self, gateway: FakeGateway) -> None:
+            self._gateway = gateway
+
+        async def authenticate(self, *args, **kwargs) -> str:
+            raise BrowserSignInError(
+                "Browser sign-in polling failed.",
+                code=BrowserSignInErrorCode.POLL_FAILED,
+            )
+
+        async def aclose(self) -> None:
+            return None
+
+    monkeypatch.setattr(onboarding_module, "HttpBrowserSignInGateway", FakeGateway)
+    monkeypatch.setattr(onboarding_module, "BrowserSignInService", FakeService)
+
+
+def _saved_env_contents() -> str:
+    return GLOBAL_ENV_FILE.path.read_text(encoding="utf-8")
+
+
+def _build_unexpected_browser_sign_in_service_factory(
+    outcomes: list[str],
+    *,
+    api_key: str = "sk-browser-onboarding-test-key",
+    close_blocker: asyncio.Event | None = None,
+    close_started: asyncio.Event | None = None,
+    close_finished: asyncio.Event | None = None,
+    close_cancelled: asyncio.Event | None = None,
+) -> Callable[[], BrowserSignInService]:
+    remaining_outcomes = list(outcomes)
+
+    class UnexpectedBrowserSignInService:
+        def __init__(self, outcome: str) -> None:
+            self._outcome = outcome
+
+        async def authenticate(
+            self, status_callback: Callable[[BrowserSignInStatus], None] | None = None
+        ) -> str:
+            if self._outcome == "completed":
+                if status_callback is not None:
+                    status_callback(BrowserSignInStatus.COMPLETED)
+                return api_key
+            if self._outcome == "runtime_error":
+                raise RuntimeError("boom")
+            msg = f"Unsupported browser sign-in outcome: {self._outcome}"
             raise AssertionError(msg)
 
+        async def aclose(self) -> None:
+            try:
+                if close_started is not None:
+                    close_started.set()
+                if close_blocker is not None:
+                    await close_blocker.wait()
+            except asyncio.CancelledError:
+                if close_cancelled is not None:
+                    close_cancelled.set()
+                raise
+            finally:
+                if close_finished is not None:
+                    close_finished.set()
 
-async def pass_welcome_screen(pilot: Pilot) -> None:
+            return None
+
+    def build_service() -> BrowserSignInService:
+        if not remaining_outcomes:
+            msg = (
+                "Unexpected browser sign-in service factory requires scripted outcomes."
+            )
+            raise AssertionError(msg)
+        return cast(
+            BrowserSignInService,
+            UnexpectedBrowserSignInService(remaining_outcomes.pop(0)),
+        )
+
+    return build_service
+
+
+async def _pass_welcome_screen(pilot: Pilot) -> None:
     welcome_screen = pilot.app.get_screen("welcome")
     await _wait_for(
         lambda: not welcome_screen.query_one("#enter-hint").has_class("hidden"), pilot
     )
     await pilot.press("enter")
-    await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+
+
+async def _show_auth_method(pilot: Pilot) -> None:
+    await _pass_welcome_screen(pilot)
+    await _wait_for(lambda: isinstance(pilot.app.screen, AuthMethodScreen), pilot)
+
+
+async def _show_browser_sign_in(pilot: Pilot) -> None:
+    await _show_auth_method(pilot)
+    await pilot.press("enter")
+    await _wait_for(lambda: isinstance(pilot.app.screen, BrowserSignInScreen), pilot)
 
 
 @pytest.mark.asyncio
-async def test_ui_gets_through_the_onboarding_successfully() -> None:
-    app = OnboardingApp()
+async def test_ui_keeps_manual_flow_when_browser_sign_in_is_unsupported() -> None:
+    app = OnboardingApp(
+        config=_build_onboarding_config(
+            browser_auth_base_url="",
+            browser_auth_api_base_url="",
+            enable_experimental_browser_sign_in=True,
+        )
+    )
     api_key_value = "sk-onboarding-test-key"
 
     async with app.run_test() as pilot:
-        await pass_welcome_screen(pilot)
-        api_screen = app.screen
-        input_widget = api_screen.query_one("#key", Input)
+        await _pass_welcome_screen(pilot)
+        await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+        input_widget = app.screen.query_one("#key", Input)
         await pilot.press(*api_key_value)
         assert input_widget.value == api_key_value
-
         await pilot.press("enter")
         await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
 
     assert app.return_value == "completed"
+    assert api_key_value in _saved_env_contents()
 
-    assert GLOBAL_ENV_FILE.path.is_file()
-    env_contents = GLOBAL_ENV_FILE.path.read_text(encoding="utf-8")
+
+@pytest.mark.asyncio
+async def test_ui_hides_browser_sign_in_when_experimental_flag_is_disabled() -> None:
+    _, browser_sign_in_service_factory, _ = build_browser_sign_in_service_factory(
+        outcomes=["completed"]
+    )
+    app = OnboardingApp(
+        config=_build_onboarding_config(
+            browser_auth_base_url=CONSOLE_URL, browser_auth_api_base_url=API_URL
+        ),
+        browser_sign_in_service_factory=browser_sign_in_service_factory,
+    )
+
+    assert app.supports_browser_sign_in is False
+
+    async with app.run_test() as pilot:
+        await _pass_welcome_screen(pilot)
+        await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+
+
+@pytest.mark.asyncio
+async def test_ui_supports_browser_sign_in_when_experimental_flag_is_enabled() -> None:
+    _, browser_sign_in_service_factory, _ = build_browser_sign_in_service_factory(
+        outcomes=["completed"]
+    )
+    app = OnboardingApp(
+        config=_build_onboarding_config(
+            browser_auth_base_url=CONSOLE_URL,
+            browser_auth_api_base_url=API_URL,
+            enable_experimental_browser_sign_in=True,
+        ),
+        browser_sign_in_service_factory=browser_sign_in_service_factory,
+    )
+
+    assert app.supports_browser_sign_in is True
+
+    async with app.run_test() as pilot:
+        await _show_auth_method(pilot)
+
+
+@pytest.mark.asyncio
+async def test_ui_offers_browser_sign_in_for_renamed_mistral_provider() -> None:
+    app = OnboardingApp(
+        config=_build_onboarding_config(
+            provider_name="customer-mistral",
+            backend=Backend.MISTRAL,
+            browser_auth_base_url=CONSOLE_URL,
+            browser_auth_api_base_url=API_URL,
+            enable_experimental_browser_sign_in=True,
+        )
+    )
+
+    assert app.supports_browser_sign_in is True
+
+    async with app.run_test() as pilot:
+        await _show_auth_method(pilot)
+
+
+@pytest.mark.asyncio
+async def test_ui_allows_manual_path_when_browser_sign_in_is_supported() -> None:
+    app = _build_browser_onboarding_app()
+    api_key_value = "sk-manual-onboarding-test-key"
+
+    async with app.run_test() as pilot:
+        await _show_auth_method(pilot)
+        await pilot.press("down", "enter")
+        await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+        input_widget = app.screen.query_one("#key", Input)
+        await pilot.press(*api_key_value)
+        await pilot.press("enter")
+        await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
+        assert input_widget.value == api_key_value
+
+    assert app.return_value == "completed"
+    assert api_key_value in _saved_env_contents()
+
+
+@pytest.mark.asyncio
+async def test_ui_completes_browser_sign_in_and_retries_after_failure() -> None:
+    gateway, browser_sign_in_service_factory, created_services = (
+        build_browser_sign_in_service_factory(outcomes=["expired", "completed"])
+    )
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=browser_sign_in_service_factory
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(
+            lambda: (
+                "expired"
+                in str(app.screen.query_one("#browser-sign-in-status").render())
+            ),
+            pilot,
+        )
+        await pilot.press("r")
+        await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
+
+    assert gateway.process_number == 2
+    assert len(created_services) == 2
+    assert created_services[0] is not created_services[1]
+    assert app.return_value == "completed"
+    assert "sk-browser-onboarding-test-key" in _saved_env_contents()
+
+
+@pytest.mark.asyncio
+async def test_ui_browser_sign_in_falls_back_to_mistral_env_var_when_missing() -> None:
+    _, browser_sign_in_service_factory, _ = build_browser_sign_in_service_factory(
+        outcomes=["completed"]
+    )
+    app = OnboardingApp(
+        config=_build_onboarding_config(
+            provider_name="custom-mistral",
+            api_key_env_var="",
+            browser_auth_base_url=CONSOLE_URL,
+            browser_auth_api_base_url=API_URL,
+            enable_experimental_browser_sign_in=True,
+        ),
+        browser_sign_in_service_factory=browser_sign_in_service_factory,
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
+
+    assert app.return_value == "completed"
+    env_contents = _saved_env_contents()
     assert "MISTRAL_API_KEY" in env_contents
+    assert "sk-browser-onboarding-test-key" in env_contents
+
+
+@pytest.mark.asyncio
+async def test_ui_shows_human_message_when_polling_fails() -> None:
+    _, browser_sign_in_service_factory, _ = build_browser_sign_in_service_factory(
+        outcomes=["poll_failed"]
+    )
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=browser_sign_in_service_factory
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(
+            lambda: (
+                "We couldn't complete sign-in. Please try again."
+                in str(app.screen.query_one("#browser-sign-in-status").render())
+            ),
+            pilot,
+        )
+
+
+@pytest.mark.asyncio
+async def test_ui_shows_retryable_error_when_browser_sign_in_fails_unexpectedly() -> (
+    None
+):
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=_build_unexpected_browser_sign_in_service_factory([
+            "runtime_error"
+        ])
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(
+            lambda: (
+                UNEXPECTED_ERROR_MESSAGE
+                in str(app.screen.query_one("#browser-sign-in-status").render())
+            ),
+            pilot,
+        )
+
+        assert isinstance(app.screen, BrowserSignInScreen)
+        status_widget = app.screen.query_one("#browser-sign-in-status")
+        assert status_widget.has_class("error")
+        assert not status_widget.has_class("pending")
+        assert ERROR_HINT in str(app.screen.query_one("#browser-sign-in-hint").render())
+        assert app.return_value is None
+
+
+@pytest.mark.asyncio
+async def test_ui_retries_after_unexpected_browser_sign_in_failure() -> None:
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=_build_unexpected_browser_sign_in_service_factory([
+            "runtime_error",
+            "completed",
+        ])
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(
+            lambda: (
+                UNEXPECTED_ERROR_MESSAGE
+                in str(app.screen.query_one("#browser-sign-in-status").render())
+            ),
+            pilot,
+        )
+        await pilot.press("r")
+        await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
+
+    assert app.return_value == "completed"
+    assert "sk-browser-onboarding-test-key" in _saved_env_contents()
+
+
+@pytest.mark.asyncio
+async def test_ui_waits_for_browser_sign_in_cleanup_before_retrying() -> None:
+    close_started = asyncio.Event()
+    close_blocker = asyncio.Event()
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=_build_unexpected_browser_sign_in_service_factory(
+            ["runtime_error", "completed"],
+            close_blocker=close_blocker,
+            close_started=close_started,
+        )
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(close_started.is_set, pilot)
+
+        status_widget = app.screen.query_one("#browser-sign-in-status")
+        hint_widget = app.screen.query_one("#browser-sign-in-hint")
+        await _wait_for(
+            lambda: "Getting things ready..." in str(status_widget.render()), pilot
+        )
+        await _wait_for(lambda: PENDING_HINT in str(hint_widget.render()), pilot)
+
+        await pilot.press("r")
+        await _wait_for(
+            lambda: "Getting things ready..." in str(status_widget.render()), pilot
+        )
+        await _wait_for(lambda: PENDING_HINT in str(hint_widget.render()), pilot)
+        assert app.return_value is None
+
+        close_blocker.set()
+        await _wait_for(
+            lambda: (
+                UNEXPECTED_ERROR_MESSAGE
+                in str(app.screen.query_one("#browser-sign-in-status").render())
+            ),
+            pilot,
+        )
+
+        await pilot.press("r")
+        await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
+
+    assert app.return_value == "completed"
+    assert "sk-browser-onboarding-test-key" in _saved_env_contents()
+
+
+@pytest.mark.asyncio
+async def test_ui_switches_to_manual_path_without_cancelling_browser_sign_in_cleanup() -> (
+    None
+):
+    close_started = asyncio.Event()
+    close_blocker = asyncio.Event()
+    close_finished = asyncio.Event()
+    close_cancelled = asyncio.Event()
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=_build_unexpected_browser_sign_in_service_factory(
+            ["runtime_error"],
+            close_blocker=close_blocker,
+            close_started=close_started,
+            close_finished=close_finished,
+            close_cancelled=close_cancelled,
+        )
+    )
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(close_started.is_set, pilot)
+
+        await pilot.press("m")
+        await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+
+        close_blocker.set()
+        await _wait_for(close_finished.is_set, pilot)
+
+    assert close_cancelled.is_set() is False
+
+
+@pytest.mark.asyncio
+async def test_ui_switches_to_manual_path_while_browser_sign_in_is_running() -> None:
+    blocker = asyncio.Event()
+
+    async def wait_forever(_: float) -> None:
+        await blocker.wait()
+
+    gateway, browser_sign_in_service_factory, _ = build_browser_sign_in_service_factory(
+        outcomes=["completed"], sleep=wait_forever
+    )
+    app = _build_browser_onboarding_app(
+        browser_sign_in_service_factory=browser_sign_in_service_factory
+    )
+    api_key_value = "sk-manual-after-browser-cancel"
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(
+            lambda: (
+                "Waiting for you to finish signing in..."
+                in str(app.screen.query_one("#browser-sign-in-status").render())
+            ),
+            pilot,
+        )
+        status_widget = app.screen.query_one("#browser-sign-in-status")
+        assert status_widget.has_class("pending")
+        assert not status_widget.has_class("error")
+        step_widgets = list(app.screen.query(".browser-sign-in-step"))
+        assert len(step_widgets) == 3
+        assert step_widgets[0].has_class("done")
+        assert "Open your browser" in str(step_widgets[0].render())
+        assert step_widgets[1].has_class("active")
+        assert "Sign in and return here" in str(step_widgets[1].render())
+        assert step_widgets[2].has_class("idle")
+        assert "Finish setup" in str(step_widgets[2].render())
+        await pilot.press("m")
+        await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+        await pilot.press(*api_key_value)
+        await pilot.press("enter")
+        await _wait_for(lambda: app.return_value is not None, pilot, timeout=2.0)
+
+    assert app.return_value == "completed"
+    assert gateway.closed is True
+    assert gateway.exchange_requests == []
+    env_contents = _saved_env_contents()
     assert api_key_value in env_contents
+    assert "sk-browser-onboarding-test-key" not in env_contents
+
+
+@pytest.mark.asyncio
+async def test_ui_uses_default_mistral_browser_auth_urls_when_experiment_is_enabled(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    captured_base_urls: list[tuple[str, str]] = []
+    _patch_failing_browser_sign_in_service(monkeypatch, captured_base_urls)
+
+    app = OnboardingApp(
+        config=build_test_vibe_config(enable_experimental_browser_sign_in=True)
+    )
+
+    assert app.supports_browser_sign_in is True
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(lambda: bool(captured_base_urls), pilot)
+
+    assert captured_base_urls == [
+        (
+            DEFAULT_MISTRAL_BROWSER_AUTH_BASE_URL,
+            DEFAULT_MISTRAL_BROWSER_AUTH_API_BASE_URL,
+        )
+    ]
+
+
+@pytest.mark.asyncio
+async def test_ui_enables_browser_sign_in_from_env_var(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setenv("VIBE_ENABLE_EXPERIMENTAL_BROWSER_SIGN_IN", "true")
+    captured_base_urls: list[tuple[str, str]] = []
+    _patch_failing_browser_sign_in_service(monkeypatch, captured_base_urls)
+
+    app = OnboardingApp()
+
+    assert app.supports_browser_sign_in is True
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(lambda: bool(captured_base_urls), pilot)
+
+    assert captured_base_urls == [
+        (
+            DEFAULT_MISTRAL_BROWSER_AUTH_BASE_URL,
+            DEFAULT_MISTRAL_BROWSER_AUTH_API_BASE_URL,
+        )
+    ]
+
+
+@pytest.mark.asyncio
+async def test_ui_keeps_browser_sign_in_disabled_from_false_env_var(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setenv("VIBE_ENABLE_EXPERIMENTAL_BROWSER_SIGN_IN", "false")
+
+    app = OnboardingApp()
+
+    assert app.supports_browser_sign_in is False
+
+    async with app.run_test() as pilot:
+        await _pass_welcome_screen(pilot)
+        await _wait_for(lambda: isinstance(pilot.app.screen, ApiKeyScreen), pilot)
+
+
+@pytest.mark.asyncio
+async def test_ui_preserves_custom_browser_auth_urls_when_api_key_is_missing(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+    monkeypatch.setenv("VIBE_HOME", str(tmp_path))
+    config_file = tmp_path / "config.toml"
+    config_file.write_text(
+        "\n".join([
+            'active_model = "devstral-2"',
+            "enable_experimental_browser_sign_in = true",
+            "[[providers]]",
+            'name = "mistral"',
+            'api_base = "https://api.mistral.ai/v1"',
+            'api_key_env_var = "MISTRAL_API_KEY"',
+            'browser_auth_base_url = "http://127.0.0.1:8787"',
+            'browser_auth_api_base_url = "http://127.0.0.1:8787"',
+            'backend = "mistral"',
+            "",
+            "[[models]]",
+            'name = "mistral-vibe-cli-latest"',
+            'provider = "mistral"',
+            'alias = "devstral-2"',
+            "",
+        ]),
+        encoding="utf-8",
+    )
+    reset_harness_files_manager()
+    init_harness_files_manager("user")
+    captured_base_urls: list[tuple[str, str]] = []
+    _patch_failing_browser_sign_in_service(monkeypatch, captured_base_urls)
+
+    app = OnboardingApp()
+
+    assert app.supports_browser_sign_in is True
+
+    async with app.run_test() as pilot:
+        await _show_browser_sign_in(pilot)
+        await _wait_for(lambda: bool(captured_base_urls), pilot)
+
+    assert captured_base_urls == [("http://127.0.0.1:8787", "http://127.0.0.1:8787")]
+
+
+@pytest.mark.asyncio
+async def test_ui_falls_back_to_default_onboarding_context_with_invalid_active_model(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+    monkeypatch.setenv("VIBE_HOME", str(tmp_path))
+    config_file = tmp_path / "config.toml"
+    config_file.write_text(
+        "\n".join([
+            'active_model = "does-not-exist"',
+            "enable_experimental_browser_sign_in = true",
+            "",
+            "[[providers]]",
+            'name = "mistral"',
+            'api_base = "https://api.mistral.ai/v1"',
+            'api_key_env_var = "MISTRAL_API_KEY"',
+            'browser_auth_base_url = "https://console.mistral.ai"',
+            'browser_auth_api_base_url = "https://api.mistral.ai"',
+            'backend = "mistral"',
+            "",
+            "[[models]]",
+            'name = "mistral-vibe-cli-latest"',
+            'provider = "mistral"',
+            'alias = "devstral-2"',
+            "",
+        ]),
+        encoding="utf-8",
+    )
+    reset_harness_files_manager()
+    init_harness_files_manager("user")
+
+    app = OnboardingApp()
+
+    assert app.supports_browser_sign_in is True
+
+    async with app.run_test() as pilot:
+        await _show_auth_method(pilot)
 
 
 def test_api_key_screen_falls_back_to_mistral_for_provider_without_env_key() -> None:

tests/session/test_resume_sessions.py

@@ -87,13 +87,6 @@ class TestListRemoteResumeSessions:
             start_time=datetime(2026, 1, 1),
             end_time=None,
         )
-        retrying = WorkflowExecutionWithoutResultResponse(
-            workflow_name="vibe",
-            execution_id="exec-retrying",
-            status=WorkflowExecutionStatus.RETRYING_AFTER_ERROR,
-            start_time=datetime(2026, 1, 1),
-            end_time=None,
-        )
         continued = WorkflowExecutionWithoutResultResponse(
             workflow_name="vibe",
             execution_id="exec-continued",
@@ -102,9 +95,7 @@ class TestListRemoteResumeSessions:
             end_time=None,
         )
 
-        mock_response = WorkflowExecutionListResponse(
-            executions=[running, retrying, continued]
-        )
+        mock_response = WorkflowExecutionListResponse(executions=[running, continued])
 
         config = MagicMock()
         config.vibe_code_enabled = True
@@ -121,10 +112,9 @@ class TestListRemoteResumeSessions:
 
             result = await list_remote_resume_sessions(config)
 
-        assert len(result) == 3
+        assert len(result) == 2
         session_ids = {s.session_id for s in result}
         assert "exec-running" in session_ids
-        assert "exec-retrying" in session_ids
         assert "exec-continued" in session_ids
         assert all(s.source == "remote" for s in result)
 
@@ -133,7 +123,6 @@ class TestListRemoteResumeSessions:
             page_size=50,
             status=[
                 WorkflowExecutionStatus.RUNNING,
-                WorkflowExecutionStatus.RETRYING_AFTER_ERROR,
                 WorkflowExecutionStatus.CONTINUED_AS_NEW,
             ],
         )
@@ -207,7 +196,7 @@ class TestListRemoteResumeSessions:
         previous = WorkflowExecutionWithoutResultResponse(
             workflow_name="vibe",
             execution_id="exec-1",
-            status=WorkflowExecutionStatus.RETRYING_AFTER_ERROR,
+            status=WorkflowExecutionStatus.FAILED,
             start_time=datetime(2026, 1, 1),
             end_time=datetime(2026, 1, 10),
         )

tests/session/test_session_loader.py

@@ -346,7 +346,10 @@ class TestSessionLoaderFindLatestSession:
         assert result == valid_session
 
     def test_find_latest_session_skips_unreadable_messages_file(
-        self, session_config: SessionLoggingConfig, create_test_session
+        self,
+        session_config: SessionLoggingConfig,
+        create_test_session,
+        monkeypatch: pytest.MonkeyPatch,
     ) -> None:
         session_dir = Path(session_config.save_dir)
 
@@ -355,14 +358,28 @@ class TestSessionLoaderFindLatestSession:
 
         unreadable_session = create_test_session(session_dir, "unreadab-session")
         unreadable_messages = unreadable_session / "messages.jsonl"
-        unreadable_messages.chmod(0)
+
+        # chmod doesn't restrict root, so simulate an unreadable file by
+        # patching Path.read_bytes (used under the hood by read_safe). This
+        # keeps the test working in CI environments running as root.
+        original_read_bytes = Path.read_bytes
+
+        def fake_read_bytes(self: Path) -> bytes:
+            if self == unreadable_messages:
+                raise PermissionError(f"Permission denied: {self}")
+            return original_read_bytes(self)
+
+        monkeypatch.setattr(Path, "read_bytes", fake_read_bytes)
 
         result = SessionLoader.find_latest_session(session_config)
         assert result is not None
         assert result == valid_session
 
     def test_find_latest_session_skips_unreadable_metadata_file(
-        self, session_config: SessionLoggingConfig, create_test_session
+        self,
+        session_config: SessionLoggingConfig,
+        create_test_session,
+        monkeypatch: pytest.MonkeyPatch,
     ) -> None:
         session_dir = Path(session_config.save_dir)
 
@@ -371,7 +388,15 @@ class TestSessionLoaderFindLatestSession:
 
         unreadable_session = create_test_session(session_dir, "unreadab-session")
         unreadable_metadata = unreadable_session / "meta.json"
-        unreadable_metadata.chmod(0)
+
+        original_read_bytes = Path.read_bytes
+
+        def fake_read_bytes(self: Path) -> bytes:
+            if self == unreadable_metadata:
+                raise PermissionError(f"Permission denied: {self}")
+            return original_read_bytes(self)
+
+        monkeypatch.setattr(Path, "read_bytes", fake_read_bytes)
 
         result = SessionLoader.find_latest_session(session_config)
         assert result is not None
@@ -436,6 +461,41 @@ class TestSessionLoaderFindSessionById:
         assert result is not None
         assert result == session_2
 
+    def test_find_session_by_id_filters_by_working_directory(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        session_dir = Path(session_config.save_dir)
+
+        session_a = create_test_session(
+            session_dir,
+            "abcd1234-session",
+            working_directory=Path("/home/user/project-a"),
+        )
+        time.sleep(0.01)
+        session_b = create_test_session(
+            session_dir,
+            "abcd1234-session",
+            working_directory=Path("/home/user/project-b"),
+        )
+
+        assert (
+            SessionLoader.find_session_by_id(
+                "abcd1234",
+                session_config,
+                working_directory=Path("/home/user/project-a"),
+            )
+            == session_a
+        )
+        assert (
+            SessionLoader.find_session_by_id(
+                "abcd1234",
+                session_config,
+                working_directory=Path("/home/user/project-c"),
+            )
+            is None
+        )
+        assert SessionLoader.find_session_by_id("abcd1234", session_config) == session_b
+
     def test_find_session_by_id_no_match(
         self, session_config: SessionLoggingConfig, create_test_session
     ) -> None:

tests/session/test_session_logger.py

@@ -11,6 +11,7 @@ import pytest
 from tests.conftest import build_test_vibe_config
 from vibe.core.agents.models import AgentProfile, AgentSafety
 from vibe.core.config import SessionLoggingConfig, VibeConfig
+from vibe.core.experiments.models import EvalResponse
 from vibe.core.loop import ScheduledLoop
 from vibe.core.session.session_logger import SessionLogger
 from vibe.core.tools.manager import ToolManager
@@ -114,16 +115,12 @@ class TestSessionLoggerMetadata:
         self, mock_getuser, mock_subprocess, session_config: SessionLoggingConfig
     ) -> None:
         """Test that session metadata is correctly initialized."""
-        # Mock git commands
-        git_commit_mock = MagicMock()
-        git_commit_mock.returncode = 0
-        git_commit_mock.stdout = "abc123\n"
+        # Mock combined git command
+        git_mock = MagicMock()
+        git_mock.returncode = 0
+        git_mock.stdout = "abc123\nmain\n"
 
-        git_branch_mock = MagicMock()
-        git_branch_mock.returncode = 0
-        git_branch_mock.stdout = "main\n"
-
-        mock_subprocess.side_effect = [git_commit_mock, git_branch_mock]
+        mock_subprocess.return_value = git_mock
         mock_getuser.return_value = "testuser"
 
         session_id = "test-session-123"
@@ -149,7 +146,7 @@ class TestSessionLoggerMetadata:
         self, mock_getuser, mock_subprocess, session_config: SessionLoggingConfig
     ) -> None:
         """Test that session metadata handles git command errors gracefully."""
-        # Mock git commands to fail
+        # Mock combined git command to fail
         mock_subprocess.side_effect = FileNotFoundError("git not found")
         mock_getuser.return_value = "testuser"
 
@@ -1002,3 +999,153 @@ class TestPersistLoops:
             metadata = json.load(f)
         assert len(metadata["loops"]) == 1
         assert metadata["loops"][0]["id"] == "aabbccdd"
+
+
+class TestPersistExperiments:
+    @pytest.fixture
+    def sample_response(self) -> EvalResponse:
+        return EvalResponse.model_validate({
+            "features": {
+                "vibe_code_cli_test_ab": {
+                    "defaultValue": "cli",
+                    "rules": [
+                        {
+                            "force": "cli_v2",
+                            "tracks": [
+                                {
+                                    "experiment": {"key": "vibe_code_cli_test_ab"},
+                                    "result": {
+                                        "key": "1",
+                                        "variationId": 1,
+                                        "inExperiment": True,
+                                    },
+                                }
+                            ],
+                        }
+                    ],
+                }
+            }
+        })
+
+    @pytest.mark.asyncio
+    async def test_writes_field_into_existing_metadata(
+        self,
+        session_config: SessionLoggingConfig,
+        mock_vibe_config: VibeConfig,
+        mock_tool_manager: ToolManager,
+        mock_agent_profile: AgentProfile,
+        sample_response: EvalResponse,
+    ) -> None:
+        logger = SessionLogger(session_config, "exp-session")
+        await logger.save_interaction(
+            messages=[
+                LLMMessage(role=Role.system, content="System prompt"),
+                LLMMessage(role=Role.user, content="Hello"),
+            ],
+            stats=AgentStats(steps=1),
+            base_config=mock_vibe_config,
+            tool_manager=mock_tool_manager,
+            agent_profile=mock_agent_profile,
+        )
+
+        await logger.persist_experiments(sample_response)
+
+        assert logger.session_dir is not None
+        with open(logger.session_dir / "meta.json") as f:
+            metadata = json.load(f)
+        assert "experiments" in metadata
+        assert (
+            metadata["experiments"]["features"]["vibe_code_cli_test_ab"]["defaultValue"]
+            == "cli"
+        )
+
+    @pytest.mark.asyncio
+    async def test_persists_none_as_null(
+        self,
+        session_config: SessionLoggingConfig,
+        mock_vibe_config: VibeConfig,
+        mock_tool_manager: ToolManager,
+        mock_agent_profile: AgentProfile,
+    ) -> None:
+        logger = SessionLogger(session_config, "exp-none")
+        await logger.save_interaction(
+            messages=[
+                LLMMessage(role=Role.system, content="x"),
+                LLMMessage(role=Role.user, content="y"),
+            ],
+            stats=AgentStats(steps=1),
+            base_config=mock_vibe_config,
+            tool_manager=mock_tool_manager,
+            agent_profile=mock_agent_profile,
+        )
+
+        await logger.persist_experiments(None)
+
+        assert logger.session_dir is not None
+        with open(logger.session_dir / "meta.json") as f:
+            metadata = json.load(f)
+        assert metadata.get("experiments") is None
+
+    @pytest.mark.asyncio
+    async def test_does_not_create_metadata_file_when_missing(
+        self, session_config: SessionLoggingConfig, sample_response: EvalResponse
+    ) -> None:
+        # Sessions without any message must not be persisted at all —
+        # persist_experiments updates only in-memory state when meta.json is
+        # absent, and lets the eventual save_interaction write it.
+        logger = SessionLogger(session_config, "fresh-session")
+        assert logger.session_dir is not None
+        assert not (logger.session_dir / "meta.json").exists()
+
+        await logger.persist_experiments(sample_response)
+
+        assert not (logger.session_dir / "meta.json").exists()
+        assert logger.session_metadata is not None
+        assert logger.session_metadata.experiments == sample_response
+
+    @pytest.mark.asyncio
+    async def test_noop_when_logging_disabled(
+        self,
+        disabled_session_config: SessionLoggingConfig,
+        sample_response: EvalResponse,
+    ) -> None:
+        logger = SessionLogger(disabled_session_config, "ignored")
+        await logger.persist_experiments(sample_response)
+
+    @pytest.mark.asyncio
+    async def test_first_save_interaction_includes_in_memory_experiments(
+        self,
+        session_config: SessionLoggingConfig,
+        mock_vibe_config: VibeConfig,
+        mock_tool_manager: ToolManager,
+        mock_agent_profile: AgentProfile,
+        sample_response: EvalResponse,
+    ) -> None:
+        # Real flow: persist_experiments at session start (no meta.json yet,
+        # in-memory only). The first save_interaction must succeed AND
+        # include the experiments snapshot in the eventual meta.json.
+        logger = SessionLogger(session_config, "first-save-after-experiments")
+        await logger.persist_experiments(sample_response)
+
+        assert logger.session_dir is not None
+        assert not (logger.session_dir / "meta.json").exists()
+
+        await logger.save_interaction(
+            messages=[
+                LLMMessage(role=Role.system, content="System prompt"),
+                LLMMessage(role=Role.user, content="Hello"),
+                LLMMessage(role=Role.assistant, content="Hi"),
+            ],
+            stats=AgentStats(steps=1),
+            base_config=mock_vibe_config,
+            tool_manager=mock_tool_manager,
+            agent_profile=mock_agent_profile,
+        )
+
+        with open(logger.session_dir / "meta.json") as f:
+            metadata = json.load(f)
+        assert metadata["total_messages"] == 2
+        assert (
+            metadata["experiments"]["features"]["vibe_code_cli_test_ab"]["defaultValue"]
+            == "cli"
+        )

tests/snapshots/__snapshots__/test_ui_snapshot_basic_conversation/test_snapshot_shows_basic_conversation.svg

@@ -187,7 +187,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="305" clip-path="url(#terminal-line-25)">Hello&#160;there,&#160;who&#160;are&#160;you?</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="500.2" clip-path="url(#terminal-line-27)">I&#x27;m&#160;the&#160;Vibe&#160;agent&#160;and&#160;I&#x27;m&#160;ready&#160;to&#160;help.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="500.2" clip-path="url(#terminal-line-27)">I&#x27;m&#160;the&#160;Vibe&#160;agent&#160;and&#160;I&#x27;m&#160;ready&#160;to&#160;help.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r5" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_code_block_horizontal_scrolling/test_snapshot_allows_horizontal_scrolling_for_long_code_blocks.svg

@@ -163,7 +163,7 @@
             </g>
         
     <g transform="translate(9, 41)" clip-path="url(#terminal-clip-terminal)">
-    <rect fill="#4b4e55" x="12.2" y="611.5" width="158.6" height="24.65" shape-rendering="crispEdges"/>
+    <rect fill="#4b4e55" x="36.6" y="611.5" width="146.4" height="24.65" shape-rendering="crispEdges"/>
     <g class="terminal-matrix">
     <text class="terminal-r1" x="1464" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
 </text><text class="terminal-r1" x="1464" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
@@ -185,14 +185,14 @@
 </text><text class="terminal-r1" x="0" y="434.8" textLength="134.2" clip-path="url(#terminal-line-17)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="434.8" textLength="61" clip-path="url(#terminal-line-17)">Type&#160;</text><text class="terminal-r3" x="231.8" y="434.8" textLength="61" clip-path="url(#terminal-line-17)">/help</text><text class="terminal-r1" x="292.8" y="434.8" textLength="256.2" clip-path="url(#terminal-line-17)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
 </text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
 </text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r1" x="0" y="508" textLength="451.4" clip-path="url(#terminal-line-20)">Here&#x27;s&#160;a&#160;very&#160;long&#160;print&#160;instruction:</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r1" x="24.4" y="508" textLength="451.4" clip-path="url(#terminal-line-20)">Here&#x27;s&#160;a&#160;very&#160;long&#160;print&#160;instruction:</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
-</text><text class="terminal-r1" x="0" y="556.8" textLength="36.6" clip-path="url(#terminal-line-22)">():</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r4" x="0" y="581.2" textLength="329.4" clip-path="url(#terminal-line-23)">ery&#160;long&#160;line&#160;(Lorem&#160;Ipsum)</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
-</text><text class="terminal-r5" x="0" y="605.6" textLength="1390.8" clip-path="url(#terminal-line-24)">m&#160;ipsum&#160;dolor&#160;sit&#160;amet,&#160;consectetur&#160;adipiscing&#160;elit.&#160;Sed&#160;do&#160;eiusmod&#160;tempor&#160;incididunt&#160;ut&#160;labore&#160;et&#160;dolore&#160;magna&#160;al</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
-</text><text class="terminal-r7" x="12.2" y="630" textLength="158.6" clip-path="url(#terminal-line-25)">▋&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;</text><text class="terminal-r8" x="170.8" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">▍</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
+</text><text class="terminal-r1" x="24.4" y="556.8" textLength="36.6" clip-path="url(#terminal-line-22)">():</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
+</text><text class="terminal-r4" x="24.4" y="581.2" textLength="329.4" clip-path="url(#terminal-line-23)">ery&#160;long&#160;line&#160;(Lorem&#160;Ipsum)</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r5" x="24.4" y="605.6" textLength="1366.4" clip-path="url(#terminal-line-24)">m&#160;ipsum&#160;dolor&#160;sit&#160;amet,&#160;consectetur&#160;adipiscing&#160;elit.&#160;Sed&#160;do&#160;eiusmod&#160;tempor&#160;incididunt&#160;ut&#160;labore&#160;et&#160;dolore&#160;magna&#160;</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
+</text><text class="terminal-r7" x="36.6" y="630" textLength="146.4" clip-path="url(#terminal-line-25)">▋&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;</text><text class="terminal-r8" x="183" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">▉</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="48.8" clip-path="url(#terminal-line-27)">The&#160;</text><text class="terminal-r9" x="48.8" y="678.8" textLength="61" clip-path="url(#terminal-line-27)">print</text><text class="terminal-r1" x="109.8" y="678.8" textLength="1085.8" clip-path="url(#terminal-line-27)">&#160;statement&#160;includes&#160;a&#160;very&#160;long&#160;line&#160;of&#160;Lorem&#160;Ipsum&#160;text&#160;to&#160;demonstrate&#160;a&#160;lengthy&#160;output.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="48.8" clip-path="url(#terminal-line-27)">The&#160;</text><text class="terminal-r9" x="73.2" y="678.8" textLength="61" clip-path="url(#terminal-line-27)">print</text><text class="terminal-r1" x="134.2" y="678.8" textLength="1085.8" clip-path="url(#terminal-line-27)">&#160;statement&#160;includes&#160;a&#160;very&#160;long&#160;line&#160;of&#160;Lorem&#160;Ipsum&#160;text&#160;to&#160;demonstrate&#160;a&#160;lengthy&#160;output.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r10" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_empty_assistant_before_reasoning/test_snapshot_empty_assistant_removed_when_reasoning_starts.svg

@@ -189,7 +189,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r5" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">✓</text><text class="terminal-r6" x="24.4" y="630" textLength="85.4" clip-path="url(#terminal-line-25)">Thought</text><text class="terminal-r6" x="122" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">▶</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="207.4" clip-path="url(#terminal-line-27)">The&#160;answer&#160;is&#160;42.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="207.4" clip-path="url(#terminal-line-27)">The&#160;answer&#160;is&#160;42.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_feedback_bar/test_snapshot_feedback_bar_visible.svg

@@ -188,7 +188,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="61" clip-path="url(#terminal-line-25)">Hello</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="329.4" clip-path="url(#terminal-line-27)">Sure,&#160;I&#160;can&#160;help&#160;with&#160;that.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="329.4" clip-path="url(#terminal-line-27)">Sure,&#160;I&#160;can&#160;help&#160;with&#160;that.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="841.8" y="727.6" textLength="329.4" clip-path="url(#terminal-line-29)">How&#160;is&#160;Vibe&#160;doing&#160;so&#160;far?&#160;&#160;</text><text class="terminal-r5" x="1171.2" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">1</text><text class="terminal-r1" x="1183.4" y="727.6" textLength="97.6" clip-path="url(#terminal-line-29)">:&#160;good&#160;&#160;</text><text class="terminal-r5" x="1281" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">2</text><text class="terminal-r1" x="1293.2" y="727.6" textLength="97.6" clip-path="url(#terminal-line-29)">:&#160;fine&#160;&#160;</text><text class="terminal-r5" x="1390.8" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">3</text><text class="terminal-r1" x="1403" y="727.6" textLength="61" clip-path="url(#terminal-line-29)">:&#160;bad</text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r6" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_mcp_command/test_snapshot_mcp_no_servers.svg

@@ -19,183 +19,183 @@
         font-weight: 700;
     }
 
-    .terminal-9108451134-matrix {
+    .terminal-matrix {
         font-family: Fira Code, monospace;
         font-size: 20px;
         line-height: 24.4px;
         font-variant-east-asian: full-width;
     }
 
-    .terminal-9108451134-title {
+    .terminal-title {
         font-size: 18px;
         font-weight: bold;
         font-family: arial;
     }
 
-    .terminal-9108451134-r1 { fill: #c5c8c6 }
-.terminal-9108451134-r2 { fill: #ff8205;font-weight: bold }
-.terminal-9108451134-r3 { fill: #68a0b3 }
-.terminal-9108451134-r4 { fill: #ff8205 }
-.terminal-9108451134-r5 { fill: #9a9b99 }
+    .terminal-r1 { fill: #c5c8c6 }
+.terminal-r2 { fill: #ff8205;font-weight: bold }
+.terminal-r3 { fill: #68a0b3 }
+.terminal-r4 { fill: #ff8205 }
+.terminal-r5 { fill: #9a9b99 }
     </style>
 
     <defs>
-    <clipPath id="terminal-9108451134-clip-terminal">
+    <clipPath id="terminal-clip-terminal">
       <rect x="0" y="0" width="1463.0" height="877.4" />
     </clipPath>
-    <clipPath id="terminal-9108451134-line-0">
+    <clipPath id="terminal-line-0">
     <rect x="0" y="1.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-1">
+<clipPath id="terminal-line-1">
     <rect x="0" y="25.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-2">
+<clipPath id="terminal-line-2">
     <rect x="0" y="50.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-3">
+<clipPath id="terminal-line-3">
     <rect x="0" y="74.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-4">
+<clipPath id="terminal-line-4">
     <rect x="0" y="99.1" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-5">
+<clipPath id="terminal-line-5">
     <rect x="0" y="123.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-6">
+<clipPath id="terminal-line-6">
     <rect x="0" y="147.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-7">
+<clipPath id="terminal-line-7">
     <rect x="0" y="172.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-8">
+<clipPath id="terminal-line-8">
     <rect x="0" y="196.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-9">
+<clipPath id="terminal-line-9">
     <rect x="0" y="221.1" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-10">
+<clipPath id="terminal-line-10">
     <rect x="0" y="245.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-11">
+<clipPath id="terminal-line-11">
     <rect x="0" y="269.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-12">
+<clipPath id="terminal-line-12">
     <rect x="0" y="294.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-13">
+<clipPath id="terminal-line-13">
     <rect x="0" y="318.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-14">
+<clipPath id="terminal-line-14">
     <rect x="0" y="343.1" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-15">
+<clipPath id="terminal-line-15">
     <rect x="0" y="367.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-16">
+<clipPath id="terminal-line-16">
     <rect x="0" y="391.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-17">
+<clipPath id="terminal-line-17">
     <rect x="0" y="416.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-18">
+<clipPath id="terminal-line-18">
     <rect x="0" y="440.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-19">
+<clipPath id="terminal-line-19">
     <rect x="0" y="465.1" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-20">
+<clipPath id="terminal-line-20">
     <rect x="0" y="489.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-21">
+<clipPath id="terminal-line-21">
     <rect x="0" y="513.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-22">
+<clipPath id="terminal-line-22">
     <rect x="0" y="538.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-23">
+<clipPath id="terminal-line-23">
     <rect x="0" y="562.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-24">
+<clipPath id="terminal-line-24">
     <rect x="0" y="587.1" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-25">
+<clipPath id="terminal-line-25">
     <rect x="0" y="611.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-26">
+<clipPath id="terminal-line-26">
     <rect x="0" y="635.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-27">
+<clipPath id="terminal-line-27">
     <rect x="0" y="660.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-28">
+<clipPath id="terminal-line-28">
     <rect x="0" y="684.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-29">
+<clipPath id="terminal-line-29">
     <rect x="0" y="709.1" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-30">
+<clipPath id="terminal-line-30">
     <rect x="0" y="733.5" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-31">
+<clipPath id="terminal-line-31">
     <rect x="0" y="757.9" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-32">
+<clipPath id="terminal-line-32">
     <rect x="0" y="782.3" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-33">
+<clipPath id="terminal-line-33">
     <rect x="0" y="806.7" width="1464" height="24.65"/>
             </clipPath>
-<clipPath id="terminal-9108451134-line-34">
+<clipPath id="terminal-line-34">
     <rect x="0" y="831.1" width="1464" height="24.65"/>
             </clipPath>
     </defs>
 
-    <rect fill="#292929" stroke="rgba(255,255,255,0.35)" stroke-width="1" x="1" y="1" width="1480" height="926.4" rx="8"/><text class="terminal-9108451134-title" fill="#c5c8c6" text-anchor="middle" x="740" y="27">SnapshotTestAppNoMcpServers</text>
+    <rect fill="#292929" stroke="rgba(255,255,255,0.35)" stroke-width="1" x="1" y="1" width="1480" height="926.4" rx="8"/><text class="terminal-title" fill="#c5c8c6" text-anchor="middle" x="740" y="27">SnapshotTestAppNoMcpServers</text>
             <g transform="translate(26,22)">
             <circle cx="0" cy="0" r="7" fill="#ff5f57"/>
             <circle cx="22" cy="0" r="7" fill="#febc2e"/>
             <circle cx="44" cy="0" r="7" fill="#28c840"/>
             </g>
         
-    <g transform="translate(9, 41)" clip-path="url(#terminal-9108451134-clip-terminal)">
+    <g transform="translate(9, 41)" clip-path="url(#terminal-clip-terminal)">
     
-    <g class="terminal-9108451134-matrix">
-    <text class="terminal-9108451134-r1" x="1464" y="20" textLength="12.2" clip-path="url(#terminal-9108451134-line-0)">
-</text><text class="terminal-9108451134-r1" x="1464" y="44.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-1)">
-</text><text class="terminal-9108451134-r1" x="1464" y="68.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-2)">
-</text><text class="terminal-9108451134-r1" x="1464" y="93.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-3)">
-</text><text class="terminal-9108451134-r1" x="1464" y="117.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-4)">
-</text><text class="terminal-9108451134-r1" x="1464" y="142" textLength="12.2" clip-path="url(#terminal-9108451134-line-5)">
-</text><text class="terminal-9108451134-r1" x="1464" y="166.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-6)">
-</text><text class="terminal-9108451134-r1" x="1464" y="190.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-7)">
-</text><text class="terminal-9108451134-r1" x="1464" y="215.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-8)">
-</text><text class="terminal-9108451134-r1" x="1464" y="239.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-9)">
-</text><text class="terminal-9108451134-r1" x="1464" y="264" textLength="12.2" clip-path="url(#terminal-9108451134-line-10)">
-</text><text class="terminal-9108451134-r1" x="1464" y="288.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-11)">
-</text><text class="terminal-9108451134-r1" x="1464" y="312.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-12)">
-</text><text class="terminal-9108451134-r1" x="1464" y="337.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-13)">
-</text><text class="terminal-9108451134-r1" x="1464" y="361.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-14)">
-</text><text class="terminal-9108451134-r1" x="1464" y="386" textLength="12.2" clip-path="url(#terminal-9108451134-line-15)">
-</text><text class="terminal-9108451134-r1" x="1464" y="410.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-16)">
-</text><text class="terminal-9108451134-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-17)">
-</text><text class="terminal-9108451134-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-18)">
-</text><text class="terminal-9108451134-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-19)">
-</text><text class="terminal-9108451134-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-9108451134-line-20)">
-</text><text class="terminal-9108451134-r1" x="0" y="532.4" textLength="134.2" clip-path="url(#terminal-9108451134-line-21)">&#160;&#160;⡠⣒⠄&#160;&#160;⡔⢄⠔⡄</text><text class="terminal-9108451134-r2" x="170.8" y="532.4" textLength="146.4" clip-path="url(#terminal-9108451134-line-21)">Mistral&#160;Vibe</text><text class="terminal-9108451134-r1" x="317.2" y="532.4" textLength="122" clip-path="url(#terminal-9108451134-line-21)">&#160;v0.0.0&#160;·&#160;</text><text class="terminal-9108451134-r3" x="439.2" y="532.4" textLength="244" clip-path="url(#terminal-9108451134-line-21)">devstral-latest[off]</text><text class="terminal-9108451134-r1" x="683.2" y="532.4" textLength="256.2" clip-path="url(#terminal-9108451134-line-21)">&#160;·&#160;[Subscription]&#160;Pro</text><text class="terminal-9108451134-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-21)">
-</text><text class="terminal-9108451134-r1" x="0" y="556.8" textLength="134.2" clip-path="url(#terminal-9108451134-line-22)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-9108451134-r1" x="170.8" y="556.8" textLength="414.8" clip-path="url(#terminal-9108451134-line-22)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-9108451134-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-22)">
-</text><text class="terminal-9108451134-r1" x="0" y="581.2" textLength="134.2" clip-path="url(#terminal-9108451134-line-23)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-9108451134-r1" x="170.8" y="581.2" textLength="61" clip-path="url(#terminal-9108451134-line-23)">Type&#160;</text><text class="terminal-9108451134-r3" x="231.8" y="581.2" textLength="61" clip-path="url(#terminal-9108451134-line-23)">/help</text><text class="terminal-9108451134-r1" x="292.8" y="581.2" textLength="256.2" clip-path="url(#terminal-9108451134-line-23)">&#160;for&#160;more&#160;information</text><text class="terminal-9108451134-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-23)">
-</text><text class="terminal-9108451134-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-24)">
-</text><text class="terminal-9108451134-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-9108451134-line-25)">
-</text><text class="terminal-9108451134-r4" x="0" y="654.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-26)">┃</text><text class="terminal-9108451134-r2" x="24.4" y="654.4" textLength="48.8" clip-path="url(#terminal-9108451134-line-26)">/mcp</text><text class="terminal-9108451134-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-26)">
-</text><text class="terminal-9108451134-r5" x="24.4" y="678.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-27)">⎣</text><text class="terminal-9108451134-r1" x="48.8" y="678.8" textLength="317.2" clip-path="url(#terminal-9108451134-line-27)">No&#160;MCP&#160;servers&#160;configured.</text><text class="terminal-9108451134-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-27)">
-</text><text class="terminal-9108451134-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-28)">
-</text><text class="terminal-9108451134-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-29)">
-</text><text class="terminal-9108451134-r5" x="0" y="752" textLength="1464" clip-path="url(#terminal-9108451134-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-9108451134-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-9108451134-line-30)">
-</text><text class="terminal-9108451134-r5" x="0" y="776.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-31)">│</text><text class="terminal-9108451134-r2" x="24.4" y="776.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-31)">&gt;</text><text class="terminal-9108451134-r5" x="1451.8" y="776.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-31)">│</text><text class="terminal-9108451134-r1" x="1464" y="776.4" textLength="12.2" clip-path="url(#terminal-9108451134-line-31)">
-</text><text class="terminal-9108451134-r5" x="0" y="800.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-32)">│</text><text class="terminal-9108451134-r5" x="1451.8" y="800.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-32)">│</text><text class="terminal-9108451134-r1" x="1464" y="800.8" textLength="12.2" clip-path="url(#terminal-9108451134-line-32)">
-</text><text class="terminal-9108451134-r5" x="0" y="825.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-33)">│</text><text class="terminal-9108451134-r5" x="1451.8" y="825.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-33)">│</text><text class="terminal-9108451134-r1" x="1464" y="825.2" textLength="12.2" clip-path="url(#terminal-9108451134-line-33)">
-</text><text class="terminal-9108451134-r5" x="0" y="849.6" textLength="1464" clip-path="url(#terminal-9108451134-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-9108451134-r1" x="1464" y="849.6" textLength="12.2" clip-path="url(#terminal-9108451134-line-34)">
-</text><text class="terminal-9108451134-r5" x="0" y="874" textLength="158.6" clip-path="url(#terminal-9108451134-line-35)">/test/workdir</text><text class="terminal-9108451134-r5" x="1256.6" y="874" textLength="207.4" clip-path="url(#terminal-9108451134-line-35)">0%&#160;of&#160;200k&#160;tokens</text>
+    <g class="terminal-matrix">
+    <text class="terminal-r1" x="1464" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r1" x="1464" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+</text><text class="terminal-r1" x="1464" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
+</text><text class="terminal-r1" x="1464" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r1" x="1464" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
+</text><text class="terminal-r1" x="1464" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
+</text><text class="terminal-r1" x="1464" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="1464" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="1464" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text><text class="terminal-r1" x="1464" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
+</text><text class="terminal-r1" x="1464" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
+</text><text class="terminal-r1" x="1464" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
+</text><text class="terminal-r1" x="1464" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
+</text><text class="terminal-r1" x="1464" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
+</text><text class="terminal-r1" x="1464" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
+</text><text class="terminal-r1" x="1464" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
+</text><text class="terminal-r1" x="1464" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
+</text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
+</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
+</text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
+</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r1" x="0" y="532.4" textLength="134.2" clip-path="url(#terminal-line-21)">&#160;&#160;⡠⣒⠄&#160;&#160;⡔⢄⠔⡄</text><text class="terminal-r2" x="170.8" y="532.4" textLength="146.4" clip-path="url(#terminal-line-21)">Mistral&#160;Vibe</text><text class="terminal-r1" x="317.2" y="532.4" textLength="122" clip-path="url(#terminal-line-21)">&#160;v0.0.0&#160;·&#160;</text><text class="terminal-r3" x="439.2" y="532.4" textLength="244" clip-path="url(#terminal-line-21)">devstral-latest[off]</text><text class="terminal-r1" x="683.2" y="532.4" textLength="256.2" clip-path="url(#terminal-line-21)">&#160;·&#160;[Subscription]&#160;Pro</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
+</text><text class="terminal-r1" x="0" y="556.8" textLength="134.2" clip-path="url(#terminal-line-22)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="556.8" textLength="414.8" clip-path="url(#terminal-line-22)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
+</text><text class="terminal-r1" x="0" y="581.2" textLength="134.2" clip-path="url(#terminal-line-23)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="581.2" textLength="61" clip-path="url(#terminal-line-23)">Type&#160;</text><text class="terminal-r3" x="231.8" y="581.2" textLength="61" clip-path="url(#terminal-line-23)">/help</text><text class="terminal-r1" x="292.8" y="581.2" textLength="256.2" clip-path="url(#terminal-line-23)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
+</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
+</text><text class="terminal-r4" x="0" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">┃</text><text class="terminal-r2" x="24.4" y="654.4" textLength="48.8" clip-path="url(#terminal-line-26)">/mcp</text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
+</text><text class="terminal-r5" x="24.4" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">⎣</text><text class="terminal-r1" x="48.8" y="678.8" textLength="488" clip-path="url(#terminal-line-27)">No&#160;MCP&#160;servers&#160;or&#160;connectors&#160;configured.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
+</text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
+</text><text class="terminal-r5" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">
+</text><text class="terminal-r5" x="0" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r2" x="24.4" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">&gt;</text><text class="terminal-r5" x="1451.8" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r1" x="1464" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">
+</text><text class="terminal-r5" x="0" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r5" x="1451.8" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r1" x="1464" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">
+</text><text class="terminal-r5" x="0" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r5" x="1451.8" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r1" x="1464" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">
+</text><text class="terminal-r5" x="0" y="849.6" textLength="1464" clip-path="url(#terminal-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1464" y="849.6" textLength="12.2" clip-path="url(#terminal-line-34)">
+</text><text class="terminal-r5" x="0" y="874" textLength="158.6" clip-path="url(#terminal-line-35)">/test/workdir</text><text class="terminal-r5" x="1256.6" y="874" textLength="207.4" clip-path="url(#terminal-line-35)">0%&#160;of&#160;200k&#160;tokens</text>
     </g>
     </g>
 </svg>

tests/snapshots/__snapshots__/test_ui_snapshot_narrator_flow/test_snapshot_narrator_idle_after_speaking.svg

@@ -187,7 +187,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="61" clip-path="url(#terminal-line-25)">Hello</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">Hello!&#160;I&#160;can&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">Hello!&#160;I&#160;can&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r5" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_narrator_flow/test_snapshot_narrator_speaking.svg

@@ -189,7 +189,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="61" clip-path="url(#terminal-line-25)">Hello</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">Hello!&#160;I&#160;can&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">Hello!&#160;I&#160;can&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r5" x="0" y="727.6" textLength="36.6" clip-path="url(#terminal-line-29)">▂▅▇</text><text class="terminal-r1" x="36.6" y="727.6" textLength="122" clip-path="url(#terminal-line-29)">&#160;speaking&#160;</text><text class="terminal-r6" x="158.6" y="727.6" textLength="219.6" clip-path="url(#terminal-line-29)">Esc/Ctrl+C&#160;to&#160;stop</text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_narrator_flow/test_snapshot_narrator_summarizing.svg

@@ -189,7 +189,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="61" clip-path="url(#terminal-line-25)">Hello</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">Hello!&#160;I&#160;can&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">Hello!&#160;I&#160;can&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r5" x="0" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">█</text><text class="terminal-r1" x="12.2" y="727.6" textLength="158.6" clip-path="url(#terminal-line-29)">&#160;summarizing&#160;</text><text class="terminal-r6" x="170.8" y="727.6" textLength="219.6" clip-path="url(#terminal-line-29)">Esc/Ctrl+C&#160;to&#160;stop</text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_plan_file_message/test_snapshot_plan_file_message.svg

@@ -0,0 +1,202 @@
+<svg class="rich-terminal" viewBox="0 0 1482 928.4" xmlns="http://www.w3.org/2000/svg">
+    <!-- Generated with Rich https://www.textualize.io -->
+    <style>
+
+    @font-face {
+        font-family: "Fira Code";
+        src: local("FiraCode-Regular"),
+                url("https://cdnjs.cloudflare.com/ajax/libs/firacode/6.2.0/woff2/FiraCode-Regular.woff2") format("woff2"),
+                url("https://cdnjs.cloudflare.com/ajax/libs/firacode/6.2.0/woff/FiraCode-Regular.woff") format("woff");
+        font-style: normal;
+        font-weight: 400;
+    }
+    @font-face {
+        font-family: "Fira Code";
+        src: local("FiraCode-Bold"),
+                url("https://cdnjs.cloudflare.com/ajax/libs/firacode/6.2.0/woff2/FiraCode-Bold.woff2") format("woff2"),
+                url("https://cdnjs.cloudflare.com/ajax/libs/firacode/6.2.0/woff/FiraCode-Bold.woff") format("woff");
+        font-style: bold;
+        font-weight: 700;
+    }
+
+    .terminal-matrix {
+        font-family: Fira Code, monospace;
+        font-size: 20px;
+        line-height: 24.4px;
+        font-variant-east-asian: full-width;
+    }
+
+    .terminal-title {
+        font-size: 18px;
+        font-weight: bold;
+        font-family: arial;
+    }
+
+    .terminal-r1 { fill: #c5c8c6 }
+.terminal-r2 { fill: #ff8205;font-weight: bold }
+.terminal-r3 { fill: #68a0b3 }
+.terminal-r4 { fill: #9a9b99 }
+.terminal-r5 { fill: #608ab1;font-weight: bold }
+.terminal-r6 { fill: #608ab1;text-decoration: underline; }
+    </style>
+
+    <defs>
+    <clipPath id="terminal-clip-terminal">
+      <rect x="0" y="0" width="1463.0" height="877.4" />
+    </clipPath>
+    <clipPath id="terminal-line-0">
+    <rect x="0" y="1.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-1">
+    <rect x="0" y="25.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-2">
+    <rect x="0" y="50.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-3">
+    <rect x="0" y="74.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-4">
+    <rect x="0" y="99.1" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-5">
+    <rect x="0" y="123.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-6">
+    <rect x="0" y="147.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-7">
+    <rect x="0" y="172.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-8">
+    <rect x="0" y="196.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-9">
+    <rect x="0" y="221.1" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-10">
+    <rect x="0" y="245.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-11">
+    <rect x="0" y="269.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-12">
+    <rect x="0" y="294.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-13">
+    <rect x="0" y="318.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-14">
+    <rect x="0" y="343.1" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-15">
+    <rect x="0" y="367.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-16">
+    <rect x="0" y="391.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-17">
+    <rect x="0" y="416.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-18">
+    <rect x="0" y="440.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-19">
+    <rect x="0" y="465.1" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-20">
+    <rect x="0" y="489.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-21">
+    <rect x="0" y="513.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-22">
+    <rect x="0" y="538.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-23">
+    <rect x="0" y="562.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-24">
+    <rect x="0" y="587.1" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-25">
+    <rect x="0" y="611.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-26">
+    <rect x="0" y="635.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-27">
+    <rect x="0" y="660.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-28">
+    <rect x="0" y="684.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-29">
+    <rect x="0" y="709.1" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-30">
+    <rect x="0" y="733.5" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-31">
+    <rect x="0" y="757.9" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-32">
+    <rect x="0" y="782.3" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-33">
+    <rect x="0" y="806.7" width="1464" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-34">
+    <rect x="0" y="831.1" width="1464" height="24.65"/>
+            </clipPath>
+    </defs>
+
+    <rect fill="#292929" stroke="rgba(255,255,255,0.35)" stroke-width="1" x="1" y="1" width="1480" height="926.4" rx="8"/><text class="terminal-title" fill="#c5c8c6" text-anchor="middle" x="740" y="27">PlanFileMessageTestApp</text>
+            <g transform="translate(26,22)">
+            <circle cx="0" cy="0" r="7" fill="#ff5f57"/>
+            <circle cx="22" cy="0" r="7" fill="#febc2e"/>
+            <circle cx="44" cy="0" r="7" fill="#28c840"/>
+            </g>
+        
+    <g transform="translate(9, 41)" clip-path="url(#terminal-clip-terminal)">
+    
+    <g class="terminal-matrix">
+    <text class="terminal-r1" x="1464" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r1" x="1464" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+</text><text class="terminal-r1" x="1464" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
+</text><text class="terminal-r1" x="1464" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r1" x="1464" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
+</text><text class="terminal-r1" x="1464" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
+</text><text class="terminal-r1" x="1464" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="1464" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="0" y="215.2" textLength="134.2" clip-path="url(#terminal-line-8)">&#160;&#160;⡠⣒⠄&#160;&#160;⡔⢄⠔⡄</text><text class="terminal-r2" x="170.8" y="215.2" textLength="146.4" clip-path="url(#terminal-line-8)">Mistral&#160;Vibe</text><text class="terminal-r1" x="317.2" y="215.2" textLength="122" clip-path="url(#terminal-line-8)">&#160;v0.0.0&#160;·&#160;</text><text class="terminal-r3" x="439.2" y="215.2" textLength="244" clip-path="url(#terminal-line-8)">devstral-latest[off]</text><text class="terminal-r1" x="683.2" y="215.2" textLength="256.2" clip-path="url(#terminal-line-8)">&#160;·&#160;[Subscription]&#160;Pro</text><text class="terminal-r1" x="1464" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text><text class="terminal-r1" x="0" y="239.6" textLength="134.2" clip-path="url(#terminal-line-9)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="239.6" textLength="414.8" clip-path="url(#terminal-line-9)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-r1" x="1464" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
+</text><text class="terminal-r1" x="0" y="264" textLength="134.2" clip-path="url(#terminal-line-10)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="264" textLength="61" clip-path="url(#terminal-line-10)">Type&#160;</text><text class="terminal-r3" x="231.8" y="264" textLength="61" clip-path="url(#terminal-line-10)">/help</text><text class="terminal-r1" x="292.8" y="264" textLength="256.2" clip-path="url(#terminal-line-10)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1464" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
+</text><text class="terminal-r1" x="1464" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
+</text><text class="terminal-r1" x="1464" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
+</text><text class="terminal-r4" x="0" y="337.2" textLength="1464" clip-path="url(#terminal-line-13)">┌──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐</text><text class="terminal-r1" x="1464" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
+</text><text class="terminal-r4" x="0" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">│</text><text class="terminal-r5" x="610" y="361.6" textLength="231.8" clip-path="url(#terminal-line-14)">Implementation&#160;Plan</text><text class="terminal-r4" x="1451.8" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">│</text><text class="terminal-r1" x="1464" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
+</text><text class="terminal-r4" x="0" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">│</text><text class="terminal-r4" x="1451.8" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">│</text><text class="terminal-r1" x="1464" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
+</text><text class="terminal-r4" x="0" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">│</text><text class="terminal-r4" x="1451.8" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">│</text><text class="terminal-r1" x="1464" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
+</text><text class="terminal-r4" x="0" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">│</text><text class="terminal-r6" x="24.4" y="434.8" textLength="317.2" clip-path="url(#terminal-line-17)">1.&#160;Add&#160;user&#160;authentication</text><text class="terminal-r4" x="1451.8" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">│</text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
+</text><text class="terminal-r4" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">│</text><text class="terminal-r4" x="1451.8" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">│</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
+</text><text class="terminal-r4" x="0" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">│</text><text class="terminal-r1" x="24.4" y="483.6" textLength="268.4" clip-path="url(#terminal-line-19)">•&#160;JWT&#160;token&#160;validation</text><text class="terminal-r4" x="1451.8" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">│</text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
+</text><text class="terminal-r4" x="0" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r1" x="24.4" y="508" textLength="244" clip-path="url(#terminal-line-20)">•&#160;Session&#160;management</text><text class="terminal-r4" x="1451.8" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r4" x="0" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">│</text><text class="terminal-r4" x="1451.8" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">│</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
+</text><text class="terminal-r4" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r4" x="1451.8" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
+</text><text class="terminal-r4" x="0" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r6" x="24.4" y="581.2" textLength="268.4" clip-path="url(#terminal-line-23)">2.&#160;Database&#160;migrations</text><text class="terminal-r4" x="1451.8" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r4" x="0" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r4" x="1451.8" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
+</text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">│</text><text class="terminal-r1" x="24.4" y="630" textLength="244" clip-path="url(#terminal-line-25)">•&#160;Create&#160;users&#160;table</text><text class="terminal-r4" x="1451.8" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">│</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
+</text><text class="terminal-r4" x="0" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r1" x="24.4" y="654.4" textLength="158.6" clip-path="url(#terminal-line-26)">•&#160;Add&#160;indexes</text><text class="terminal-r4" x="1451.8" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
+</text><text class="terminal-r4" x="0" y="678.8" textLength="1464" clip-path="url(#terminal-line-27)">└──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
+</text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
+</text><text class="terminal-r4" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">
+</text><text class="terminal-r4" x="0" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r2" x="24.4" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">&gt;</text><text class="terminal-r4" x="1451.8" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r1" x="1464" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">
+</text><text class="terminal-r4" x="0" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r4" x="1451.8" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r1" x="1464" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">
+</text><text class="terminal-r4" x="0" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r4" x="1451.8" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r1" x="1464" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">
+</text><text class="terminal-r4" x="0" y="849.6" textLength="1464" clip-path="url(#terminal-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1464" y="849.6" textLength="12.2" clip-path="url(#terminal-line-34)">
+</text><text class="terminal-r4" x="0" y="874" textLength="158.6" clip-path="url(#terminal-line-35)">/test/workdir</text><text class="terminal-r4" x="1256.6" y="874" textLength="207.4" clip-path="url(#terminal-line-35)">0%&#160;of&#160;200k&#160;tokens</text>
+    </g>
+    </g>
+</svg>

tests/snapshots/__snapshots__/test_ui_snapshot_reasoning_content/test_snapshot_buffered_reasoning_yields_before_content.svg

@@ -189,7 +189,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r5" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">✓</text><text class="terminal-r6" x="24.4" y="630" textLength="85.4" clip-path="url(#terminal-line-25)">Thought</text><text class="terminal-r6" x="122" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">▶</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="707.6" clip-path="url(#terminal-line-27)">Here&#160;is&#160;my&#160;carefully&#160;considered&#160;answer.&#160;I&#160;hope&#160;this&#160;helps!</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="707.6" clip-path="url(#terminal-line-27)">Here&#160;is&#160;my&#160;carefully&#160;considered&#160;answer.&#160;I&#160;hope&#160;this&#160;helps!</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_reasoning_content/test_snapshot_shows_interleaved_reasoning.svg

@@ -185,11 +185,11 @@
 </text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r5" x="0" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">✓</text><text class="terminal-r6" x="24.4" y="532.4" textLength="85.4" clip-path="url(#terminal-line-21)">Thought</text><text class="terminal-r6" x="122" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">▶</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r1" x="0" y="581.2" textLength="439.2" clip-path="url(#terminal-line-23)">Here&#x27;s&#160;the&#160;first&#160;part&#160;of&#160;the&#160;answer.</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r1" x="24.4" y="581.2" textLength="439.2" clip-path="url(#terminal-line-23)">Here&#x27;s&#160;the&#160;first&#160;part&#160;of&#160;the&#160;answer.</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r5" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">✓</text><text class="terminal-r6" x="24.4" y="630" textLength="85.4" clip-path="url(#terminal-line-25)">Thought</text><text class="terminal-r6" x="122" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">▶</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="317.2" clip-path="url(#terminal-line-27)">And&#160;here&#x27;s&#160;the&#160;conclusion!</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="317.2" clip-path="url(#terminal-line-27)">And&#160;here&#x27;s&#160;the&#160;conclusion!</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_reasoning_content/test_snapshot_shows_reasoning_content.svg

@@ -189,7 +189,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r5" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">✓</text><text class="terminal-r6" x="24.4" y="630" textLength="85.4" clip-path="url(#terminal-line-25)">Thought</text><text class="terminal-r6" x="122" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">▶</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="768.6" clip-path="url(#terminal-line-27)">The&#160;answer&#160;to&#160;your&#160;question&#160;is&#160;42.&#160;This&#160;is&#160;the&#160;ultimate&#160;answer.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="768.6" clip-path="url(#terminal-line-27)">The&#160;answer&#160;to&#160;your&#160;question&#160;is&#160;42.&#160;This&#160;is&#160;the&#160;ultimate&#160;answer.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_reasoning_content/test_snapshot_shows_reasoning_content_expanded.svg

@@ -189,7 +189,7 @@
 </text><text class="terminal-r5" x="0" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">✓</text><text class="terminal-r6" x="24.4" y="605.6" textLength="85.4" clip-path="url(#terminal-line-24)">Thought</text><text class="terminal-r6" x="122" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">▼</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r6" x="24.4" y="630" textLength="1390.8" clip-path="url(#terminal-line-25)">Let&#160;me&#160;think&#160;about&#160;this&#160;step&#160;by&#160;step...&#160;First,&#160;I&#160;need&#160;to&#160;understand&#160;the&#160;question.&#160;Then&#160;I&#160;can&#160;formulate&#160;a&#160;response.</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="768.6" clip-path="url(#terminal-line-27)">The&#160;answer&#160;to&#160;your&#160;question&#160;is&#160;42.&#160;This&#160;is&#160;the&#160;ultimate&#160;answer.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="768.6" clip-path="url(#terminal-line-27)">The&#160;answer&#160;to&#160;your&#160;question&#160;is&#160;42.&#160;This&#160;is&#160;the&#160;ultimate&#160;answer.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_rewind/test_snapshot_rewind_error_shows_toast.svg

@@ -183,7 +183,7 @@
 </text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
 </text><text class="terminal-r4" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">┃</text><text class="terminal-r2" x="24.4" y="459.2" textLength="158.6" clip-path="url(#terminal-line-18)">first&#160;message</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
 </text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r1" x="0" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r1" x="24.4" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r4" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">┃</text><text class="terminal-r2" x="24.4" y="556.8" textLength="170.8" clip-path="url(#terminal-line-22)">second&#160;message</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
 </text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">

tests/snapshots/__snapshots__/test_ui_snapshot_rewind/test_snapshot_rewind_exit_on_escape.svg

@@ -183,7 +183,7 @@
 </text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r4" x="0" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">┃</text><text class="terminal-r2" x="24.4" y="532.4" textLength="158.6" clip-path="url(#terminal-line-21)">first&#160;message</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r1" x="0" y="581.2" textLength="317.2" clip-path="url(#terminal-line-23)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r1" x="24.4" y="581.2" textLength="317.2" clip-path="url(#terminal-line-23)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="170.8" clip-path="url(#terminal-line-25)">second&#160;message</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">

tests/snapshots/__snapshots__/test_ui_snapshot_rewind/test_snapshot_rewind_navigate_down.svg

@@ -182,7 +182,7 @@
 </text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
 </text><text class="terminal-r4" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">┃</text><text class="terminal-r2" x="24.4" y="459.2" textLength="158.6" clip-path="url(#terminal-line-18)">first&#160;message</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
 </text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r1" x="0" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r1" x="24.4" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r4" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">┃</text><text class="terminal-r2" x="24.4" y="556.8" textLength="170.8" clip-path="url(#terminal-line-22)">second&#160;message</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
 </text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">

tests/snapshots/__snapshots__/test_ui_snapshot_rewind/test_snapshot_rewind_navigate_up.svg

@@ -182,7 +182,7 @@
 </text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
 </text><text class="terminal-r4" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">┃</text><text class="terminal-r2" x="24.4" y="459.2" textLength="158.6" clip-path="url(#terminal-line-18)">first&#160;message</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
 </text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r1" x="0" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r1" x="24.4" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r4" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">┃</text><text class="terminal-r5" x="24.4" y="556.8" textLength="170.8" clip-path="url(#terminal-line-22)">second&#160;message</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
 </text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">

tests/snapshots/__snapshots__/test_ui_snapshot_rewind/test_snapshot_rewind_panel_shown.svg

@@ -182,7 +182,7 @@
 </text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
 </text><text class="terminal-r4" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">┃</text><text class="terminal-r2" x="24.4" y="459.2" textLength="158.6" clip-path="url(#terminal-line-18)">first&#160;message</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
 </text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r1" x="0" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r1" x="24.4" y="508" textLength="317.2" clip-path="url(#terminal-line-20)">Hello!&#160;How&#160;can&#160;I&#160;help&#160;you?</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r4" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">┃</text><text class="terminal-r2" x="24.4" y="556.8" textLength="170.8" clip-path="url(#terminal-line-22)">second&#160;message</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
 </text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">

tests/snapshots/__snapshots__/test_ui_snapshot_session_resume/test_snapshot_shows_resumed_session_messages.svg

@@ -186,7 +186,7 @@
 </text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
 </text><text class="terminal-r4" x="0" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">┃</text><text class="terminal-r2" x="24.4" y="581.2" textLength="231.8" clip-path="url(#terminal-line-23)">Hello,&#160;how&#160;are&#160;you?</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
-</text><text class="terminal-r1" x="0" y="630" textLength="695.4" clip-path="url(#terminal-line-25)">I&#x27;m&#160;doing&#160;well,&#160;thank&#160;you!&#160;Let&#160;me&#160;read&#160;that&#160;file&#160;for&#160;you.</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
+</text><text class="terminal-r1" x="24.4" y="630" textLength="695.4" clip-path="url(#terminal-line-25)">I&#x27;m&#160;doing&#160;well,&#160;thank&#160;you!&#160;Let&#160;me&#160;read&#160;that&#160;file&#160;for&#160;you.</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
 </text><text class="terminal-r5" x="0" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">✓</text><text class="terminal-r1" x="24.4" y="678.8" textLength="109.8" clip-path="url(#terminal-line-27)">read_file</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">

tests/snapshots/__snapshots__/test_ui_snapshot_session_resume_injected/test_snapshot_session_resume_hides_injected_messages.svg

@@ -183,11 +183,11 @@
 </text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
 </text><text class="terminal-r4" x="0" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">┃</text><text class="terminal-r2" x="24.4" y="532.4" textLength="280.6" clip-path="url(#terminal-line-21)">Hello,&#160;can&#160;you&#160;help&#160;me?</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
 </text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r1" x="0" y="581.2" textLength="280.6" clip-path="url(#terminal-line-23)">Sure!&#160;What&#160;do&#160;you&#160;need?</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r1" x="24.4" y="581.2" textLength="280.6" clip-path="url(#terminal-line-23)">Sure!&#160;What&#160;do&#160;you&#160;need?</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="329.4" clip-path="url(#terminal-line-25)">Please&#160;read&#160;my&#160;config&#160;file.</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="488" clip-path="url(#terminal-line-27)">Here&#160;is&#160;the&#160;content&#160;of&#160;your&#160;config&#160;file.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="488" clip-path="url(#terminal-line-27)">Here&#160;is&#160;the&#160;content&#160;of&#160;your&#160;config&#160;file.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r5" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_streaming_code_fence/test_snapshot_streaming_code_fence_preserved.svg

@@ -173,23 +173,23 @@
 </text><text class="terminal-r1" x="1464" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
 </text><text class="terminal-r4" x="0" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">┃</text><text class="terminal-r2" x="24.4" y="239.6" textLength="146.4" clip-path="url(#terminal-line-9)">show&#160;diagram</text><text class="terminal-r1" x="1464" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
 </text><text class="terminal-r1" x="1464" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
-</text><text class="terminal-r5" x="0" y="288.4" textLength="97.6" clip-path="url(#terminal-line-11)">&#160;&#160;Client</text><text class="terminal-r1" x="1464" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
-</text><text class="terminal-r5" x="0" y="312.8" textLength="61" clip-path="url(#terminal-line-12)">&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
-</text><text class="terminal-r5" x="0" y="337.2" textLength="280.6" clip-path="url(#terminal-line-13)">&#160;&#160;&#160;&#160;|&#160;&#160;POST&#160;/api/orders</text><text class="terminal-r1" x="1464" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
-</text><text class="terminal-r5" x="0" y="361.6" textLength="61" clip-path="url(#terminal-line-14)">&#160;&#160;&#160;&#160;v</text><text class="terminal-r1" x="1464" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
-</text><text class="terminal-r3" x="0" y="386" textLength="134.2" clip-path="url(#terminal-line-15)">+---------+</text><text class="terminal-r1" x="1464" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
-</text><text class="terminal-r5" x="0" y="410.4" textLength="134.2" clip-path="url(#terminal-line-16)">|&#160;Gateway&#160;|</text><text class="terminal-r1" x="1464" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
-</text><text class="terminal-r3" x="0" y="434.8" textLength="134.2" clip-path="url(#terminal-line-17)">+---------+</text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
-</text><text class="terminal-r5" x="0" y="459.2" textLength="61" clip-path="url(#terminal-line-18)">&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
-</text><text class="terminal-r5" x="0" y="483.6" textLength="61" clip-path="url(#terminal-line-19)">&#160;&#160;&#160;&#160;v</text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r3" x="0" y="508" textLength="134.2" clip-path="url(#terminal-line-20)">+---------+</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
-</text><text class="terminal-r5" x="0" y="532.4" textLength="134.2" clip-path="url(#terminal-line-21)">|&#160;Service&#160;|</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
-</text><text class="terminal-r3" x="0" y="556.8" textLength="134.2" clip-path="url(#terminal-line-22)">+---------+</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r5" x="0" y="581.2" textLength="61" clip-path="url(#terminal-line-23)">&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
-</text><text class="terminal-r5" x="0" y="605.6" textLength="61" clip-path="url(#terminal-line-24)">&#160;&#160;&#160;&#160;v</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
-</text><text class="terminal-r3" x="0" y="630" textLength="134.2" clip-path="url(#terminal-line-25)">+----+----+</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
-</text><text class="terminal-r5" x="0" y="654.4" textLength="134.2" clip-path="url(#terminal-line-26)">|&#160;&#160;&#160;DB&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r3" x="0" y="678.8" textLength="134.2" clip-path="url(#terminal-line-27)">+---------+</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r5" x="24.4" y="288.4" textLength="97.6" clip-path="url(#terminal-line-11)">&#160;&#160;Client</text><text class="terminal-r1" x="1464" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
+</text><text class="terminal-r5" x="24.4" y="312.8" textLength="61" clip-path="url(#terminal-line-12)">&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
+</text><text class="terminal-r5" x="24.4" y="337.2" textLength="280.6" clip-path="url(#terminal-line-13)">&#160;&#160;&#160;&#160;|&#160;&#160;POST&#160;/api/orders</text><text class="terminal-r1" x="1464" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
+</text><text class="terminal-r5" x="24.4" y="361.6" textLength="61" clip-path="url(#terminal-line-14)">&#160;&#160;&#160;&#160;v</text><text class="terminal-r1" x="1464" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
+</text><text class="terminal-r3" x="24.4" y="386" textLength="134.2" clip-path="url(#terminal-line-15)">+---------+</text><text class="terminal-r1" x="1464" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
+</text><text class="terminal-r5" x="24.4" y="410.4" textLength="134.2" clip-path="url(#terminal-line-16)">|&#160;Gateway&#160;|</text><text class="terminal-r1" x="1464" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
+</text><text class="terminal-r3" x="24.4" y="434.8" textLength="134.2" clip-path="url(#terminal-line-17)">+---------+</text><text class="terminal-r1" x="1464" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
+</text><text class="terminal-r5" x="24.4" y="459.2" textLength="61" clip-path="url(#terminal-line-18)">&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
+</text><text class="terminal-r5" x="24.4" y="483.6" textLength="61" clip-path="url(#terminal-line-19)">&#160;&#160;&#160;&#160;v</text><text class="terminal-r1" x="1464" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
+</text><text class="terminal-r3" x="24.4" y="508" textLength="134.2" clip-path="url(#terminal-line-20)">+---------+</text><text class="terminal-r1" x="1464" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">
+</text><text class="terminal-r5" x="24.4" y="532.4" textLength="134.2" clip-path="url(#terminal-line-21)">|&#160;Service&#160;|</text><text class="terminal-r1" x="1464" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
+</text><text class="terminal-r3" x="24.4" y="556.8" textLength="134.2" clip-path="url(#terminal-line-22)">+---------+</text><text class="terminal-r1" x="1464" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
+</text><text class="terminal-r5" x="24.4" y="581.2" textLength="61" clip-path="url(#terminal-line-23)">&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
+</text><text class="terminal-r5" x="24.4" y="605.6" textLength="61" clip-path="url(#terminal-line-24)">&#160;&#160;&#160;&#160;v</text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
+</text><text class="terminal-r3" x="24.4" y="630" textLength="134.2" clip-path="url(#terminal-line-25)">+----+----+</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
+</text><text class="terminal-r5" x="24.4" y="654.4" textLength="134.2" clip-path="url(#terminal-line-26)">|&#160;&#160;&#160;DB&#160;&#160;&#160;&#160;|</text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
+</text><text class="terminal-r3" x="24.4" y="678.8" textLength="134.2" clip-path="url(#terminal-line-27)">+---------+</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r7" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/__snapshots__/test_ui_snapshot_voice_mode/test_snapshot_recording_then_conversation.svg

@@ -187,7 +187,7 @@
 </text><text class="terminal-r1" x="1464" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">
 </text><text class="terminal-r4" x="0" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">┃</text><text class="terminal-r2" x="24.4" y="630" textLength="61" clip-path="url(#terminal-line-25)">hello</text><text class="terminal-r1" x="1464" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
 </text><text class="terminal-r1" x="1464" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
-</text><text class="terminal-r1" x="0" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">I&#x27;m&#160;ready&#160;to&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r1" x="24.4" y="678.8" textLength="268.4" clip-path="url(#terminal-line-27)">I&#x27;m&#160;ready&#160;to&#160;help&#160;you.</text><text class="terminal-r1" x="1464" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
 </text><text class="terminal-r1" x="1464" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
 </text><text class="terminal-r1" x="1464" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
 </text><text class="terminal-r5" x="0" y="752" textLength="1464" clip-path="url(#terminal-line-30)">┌────────────────────────────────────────────────────────────────────────────────────────────────────────────&#160;default&#160;─┐</text><text class="terminal-r1" x="1464" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">

tests/snapshots/test_ui_snapshot_mcp_command.py

@@ -12,7 +12,6 @@ from tests.stubs.fake_mcp_registry import (
     FakeMCPRegistryWithBrokenServer,
 )
 from vibe.core.config import MCPHttp, MCPStdio
-from vibe.core.tools.connectors import CONNECTORS_ENV_VAR
 from vibe.core.tools.mcp.tools import RemoteTool
 
 _MCP_PATCH = "vibe.core.agent_loop.MCPRegistry"
@@ -239,7 +238,6 @@ class SnapshotTestAppConnectorsMixedState(BaseSnapshotTestApp):
 # ---------------------------------------------------------------------------
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_mcp_with_connectors_overview(snap_compare: SnapCompare) -> None:
 
     async def run_before(pilot: Pilot) -> None:
@@ -258,7 +256,6 @@ def test_snapshot_mcp_with_connectors_overview(snap_compare: SnapCompare) -> Non
 # ---------------------------------------------------------------------------
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_connector_auth_opens_on_disconnected(
     snap_compare: SnapCompare,
 ) -> None:
@@ -280,7 +277,6 @@ def test_snapshot_connector_auth_opens_on_disconnected(
     )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_connector_auth_show_url(snap_compare: SnapCompare) -> None:
     """Selecting 'Manually show the URL' reveals the auth URL."""
 
@@ -303,7 +299,6 @@ def test_snapshot_connector_auth_show_url(snap_compare: SnapCompare) -> None:
     )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_connector_auth_back_to_mcp(snap_compare: SnapCompare) -> None:
     """Pressing backspace in the auth app returns to the /mcp menu."""
 
@@ -323,7 +318,6 @@ def test_snapshot_connector_auth_back_to_mcp(snap_compare: SnapCompare) -> None:
     )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_mcp_help_bar_shows_authenticate(snap_compare: SnapCompare) -> None:
     """Help bar shows 'Enter Authenticate' when a disconnected connector is highlighted."""
 
@@ -340,7 +334,6 @@ def test_snapshot_mcp_help_bar_shows_authenticate(snap_compare: SnapCompare) ->
     )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_mcp_connectors_only(snap_compare: SnapCompare) -> None:
 
     async def run_before(pilot: Pilot) -> None:
@@ -353,7 +346,6 @@ def test_snapshot_mcp_connectors_only(snap_compare: SnapCompare) -> None:
     )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_mcp_connectors_sorted_by_status(snap_compare: SnapCompare) -> None:
 
     async def run_before(pilot: Pilot) -> None:
@@ -366,7 +358,6 @@ def test_snapshot_mcp_connectors_sorted_by_status(snap_compare: SnapCompare) ->
     )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_mcp_drill_into_connector(snap_compare: SnapCompare) -> None:
 
     async def run_before(pilot: Pilot) -> None:
@@ -387,7 +378,6 @@ def test_snapshot_mcp_drill_into_connector(snap_compare: SnapCompare) -> None:
         )
 
 
-@patch.dict("os.environ", {CONNECTORS_ENV_VAR: "1"})
 def test_snapshot_mcp_connector_back_to_overview(snap_compare: SnapCompare) -> None:
 
     async def run_before(pilot: Pilot) -> None:

tests/snapshots/test_ui_snapshot_plan_file_message.py

@@ -0,0 +1,43 @@
+from __future__ import annotations
+
+from pathlib import Path
+from typing import cast
+
+from textual.pilot import Pilot
+
+from tests.snapshots.base_snapshot_test_app import BaseSnapshotTestApp
+from tests.snapshots.snap_compare import SnapCompare
+from vibe.cli.textual_ui.widgets.messages import PlanFileMessage
+
+PLAN_CONTENT = """\
+# Implementation Plan
+
+## 1. Add user authentication
+- JWT token validation
+- Session management
+
+## 2. Database migrations
+- Create users table
+- Add indexes
+"""
+
+
+class PlanFileMessageTestApp(BaseSnapshotTestApp):
+    pass
+
+
+def test_snapshot_plan_file_message(snap_compare: SnapCompare, tmp_path: Path) -> None:
+    plan_file = tmp_path / "plan.md"
+    plan_file.write_text(PLAN_CONTENT)
+
+    async def run_before(pilot: Pilot) -> None:
+        app = cast(PlanFileMessageTestApp, pilot.app)
+        plan_widget = PlanFileMessage(file_path=plan_file)
+        await app._mount_and_scroll(plan_widget)
+        await pilot.pause(0.3)
+
+    assert snap_compare(
+        "test_ui_snapshot_plan_file_message.py:PlanFileMessageTestApp",
+        terminal_size=(120, 36),
+        run_before=run_before,
+    )

tests/test_agent_backend.py

@@ -146,7 +146,7 @@ async def test_passes_parent_session_id_to_backend_after_reset(vibe_config: Vibe
     [_ async for _ in agent.act("Hello")]
     first_session_id = agent.session_id
 
-    agent._reset_session()
+    await agent._reset_session()
     [_ async for _ in agent.act("Hello again")]
 
     assert len(backend.requests_metadata) >= 2

tests/test_agent_loop_refresh_system_prompt.py

@@ -0,0 +1,33 @@
+from __future__ import annotations
+
+import pytest
+
+from tests.conftest import build_test_agent_loop, build_test_vibe_config
+
+
+@pytest.mark.asyncio
+async def test_refresh_system_prompt_preserves_scratchpad_section() -> None:
+    # Regression: refresh_system_prompt must pass scratchpad_dir, otherwise
+    # it silently drops the scratchpad instructions from the system prompt.
+    # This fires on every session start/resume via initialize_experiments
+    # and hydrate_experiments_from_session, so the LLM would lose awareness
+    # of the scratchpad on the very first turn for any user with telemetry
+    # enabled and a Mistral API key.
+    config = build_test_vibe_config(
+        include_project_context=False,
+        include_prompt_detail=True,
+        include_model_info=False,
+        include_commit_signature=False,
+    )
+    agent = build_test_agent_loop(config=config)
+
+    initial_prompt = agent.messages[0].content or ""
+    assert "Scratchpad Directory" in initial_prompt
+    assert agent.scratchpad_dir is not None
+    assert str(agent.scratchpad_dir) in initial_prompt
+
+    await agent.refresh_system_prompt()
+
+    refreshed_prompt = agent.messages[0].content or ""
+    assert "Scratchpad Directory" in refreshed_prompt
+    assert str(agent.scratchpad_dir) in refreshed_prompt

tests/test_agent_tool_call.py

@@ -864,3 +864,34 @@ async def test_parallel_conversation_history_has_all_tool_messages() -> None:
         "call_h3",
     }
     assert agent_loop.stats.tool_calls_succeeded == 4
+
+
+@pytest.mark.asyncio
+async def test_pending_injected_message_continues_loop_after_tool_result() -> None:
+    tool_call = make_todo_tool_call("call_inject")
+    backend = FakeBackend([
+        [mock_llm_chunk(content="Let me check.", tool_calls=[tool_call])],
+        [mock_llm_chunk(content="Acting on the injected guidance.")],
+    ])
+    agent_loop = make_agent_loop(auto_approve=True, backend=backend)
+
+    events: list[BaseEvent] = []
+    async for event in agent_loop.act("Go"):
+        events.append(event)
+        if isinstance(event, ToolResultEvent):
+            agent_loop._pending_injected_messages.append(
+                LLMMessage(role=Role.user, content="updated context", injected=True)
+            )
+
+    assistant_events = [e for e in events if isinstance(e, AssistantEvent)]
+    assert len(assistant_events) == 2
+
+    injected_msgs = [
+        m for m in agent_loop.messages if m.role == Role.user and m.injected
+    ]
+    assert any("updated context" in (m.content or "") for m in injected_msgs)
+
+    last_assistant = next(
+        m for m in reversed(agent_loop.messages) if m.role == Role.assistant
+    )
+    assert "Acting on the injected guidance" in (last_assistant.content or "")

tests/test_agents.py

@@ -259,7 +259,7 @@ class TestAgentApplyToConfig:
 
         mock_manager = _MockManager(sources=("user",))
         monkeypatch.setattr(
-            "vibe.core.config._settings.get_harness_files_manager", lambda: mock_manager
+            "vibe.core.prompts.get_harness_files_manager", lambda: mock_manager
         )
 
         base = VibeConfig(include_project_context=False, include_prompt_detail=False)
@@ -298,7 +298,7 @@ class TestAgentApplyToConfig:
 
         mock_manager = _MockManager(sources=("user",))
         monkeypatch.setattr(
-            "vibe.core.config._settings.get_harness_files_manager", lambda: mock_manager
+            "vibe.core.prompts.get_harness_files_manager", lambda: mock_manager
         )
 
         config = VibeConfig(
@@ -700,7 +700,7 @@ class TestAgentLoopInitialization:
 
         mock_manager = _MockManager(sources=("user",))
         monkeypatch.setattr(
-            "vibe.core.config._settings.get_harness_files_manager", lambda: mock_manager
+            "vibe.core.prompts.get_harness_files_manager", lambda: mock_manager
         )
 
         custom_agent = AgentProfile(

tests/test_approve_always_permanent.py

@@ -4,6 +4,7 @@ from pathlib import Path
 import tomllib
 
 from tests.conftest import build_test_agent_loop, build_test_vibe_config
+from vibe.core.tools.base import ToolPermission
 from vibe.core.tools.permissions import PermissionScope, RequiredPermission
 
 
@@ -27,7 +28,10 @@ class TestApproveAlwaysPermanentNoGranularPermissions:
 
         agent.approve_always("bash", None, save_permanently=False)
 
-        assert agent.config.tools["bash"]["permission"] == "always"
+        assert (
+            agent.tool_manager.get_tool_config("bash").permission
+            == ToolPermission.ALWAYS
+        )
         persisted = _read_persisted_config(config_dir)
         assert "bash" not in persisted.get("tools", {})
 
@@ -58,8 +62,8 @@ class TestApproveAlwaysPermanentWithGranularPermissions:
 
         agent.approve_always("bash", perms, save_permanently=True)
 
-        assert len(agent._session_rules) == 1
-        rule = agent._session_rules[0]
+        assert len(agent._permission_store._rules) == 1
+        rule = agent._permission_store._rules[0]
         assert rule.tool_name == "bash"
         assert rule.scope == PermissionScope.COMMAND_PATTERN
         assert rule.session_pattern == "npm install *"
@@ -70,7 +74,7 @@ class TestApproveAlwaysPermanentWithGranularPermissions:
 
         agent.approve_always("bash", perms, save_permanently=False)
 
-        assert len(agent._session_rules) == 1
+        assert len(agent._permission_store._rules) == 1
         persisted = _read_persisted_config(config_dir)
         assert "bash" not in persisted.get("tools", {})
 

tests/test_deferred_init.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import threading
 from unittest.mock import AsyncMock, MagicMock, patch
 
@@ -12,6 +13,7 @@ from tests.mock.utils import mock_llm_chunk
 from tests.stubs.fake_backend import FakeBackend
 from tests.stubs.fake_connector_registry import FakeConnectorRegistry
 from tests.stubs.fake_mcp_registry import FakeMCPRegistry
+from vibe.core import agent_loop as agent_loop_module
 from vibe.core.agent_loop import AgentLoop
 from vibe.core.config import MCPStdio
 from vibe.core.tools.manager import ToolManager
@@ -256,3 +258,252 @@ class TestDeferredInitPublicMethods:
 
         assert loop.is_initialized
         assert loop.messages[-1].content == "context"
+
+
+# ---------------------------------------------------------------------------
+# start_initialize_experiments / wait_until_ready experiment gating
+# ---------------------------------------------------------------------------
+
+
+class TestStartInitializeExperiments:
+    @pytest.mark.asyncio
+    async def test_does_not_block_caller(self) -> None:
+        loop = build_test_agent_loop()
+        gate = asyncio.Event()
+
+        async def slow_init() -> None:
+            await gate.wait()
+
+        with patch.object(loop, "initialize_experiments", side_effect=slow_init):
+            loop.start_initialize_experiments()
+
+            task = loop._experiments_task
+            assert task is not None
+            assert not task.done()
+
+            gate.set()
+            await task
+
+    @pytest.mark.asyncio
+    async def test_is_idempotent(self) -> None:
+        loop = build_test_agent_loop()
+        init_mock = AsyncMock()
+
+        with patch.object(loop, "initialize_experiments", new=init_mock):
+            loop.start_initialize_experiments()
+            first_task = loop._experiments_task
+            loop.start_initialize_experiments()
+            second_task = loop._experiments_task
+
+            assert first_task is second_task
+            assert first_task is not None
+            await first_task
+
+        assert init_mock.await_count == 1
+
+    @pytest.mark.asyncio
+    async def test_sets_pending_telemetry_flags(self) -> None:
+        loop = build_test_agent_loop()
+
+        with patch.object(loop, "initialize_experiments", new=AsyncMock()):
+            loop.start_initialize_experiments()
+
+            assert loop._pending_new_session_telemetry is True
+            assert loop._ready_telemetry_pending is True
+
+            task = loop._experiments_task
+            assert task is not None
+            await task
+
+    @pytest.mark.asyncio
+    async def test_refreshes_system_prompt_when_experiments_update(self) -> None:
+        loop = build_test_agent_loop()
+        refresh_mock = AsyncMock()
+
+        with (
+            patch.object(
+                agent_loop_module,
+                "session_initialize_experiments",
+                new=AsyncMock(return_value=True),
+            ),
+            patch.object(loop, "refresh_system_prompt", new=refresh_mock),
+        ):
+            await loop.initialize_experiments()
+
+        refresh_mock.assert_awaited_once()
+
+    @pytest.mark.asyncio
+    async def test_does_not_refresh_system_prompt_when_experiments_unchanged(
+        self,
+    ) -> None:
+        loop = build_test_agent_loop()
+        refresh_mock = AsyncMock()
+
+        with (
+            patch.object(
+                agent_loop_module,
+                "session_initialize_experiments",
+                new=AsyncMock(return_value=False),
+            ),
+            patch.object(loop, "refresh_system_prompt", new=refresh_mock),
+        ):
+            await loop.initialize_experiments()
+
+        refresh_mock.assert_not_awaited()
+
+
+class TestWaitUntilReadyJoinsExperiments:
+    @pytest.mark.asyncio
+    async def test_joins_in_flight_task(self) -> None:
+        loop = build_test_agent_loop()
+        gate = asyncio.Event()
+        completed = False
+
+        async def slow_init() -> None:
+            nonlocal completed
+            await gate.wait()
+            completed = True
+
+        with patch.object(loop, "initialize_experiments", side_effect=slow_init):
+            loop.start_initialize_experiments()
+
+            async def release() -> None:
+                await asyncio.sleep(0.01)
+                gate.set()
+
+            asyncio.create_task(release())
+            await loop.wait_until_ready()
+
+            assert completed is True
+            task = loop._experiments_task
+            assert task is not None
+            assert task.done()
+
+    @pytest.mark.asyncio
+    async def test_emits_new_session_telemetry_once(self) -> None:
+        loop = build_test_agent_loop()
+        emit_new_session = MagicMock()
+
+        with (
+            patch.object(loop, "initialize_experiments", new=AsyncMock()),
+            patch.object(loop, "emit_new_session_telemetry", new=emit_new_session),
+        ):
+            loop.start_initialize_experiments()
+            await loop.wait_until_ready()
+            await loop.wait_until_ready()
+
+        emit_new_session.assert_called_once()
+        assert loop._pending_new_session_telemetry is False
+
+    @pytest.mark.asyncio
+    async def test_emits_ready_telemetry_when_only_experiments_deferred(self) -> None:
+        loop = build_test_agent_loop()
+        emit_ready = MagicMock()
+
+        with (
+            patch.object(loop, "initialize_experiments", new=AsyncMock()),
+            patch.object(loop, "emit_ready_telemetry", new=emit_ready),
+        ):
+            loop.start_initialize_experiments()
+            await loop.wait_until_ready()
+            await loop.wait_until_ready()
+
+        emit_ready.assert_called_once()
+        ((duration,), _) = emit_ready.call_args
+        assert isinstance(duration, int)
+        assert duration >= 0
+        assert loop._ready_telemetry_pending is False
+
+    @pytest.mark.asyncio
+    async def test_does_not_emit_new_session_when_only_hydrating(self) -> None:
+        loop = build_test_agent_loop()
+        emit_new_session = MagicMock()
+
+        with (
+            patch.object(loop, "hydrate_experiments_from_session", new=AsyncMock()),
+            patch.object(loop, "emit_new_session_telemetry", new=emit_new_session),
+        ):
+            await loop.hydrate_experiments_from_session()
+            await loop.wait_until_ready()
+
+        emit_new_session.assert_not_called()
+        assert loop._pending_new_session_telemetry is False
+
+    @pytest.mark.asyncio
+    async def test_no_op_when_nothing_deferred(self) -> None:
+        loop = build_test_agent_loop()
+        emit_ready = MagicMock()
+        emit_new_session = MagicMock()
+
+        with (
+            patch.object(loop, "emit_ready_telemetry", new=emit_ready),
+            patch.object(loop, "emit_new_session_telemetry", new=emit_new_session),
+        ):
+            await loop.wait_until_ready()
+
+        emit_ready.assert_not_called()
+        emit_new_session.assert_not_called()
+
+
+class TestACloseCancelsExperimentsTask:
+    @pytest.mark.asyncio
+    async def test_cancels_in_flight_task(self) -> None:
+        loop = build_test_agent_loop()
+        gate = asyncio.Event()
+
+        async def never_completing() -> None:
+            await gate.wait()
+
+        with patch.object(loop, "initialize_experiments", side_effect=never_completing):
+            loop.start_initialize_experiments()
+            task = loop._experiments_task
+            assert task is not None
+            assert not task.done()
+
+            await loop.aclose()
+
+            assert task.done()
+            assert task.cancelled()
+
+    @pytest.mark.asyncio
+    async def test_does_not_cancel_completed_task(self) -> None:
+        loop = build_test_agent_loop()
+
+        with patch.object(loop, "initialize_experiments", new=AsyncMock()):
+            loop.start_initialize_experiments()
+            task = loop._experiments_task
+            assert task is not None
+            await task
+
+            await loop.aclose()
+
+            assert task.done()
+            assert not task.cancelled()
+
+
+class TestActGatesOnExperiments:
+    @pytest.mark.asyncio
+    async def test_act_awaits_experiments_before_llm_call(self) -> None:
+        loop = build_test_agent_loop(
+            backend=FakeBackend(mock_llm_chunk(content="hello"))
+        )
+        gate = asyncio.Event()
+        finished_init = False
+
+        async def slow_init() -> None:
+            nonlocal finished_init
+            await gate.wait()
+            finished_init = True
+
+        with patch.object(loop, "initialize_experiments", side_effect=slow_init):
+            loop.start_initialize_experiments()
+
+            async def release() -> None:
+                await asyncio.sleep(0.01)
+                gate.set()
+
+            asyncio.create_task(release())
+            events = [event async for event in loop.act("Hello")]
+
+            assert finished_init is True
+            assert any(getattr(event, "content", None) == "hello" for event in events)

tests/test_message_merging.py

@@ -1,47 +0,0 @@
-from __future__ import annotations
-
-from vibe.core.llm.message_utils import merge_consecutive_user_messages
-from vibe.core.types import LLMMessage, Role
-
-
-def test_merge_consecutive_user_messages() -> None:
-    messages = [
-        LLMMessage(role=Role.system, content="System"),
-        LLMMessage(role=Role.user, content="User 1"),
-        LLMMessage(role=Role.user, content="User 2"),
-        LLMMessage(role=Role.assistant, content="Assistant"),
-    ]
-    result = merge_consecutive_user_messages(messages)
-    assert len(result) == 3
-    assert result[1].content == "User 1\n\nUser 2"
-
-
-def test_preserves_non_consecutive_user_messages() -> None:
-    messages = [
-        LLMMessage(role=Role.user, content="User 1"),
-        LLMMessage(role=Role.assistant, content="Assistant"),
-        LLMMessage(role=Role.user, content="User 2"),
-    ]
-    result = merge_consecutive_user_messages(messages)
-    assert len(result) == 3
-
-
-def test_empty_messages() -> None:
-    assert merge_consecutive_user_messages([]) == []
-
-
-def test_single_message() -> None:
-    messages = [LLMMessage(role=Role.user, content="Only one")]
-    result = merge_consecutive_user_messages(messages)
-    assert len(result) == 1
-
-
-def test_three_consecutive_user_messages() -> None:
-    messages = [
-        LLMMessage(role=Role.user, content="A"),
-        LLMMessage(role=Role.user, content="B"),
-        LLMMessage(role=Role.user, content="C"),
-    ]
-    result = merge_consecutive_user_messages(messages)
-    assert len(result) == 1
-    assert result[0].content == "A\n\nB\n\nC"

tests/test_permission_store.py

@@ -0,0 +1,166 @@
+from __future__ import annotations
+
+from pydantic import BaseModel
+import pytest
+
+from tests.conftest import build_test_agent_loop, build_test_vibe_config
+from tests.mock.utils import mock_llm_chunk
+from tests.stubs.fake_backend import FakeBackend
+from vibe.core.tools.base import ToolPermission
+from vibe.core.tools.permissions import (
+    ApprovedRule,
+    PermissionScope,
+    PermissionStore,
+    RequiredPermission,
+)
+from vibe.core.types import ApprovalResponse, FunctionCall, ToolCall, ToolResultEvent
+
+
+class TestPermissionStore:
+    def test_covers_returns_false_when_empty(self):
+        store = PermissionStore()
+        rp = RequiredPermission(
+            scope=PermissionScope.COMMAND_PATTERN,
+            invocation_pattern="npm install foo",
+            session_pattern="npm install *",
+            label="npm install *",
+        )
+
+        assert not store.covers("bash", rp)
+
+    def test_covers_returns_true_after_matching_rule_added(self):
+        store = PermissionStore()
+        store.add_rule(
+            ApprovedRule(
+                tool_name="bash",
+                scope=PermissionScope.COMMAND_PATTERN,
+                session_pattern="npm install *",
+            )
+        )
+        rp = RequiredPermission(
+            scope=PermissionScope.COMMAND_PATTERN,
+            invocation_pattern="npm install foo",
+            session_pattern="npm install *",
+            label="npm install *",
+        )
+
+        assert store.covers("bash", rp)
+
+    def test_covers_isolates_by_tool_name(self):
+        store = PermissionStore()
+        store.add_rule(
+            ApprovedRule(
+                tool_name="bash",
+                scope=PermissionScope.COMMAND_PATTERN,
+                session_pattern="npm install *",
+            )
+        )
+        rp = RequiredPermission(
+            scope=PermissionScope.COMMAND_PATTERN,
+            invocation_pattern="npm install foo",
+            session_pattern="npm install *",
+            label="npm install *",
+        )
+
+        assert not store.covers("read_file", rp)
+
+    def test_tool_permission_round_trip(self):
+        store = PermissionStore()
+        assert store.get_tool_permission("bash") is None
+
+        store.set_tool_permission("bash", ToolPermission.ALWAYS)
+
+        assert store.get_tool_permission("bash") == ToolPermission.ALWAYS
+
+
+class TestAgentLoopSharesStore:
+    def test_subagent_inherits_parent_session_rules(self):
+        store = PermissionStore()
+        parent = build_test_agent_loop(permission_store=store)
+        subagent = build_test_agent_loop(permission_store=store)
+
+        parent.approve_always(
+            "bash",
+            [
+                RequiredPermission(
+                    scope=PermissionScope.COMMAND_PATTERN,
+                    invocation_pattern="npm install foo",
+                    session_pattern="npm install *",
+                    label="npm install *",
+                )
+            ],
+        )
+
+        rp = RequiredPermission(
+            scope=PermissionScope.COMMAND_PATTERN,
+            invocation_pattern="npm install bar",
+            session_pattern="npm install *",
+            label="npm install *",
+        )
+        assert subagent._permission_store.covers("bash", rp)
+
+    def test_subagent_inherits_parent_tool_permission(self):
+        store = PermissionStore()
+        parent = build_test_agent_loop(permission_store=store)
+        subagent = build_test_agent_loop(permission_store=store)
+
+        parent.approve_always("bash", None)
+
+        assert (
+            subagent._permission_store.get_tool_permission("bash")
+            == ToolPermission.ALWAYS
+        )
+
+    @pytest.mark.asyncio
+    async def test_subagent_applies_parent_tool_permission_before_resolution(self):
+        store = PermissionStore()
+        parent = build_test_agent_loop(permission_store=store)
+        parent.approve_always("bash", None)
+        approval_requested = False
+
+        async def approval_callback(
+            tool_name: str,
+            args: BaseModel,
+            tool_call_id: str,
+            required_permissions: list[RequiredPermission] | None,
+        ) -> tuple[ApprovalResponse, str | None]:
+            nonlocal approval_requested
+            approval_requested = True
+            return ApprovalResponse.NO, None
+
+        tool_call = ToolCall(
+            id="call_1",
+            index=0,
+            function=FunctionCall(name="bash", arguments='{"command":"true"}'),
+        )
+        subagent = build_test_agent_loop(
+            config=build_test_vibe_config(enabled_tools=["bash"]),
+            backend=FakeBackend([
+                [mock_llm_chunk(content="Running it.", tool_calls=[tool_call])],
+                [mock_llm_chunk(content="Done.")],
+            ]),
+            permission_store=store,
+        )
+        subagent.set_approval_callback(approval_callback)
+
+        events = [event async for event in subagent.act("run true")]
+        tool_results = [event for event in events if isinstance(event, ToolResultEvent)]
+
+        assert not approval_requested
+        assert len(tool_results) == 1
+        assert tool_results[0].skipped is False
+        assert "permission" not in subagent.config.tools.get("bash", {})
+
+    def test_default_store_is_per_loop_when_not_shared(self):
+        a = build_test_agent_loop()
+        b = build_test_agent_loop()
+
+        assert a._permission_store is not b._permission_store
+        assert a._permission_store.lock is not b._permission_store.lock
+
+    def test_subagent_shares_parent_approval_lock(self):
+        store = PermissionStore()
+        parent = build_test_agent_loop(permission_store=store)
+        subagent = build_test_agent_loop(permission_store=store)
+
+        assert parent._permission_store.lock is subagent._permission_store.lock

tests/test_prepare_release.py

@@ -0,0 +1,82 @@
+from __future__ import annotations
+
+import subprocess
+import tomllib
+
+from scripts import prepare_release
+
+
+def test_pin_dependencies_preserves_toml_format_and_refreshes_lock(
+    monkeypatch, tmp_path
+):
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text(
+        """[project]
+name = "mistral-vibe"
+version = "2.9.6"
+license = { text = "Apache-2.0" }
+authors = [{ name = "Mistral AI" }]
+dependencies = [
+    "httpx[http2]>=0.28.1",
+]
+
+[dependency-groups]
+build = ["pyinstaller>=6.17.0"]
+
+[tool.pytest.ini_options]
+filterwarnings = [
+    # Keep this comment when pinning dependencies.
+    "ignore:example",
+]
+""",
+        encoding="utf-8",
+    )
+    (tmp_path / "uv.lock").write_text("", encoding="utf-8")
+    monkeypatch.chdir(tmp_path)
+
+    def fake_get_pinned_dependencies(group=None):
+        if group == "build":
+            return ["pyinstaller==6.17.0", "truststore==0.10.4"]
+        return ["anyio==4.12.0", 'httpx[http2]==0.28.1 ; python_version >= "3.12"']
+
+    uv_commands = []
+
+    def fake_subprocess_run(args, **kwargs):
+        uv_commands.append(args)
+        return subprocess.CompletedProcess(args=args, returncode=0, stdout="")
+
+    git_commands = []
+
+    def fake_run_git_command(*args, **kwargs):
+        git_commands.append(args)
+        return subprocess.CompletedProcess(args=args, returncode=0, stdout="")
+
+    monkeypatch.setattr(
+        prepare_release, "get_pinned_dependencies", fake_get_pinned_dependencies
+    )
+    monkeypatch.setattr(prepare_release.subprocess, "run", fake_subprocess_run)
+    monkeypatch.setattr(prepare_release, "run_git_command", fake_run_git_command)
+
+    prepare_release.pin_dependencies("2.9.6")
+
+    updated_pyproject = pyproject_path.read_text(encoding="utf-8")
+    updated_data = tomllib.loads(updated_pyproject)
+    assert 'license = { text = "Apache-2.0" }' in updated_pyproject
+    assert 'authors = [{ name = "Mistral AI" }]' in updated_pyproject
+    assert "# Keep this comment when pinning dependencies." in updated_pyproject
+    assert updated_data["project"]["dependencies"] == [
+        "anyio==4.12.0",
+        'httpx[http2]==0.28.1 ; python_version >= "3.12"',
+    ]
+    assert updated_data["dependency-groups"]["build"] == [
+        "pyinstaller==6.17.0",
+        "truststore==0.10.4",
+    ]
+    assert uv_commands == [["uv", "lock"]]
+    assert ("add", "pyproject.toml", "uv.lock") in git_commands
+    assert (
+        "commit",
+        "--allow-empty",
+        "-m",
+        "chore: pin dependencies for v2.9.6",
+    ) in git_commands

tests/test_ui_input_history.py

@@ -61,10 +61,15 @@ async def test_ui_navigation_restores_partially_typed_draft_after_round_trip(
     async with vibe_app.run_test() as pilot:
         inject_history_file(vibe_app, history_file)
         chat_input = vibe_app.query_one(ChatInputContainer)
+        textarea = chat_input.input_widget
+        assert textarea is not None
 
         await pilot.press(*"he")
         assert chat_input.value == "he"
 
+        await pilot.press("up")
+        assert chat_input.value == "he"
+        assert textarea.cursor_location == (0, 0)
         await pilot.press("up")
         assert chat_input.value == "how are you?"
         await pilot.press("down")
@@ -207,6 +212,7 @@ async def test_ui_intercepts_arrow_up_only_on_first_wrapped_row(
         while textarea.get_cursor_up_location() != textarea.cursor_location:
             textarea.action_cursor_up()
 
+        assert textarea.cursor_location == (0, 0)
         await pilot.press("up")
         assert chat_input.value == "how are you?"
 

tests/tools/test_connectors.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import os
 from typing import Any, cast
 from unittest.mock import AsyncMock, patch
 
@@ -8,11 +7,11 @@ import httpx
 import pytest
 import respx
 
+from tests.conftest import build_test_vibe_config
 from tests.stubs.fake_connector_registry import FakeConnectorRegistry
 from tests.stubs.fake_mcp_registry import FakeMCPRegistry
 from vibe.core.config import ConnectorConfig, VibeConfig
 from vibe.core.tools.base import BaseToolConfig, ToolError
-from vibe.core.tools.connectors import CONNECTORS_ENV_VAR
 from vibe.core.tools.connectors.connector_registry import (
     ConnectorRegistry,
     RemoteTool,
@@ -153,22 +152,7 @@ class TestFakeConnectorRegistry:
 class TestToolManagerConnectorIntegration:
     @staticmethod
     def _make_config(connectors: list[ConnectorConfig] | None = None) -> VibeConfig:
-        """Minimal VibeConfig-like stub for ToolManager."""
-        return cast(
-            VibeConfig,
-            type(
-                "_Cfg",
-                (),
-                {
-                    "mcp_servers": [],
-                    "connectors": connectors or [],
-                    "enabled_tools": [],
-                    "disabled_tools": [],
-                    "tools": {},
-                    "tool_paths": [],
-                },
-            )(),
-        )
+        return build_test_vibe_config(connectors=connectors or [])
 
     def test_connector_tools_registered(self) -> None:
         registry = FakeConnectorRegistry(
@@ -198,22 +182,6 @@ class TestToolManagerConnectorIntegration:
         assert connector_tools == []
 
 
-# ---------------------------------------------------------------------------
-# ConnectorRegistry env var gating (tested via agent_loop helper logic)
-# ---------------------------------------------------------------------------
-
-
-class TestConnectorRegistryEnvGating:
-    def test_disabled_without_env_var(self) -> None:
-        with patch.dict(os.environ, {}, clear=False):
-            os.environ.pop(CONNECTORS_ENV_VAR, None)
-            assert os.getenv(CONNECTORS_ENV_VAR) != "1"
-
-    def test_enabled_with_env_var(self) -> None:
-        with patch.dict(os.environ, {CONNECTORS_ENV_VAR: "1"}):
-            assert os.getenv(CONNECTORS_ENV_VAR) == "1"
-
-
 # ---------------------------------------------------------------------------
 # Error message helpers
 # ---------------------------------------------------------------------------
@@ -339,7 +307,7 @@ class TestConnectorProxyToolRun:
 
         mock_call.assert_awaited_once()
         call_args = mock_call.call_args
-        assert "/v1/experimental/connectors/conn-123/mcp" in call_args.args[0]
+        assert "/v1/connectors-gateway/conn-123/mcp" in call_args.args[0]
         assert call_args.args[1] == "search"
         assert call_args.kwargs["headers"]["Authorization"] == "Bearer test-key"
 
@@ -412,21 +380,7 @@ class TestConnectorProxyToolRun:
 class TestConnectorDisableFiltering:
     @staticmethod
     def _make_config(connectors: list[ConnectorConfig] | None = None) -> VibeConfig:
-        return cast(
-            VibeConfig,
-            type(
-                "_Cfg",
-                (),
-                {
-                    "mcp_servers": [],
-                    "connectors": connectors or [],
-                    "enabled_tools": [],
-                    "disabled_tools": [],
-                    "tools": {},
-                    "tool_paths": [],
-                },
-            )(),
-        )
+        return build_test_vibe_config(connectors=connectors or [])
 
     def test_disabled_connector_excludes_all_tools(self) -> None:
         registry = FakeConnectorRegistry(

tests/tools/test_exit_plan_mode.py

@@ -232,7 +232,7 @@ class TestAnswerHandling:
 
 class TestPlanFile:
     @pytest.mark.asyncio
-    async def test_content_passed_as_preview(
+    async def test_keybinding_hint_shown_as_preview(
         self, tool: ExitPlanMode, plan_manager: MockAgentManager, tmp_path: Path
     ) -> None:
         plan_file = tmp_path / "plan.md"
@@ -247,21 +247,36 @@ class TestPlanFile:
         )
         await collect_result(tool.run(ExitPlanModeArgs(), ctx))
         assert isinstance(cb.received_args, AskUserQuestionArgs)
-        assert cb.received_args.content_preview == "# My Plan\n\n- Step 1\n- Step 2\n"
+        assert cb.received_args.footer_note is not None
+        assert "Ctrl+G" in cb.received_args.footer_note
+        assert str(plan_file) in cb.received_args.footer_note
 
     @pytest.mark.asyncio
-    async def test_missing_file_means_none_preview(
+    async def test_result_does_not_include_plan_content(
         self, tool: ExitPlanMode, plan_manager: MockAgentManager, tmp_path: Path
     ) -> None:
-        plan_file = tmp_path / "nonexistent.md"
+        plan_file = tmp_path / "plan.md"
+        plan_file.write_text("# My Plan\n\n- Step 1\n- Step 2\n")
 
-        cb = MockCallback(AskUserQuestionResult(answers=[], cancelled=True))
+        cb = MockCallback(
+            AskUserQuestionResult(
+                answers=[
+                    Answer(
+                        question="q",
+                        answer="Yes, and auto approve edits",
+                        is_other=False,
+                    )
+                ],
+                cancelled=False,
+            )
+        )
         ctx = InvokeContext(
             tool_call_id="t1",
             agent_manager=plan_manager,  # type: ignore[arg-type]
             user_input_callback=cb,
             plan_file_path=plan_file,
         )
-        await collect_result(tool.run(ExitPlanModeArgs(), ctx))
-        assert isinstance(cb.received_args, AskUserQuestionArgs)
-        assert cb.received_args.content_preview is None
+        result = await collect_result(tool.run(ExitPlanModeArgs(), ctx))
+        assert result.switched is True
+        assert "# My Plan" not in result.message
+        assert "source of truth" not in result.message

tests/tools/test_granular_permissions.py

@@ -34,8 +34,8 @@ from vibe.core.tools.permissions import (
     PermissionContext,
     PermissionScope,
     RequiredPermission,
+    wildcard_match,
 )
-from vibe.core.tools.utils import wildcard_match
 
 
 class TestBashGranularPermissions:

tests/tools/test_mcp.py

@@ -10,6 +10,7 @@ from unittest.mock import AsyncMock, MagicMock, patch
 from pydantic import ValidationError
 import pytest
 
+from tests.conftest import build_test_vibe_config
 from tests.stubs.fake_mcp_registry import FakeMCPRegistry
 from vibe.core.config import MCPHttp, MCPStdio, MCPStreamableHttp, VibeConfig
 from vibe.core.tools.mcp import (
@@ -763,21 +764,7 @@ class TestMCPDisableFiltering:
     def _make_config(
         mcp_servers: list[MCPHttp | MCPStdio | MCPStreamableHttp] | None = None,
     ) -> VibeConfig:
-        return cast(
-            VibeConfig,
-            type(
-                "_Cfg",
-                (),
-                {
-                    "mcp_servers": mcp_servers or [],
-                    "connectors": [],
-                    "enabled_tools": [],
-                    "disabled_tools": [],
-                    "tools": {},
-                    "tool_paths": [],
-                },
-            )(),
-        )
+        return build_test_vibe_config(mcp_servers=mcp_servers or [])
 
     def test_disabled_server_excludes_all_tools(self):
         srv = MCPHttp(

tests/tools/test_websearch.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
-from unittest.mock import AsyncMock, MagicMock, patch
+from typing import TYPE_CHECKING, cast
+from unittest.mock import AsyncMock, patch
 
 from mistralai.client import Mistral
 from mistralai.client.errors import SDKError
@@ -13,12 +14,47 @@ from mistralai.client.models import (
 )
 import pytest
 
+from tests.conftest import build_test_vibe_config
 from tests.mock.utils import collect_result
-from vibe.core.config import ProviderConfig
+from vibe.core.config import ProviderConfig, VibeConfig
 from vibe.core.tools.base import BaseToolState, InvokeContext, ToolError
 from vibe.core.tools.builtins.websearch import WebSearch, WebSearchArgs, WebSearchConfig
+from vibe.core.tools.manager import ToolManager
 from vibe.core.types import Backend
 
+if TYPE_CHECKING:
+    from vibe.core.agents.manager import AgentManager
+
+
+class InMemoryAgentManager:
+    def __init__(self, config: VibeConfig) -> None:
+        self.config = config
+
+
+def _ctx_with_config(config: VibeConfig) -> InvokeContext:
+    return InvokeContext(
+        tool_call_id="t1",
+        agent_manager=cast("AgentManager", InMemoryAgentManager(config)),
+    )
+
+
+def _mistral_provider(
+    api_key_env_var: str = "MISTRAL_API_KEY",
+    api_base: str = "https://on-prem.example.com/v1",
+) -> ProviderConfig:
+    return ProviderConfig(
+        name="mistral",
+        api_base=api_base,
+        api_key_env_var=api_key_env_var,
+        backend=Backend.MISTRAL,
+    )
+
+
+def _llamacpp_provider() -> ProviderConfig:
+    return ProviderConfig(
+        name="llamacpp", api_base="http://127.0.0.1:8080/v1", backend=Backend.GENERIC
+    )
+
 
 def _make_response(
     content: list | None = None, outputs: list | None = None
@@ -109,6 +145,68 @@ async def test_run_missing_api_key(monkeypatch):
         await collect_result(ws.run(WebSearchArgs(query="test")))
 
 
+@pytest.mark.asyncio
+async def test_run_uses_mistral_provider_api_key_env_var(monkeypatch):
+    monkeypatch.setenv("MISTRAL_API_KEY", "wrong-key")
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    config = WebSearchConfig()
+    ws = WebSearch(config_getter=lambda: config, state=BaseToolState())
+    ctx = _ctx_with_config(
+        build_test_vibe_config(providers=[_mistral_provider("TEST_API_KEY")])
+    )
+    response = _make_response(content=[TextChunk(text="The answer")])
+
+    with patch("vibe.core.tools.builtins.websearch.Mistral") as mistral_cls:
+        client = mistral_cls.return_value
+        client.__aenter__ = AsyncMock(return_value=client)
+        client.__aexit__ = AsyncMock(return_value=None)
+        client.beta.conversations.start_async = AsyncMock(return_value=response)
+
+        result = await collect_result(ws.run(WebSearchArgs(query="test query"), ctx))
+
+    assert result.answer == "The answer"
+    assert mistral_cls.call_args.kwargs["api_key"] == "provider-key"
+    assert mistral_cls.call_args.kwargs["server_url"] == "https://on-prem.example.com"
+    assert mistral_cls.call_args.kwargs["timeout_ms"] == 120000
+
+
+@pytest.mark.asyncio
+async def test_run_falls_back_to_default_api_key_env_var_when_provider_env_var_empty(
+    monkeypatch,
+):
+    monkeypatch.setenv("MISTRAL_API_KEY", "fallback-key")
+    config = WebSearchConfig()
+    ws = WebSearch(config_getter=lambda: config, state=BaseToolState())
+    ctx = _ctx_with_config(build_test_vibe_config(providers=[_mistral_provider("")]))
+    response = _make_response(content=[TextChunk(text="The answer")])
+
+    with patch("vibe.core.tools.builtins.websearch.Mistral") as mistral_cls:
+        client = mistral_cls.return_value
+        client.__aenter__ = AsyncMock(return_value=client)
+        client.__aexit__ = AsyncMock(return_value=None)
+        client.beta.conversations.start_async = AsyncMock(return_value=response)
+
+        result = await collect_result(ws.run(WebSearchArgs(query="test query"), ctx))
+
+    assert result.answer == "The answer"
+    assert mistral_cls.call_args.kwargs["api_key"] == "fallback-key"
+
+
+@pytest.mark.asyncio
+async def test_run_reports_configured_api_key_env_var_when_missing(monkeypatch):
+    monkeypatch.setenv("MISTRAL_API_KEY", "fallback-key")
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    ctx = _ctx_with_config(
+        build_test_vibe_config(providers=[_mistral_provider("TEST_API_KEY")])
+    )
+    monkeypatch.delenv("TEST_API_KEY", raising=False)
+    config = WebSearchConfig()
+    ws = WebSearch(config_getter=lambda: config, state=BaseToolState())
+
+    with pytest.raises(ToolError, match="TEST_API_KEY"):
+        await collect_result(ws.run(WebSearchArgs(query="test"), ctx))
+
+
 @pytest.mark.asyncio
 async def test_run_returns_parsed_result(websearch):
     response = _make_response(
@@ -165,48 +263,23 @@ def test_resolve_server_url_no_agent_manager(websearch):
 
 
 def test_resolve_server_url_with_mistral_provider(websearch):
-    config = MagicMock()
-    config.providers = [
-        ProviderConfig(
-            name="mistral",
-            api_base="https://on-prem.example.com/v1",
-            api_key_env_var="MISTRAL_API_KEY",
-            backend=Backend.MISTRAL,
-        )
-    ]
-    agent_manager = MagicMock()
-    agent_manager.config = config
-
-    ctx = InvokeContext(tool_call_id="t1", agent_manager=agent_manager)
+    ctx = _ctx_with_config(build_test_vibe_config(providers=[_mistral_provider()]))
     assert websearch._resolve_server_url(ctx) == "https://on-prem.example.com"
 
 
 def test_resolve_server_url_with_default_provider(websearch):
-    config = MagicMock()
-    config.providers = [
-        ProviderConfig(
-            name="mistral",
-            api_base="https://api.mistral.ai/v1",
-            api_key_env_var="MISTRAL_API_KEY",
-            backend=Backend.MISTRAL,
+    ctx = _ctx_with_config(
+        build_test_vibe_config(
+            providers=[_mistral_provider(api_base="https://api.mistral.ai/v1")]
         )
-    ]
-    agent_manager = MagicMock()
-    agent_manager.config = config
-
-    ctx = InvokeContext(tool_call_id="t1", agent_manager=agent_manager)
+    )
     assert websearch._resolve_server_url(ctx) == "https://api.mistral.ai"
 
 
 def test_resolve_server_url_no_mistral_provider(websearch):
-    config = MagicMock()
-    config.providers = [
-        ProviderConfig(name="llamacpp", api_base="http://127.0.0.1:8080/v1")
-    ]
-    agent_manager = MagicMock()
-    agent_manager.config = config
-
-    ctx = InvokeContext(tool_call_id="t1", agent_manager=agent_manager)
+    ctx = _ctx_with_config(
+        build_test_vibe_config(active_model="local", providers=[_llamacpp_provider()])
+    )
     assert websearch._resolve_server_url(ctx) is None
 
 
@@ -220,5 +293,86 @@ def test_is_available_without_key(monkeypatch):
     assert WebSearch.is_available() is False
 
 
+def test_is_available_uses_mistral_provider_api_key_env_var(monkeypatch):
+    monkeypatch.setenv("MISTRAL_API_KEY", "fallback-key")
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    config = build_test_vibe_config(providers=[_mistral_provider("TEST_API_KEY")])
+    monkeypatch.delenv("TEST_API_KEY", raising=False)
+
+    assert WebSearch.is_available(config) is False
+
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    assert WebSearch.is_available(config) is True
+
+
+def test_is_available_uses_non_active_mistral_provider(monkeypatch):
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    config = build_test_vibe_config(
+        active_model="local",
+        providers=[_llamacpp_provider(), _mistral_provider("TEST_API_KEY")],
+    )
+    monkeypatch.delenv("TEST_API_KEY", raising=False)
+
+    assert WebSearch.is_available(config) is False
+
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    assert WebSearch.is_available(config) is True
+
+
+def test_is_available_falls_back_to_default_api_key_env_var_without_mistral_provider(
+    monkeypatch,
+):
+    monkeypatch.setenv("MISTRAL_API_KEY", "fallback-key")
+    config = build_test_vibe_config(
+        active_model="local", providers=[_llamacpp_provider()]
+    )
+
+    assert WebSearch.is_available(config) is True
+
+    monkeypatch.delenv("MISTRAL_API_KEY")
+
+    assert WebSearch.is_available(config) is False
+
+
+def test_is_available_falls_back_to_default_api_key_env_var_when_provider_env_var_empty(
+    monkeypatch,
+):
+    monkeypatch.setenv("MISTRAL_API_KEY", "fallback-key")
+    config = build_test_vibe_config(providers=[_mistral_provider("")])
+
+    assert WebSearch.is_available(config) is True
+
+    monkeypatch.delenv("MISTRAL_API_KEY")
+
+    assert WebSearch.is_available(config) is False
+
+
+def test_tool_manager_websearch_availability_uses_provider_api_key_env_var(monkeypatch):
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+    monkeypatch.setenv("TEST_API_KEY", "provider-key")
+    config = build_test_vibe_config(providers=[_mistral_provider("TEST_API_KEY")])
+    manager = ToolManager(lambda: config)
+
+    assert "web_search" in manager.available_tools
+
+    monkeypatch.delenv("TEST_API_KEY")
+    assert "web_search" not in manager.available_tools
+
+
+def test_tool_manager_websearch_availability_falls_back_without_mistral_provider(
+    monkeypatch,
+):
+    monkeypatch.setenv("MISTRAL_API_KEY", "fallback-key")
+    config = build_test_vibe_config(
+        active_model="local", providers=[_llamacpp_provider()]
+    )
+    manager = ToolManager(lambda: config)
+
+    assert "web_search" in manager.available_tools
+
+    monkeypatch.delenv("MISTRAL_API_KEY")
+    assert "web_search" not in manager.available_tools
+
+
 def test_get_status_text():
     assert WebSearch.get_status_text() == "Searching the web"

tests/tools/test_wildcard_match.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from vibe.core.tools.utils import wildcard_match
+from vibe.core.tools.permissions import wildcard_match
 
 
 class TestWildcardMatch: