v2.3.0 (#429)

Co-authored-by: Carlo <carloantonio.patti@mistral.ai>
Co-authored-by: Mathias Gesbert <mathias.gesbert@mistral.ai>
Co-authored-by: Clement Sirieix <clem.sirieix@gmail.com>
Co-authored-by: Michel Thomazo <michel.thomazo@mistral.ai>
Co-authored-by: Clément Drouin <clement.drouin@mistral.ai>
Co-authored-by: Vincent Guilloux <vincent.guilloux@mistral.ai>
Co-authored-by: Thomas Kenbeek <thomas.kenbeek@mistral.ai>
Co-authored-by: Mistral Vibe <vibe@mistral.ai>

tests/acp/test_initialize.py

@@ -28,7 +28,7 @@ class TestACPInitialize:
             session_capabilities=SessionCapabilities(list=SessionListCapabilities()),
         )
         assert response.agent_info == Implementation(
-            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.2.1"
+            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.3.0"
         )
 
         assert response.auth_methods == []
@@ -52,7 +52,7 @@ class TestACPInitialize:
             session_capabilities=SessionCapabilities(list=SessionListCapabilities()),
         )
         assert response.agent_info == Implementation(
-            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.2.1"
+            name="@mistralai/mistral-vibe", title="Mistral Vibe", version="2.3.0"
         )
 
         assert response.auth_methods is not None

tests/acp/test_load_session.py

@@ -34,7 +34,10 @@ def acp_agent_with_session_config(
         models=[
             ModelConfig(
                 name="devstral-latest", provider="mistral", alias="devstral-latest"
-            )
+            ),
+            ModelConfig(
+                name="devstral-small", provider="mistral", alias="devstral-small"
+            ),
         ],
         session_logging=session_config,
     )
@@ -58,7 +61,7 @@ def acp_agent_with_session_config(
 
 class TestLoadSession:
     @pytest.mark.asyncio
-    async def test_load_session_returns_response_with_models_and_modes(
+    async def test_load_session_response_structure(
         self,
         acp_agent_with_session_config: tuple[VibeAcpAgentLoop, FakeClient],
         temp_session_dir: Path,
@@ -76,9 +79,49 @@ class TestLoadSession:
 
         assert response is not None
         assert response.models is not None
+        assert len(response.models.available_models) == 2
+
         assert response.models.current_model_id == "devstral-latest"
+        assert response.models.available_models[0].model_id == "devstral-latest"
+        assert response.models.available_models[0].name == "devstral-latest"
+        assert response.models.available_models[1].model_id == "devstral-small"
+        assert response.models.available_models[1].name == "devstral-small"
+
         assert response.modes is not None
         assert response.modes.current_mode_id == BuiltinAgentName.DEFAULT
+        modes_ids = {m.id for m in response.modes.available_modes}
+        assert modes_ids == {
+            BuiltinAgentName.DEFAULT,
+            BuiltinAgentName.CHAT,
+            BuiltinAgentName.AUTO_APPROVE,
+            BuiltinAgentName.PLAN,
+            BuiltinAgentName.ACCEPT_EDITS,
+        }
+
+        assert response.config_options is not None
+        assert len(response.config_options) == 2
+        assert response.config_options[0].root.id == "mode"
+        assert response.config_options[0].root.category == "mode"
+        assert response.config_options[0].root.current_value == BuiltinAgentName.DEFAULT
+        assert len(response.config_options[0].root.options) == 5
+        mode_option_values = {
+            opt.value for opt in response.config_options[0].root.options
+        }
+        assert mode_option_values == {
+            BuiltinAgentName.DEFAULT,
+            BuiltinAgentName.CHAT,
+            BuiltinAgentName.AUTO_APPROVE,
+            BuiltinAgentName.PLAN,
+            BuiltinAgentName.ACCEPT_EDITS,
+        }
+        assert response.config_options[1].root.id == "model"
+        assert response.config_options[1].root.category == "model"
+        assert response.config_options[1].root.current_value == "devstral-latest"
+        assert len(response.config_options[1].root.options) == 2
+        model_option_values = {
+            opt.value for opt in response.config_options[1].root.options
+        }
+        assert model_option_values == {"devstral-latest", "devstral-small"}
 
     @pytest.mark.asyncio
     async def test_load_session_registers_session_with_original_id(

tests/acp/test_new_session.py

@@ -48,7 +48,7 @@ class TestACPNewSession:
         )
 
         new_session_events = [
-            e for e in telemetry_events if e.get("event_name") == "vibe/new_session"
+            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"
@@ -84,18 +84,47 @@ class TestACPNewSession:
         assert session_response.modes is not None
         assert session_response.modes.current_mode_id is not None
         assert session_response.modes.available_modes is not None
-        assert len(session_response.modes.available_modes) == 4
+        assert len(session_response.modes.available_modes) == 5
 
         assert session_response.modes.current_mode_id == BuiltinAgentName.DEFAULT
         # Check that all primary agents are available (order may vary)
         mode_ids = {m.id for m in session_response.modes.available_modes}
         assert mode_ids == {
             BuiltinAgentName.DEFAULT,
+            BuiltinAgentName.CHAT,
             BuiltinAgentName.AUTO_APPROVE,
             BuiltinAgentName.PLAN,
             BuiltinAgentName.ACCEPT_EDITS,
         }
 
+        # Check config_options
+        assert session_response.config_options is not None
+        assert len(session_response.config_options) == 2
+
+        # Mode config option
+        mode_config = session_response.config_options[0]
+        assert mode_config.root.id == "mode"
+        assert mode_config.root.category == "mode"
+        assert mode_config.root.current_value == BuiltinAgentName.DEFAULT
+        assert len(mode_config.root.options) == 5
+        mode_option_values = {opt.value for opt in mode_config.root.options}
+        assert mode_option_values == {
+            BuiltinAgentName.DEFAULT,
+            BuiltinAgentName.CHAT,
+            BuiltinAgentName.AUTO_APPROVE,
+            BuiltinAgentName.PLAN,
+            BuiltinAgentName.ACCEPT_EDITS,
+        }
+
+        # Model config option
+        model_config = session_response.config_options[1]
+        assert model_config.root.id == "model"
+        assert model_config.root.category == "model"
+        assert model_config.root.current_value == "devstral-latest"
+        assert len(model_config.root.options) == 2
+        model_option_values = {opt.value for opt in model_config.root.options}
+        assert model_option_values == {"devstral-latest", "devstral-small"}
+
     @pytest.mark.skip(reason="TODO: Fix this test")
     @pytest.mark.asyncio
     async def test_new_session_preserves_model_after_set_model(

tests/acp/test_set_config_option.py

@@ -0,0 +1,287 @@
+from __future__ import annotations
+
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+from tests.acp.conftest import _create_acp_agent
+from tests.conftest import build_test_vibe_config
+from vibe.acp.acp_agent_loop import VibeAcpAgentLoop
+from vibe.core.agent_loop import AgentLoop
+from vibe.core.agents.models import BuiltinAgentName
+from vibe.core.config import ModelConfig, VibeConfig
+
+
+@pytest.fixture
+def acp_agent_loop(backend) -> VibeAcpAgentLoop:
+    config = build_test_vibe_config(
+        active_model="devstral-latest",
+        models=[
+            ModelConfig(
+                name="devstral-latest",
+                provider="mistral",
+                alias="devstral-latest",
+                input_price=0.4,
+                output_price=2.0,
+            ),
+            ModelConfig(
+                name="devstral-small",
+                provider="mistral",
+                alias="devstral-small",
+                input_price=0.1,
+                output_price=0.3,
+            ),
+        ],
+    )
+
+    VibeConfig.dump_config(config.model_dump())
+
+    class PatchedAgentLoop(AgentLoop):
+        def __init__(self, *args, **kwargs) -> None:
+            super().__init__(*args, **{**kwargs, "backend": backend})
+            self._base_config = config
+            self.agent_manager.invalidate_config()
+            try:
+                active_model = config.get_active_model()
+                self.stats.input_price_per_million = active_model.input_price
+                self.stats.output_price_per_million = active_model.output_price
+            except ValueError:
+                pass
+
+    patch("vibe.acp.acp_agent_loop.AgentLoop", side_effect=PatchedAgentLoop).start()
+
+    return _create_acp_agent()
+
+
+class TestACPSetConfigOptionMode:
+    @pytest.mark.asyncio
+    async def test_set_config_option_mode_to_auto_approve(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+        assert acp_session.agent_loop.agent_profile.name == BuiltinAgentName.DEFAULT
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="mode", value=BuiltinAgentName.AUTO_APPROVE
+        )
+
+        assert response is not None
+        assert response.config_options is not None
+        assert len(response.config_options) == 2
+        assert (
+            acp_session.agent_loop.agent_profile.name == BuiltinAgentName.AUTO_APPROVE
+        )
+        assert acp_session.agent_loop.auto_approve is True
+
+        # Verify config_options reflect the new state
+        mode_config = response.config_options[0]
+        assert mode_config.root.id == "mode"
+        assert mode_config.root.current_value == BuiltinAgentName.AUTO_APPROVE
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_mode_to_plan(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="mode", value=BuiltinAgentName.PLAN
+        )
+
+        assert response is not None
+        assert acp_session.agent_loop.agent_profile.name == BuiltinAgentName.PLAN
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_mode_invalid_returns_none(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+        initial_mode = acp_session.agent_loop.agent_profile.name
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="mode", value="invalid-mode"
+        )
+
+        assert response is None
+        assert acp_session.agent_loop.agent_profile.name == initial_mode
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_mode_empty_string_returns_none(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+        initial_mode = acp_session.agent_loop.agent_profile.name
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="mode", value=""
+        )
+
+        assert response is None
+        assert acp_session.agent_loop.agent_profile.name == initial_mode
+
+
+class TestACPSetConfigOptionModel:
+    @pytest.mark.asyncio
+    async def test_set_config_option_model_success(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+        assert acp_session.agent_loop.config.active_model == "devstral-latest"
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="model", value="devstral-small"
+        )
+
+        assert response is not None
+        assert response.config_options is not None
+        assert len(response.config_options) == 2
+        assert acp_session.agent_loop.config.active_model == "devstral-small"
+
+        # Verify config_options reflect the new state
+        model_config = response.config_options[1]
+        assert model_config.root.id == "model"
+        assert model_config.root.current_value == "devstral-small"
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_model_invalid_returns_none(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+        initial_model = acp_session.agent_loop.config.active_model
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="model", value="non-existent-model"
+        )
+
+        assert response is None
+        assert acp_session.agent_loop.config.active_model == initial_model
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_model_empty_string_returns_none(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+        initial_model = acp_session.agent_loop.config.active_model
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="model", value=""
+        )
+
+        assert response is None
+        assert acp_session.agent_loop.config.active_model == initial_model
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_model_saves_to_config(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+
+        with patch("vibe.acp.acp_agent_loop.VibeConfig.save_updates") as mock_save:
+            response = await acp_agent_loop.set_config_option(
+                session_id=session_id, config_id="model", value="devstral-small"
+            )
+
+            assert response is not None
+            mock_save.assert_called_once_with({"active_model": "devstral-small"})
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_model_does_not_save_on_invalid(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+
+        with patch("vibe.acp.acp_agent_loop.VibeConfig.save_updates") as mock_save:
+            response = await acp_agent_loop.set_config_option(
+                session_id=session_id, config_id="model", value="non-existent-model"
+            )
+
+            assert response is None
+            mock_save.assert_not_called()
+
+
+class TestACPSetConfigOptionInvalidConfigId:
+    @pytest.mark.asyncio
+    async def test_set_config_option_invalid_config_id_returns_none(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="invalid_config", value="some_value"
+        )
+
+        assert response is None
+
+    @pytest.mark.asyncio
+    async def test_set_config_option_empty_config_id_returns_none(
+        self, acp_agent_loop: VibeAcpAgentLoop
+    ) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+
+        response = await acp_agent_loop.set_config_option(
+            session_id=session_id, config_id="", value="some_value"
+        )
+
+        assert response is None

tests/acp/test_set_mode.py

@@ -106,6 +106,29 @@ class TestACPSetMode:
             acp_session.agent_loop.auto_approve is False
         )  # Accept Edits mode doesn't auto-approve all
 
+    @pytest.mark.asyncio
+    async def test_set_mode_to_chat(self, acp_agent_loop: VibeAcpAgentLoop) -> None:
+        session_response = await acp_agent_loop.new_session(
+            cwd=str(Path.cwd()), mcp_servers=[]
+        )
+        session_id = session_response.session_id
+        acp_session = next(
+            (s for s in acp_agent_loop.sessions.values() if s.id == session_id), None
+        )
+        assert acp_session is not None
+
+        assert acp_session.agent_loop.agent_profile.name == BuiltinAgentName.DEFAULT
+
+        response = await acp_agent_loop.set_session_mode(
+            session_id=session_id, mode_id=BuiltinAgentName.CHAT
+        )
+
+        assert response is not None
+        assert acp_session.agent_loop.agent_profile.name == BuiltinAgentName.CHAT
+        assert (
+            acp_session.agent_loop.auto_approve is True
+        )  # Chat mode auto-approves read-only tools
+
     @pytest.mark.asyncio
     async def test_set_mode_invalid_mode_returns_none(
         self, acp_agent_loop: VibeAcpAgentLoop

tests/acp/test_tool_call_session_update.py

@@ -0,0 +1,44 @@
+from __future__ import annotations
+
+from acp.schema import ToolCallStart
+
+from vibe.acp.tools.session_update import tool_call_session_update
+from vibe.core.tools.builtins.read_file import ReadFile, ReadFileArgs
+from vibe.core.types import ToolCallEvent
+
+
+class TestToolCallSessionUpdate:
+    def _create_event(self) -> ToolCallEvent:
+        return ToolCallEvent(
+            tool_name="read_file",
+            tool_call_id="test_call_123",
+            args=ReadFileArgs(path="/tmp/test.txt"),
+            tool_class=ReadFile,
+        )
+
+    def test_returns_tool_call_start(self) -> None:
+        event = self._create_event()
+
+        update = tool_call_session_update(event)
+
+        assert update is not None
+        assert isinstance(update, ToolCallStart)
+        assert update.session_update == "tool_call"
+        assert update.tool_call_id == "test_call_123"
+
+    def test_returns_tool_call_start_for_streaming_event(self) -> None:
+        event = ToolCallEvent(
+            tool_name="read_file",
+            tool_call_id="test_call_123",
+            tool_class=ReadFile,
+            args=None,
+        )
+
+        update = tool_call_session_update(event)
+
+        assert update is not None
+        assert isinstance(update, ToolCallStart)
+        assert update.session_update == "tool_call"
+        assert update.tool_call_id == "test_call_123"
+        assert update.kind == "read"
+        assert update.raw_input is None

tests/autocompletion/test_file_indexer.py

@@ -89,11 +89,15 @@ def test_updates_index_on_file_rename(
     old_file.rename(new_file)
 
     assert _wait_for(
-        lambda: all(
-            entry.rel != old_file.name for entry in file_indexer.get_index(Path("."))
-        )
-        and any(
-            entry.rel == new_file.name for entry in file_indexer.get_index(Path("."))
+        lambda: (
+            all(
+                entry.rel != old_file.name
+                for entry in file_indexer.get_index(Path("."))
+            )
+            and any(
+                entry.rel == new_file.name
+                for entry in file_indexer.get_index(Path("."))
+            )
         )
     )
 

tests/autocompletion/test_ui_chat_autocompletion.py

@@ -118,7 +118,7 @@ async def test_pressing_enter_submits_selected_command_and_hides_popup(
         slash_used = [
             e
             for e in telemetry_events
-            if e.get("event_name") == "vibe/slash_command_used"
+            if e.get("event_name") == "vibe.slash_command_used"
         ]
         assert any(
             e.get("properties", {}).get("command") == "help"

tests/cli/test_bell_notifications.py

@@ -0,0 +1,151 @@
+from __future__ import annotations
+
+from unittest.mock import MagicMock, PropertyMock
+
+import pytest
+
+from vibe.cli.textual_ui.notifications import (
+    NotificationContext,
+    TextualNotificationAdapter,
+)
+
+
+def _make_fake_app(*, is_headless: bool = False) -> MagicMock:
+    app = MagicMock()
+    type(app).is_headless = PropertyMock(return_value=is_headless)
+    return app
+
+
+@pytest.fixture
+def fake_app() -> MagicMock:
+    return _make_fake_app()
+
+
+@pytest.fixture
+def adapter_enabled(fake_app: MagicMock) -> TextualNotificationAdapter:
+    return TextualNotificationAdapter(
+        fake_app, get_enabled=lambda: True, default_title="Vibe"
+    )
+
+
+@pytest.fixture
+def adapter_disabled(fake_app: MagicMock) -> TextualNotificationAdapter:
+    return TextualNotificationAdapter(
+        fake_app, get_enabled=lambda: False, default_title="Vibe"
+    )
+
+
+class TestTextualNotificationAdapter:
+    def test_no_notification_when_disabled(
+        self, adapter_disabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_disabled.on_blur()
+
+        adapter_disabled.notify(NotificationContext.ACTION_REQUIRED)
+
+        fake_app.bell.assert_not_called()
+
+    def test_no_notification_when_focused(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.notify(NotificationContext.ACTION_REQUIRED)
+
+        fake_app.bell.assert_not_called()
+
+    def test_notification_sent_when_unfocused_and_enabled(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.on_blur()
+
+        adapter_enabled.notify(NotificationContext.ACTION_REQUIRED)
+
+        fake_app.bell.assert_called_once()
+        fake_app._driver.write.assert_called_once_with(
+            "\x1b]0;Vibe - Action Required\x07"
+        )
+
+    def test_throttle_prevents_rapid_notifications(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.on_blur()
+
+        adapter_enabled.notify(NotificationContext.ACTION_REQUIRED)
+        assert fake_app.bell.call_count == 1
+
+        adapter_enabled.notify(NotificationContext.ACTION_REQUIRED)
+        assert fake_app.bell.call_count == 1
+
+    def test_contextual_title_for_action_required(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.on_blur()
+
+        adapter_enabled.notify(NotificationContext.ACTION_REQUIRED)
+
+        fake_app._driver.write.assert_called_once_with(
+            "\x1b]0;Vibe - Action Required\x07"
+        )
+
+    def test_contextual_title_for_complete(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.on_blur()
+
+        adapter_enabled.notify(NotificationContext.COMPLETE)
+
+        fake_app._driver.write.assert_called_once_with(
+            "\x1b]0;Vibe - Task Complete\x07"
+        )
+
+    def test_restore_sets_default_title(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.restore()
+
+        fake_app._driver.write.assert_called_once_with("\x1b]0;Vibe\x07")
+
+    def test_on_focus_restores_title(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.on_blur()
+        adapter_enabled.on_focus()
+
+        fake_app._driver.write.assert_called_once_with("\x1b]0;Vibe\x07")
+
+    def test_on_focus_prevents_notifications(
+        self, adapter_enabled: TextualNotificationAdapter, fake_app: MagicMock
+    ) -> None:
+        adapter_enabled.on_blur()
+        adapter_enabled.on_focus()
+        fake_app.reset_mock()
+
+        adapter_enabled.notify(NotificationContext.ACTION_REQUIRED)
+
+        fake_app.bell.assert_not_called()
+
+    def test_no_title_write_when_headless(self) -> None:
+        app = _make_fake_app(is_headless=True)
+        adapter = TextualNotificationAdapter(
+            app, get_enabled=lambda: True, default_title="Vibe"
+        )
+        adapter.on_blur()
+
+        adapter.notify(NotificationContext.ACTION_REQUIRED)
+
+        app.bell.assert_called_once()
+        app._driver.write.assert_not_called()
+
+    def test_enabled_callback_reads_live_value(self, fake_app: MagicMock) -> None:
+        enabled = True
+        adapter = TextualNotificationAdapter(
+            fake_app, get_enabled=lambda: enabled, default_title="Vibe"
+        )
+        adapter.on_blur()
+
+        adapter.notify(NotificationContext.ACTION_REQUIRED)
+        assert fake_app.bell.call_count == 1
+
+        enabled = False
+        adapter._last_notification_time = 0.0
+        adapter.notify(NotificationContext.ACTION_REQUIRED)
+        assert fake_app.bell.call_count == 1

tests/cli/test_commands.py

@@ -52,3 +52,11 @@ class TestCommandRegistry:
         assert registry.get_command_name("/exit") is None
         assert registry.find_command("/exit") is None
         assert registry.get_command_name("/help") == "help"
+
+    def test_resume_command_registration(self) -> None:
+        registry = CommandRegistry()
+        assert registry.get_command_name("/resume") == "resume"
+        assert registry.get_command_name("/continue") == "resume"
+        cmd = registry.find_command("/resume")
+        assert cmd is not None
+        assert cmd.handler == "_show_session_picker"

tests/cli/test_switching_mode.py

@@ -0,0 +1,109 @@
+from __future__ import annotations
+
+import pytest
+
+from tests.conftest import build_test_vibe_app
+from vibe.cli.textual_ui.widgets.chat_input import ChatInputContainer
+from vibe.cli.textual_ui.widgets.chat_input.body import ChatInputBody, _PromptSpinner
+from vibe.cli.textual_ui.widgets.messages import UserMessage
+
+
+@pytest.mark.asyncio
+async def test_submit_ignored_while_switching_mode() -> None:
+    """Enter press during mode switch must not clear input or send a message."""
+    app = build_test_vibe_app()
+    async with app.run_test() as pilot:
+        await pilot.pause(0.1)
+
+        body = app.query_one(ChatInputBody)
+        body.switching_mode = True
+        await pilot.pause(0.1)
+
+        # Type some text and press enter
+        app.query_one(ChatInputContainer).value = "hello world"
+        await pilot.press("enter")
+        await pilot.pause(0.1)
+
+        # Text must remain in the input
+        assert app.query_one(ChatInputContainer).value == "hello world"
+        # No user message should have been posted
+        assert len(app.query(UserMessage)) == 0
+
+
+@pytest.mark.asyncio
+async def test_submit_works_after_switching_mode_ends() -> None:
+    """After switching_mode is set back to False, Enter should work normally."""
+    app = build_test_vibe_app()
+    async with app.run_test() as pilot:
+        await pilot.pause(0.1)
+
+        body = app.query_one(ChatInputBody)
+
+        # Enable then disable switching mode
+        body.switching_mode = True
+        await pilot.pause(0.1)
+        body.switching_mode = False
+        await pilot.pause(0.1)
+
+        # Now submit should work
+        app.query_one(ChatInputContainer).value = "hello"
+        await pilot.press("enter")
+        await pilot.pause(0.1)
+
+        assert app.query_one(ChatInputContainer).value == ""
+
+
+@pytest.mark.asyncio
+async def test_spinner_shown_while_switching_mode() -> None:
+    """Prompt widget is hidden and spinner is mounted when switching_mode is True."""
+    app = build_test_vibe_app()
+    async with app.run_test() as pilot:
+        await pilot.pause(0.1)
+
+        body = app.query_one(ChatInputBody)
+        prompt = body.prompt_widget
+        assert prompt is not None
+        assert prompt.display is True
+        assert len(body.query(_PromptSpinner)) == 0
+
+        body.switching_mode = True
+        await pilot.pause(0.1)
+
+        assert prompt.display is False
+        assert len(body.query(_PromptSpinner)) == 1
+
+
+@pytest.mark.asyncio
+async def test_spinner_removed_after_switching_mode_ends() -> None:
+    """Prompt is restored and spinner removed when switching_mode becomes False."""
+    app = build_test_vibe_app()
+    async with app.run_test() as pilot:
+        await pilot.pause(0.1)
+
+        body = app.query_one(ChatInputBody)
+        body.switching_mode = True
+        await pilot.pause(0.1)
+        body.switching_mode = False
+        await pilot.pause(0.1)
+
+        assert body.prompt_widget is not None
+        assert body.prompt_widget.display is True
+        assert len(body.query(_PromptSpinner)) == 0
+
+
+@pytest.mark.asyncio
+async def test_rapid_switching_mode_no_duplicate_spinners() -> None:
+    """Rapidly toggling switching_mode must never produce duplicate spinners."""
+    app = build_test_vibe_app()
+    async with app.run_test() as pilot:
+        await pilot.pause(0.1)
+
+        body = app.query_one(ChatInputBody)
+
+        # Rapidly toggle several times
+        for _ in range(5):
+            body.switching_mode = True
+            body.switching_mode = True  # double set
+        await pilot.pause(0.1)
+
+        assert len(body.query(_PromptSpinner)) == 1

tests/cli/test_ui_session_incremental_renderer.py

@@ -144,8 +144,9 @@ async def test_ui_session_incremental_loader_load_more_batches_until_done(
             app.post_message(HistoryLoadMoreRequested())
             await _wait_until(
                 pilot.pause,
-                lambda current_count=current_count: len(app.query(UserMessage))
-                > current_count,
+                lambda current_count=current_count: (
+                    len(app.query(UserMessage)) > current_count
+                ),
             )
 
         await _wait_until(

tests/cli/textual_ui/test_session_picker.py

@@ -0,0 +1,131 @@
+from __future__ import annotations
+
+from datetime import UTC, datetime, timedelta
+
+import pytest
+
+from vibe.cli.textual_ui.widgets.session_picker import (
+    SessionPickerApp,
+    _format_relative_time,
+)
+from vibe.core.session.session_loader import SessionInfo
+
+
+@pytest.fixture
+def sample_sessions() -> list[SessionInfo]:
+    return [
+        SessionInfo(
+            session_id="session-a",
+            cwd="/test",
+            title="Session A",
+            end_time=(datetime.now(UTC) - timedelta(minutes=5)).isoformat(),
+        ),
+        SessionInfo(
+            session_id="session-b",
+            cwd="/test",
+            title="Session B",
+            end_time=(datetime.now(UTC) - timedelta(hours=1)).isoformat(),
+        ),
+        SessionInfo(
+            session_id="session-c",
+            cwd="/test",
+            title="Session C",
+            end_time=(datetime.now(UTC) - timedelta(days=1)).isoformat(),
+        ),
+    ]
+
+
+@pytest.fixture
+def sample_latest_messages() -> dict[str, str]:
+    return {
+        "session-a": "Help me fix this bug",
+        "session-b": "Refactor the authentication module",
+        "session-c": "Add unit tests for the API",
+    }
+
+
+class TestFormatRelativeTime:
+    def test_just_now(self) -> None:
+        now = datetime.now(UTC).isoformat()
+        assert _format_relative_time(now) == "just now"
+
+    def test_minutes_ago(self) -> None:
+        time_5m_ago = (datetime.now(UTC) - timedelta(minutes=5)).isoformat()
+        assert _format_relative_time(time_5m_ago) == "5m ago"
+
+    def test_hours_ago(self) -> None:
+        time_2h_ago = (datetime.now(UTC) - timedelta(hours=2)).isoformat()
+        assert _format_relative_time(time_2h_ago) == "2h ago"
+
+    def test_days_ago(self) -> None:
+        time_3d_ago = (datetime.now(UTC) - timedelta(days=3)).isoformat()
+        assert _format_relative_time(time_3d_ago) == "3d ago"
+
+    def test_weeks_ago(self) -> None:
+        time_2w_ago = (datetime.now(UTC) - timedelta(weeks=2)).isoformat()
+        assert _format_relative_time(time_2w_ago) == "2w ago"
+
+    def test_none_returns_unknown(self) -> None:
+        assert _format_relative_time(None) == "unknown"
+
+    def test_invalid_format_returns_unknown(self) -> None:
+        assert _format_relative_time("not-a-date") == "unknown"
+
+    def test_handles_z_suffix(self) -> None:
+        time_str = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%SZ")
+        assert _format_relative_time(time_str) == "just now"
+
+    def test_boundary_59_seconds(self) -> None:
+        time_59s_ago = (datetime.now(UTC) - timedelta(seconds=59)).isoformat()
+        assert _format_relative_time(time_59s_ago) == "just now"
+
+    def test_boundary_60_seconds(self) -> None:
+        time_60s_ago = (datetime.now(UTC) - timedelta(seconds=60)).isoformat()
+        assert _format_relative_time(time_60s_ago) == "1m ago"
+
+
+class TestSessionPickerAppInit:
+    def test_init_sets_properties(
+        self, sample_sessions: list[SessionInfo], sample_latest_messages: dict[str, str]
+    ) -> None:
+        picker = SessionPickerApp(
+            sessions=sample_sessions, latest_messages=sample_latest_messages
+        )
+        assert picker._sessions == sample_sessions
+        assert picker._latest_messages == sample_latest_messages
+
+    def test_id_is_sessionpicker_app(self) -> None:
+        picker = SessionPickerApp(sessions=[], latest_messages={})
+        assert picker.id == "sessionpicker-app"
+
+    def test_can_focus_children_is_true(self) -> None:
+        assert SessionPickerApp.can_focus_children is True
+
+
+class TestSessionPickerMessages:
+    def test_session_selected_stores_session_id(self) -> None:
+        msg = SessionPickerApp.SessionSelected("test-session-id")
+        assert msg.session_id == "test-session-id"
+
+    def test_session_selected_with_full_uuid(self) -> None:
+        session_id = "abc12345-6789-0123-4567-89abcdef0123"
+        msg = SessionPickerApp.SessionSelected(session_id)
+        assert msg.session_id == session_id
+
+    def test_cancelled_can_be_instantiated(self) -> None:
+        msg = SessionPickerApp.Cancelled()
+        assert isinstance(msg, SessionPickerApp.Cancelled)
+
+
+class TestSessionPickerAppBindings:
+    def _get_binding_keys(self) -> list[str]:
+        keys = []
+        for binding in SessionPickerApp.BINDINGS:
+            if isinstance(binding, tuple) and len(binding) >= 1:
+                keys.append(binding[0])
+            else:
+                keys.append(binding.key)
+        return keys
+
+    def test_has_escape_binding(self) -> None:
+        assert "escape" in self._get_binding_keys()

tests/core/test_config_paths.py

@@ -3,29 +3,33 @@ from __future__ import annotations
 from pathlib import Path
 from unittest.mock import patch
 
-from vibe.core.paths.config_paths import resolve_local_skills_dirs
+from vibe.core.paths.config_paths import (
+    discover_local_agents_dirs,
+    discover_local_skills_dirs,
+    discover_local_tools_dirs,
+)
 
 
-class TestResolveLocalSkillsDirs:
+class TestDiscoverLocalSkillsDirs:
     def test_returns_empty_list_when_dir_not_trusted(self, tmp_path: Path) -> None:
         (tmp_path / ".vibe" / "skills").mkdir(parents=True)
         with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
             mock_tfm.is_trusted.return_value = False
-            assert resolve_local_skills_dirs(tmp_path) == []
+            assert discover_local_skills_dirs(tmp_path) == []
 
     def test_returns_empty_list_when_trusted_but_no_skills_dirs(
         self, tmp_path: Path
     ) -> None:
         with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
             mock_tfm.is_trusted.return_value = True
-            assert resolve_local_skills_dirs(tmp_path) == []
+            assert discover_local_skills_dirs(tmp_path) == []
 
     def test_returns_vibe_skills_only_when_only_it_exists(self, tmp_path: Path) -> None:
         vibe_skills = tmp_path / ".vibe" / "skills"
         vibe_skills.mkdir(parents=True)
         with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
             mock_tfm.is_trusted.return_value = True
-            result = resolve_local_skills_dirs(tmp_path)
+            result = discover_local_skills_dirs(tmp_path)
         assert result == [vibe_skills]
 
     def test_returns_agents_skills_only_when_only_it_exists(
@@ -35,7 +39,7 @@ class TestResolveLocalSkillsDirs:
         agents_skills.mkdir(parents=True)
         with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
             mock_tfm.is_trusted.return_value = True
-            result = resolve_local_skills_dirs(tmp_path)
+            result = discover_local_skills_dirs(tmp_path)
         assert result == [agents_skills]
 
     def test_returns_both_in_order_when_both_exist(self, tmp_path: Path) -> None:
@@ -45,7 +49,7 @@ class TestResolveLocalSkillsDirs:
         agents_skills.mkdir(parents=True)
         with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
             mock_tfm.is_trusted.return_value = True
-            result = resolve_local_skills_dirs(tmp_path)
+            result = discover_local_skills_dirs(tmp_path)
         assert result == [vibe_skills, agents_skills]
 
     def test_ignores_vibe_skills_when_file_not_dir(self, tmp_path: Path) -> None:
@@ -53,5 +57,128 @@ class TestResolveLocalSkillsDirs:
         (tmp_path / ".vibe" / "skills").write_text("", encoding="utf-8")
         with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
             mock_tfm.is_trusted.return_value = True
-            result = resolve_local_skills_dirs(tmp_path)
+            result = discover_local_skills_dirs(tmp_path)
         assert result == []
+
+    def test_finds_skills_dirs_recursively_in_trusted_folder(
+        self, tmp_path: Path
+    ) -> None:
+        (tmp_path / ".vibe" / "skills").mkdir(parents=True)
+        (tmp_path / "sub" / ".agents" / "skills").mkdir(parents=True)
+        (tmp_path / "sub" / "deep" / ".vibe" / "skills").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_skills_dirs(tmp_path)
+        assert result == [
+            tmp_path / ".vibe" / "skills",
+            tmp_path / "sub" / ".agents" / "skills",
+            tmp_path / "sub" / "deep" / ".vibe" / "skills",
+        ]
+
+    def test_does_not_descend_into_ignored_dirs(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "skills").mkdir(parents=True)
+        (tmp_path / "node_modules" / ".vibe" / "skills").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_skills_dirs(tmp_path)
+        assert result == [tmp_path / ".vibe" / "skills"]
+
+
+class TestDiscoverLocalToolsDirs:
+    def test_returns_empty_list_when_dir_not_trusted(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "tools").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = False
+            assert discover_local_tools_dirs(tmp_path) == []
+
+    def test_returns_empty_list_when_trusted_but_no_tools_dir(
+        self, tmp_path: Path
+    ) -> None:
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            assert discover_local_tools_dirs(tmp_path) == []
+
+    def test_returns_tools_dir_when_exists(self, tmp_path: Path) -> None:
+        vibe_tools = tmp_path / ".vibe" / "tools"
+        vibe_tools.mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_tools_dirs(tmp_path)
+        assert result == [vibe_tools]
+
+    def test_ignores_tools_when_file_not_dir(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe").mkdir()
+        (tmp_path / ".vibe" / "tools").write_text("", encoding="utf-8")
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_tools_dirs(tmp_path)
+        assert result == []
+
+    def test_finds_tools_dirs_recursively(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "tools").mkdir(parents=True)
+        (tmp_path / "sub" / ".vibe" / "tools").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_tools_dirs(tmp_path)
+        assert result == [
+            tmp_path / ".vibe" / "tools",
+            tmp_path / "sub" / ".vibe" / "tools",
+        ]
+
+    def test_does_not_descend_into_ignored_dirs(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "tools").mkdir(parents=True)
+        (tmp_path / ".git" / ".vibe" / "tools").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_tools_dirs(tmp_path)
+        assert result == [tmp_path / ".vibe" / "tools"]
+
+
+class TestDiscoverLocalAgentsDirs:
+    def test_returns_empty_list_when_dir_not_trusted(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "agents").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = False
+            assert discover_local_agents_dirs(tmp_path) == []
+
+    def test_returns_empty_list_when_trusted_but_no_agents_dir(
+        self, tmp_path: Path
+    ) -> None:
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            assert discover_local_agents_dirs(tmp_path) == []
+
+    def test_returns_agents_dir_when_exists(self, tmp_path: Path) -> None:
+        vibe_agents = tmp_path / ".vibe" / "agents"
+        vibe_agents.mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_agents_dirs(tmp_path)
+        assert result == [vibe_agents]
+
+    def test_ignores_agents_when_file_not_dir(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe").mkdir()
+        (tmp_path / ".vibe" / "agents").write_text("", encoding="utf-8")
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_agents_dirs(tmp_path)
+        assert result == []
+
+    def test_finds_agents_dirs_recursively(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "agents").mkdir(parents=True)
+        (tmp_path / "sub" / "deep" / ".vibe" / "agents").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_agents_dirs(tmp_path)
+        assert result == [
+            tmp_path / ".vibe" / "agents",
+            tmp_path / "sub" / "deep" / ".vibe" / "agents",
+        ]
+
+    def test_does_not_descend_into_ignored_dirs(self, tmp_path: Path) -> None:
+        (tmp_path / ".vibe" / "agents").mkdir(parents=True)
+        (tmp_path / "__pycache__" / ".vibe" / "agents").mkdir(parents=True)
+        with patch("vibe.core.paths.config_paths.trusted_folders_manager") as mock_tfm:
+            mock_tfm.is_trusted.return_value = True
+            result = discover_local_agents_dirs(tmp_path)
+        assert result == [tmp_path / ".vibe" / "agents"]

tests/core/test_file_logging.py

@@ -8,7 +8,7 @@ from unittest.mock import MagicMock, patch
 
 import pytest
 
-from vibe.core.utils import StructuredLogFormatter, apply_logging_config
+from vibe.core.logger import StructuredLogFormatter, apply_logging_config
 
 
 @pytest.fixture
@@ -19,8 +19,8 @@ def mock_log_dir(tmp_path: Path):
     mock_file = MagicMock()
     mock_file.path = tmp_path / "vibe.log"
     with (
-        patch("vibe.core.utils.LOG_DIR", mock_dir),
-        patch("vibe.core.utils.LOG_FILE", mock_file),
+        patch("vibe.core.logger.LOG_DIR", mock_dir),
+        patch("vibe.core.logger.LOG_FILE", mock_file),
     ):
         yield tmp_path
 
@@ -222,8 +222,8 @@ class TestApplyLoggingConfig:
         mock_file = MagicMock()
         mock_file.path = log_dir / "vibe.log"
         with (
-            patch("vibe.core.utils.LOG_DIR", mock_dir),
-            patch("vibe.core.utils.LOG_FILE", mock_file),
+            patch("vibe.core.logger.LOG_DIR", mock_dir),
+            patch("vibe.core.logger.LOG_FILE", mock_file),
         ):
             monkeypatch.setenv("LOG_LEVEL", "DEBUG")
             test_logger = logging.getLogger("test_mkdir")

tests/core/test_telemetry_send.py

@@ -47,7 +47,7 @@ class TestTelemetryClient:
     def test_send_telemetry_event_does_nothing_when_api_key_is_none(
         self, monkeypatch: pytest.MonkeyPatch
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         env_key = config.get_provider_for_model(
             config.get_active_model()
         ).api_key_env_var
@@ -57,7 +57,7 @@ class TestTelemetryClient:
         client._client = MagicMock()
         client._client.post = AsyncMock()
 
-        client.send_telemetry_event("vibe/test", {})
+        client.send_telemetry_event("vibe.test", {})
         _run_telemetry_tasks()
 
         client._client.post.assert_not_called()
@@ -65,7 +65,10 @@ class TestTelemetryClient:
     def test_send_telemetry_event_does_nothing_when_disabled(
         self, monkeypatch: pytest.MonkeyPatch
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=True)
+        monkeypatch.setattr(
+            TelemetryClient, "send_telemetry_event", _original_send_telemetry_event
+        )
+        config = build_test_vibe_config(enable_telemetry=False)
         env_key = config.get_provider_for_model(
             config.get_active_model()
         ).api_key_env_var
@@ -74,7 +77,7 @@ class TestTelemetryClient:
         client._client = MagicMock()
         client._client.post = AsyncMock()
 
-        client.send_telemetry_event("vibe/test", {})
+        client.send_telemetry_event("vibe.test", {})
         _run_telemetry_tasks()
 
         client._client.post.assert_not_called()
@@ -86,7 +89,7 @@ class TestTelemetryClient:
         monkeypatch.setattr(
             TelemetryClient, "send_telemetry_event", _original_send_telemetry_event
         )
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         env_key = config.get_provider_for_model(
             config.get_active_model()
         ).api_key_env_var
@@ -97,12 +100,12 @@ class TestTelemetryClient:
         client._client.post = mock_post
         client._client.aclose = AsyncMock()
 
-        client.send_telemetry_event("vibe/test_event", {"key": "value"})
+        client.send_telemetry_event("vibe.test_event", {"key": "value"})
         await client.aclose()
 
         mock_post.assert_called_once_with(
             DATALAKE_EVENTS_URL,
-            json={"event": "vibe/test_event", "properties": {"key": "value"}},
+            json={"event": "vibe.test_event", "properties": {"key": "value"}},
             headers={
                 "Content-Type": "application/json",
                 "Authorization": "Bearer sk-test",
@@ -113,7 +116,7 @@ class TestTelemetryClient:
     def test_send_tool_call_finished_payload_shape(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
         tool_call = _make_resolved_tool_call("todo", {})
         decision = ToolDecision(
@@ -129,7 +132,7 @@ class TestTelemetryClient:
 
         assert len(telemetry_events) == 1
         event_name = telemetry_events[0]["event_name"]
-        assert event_name == "vibe/tool_call_finished"
+        assert event_name == "vibe.tool_call_finished"
         properties = telemetry_events[0]["properties"]
         assert properties["tool_name"] == "todo"
         assert properties["status"] == "success"
@@ -142,7 +145,7 @@ class TestTelemetryClient:
     def test_send_tool_call_finished_nb_files_created_write_file_new(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
         tool_call = _make_resolved_tool_call("write_file", {"overwrite": False})
 
@@ -160,7 +163,7 @@ class TestTelemetryClient:
     def test_send_tool_call_finished_nb_files_modified_write_file_overwrite(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
         tool_call = _make_resolved_tool_call("write_file", {"overwrite": True})
 
@@ -178,7 +181,7 @@ class TestTelemetryClient:
     def test_send_tool_call_finished_decision_none(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
         tool_call = _make_resolved_tool_call("todo", {})
 
@@ -195,49 +198,49 @@ class TestTelemetryClient:
     def test_send_user_copied_text_payload(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
 
         client.send_user_copied_text("hello world")
 
         assert len(telemetry_events) == 1
-        assert telemetry_events[0]["event_name"] == "vibe/user_copied_text"
+        assert telemetry_events[0]["event_name"] == "vibe.user_copied_text"
         assert telemetry_events[0]["properties"]["text_length"] == 11
 
     def test_send_user_cancelled_action_payload(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
 
         client.send_user_cancelled_action("interrupt_agent")
 
         assert len(telemetry_events) == 1
-        assert telemetry_events[0]["event_name"] == "vibe/user_cancelled_action"
+        assert telemetry_events[0]["event_name"] == "vibe.user_cancelled_action"
         assert telemetry_events[0]["properties"]["action"] == "interrupt_agent"
 
     def test_send_auto_compact_triggered_payload(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
 
         client.send_auto_compact_triggered()
 
         assert len(telemetry_events) == 1
-        assert telemetry_events[0]["event_name"] == "vibe/auto_compact_triggered"
+        assert telemetry_events[0]["event_name"] == "vibe.auto_compact_triggered"
 
     def test_send_slash_command_used_payload(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
 
         client.send_slash_command_used("help", "builtin")
         client.send_slash_command_used("my_skill", "skill")
 
         assert len(telemetry_events) == 2
-        assert telemetry_events[0]["event_name"] == "vibe/slash_command_used"
+        assert telemetry_events[0]["event_name"] == "vibe.slash_command_used"
         assert telemetry_events[0]["properties"]["command"] == "help"
         assert telemetry_events[0]["properties"]["command_type"] == "builtin"
         assert telemetry_events[1]["properties"]["command"] == "my_skill"
@@ -246,7 +249,7 @@ class TestTelemetryClient:
     def test_send_new_session_payload(
         self, telemetry_events: list[dict[str, Any]]
     ) -> None:
-        config = build_test_vibe_config(disable_telemetry=False)
+        config = build_test_vibe_config(enable_telemetry=True)
         client = TelemetryClient(config_getter=lambda: config)
 
         client.send_new_session(
@@ -255,15 +258,17 @@ class TestTelemetryClient:
             nb_mcp_servers=1,
             nb_models=3,
             entrypoint="cli",
+            terminal_emulator="vscode",
         )
 
         assert len(telemetry_events) == 1
         event_name = telemetry_events[0]["event_name"]
-        assert event_name == "vibe/new_session"
+        assert event_name == "vibe.new_session"
         properties = telemetry_events[0]["properties"]
         assert properties["has_agents_md"] is True
         assert properties["nb_skills"] == 2
         assert properties["nb_mcp_servers"] == 1
         assert properties["nb_models"] == 3
         assert properties["entrypoint"] == "cli"
+        assert properties["terminal_emulator"] == "vscode"
         assert "version" in properties

tests/core/test_trusted_folders.py

@@ -237,11 +237,11 @@ class TestHasAgentsMdFile:
 
 class TestHasTrustableContent:
     def test_returns_true_when_vibe_dir_exists(self, tmp_path: Path) -> None:
-        (tmp_path / ".vibe").mkdir()
+        (tmp_path / ".vibe" / "skills").mkdir(parents=True)
         assert has_trustable_content(tmp_path) is True
 
     def test_returns_true_when_agents_dir_exists(self, tmp_path: Path) -> None:
-        (tmp_path / ".agents").mkdir()
+        (tmp_path / ".agents" / "skills").mkdir(parents=True)
         assert has_trustable_content(tmp_path) is True
 
     def test_returns_true_when_agents_md_filename_exists(self, tmp_path: Path) -> None:
@@ -253,3 +253,17 @@ class TestHasTrustableContent:
     def test_returns_false_when_no_trustable_content(self, tmp_path: Path) -> None:
         (tmp_path / "other.txt").write_text("", encoding="utf-8")
         assert has_trustable_content(tmp_path) is False
+
+    def test_returns_true_when_vibe_config_in_subfolder(self, tmp_path: Path) -> None:
+        (tmp_path / "sub" / ".vibe" / "skills").mkdir(parents=True)
+        assert has_trustable_content(tmp_path) is True
+
+    def test_returns_true_when_agents_skills_in_subfolder(self, tmp_path: Path) -> None:
+        (tmp_path / "deep" / "nested" / ".agents" / "skills").mkdir(parents=True)
+        assert has_trustable_content(tmp_path) is True
+
+    def test_returns_false_when_config_only_inside_ignored_dir(
+        self, tmp_path: Path
+    ) -> None:
+        (tmp_path / "node_modules" / ".vibe" / "skills").mkdir(parents=True)
+        assert has_trustable_content(tmp_path) is False

tests/e2e/common.py

@@ -0,0 +1,86 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from contextlib import AbstractContextManager
+import io
+from pathlib import Path
+import re
+import time
+from typing import Protocol
+
+import pexpect
+
+
+class SpawnedVibeProcessFixture(Protocol):
+    def __call__(
+        self, workdir: Path
+    ) -> AbstractContextManager[tuple[pexpect.spawn, io.StringIO]]: ...
+
+
+def ansi_tolerant_pattern(text: str) -> re.Pattern[str]:
+    ansi = r"(?:\x1b\[[0-9;?]*[ -/]*[@-~]|\x1b\][^\x07]*\x07|\r|\n)*"
+    return re.compile(ansi.join(re.escape(char) for char in text))
+
+
+def write_e2e_config(vibe_home: Path, api_base: str) -> None:
+    vibe_home.mkdir(parents=True, exist_ok=True)
+    (vibe_home / "config.toml").write_text(
+        "\n".join([
+            'active_model = "mock-model"',
+            "enable_update_checks = false",
+            "enable_auto_update = false",
+            "",
+            "[[providers]]",
+            'name = "mock-provider"',
+            f'api_base = "{api_base}"',
+            'api_key_env_var = "MISTRAL_API_KEY"',
+            'backend = "generic"',
+            "",
+            "[[models]]",
+            'name = "mock-model"',
+            'provider = "mock-provider"',
+            'alias = "mock-model"',
+        ]),
+        encoding="utf-8",
+    )
+
+
+def strip_ansi(text: str) -> str:
+    return re.sub(r"\x1b\[[0-9;?]*[ -/]*[@-~]|\x1b\][^\x07]*\x07", "", text)
+
+
+def wait_for_request_count(
+    request_count_getter: Callable[[], int], expected_count: int, timeout: float
+) -> None:
+    start = time.monotonic()
+    while time.monotonic() - start < timeout:
+        if request_count_getter() >= expected_count:
+            return
+        time.sleep(0.05)
+    raise AssertionError(f"Timed out waiting for {expected_count} backend request(s).")
+
+
+def wait_for_main_screen(child: pexpect.spawn, timeout: float = 20.0) -> None:
+    child.expect(ansi_tolerant_pattern("Mistral Vibe v"), timeout=timeout)
+
+
+def wait_for_rendered_text(
+    child: pexpect.spawn, captured: io.StringIO, needle: str, timeout: float
+) -> None:
+    start = time.monotonic()
+    while time.monotonic() - start < timeout:
+        if needle in strip_ansi(captured.getvalue()):
+            return
+        try:
+            child.expect(r"\S", timeout=0.1)
+        except pexpect.TIMEOUT:
+            pass
+        except pexpect.EOF as exc:
+            rendered_tail = strip_ansi(captured.getvalue())[-1200:]
+            raise AssertionError(
+                f"Child exited while waiting for rendered text: {needle!r}\n\nRendered tail:\n{rendered_tail}"
+            ) from exc
+    rendered_tail = strip_ansi(captured.getvalue())[-1200:]
+    raise AssertionError(
+        f"Timed out waiting for rendered text: {needle!r}\n\nRendered tail:\n{rendered_tail}"
+    )

tests/e2e/conftest.py

@@ -0,0 +1,82 @@
+from __future__ import annotations
+
+from collections.abc import Callable, Iterator
+from contextlib import AbstractContextManager, contextmanager
+import io
+import os
+from pathlib import Path
+from typing import cast
+
+import pexpect
+import pytest
+
+from tests import TESTS_ROOT
+from tests.e2e.common import write_e2e_config
+from tests.e2e.mock_server import ChunkFactory, StreamingMockServer
+
+
+@pytest.fixture
+def streaming_mock_server(
+    request: pytest.FixtureRequest,
+) -> Iterator[StreamingMockServer]:
+    chunk_factory = cast(ChunkFactory | None, getattr(request, "param", None))
+    server = StreamingMockServer(chunk_factory=chunk_factory)
+    server.start()
+    try:
+        yield server
+    finally:
+        server.stop()
+
+
+@pytest.fixture
+def setup_e2e_env(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+    streaming_mock_server: StreamingMockServer,
+) -> None:
+    vibe_home = tmp_path / "vibe-home"
+    write_e2e_config(vibe_home, streaming_mock_server.api_base)
+    monkeypatch.setenv("MISTRAL_API_KEY", "fake-key")
+    monkeypatch.setenv("VIBE_HOME", str(vibe_home))
+    monkeypatch.setenv("TERM", "xterm-256color")
+
+
+@pytest.fixture
+def e2e_workdir(tmp_path: Path) -> Path:
+    workdir = tmp_path / "workdir"
+    workdir.mkdir()
+    return workdir
+
+
+type SpawnedVibeContext = Iterator[tuple[pexpect.spawn, io.StringIO]]
+type SpawnedVibeContextManager = AbstractContextManager[
+    tuple[pexpect.spawn, io.StringIO]
+]
+type SpawnedVibeFactory = Callable[[Path], SpawnedVibeContextManager]
+
+
+@pytest.fixture
+def spawned_vibe_process() -> SpawnedVibeFactory:
+    @contextmanager
+    def spawn(workdir: Path) -> SpawnedVibeContext:
+        captured = io.StringIO()
+        child = pexpect.spawn(
+            "uv",
+            ["run", "vibe", "--workdir", str(workdir)],
+            cwd=str(TESTS_ROOT.parent),
+            env=os.environ,
+            encoding="utf-8",
+            timeout=30,
+            dimensions=(36, 120),
+        )
+        child.logfile_read = captured
+
+        try:
+            yield child, captured
+        finally:
+            if child.isalive():
+                child.terminate(force=True)
+            if not child.closed:
+                child.close()
+
+    return spawn

tests/e2e/mock_server.py

@@ -0,0 +1,149 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+import json
+import threading
+import time
+from typing import TypedDict, cast
+
+
+class StreamOptionsPayload(TypedDict, total=False):
+    include_usage: bool
+    stream_tool_calls: bool
+
+
+class ChatMessagePayload(TypedDict, total=False):
+    role: str
+    content: str
+
+
+class ChatCompletionsRequestPayload(TypedDict, total=False):
+    model: str
+    messages: list[ChatMessagePayload]
+    stream: bool
+    stream_options: StreamOptionsPayload
+
+
+type StreamChunk = dict[str, object]
+type ChunkFactory = Callable[[int, ChatCompletionsRequestPayload], list[StreamChunk]]
+
+
+class StreamingMockServer:
+    @staticmethod
+    def build_chunk(
+        *,
+        created: int,
+        delta: dict[str, object],
+        finish_reason: str | None,
+        usage: dict[str, int] | None = None,
+    ) -> StreamChunk:
+        chunk: dict[str, object] = {
+            "id": "mock-id",
+            "object": "chat.completion.chunk",
+            "created": created,
+            "model": "mock-model",
+            "choices": [{"index": 0, "delta": delta, "finish_reason": finish_reason}],
+        }
+        if usage is not None:
+            chunk["usage"] = usage
+        return chunk
+
+    @staticmethod
+    def build_tool_call_delta(
+        *, call_id: str, tool_name: str, arguments: str, index: int = 0
+    ) -> dict[str, object]:
+        return {
+            "role": "assistant",
+            "tool_calls": [
+                {
+                    "index": index,
+                    "id": call_id,
+                    "type": "function",
+                    "function": {"name": tool_name, "arguments": arguments},
+                }
+            ],
+        }
+
+    @staticmethod
+    def _stream_chunks() -> list[StreamChunk]:
+        return [
+            StreamingMockServer.build_chunk(
+                created=123,
+                delta={"role": "assistant", "content": "Hello"},
+                finish_reason=None,
+            ),
+            StreamingMockServer.build_chunk(
+                created=124, delta={"content": " from mock server"}, finish_reason=None
+            ),
+            StreamingMockServer.build_chunk(
+                created=125,
+                delta={},
+                finish_reason="stop",
+                usage={"prompt_tokens": 3, "completion_tokens": 4},
+            ),
+        ]
+
+    def __init__(self, *, chunk_factory: ChunkFactory | None = None) -> None:
+        self.requests: list[ChatCompletionsRequestPayload] = []
+        self._lock = threading.Lock()
+        self._chunk_factory = chunk_factory
+        self._server = ThreadingHTTPServer(("127.0.0.1", 0), self._build_handler())
+        self._thread = threading.Thread(target=self._server.serve_forever, daemon=True)
+
+    def _build_handler(self) -> type[BaseHTTPRequestHandler]:
+        parent = self
+
+        class Handler(BaseHTTPRequestHandler):
+            def log_message(self, format, *args):
+                return
+
+            def do_POST(self) -> None:
+                if self.path != "/v1/chat/completions":
+                    self.send_response(404)
+                    self.end_headers()
+                    return
+
+                length = int(self.headers.get("Content-Length", "0"))
+                body = self.rfile.read(length)
+                payload = cast(
+                    ChatCompletionsRequestPayload, json.loads(body.decode("utf-8"))
+                )
+
+                with parent._lock:
+                    parent.requests.append(payload)
+                    request_index = len(parent.requests) - 1
+
+                self.send_response(200)
+                self.send_header("Content-Type", "text/event-stream")
+                self.send_header("Cache-Control", "no-cache")
+                self.end_headers()
+
+                chunks = (
+                    parent._chunk_factory(request_index, payload)
+                    if parent._chunk_factory is not None
+                    else parent._stream_chunks()
+                )
+
+                for chunk in chunks:
+                    data = json.dumps(chunk, ensure_ascii=False)
+                    self.wfile.write(f"data: {data}\n\n".encode())
+                    self.wfile.flush()
+                    time.sleep(0.03)
+
+                self.wfile.write(b"data: [DONE]\n\n")
+                self.wfile.flush()
+
+        return Handler
+
+    @property
+    def api_base(self) -> str:
+        return f"http://127.0.0.1:{self._server.server_port}/v1"
+
+    def start(self) -> None:
+        self._thread.start()
+
+    def stop(self) -> None:
+        self._server.shutdown()
+        self._server.server_close()
+        self._thread.join(timeout=1)

tests/e2e/test_cli_tui_onboarding.py

@@ -0,0 +1,31 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import pexpect
+import pytest
+
+from tests.e2e.common import SpawnedVibeProcessFixture, ansi_tolerant_pattern
+
+
+@pytest.mark.timeout(15)
+def test_spawn_cli_shows_onboarding_when_api_key_missing(
+    tmp_path: Path,
+    e2e_workdir: Path,
+    spawned_vibe_process: SpawnedVibeProcessFixture,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    vibe_home = tmp_path / "vibe-home-onboarding"
+    vibe_home.mkdir(parents=True, exist_ok=True)
+
+    monkeypatch.setenv("VIBE_HOME", str(vibe_home))
+    monkeypatch.setenv("TERM", "xterm-256color")
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+
+    with spawned_vibe_process(e2e_workdir) as (child, captured):
+        child.expect(ansi_tolerant_pattern("Welcome to Mistral Vibe"), timeout=15)
+        child.sendcontrol("c")
+        child.expect(pexpect.EOF, timeout=10)
+
+    output = captured.getvalue()
+    assert "Setup cancelled" in output

tests/e2e/test_cli_tui_streaming.py

@@ -0,0 +1,51 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import pexpect
+import pytest
+
+from tests.e2e.common import (
+    SpawnedVibeProcessFixture,
+    ansi_tolerant_pattern,
+    wait_for_main_screen,
+    wait_for_request_count,
+)
+from tests.e2e.mock_server import StreamingMockServer
+
+
+@pytest.mark.timeout(15)
+def test_spawn_cli_to_send_and_receive_message(
+    streaming_mock_server: StreamingMockServer,
+    setup_e2e_env: None,
+    e2e_workdir: Path,
+    spawned_vibe_process: SpawnedVibeProcessFixture,
+) -> None:
+    with spawned_vibe_process(e2e_workdir) as (child, captured):
+        wait_for_main_screen(child, timeout=15)
+        child.send("Greet")
+        child.send("\r")
+
+        wait_for_request_count(
+            lambda: len(streaming_mock_server.requests), expected_count=1, timeout=10
+        )
+        child.expect(ansi_tolerant_pattern("Hello from mock server"), timeout=10)
+
+        child.sendcontrol("c")
+        child.expect(pexpect.EOF, timeout=10)
+
+    output = captured.getvalue()
+    assert "Welcome to Mistral Vibe" not in output
+
+    request_payload = streaming_mock_server.requests[-1]
+    assert request_payload.get("stream") is True
+    assert request_payload.get("model") == "mock-model"
+    stream_options = request_payload.get("stream_options")
+    assert stream_options is not None
+    assert stream_options.get("include_usage") is True
+    messages = request_payload.get("messages")
+    assert messages is not None
+    assert any(
+        message.get("role") == "user" and message.get("content") == "Greet"
+        for message in messages
+    )

tests/e2e/test_cli_tui_tool_approval.py

@@ -0,0 +1,87 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import pexpect
+import pytest
+
+from tests.e2e.common import (
+    SpawnedVibeProcessFixture,
+    wait_for_main_screen,
+    wait_for_rendered_text,
+    wait_for_request_count,
+)
+from tests.e2e.mock_server import ChatCompletionsRequestPayload, StreamingMockServer
+
+PREDICTABLE_OUTPUT = "__E2E_BASH_OK__"
+TOOL_ARGUMENTS = f'{{"command":"printf \\"{PREDICTABLE_OUTPUT}\\\\n\\""}}'
+
+
+def _tool_call_factory(
+    request_index: int, _payload: ChatCompletionsRequestPayload
+) -> list[dict[str, object]]:
+    if request_index == 0:
+        return [
+            StreamingMockServer.build_chunk(
+                created=1,
+                delta=StreamingMockServer.build_tool_call_delta(
+                    call_id="call_bash_1", tool_name="bash", arguments=TOOL_ARGUMENTS
+                ),
+                finish_reason=None,
+            ),
+            StreamingMockServer.build_chunk(
+                created=2,
+                delta={},
+                finish_reason="tool_calls",
+                usage={"prompt_tokens": 3, "completion_tokens": 4},
+            ),
+        ]
+
+    return [
+        StreamingMockServer.build_chunk(
+            created=3,
+            delta={
+                "role": "assistant",
+                "content": f"The string {PREDICTABLE_OUTPUT} has been printed successfully.",
+            },
+            finish_reason=None,
+        ),
+        StreamingMockServer.build_chunk(
+            created=4, delta={"content": PREDICTABLE_OUTPUT}, finish_reason=None
+        ),
+        StreamingMockServer.build_chunk(
+            created=5,
+            delta={},
+            finish_reason="stop",
+            usage={"prompt_tokens": 3, "completion_tokens": 4},
+        ),
+    ]
+
+
+@pytest.mark.timeout(25)
+@pytest.mark.parametrize(
+    "streaming_mock_server",
+    [pytest.param(_tool_call_factory, id="tool-call-stream")],
+    indirect=True,
+)
+def test_spawn_cli_asks_bash_permission_and_shows_tool_output_after_approval(
+    streaming_mock_server: StreamingMockServer,
+    setup_e2e_env: None,
+    e2e_workdir: Path,
+    spawned_vibe_process: SpawnedVibeProcessFixture,
+) -> None:
+    with spawned_vibe_process(e2e_workdir) as (child, captured):
+        wait_for_main_screen(child, timeout=15)
+        child.send("Run a shell command")
+        child.send("\r")
+
+        wait_for_request_count(
+            lambda: len(streaming_mock_server.requests), expected_count=1, timeout=10
+        )
+        wait_for_rendered_text(child, captured, needle="bash command", timeout=10)
+        child.send("y")
+        child.send("\r")
+        wait_for_rendered_text(child, captured, needle=PREDICTABLE_OUTPUT, timeout=10)
+
+        child.sendcontrol("c")
+        child.expect(pexpect.EOF, timeout=10)

tests/session/test_session_loader.py

@@ -822,3 +822,133 @@ class TestSessionLoaderListSessions:
         assert len(result) == 1
         assert result[0]["session_id"] == "notitle0"
         assert result[0]["title"] is None
+
+
+class TestSessionLoaderGetFirstUserMessage:
+    """Tests for SessionLoader.get_first_user_message method."""
+
+    def test_returns_first_user_message(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        """Test that get_first_user_message returns the first user message."""
+        session_dir = Path(session_config.save_dir)
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.user, content="First user message"),
+            LLMMessage(role=Role.assistant, content="First response"),
+            LLMMessage(role=Role.user, content="Second user message"),
+            LLMMessage(role=Role.assistant, content="Second response"),
+        ]
+        create_test_session(session_dir, "test-sess", messages=messages)
+
+        result = SessionLoader.get_first_user_message("test-sess", session_config)
+
+        assert result == "First user message"
+
+    def test_returns_fallback_for_missing_session(
+        self, session_config: SessionLoggingConfig
+    ) -> None:
+        """Test that get_first_user_message returns fallback when session not found."""
+        result = SessionLoader.get_first_user_message("nonexistent", session_config)
+
+        assert result == "(session not found)"
+
+    def test_returns_no_user_messages_fallback(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        """Test fallback when session has no user messages."""
+        session_dir = Path(session_config.save_dir)
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.assistant, content="Assistant only"),
+        ]
+        create_test_session(session_dir, "no-user0", messages=messages)
+
+        result = SessionLoader.get_first_user_message("no-user0", session_config)
+
+        assert result == "(no user messages)"
+
+    def test_replaces_newlines_with_spaces(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        """Test that newlines in messages are replaced with spaces."""
+        session_dir = Path(session_config.save_dir)
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.user, content="Line one\nLine two\nLine three"),
+        ]
+        create_test_session(session_dir, "newline0", messages=messages)
+
+        result = SessionLoader.get_first_user_message("newline0", session_config)
+
+        assert "\n" not in result
+        assert "Line one Line two Line three" == result
+
+    def test_handles_empty_user_message(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        """Test handling of empty user message content."""
+        session_dir = Path(session_config.save_dir)
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.user, content=""),
+        ]
+        create_test_session(session_dir, "empty-ms", messages=messages)
+
+        result = SessionLoader.get_first_user_message("empty-ms", session_config)
+
+        assert result == "(no user messages)"
+
+    def test_handles_whitespace_only_message(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        """Test handling of whitespace-only user message."""
+        session_dir = Path(session_config.save_dir)
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.user, content="   \n\t  "),
+        ]
+        create_test_session(session_dir, "whitespc", messages=messages)
+
+        result = SessionLoader.get_first_user_message("whitespc", session_config)
+
+        assert result == "(empty message)"
+
+    def test_handles_invalid_session_as_not_found(
+        self, session_config: SessionLoggingConfig
+    ) -> None:
+        """Test that invalid sessions (bad JSON) are treated as not found.
+
+        Note: Sessions with invalid JSON are filtered out by _is_valid_session
+        during find_session_by_id, so they return 'not found' rather than
+        'corrupted'. This is the expected behavior.
+        """
+        session_dir = Path(session_config.save_dir)
+
+        # Create a session with invalid JSON - will fail validation
+        session_folder = session_dir / "test_20230101_120000_corrupt0"
+        session_folder.mkdir()
+        (session_folder / "messages.jsonl").write_text("{invalid json}")
+        (session_folder / "meta.json").write_text('{"session_id": "corrupt0"}')
+
+        result = SessionLoader.get_first_user_message("corrupt0", session_config)
+
+        # Invalid sessions are filtered by _is_valid_session, so not found
+        assert result == "(session not found)"
+
+    def test_skips_non_user_messages(
+        self, session_config: SessionLoggingConfig, create_test_session
+    ) -> None:
+        """Test that only user messages are considered, not assistant/system."""
+        session_dir = Path(session_config.save_dir)
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.user, content="User question"),
+            LLMMessage(role=Role.assistant, content="Assistant response"),
+        ]
+        create_test_session(session_dir, "skip-non", messages=messages)
+
+        result = SessionLoader.get_first_user_message("skip-non", session_config)
+
+        # Should return "User question", not "Assistant response"
+        assert result == "User question"

tests/session/test_session_logger.py

@@ -1,10 +1,10 @@
 from __future__ import annotations
 
-from datetime import datetime, timedelta
+from datetime import UTC, datetime, timedelta
 import json
 import os
 from pathlib import Path
-from unittest.mock import MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 
@@ -502,6 +502,61 @@ class TestSessionLoggerSaveInteraction:
         with open(messages_file) as f:
             assert len(f.readlines()) == 2
 
+    @pytest.mark.asyncio
+    async def test_save_interaction_throttles_tmp_cleanup(
+        self,
+        session_config: SessionLoggingConfig,
+        mock_vibe_config: VibeConfig,
+        mock_tool_manager: ToolManager,
+        mock_agent_profile: AgentProfile,
+    ) -> None:
+        logger = SessionLogger(session_config, "test-session-123")
+
+        messages = [
+            LLMMessage(role=Role.system, content="System prompt"),
+            LLMMessage(role=Role.user, content="Hello"),
+            LLMMessage(role=Role.assistant, content="Hi there!"),
+        ]
+
+        cleanup_spy = MagicMock()
+        with (
+            patch.object(
+                SessionLogger, "persist_messages", new_callable=AsyncMock
+            ) as persist_messages_mock,
+            patch.object(
+                SessionLogger, "persist_metadata", new_callable=AsyncMock
+            ) as persist_metadata_mock,
+            patch.object(logger, "cleanup_tmp_files", cleanup_spy),
+            patch(
+                "vibe.core.session.session_logger.utc_now",
+                # a bit brittle, but required for the call-count choregraphy...
+                side_effect=[
+                    datetime(2026, 1, 1, 10, 0, 0, tzinfo=UTC),
+                    datetime(2026, 1, 1, 10, 0, 0, tzinfo=UTC),
+                    datetime(2026, 1, 1, 10, 0, 1, tzinfo=UTC),
+                    datetime(2026, 1, 1, 10, 0, 1, tzinfo=UTC),
+                ],
+            ),
+        ):
+            await logger.save_interaction(
+                messages=messages,
+                stats=AgentStats(steps=1),
+                base_config=mock_vibe_config,
+                tool_manager=mock_tool_manager,
+                agent_profile=mock_agent_profile,
+            )
+            await logger.save_interaction(
+                messages=messages,
+                stats=AgentStats(steps=2),
+                base_config=mock_vibe_config,
+                tool_manager=mock_tool_manager,
+                agent_profile=mock_agent_profile,
+            )
+
+        assert persist_messages_mock.await_count == 2
+        assert persist_metadata_mock.await_count == 2
+        assert cleanup_spy.call_count == 1
+
 
 class TestSessionLoggerResetSession:
     def test_reset_session(self, session_config: SessionLoggingConfig) -> None:
@@ -701,3 +756,27 @@ class TestSessionLoggerCleanupTmpFiles:
 
         assert old_tmp_file.exists()
         assert not another_old_tmp_file.exists()
+
+    def test_maybe_cleanup_tmp_files_throttles_calls(
+        self, session_config: SessionLoggingConfig
+    ) -> None:
+        session_id = "test-session-123"
+        logger = SessionLogger(session_config, session_id)
+
+        cleanup_spy = MagicMock()
+        with (
+            patch.object(logger, "cleanup_tmp_files", cleanup_spy),
+            patch(
+                "vibe.core.session.session_logger.utc_now",
+                side_effect=[
+                    datetime(2026, 1, 1, 10, 0, 0, tzinfo=UTC),
+                    datetime(2026, 1, 1, 10, 0, 1, tzinfo=UTC),
+                    datetime(2026, 1, 1, 10, 0, 6, tzinfo=UTC),
+                ],
+            ),
+        ):
+            logger.maybe_cleanup_tmp_files()
+            logger.maybe_cleanup_tmp_files()
+            logger.maybe_cleanup_tmp_files()
+
+        assert cleanup_spy.call_count == 2

tests/snapshots/__snapshots__/test_ui_snapshot_empty_assistant_before_reasoning/test_snapshot_empty_assistant_removed_when_reasoning_starts.svg

@@ -0,0 +1,203 @@
+<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: #ff8205 }
+.terminal-r5 { fill: #98a84b }
+.terminal-r6 { fill: #9a9b99;font-style: italic; }
+.terminal-r7 { fill: #9a9b99 }
+    </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">SnapshotTestAppEmptyAssistantThenReasoning</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="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="0" y="459.2" textLength="134.2" clip-path="url(#terminal-line-18)">&#160;&#160;⡠⣒⠄&#160;&#160;⡔⢄⠔⡄</text><text class="terminal-r2" x="170.8" y="459.2" textLength="146.4" clip-path="url(#terminal-line-18)">Mistral&#160;Vibe</text><text class="terminal-r1" x="317.2" y="459.2" textLength="122" clip-path="url(#terminal-line-18)">&#160;v0.0.0&#160;·&#160;</text><text class="terminal-r3" x="439.2" y="459.2" textLength="183" clip-path="url(#terminal-line-18)">devstral-latest</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="0" y="483.6" textLength="134.2" clip-path="url(#terminal-line-19)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="483.6" textLength="414.8" clip-path="url(#terminal-line-19)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</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="134.2" clip-path="url(#terminal-line-20)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="508" textLength="61" clip-path="url(#terminal-line-20)">Type&#160;</text><text class="terminal-r3" x="231.8" y="508" textLength="61" clip-path="url(#terminal-line-20)">/help</text><text class="terminal-r1" x="292.8" y="508" textLength="256.2" clip-path="url(#terminal-line-20)">&#160;for&#160;more&#160;information</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="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)">What&#160;is&#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="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)">
+</text><text class="terminal-r7" 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-r7" 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-r7" x="0" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r7" 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-r7" x="0" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r7" 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-r7" 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-r7" x="0" y="874" textLength="158.6" clip-path="url(#terminal-line-35)">/test/workdir</text><text class="terminal-r7" 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_parallel_tool_calls/test_snapshot_parallel_tool_calls_pending.svg

@@ -0,0 +1,114 @@
+<svg class="rich-terminal" viewBox="0 0 994 416.0" 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: #d9d9d9 }
+    </style>
+
+    <defs>
+    <clipPath id="terminal-clip-terminal">
+      <rect x="0" y="0" width="975.0" height="365.0" />
+    </clipPath>
+    <clipPath id="terminal-line-0">
+    <rect x="0" y="1.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-1">
+    <rect x="0" y="25.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-2">
+    <rect x="0" y="50.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-3">
+    <rect x="0" y="74.7" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-4">
+    <rect x="0" y="99.1" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-5">
+    <rect x="0" y="123.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-6">
+    <rect x="0" y="147.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-7">
+    <rect x="0" y="172.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-8">
+    <rect x="0" y="196.7" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-9">
+    <rect x="0" y="221.1" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-10">
+    <rect x="0" y="245.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-11">
+    <rect x="0" y="269.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-12">
+    <rect x="0" y="294.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-13">
+    <rect x="0" y="318.7" width="976" height="24.65"/>
+            </clipPath>
+    </defs>
+
+    <rect fill="#292929" stroke="rgba(255,255,255,0.35)" stroke-width="1" x="1" y="1" width="992" height="414" rx="8"/><text class="terminal-title" fill="#c5c8c6" text-anchor="middle" x="496" y="27">ParallelToolCallsApp</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)">
+    <rect fill="#121212" x="0" y="1.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="25.9" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="50.3" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="12.2" y="50.3" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="24.4" y="50.3" width="268.4" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="292.8" y="50.3" width="683.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="74.7" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="12.2" y="74.7" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="24.4" y="74.7" width="268.4" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="292.8" y="74.7" width="683.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="99.1" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="12.2" y="99.1" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="24.4" y="99.1" width="268.4" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="292.8" y="99.1" width="683.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="123.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="147.9" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="172.3" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="196.7" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="221.1" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="245.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="269.9" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="294.3" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="318.7" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="343.1" width="976" height="24.65" shape-rendering="crispEdges"/>
+    <g class="terminal-matrix">
+    <text class="terminal-r1" x="976" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r1" x="976" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+</text><text class="terminal-r2" x="0" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">■</text><text class="terminal-r2" x="24.4" y="68.8" textLength="268.4" clip-path="url(#terminal-line-2)">Reading&#160;/src/file_0.py</text><text class="terminal-r1" x="976" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
+</text><text class="terminal-r2" x="0" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">■</text><text class="terminal-r2" x="24.4" y="93.2" textLength="268.4" clip-path="url(#terminal-line-3)">Reading&#160;/src/file_1.py</text><text class="terminal-r1" x="976" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r2" x="0" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">■</text><text class="terminal-r2" x="24.4" y="117.6" textLength="268.4" clip-path="url(#terminal-line-4)">Reading&#160;/src/file_2.py</text><text class="terminal-r1" x="976" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
+</text><text class="terminal-r1" x="976" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
+</text><text class="terminal-r1" x="976" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="976" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="976" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text><text class="terminal-r1" x="976" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
+</text><text class="terminal-r1" x="976" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
+</text><text class="terminal-r1" x="976" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
+</text><text class="terminal-r1" x="976" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
+</text><text class="terminal-r1" x="976" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
+</text>
+    </g>
+    </g>
+</svg>

tests/snapshots/__snapshots__/test_ui_snapshot_proxy_setup/test_snapshot_proxy_setup_initial_empty.svg

@@ -33,13 +33,12 @@
     }
 
     .terminal-r1 { fill: #c5c8c6 }
-.terminal-r2 { fill: #4b4e55 }
+.terminal-r2 { fill: #ff8205;font-weight: bold }
 .terminal-r3 { fill: #68a0b3 }
-.terminal-r4 { fill: #292929 }
-.terminal-r5 { fill: #9a9b99 }
-.terminal-r6 { fill: #608ab1;font-weight: bold }
+.terminal-r4 { fill: #9a9b99 }
+.terminal-r5 { fill: #608ab1;font-weight: bold }
+.terminal-r6 { fill: #949798 }
 .terminal-r7 { fill: #868887 }
-.terminal-r8 { fill: #949798 }
     </style>
 
     <defs>
@@ -161,44 +160,44 @@
             </g>
         
     <g transform="translate(9, 41)" clip-path="url(#terminal-clip-terminal)">
-    <rect fill="#4b4e55" x="1207.8" y="25.9" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="1207.8" y="50.3" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="1207.8" y="74.7" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="48.8" y="269.9" width="12.2" height="24.65" shape-rendering="crispEdges"/>
+    <rect fill="#4b4e55" x="48.8" y="538.3" width="12.2" height="24.65" shape-rendering="crispEdges"/>
     <g class="terminal-matrix">
-    <text class="terminal-r1" x="0" y="20" textLength="134.2" clip-path="url(#terminal-line-0)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="20" textLength="414.8" clip-path="url(#terminal-line-0)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-r2" x="1207.8" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">▂</text><text class="terminal-r1" x="1220" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
-</text><text class="terminal-r1" x="0" y="44.4" textLength="134.2" clip-path="url(#terminal-line-1)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="44.4" textLength="61" clip-path="url(#terminal-line-1)">Type&#160;</text><text class="terminal-r3" x="231.8" y="44.4" textLength="61" clip-path="url(#terminal-line-1)">/help</text><text class="terminal-r1" x="292.8" y="44.4" textLength="256.2" clip-path="url(#terminal-line-1)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1220" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+    <text class="terminal-r1" x="1220" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r1" x="1220" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
 </text><text class="terminal-r1" x="1220" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
-</text><text class="terminal-r5" x="24.4" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">⎣</text><text class="terminal-r1" x="48.8" y="93.2" textLength="256.2" clip-path="url(#terminal-line-3)">Proxy&#160;setup&#160;opened...</text><text class="terminal-r1" x="1220" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r1" x="1220" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
 </text><text class="terminal-r1" x="1220" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
 </text><text class="terminal-r1" x="1220" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
-</text><text class="terminal-r5" x="0" y="166.4" textLength="1220" clip-path="url(#terminal-line-6)">┌──────────────────────────────────────────────────────────────────────────────────────────────────┐</text><text class="terminal-r1" x="1220" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
-</text><text class="terminal-r5" x="0" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">│</text><text class="terminal-r6" x="24.4" y="190.8" textLength="231.8" clip-path="url(#terminal-line-7)">Proxy&#160;Configuration</text><text class="terminal-r5" x="1207.8" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">│</text><text class="terminal-r1" x="1220" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
-</text><text class="terminal-r5" x="0" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">│</text><text class="terminal-r5" x="1207.8" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">│</text><text class="terminal-r1" x="1220" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
-</text><text class="terminal-r5" x="0" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">│</text><text class="terminal-r6" x="24.4" y="239.6" textLength="122" clip-path="url(#terminal-line-9)">HTTP_PROXY</text><text class="terminal-r7" x="170.8" y="239.6" textLength="329.4" clip-path="url(#terminal-line-9)">Proxy&#160;URL&#160;for&#160;HTTP&#160;requests</text><text class="terminal-r5" x="1207.8" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">│</text><text class="terminal-r1" x="1220" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
-</text><text class="terminal-r5" x="0" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">│</text><text class="terminal-r5" x="1207.8" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">│</text><text class="terminal-r1" x="1220" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
-</text><text class="terminal-r5" x="0" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">│</text><text class="terminal-r5" x="24.4" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">▎</text><text class="terminal-r8" x="48.8" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">N</text><text class="terminal-r7" x="61" y="288.4" textLength="73.2" clip-path="url(#terminal-line-11)">OT&#160;SET</text><text class="terminal-r5" x="1207.8" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">│</text><text class="terminal-r1" x="1220" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
-</text><text class="terminal-r5" x="0" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">│</text><text class="terminal-r5" x="1207.8" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">│</text><text class="terminal-r1" x="1220" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
-</text><text class="terminal-r5" x="0" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">│</text><text class="terminal-r6" x="24.4" y="337.2" textLength="134.2" clip-path="url(#terminal-line-13)">HTTPS_PROXY</text><text class="terminal-r7" x="183" y="337.2" textLength="341.6" clip-path="url(#terminal-line-13)">Proxy&#160;URL&#160;for&#160;HTTPS&#160;requests</text><text class="terminal-r5" x="1207.8" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">│</text><text class="terminal-r1" x="1220" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
-</text><text class="terminal-r5" x="0" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">│</text><text class="terminal-r5" x="1207.8" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">│</text><text class="terminal-r1" x="1220" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
-</text><text class="terminal-r5" x="0" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">│</text><text class="terminal-r5" x="24.4" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">▎</text><text class="terminal-r7" x="48.8" y="386" textLength="85.4" clip-path="url(#terminal-line-15)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">│</text><text class="terminal-r1" x="1220" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
-</text><text class="terminal-r5" x="0" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">│</text><text class="terminal-r5" x="1207.8" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">│</text><text class="terminal-r1" x="1220" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
-</text><text class="terminal-r5" 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="109.8" clip-path="url(#terminal-line-17)">ALL_PROXY</text><text class="terminal-r7" x="158.6" y="434.8" textLength="451.4" clip-path="url(#terminal-line-17)">Proxy&#160;URL&#160;for&#160;all&#160;requests&#160;(fallback)</text><text class="terminal-r5" x="1207.8" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">│</text><text class="terminal-r1" x="1220" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
-</text><text class="terminal-r5" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">│</text><text class="terminal-r5" x="1207.8" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">│</text><text class="terminal-r1" x="1220" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
-</text><text class="terminal-r5" x="0" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">│</text><text class="terminal-r5" x="24.4" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">▎</text><text class="terminal-r7" x="48.8" y="483.6" textLength="85.4" clip-path="url(#terminal-line-19)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">│</text><text class="terminal-r1" x="1220" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r5" x="0" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r5" x="1207.8" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r1" x="1220" 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="97.6" clip-path="url(#terminal-line-21)">NO_PROXY</text><text class="terminal-r7" x="146.4" y="532.4" textLength="549" clip-path="url(#terminal-line-21)">Comma-separated&#160;list&#160;of&#160;hosts&#160;to&#160;bypass&#160;proxy</text><text class="terminal-r5" x="1207.8" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">│</text><text class="terminal-r1" x="1220" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
-</text><text class="terminal-r5" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r5" x="1207.8" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r1" x="1220" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r5" x="0" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r5" x="24.4" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">▎</text><text class="terminal-r7" x="48.8" y="581.2" textLength="85.4" clip-path="url(#terminal-line-23)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r1" x="1220" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
-</text><text class="terminal-r5" x="0" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r5" x="1207.8" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r1" x="1220" 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="158.6" clip-path="url(#terminal-line-25)">SSL_CERT_FILE</text><text class="terminal-r7" x="207.4" y="630" textLength="427" clip-path="url(#terminal-line-25)">Path&#160;to&#160;custom&#160;SSL&#160;certificate&#160;file</text><text class="terminal-r5" x="1207.8" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">│</text><text class="terminal-r1" x="1220" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
-</text><text class="terminal-r5" x="0" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r5" x="1207.8" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">▎</text><text class="terminal-r7" x="48.8" y="678.8" textLength="85.4" clip-path="url(#terminal-line-27)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">│</text><text class="terminal-r1" x="1220" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
-</text><text class="terminal-r5" x="0" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r5" x="1207.8" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r1" x="1220" 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-r6" x="24.4" y="727.6" textLength="146.4" clip-path="url(#terminal-line-29)">SSL_CERT_DIR</text><text class="terminal-r7" x="195.2" y="727.6" textLength="549" clip-path="url(#terminal-line-29)">Path&#160;to&#160;directory&#160;containing&#160;SSL&#160;certificates</text><text class="terminal-r5" x="1207.8" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">│</text><text class="terminal-r1" x="1220" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
-</text><text class="terminal-r5" x="0" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r5" x="1207.8" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">▎</text><text class="terminal-r7" x="48.8" y="776.4" textLength="85.4" clip-path="url(#terminal-line-31)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r1" x="1220" 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="1207.8" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="825.2" textLength="512.4" clip-path="url(#terminal-line-33)">↑↓&#160;navigate&#160;&#160;Enter&#160;save&#160;&amp;&#160;exit&#160;&#160;ESC&#160;cancel</text><text class="terminal-r5" x="1207.8" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r1" x="1220" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">
-</text><text class="terminal-r5" x="0" y="849.6" textLength="1220" clip-path="url(#terminal-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1220" 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="1012.6" y="874" textLength="207.4" clip-path="url(#terminal-line-35)">0%&#160;of&#160;200k&#160;tokens</text>
+</text><text class="terminal-r1" x="1220" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="1220" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="1220" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text><text class="terminal-r1" x="1220" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
+</text><text class="terminal-r1" x="1220" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
+</text><text class="terminal-r1" x="1220" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
+</text><text class="terminal-r1" x="0" y="312.8" textLength="134.2" clip-path="url(#terminal-line-12)">&#160;&#160;⡠⣒⠄&#160;&#160;⡔⢄⠔⡄</text><text class="terminal-r2" x="170.8" y="312.8" textLength="146.4" clip-path="url(#terminal-line-12)">Mistral&#160;Vibe</text><text class="terminal-r1" x="317.2" y="312.8" textLength="122" clip-path="url(#terminal-line-12)">&#160;v0.0.0&#160;·&#160;</text><text class="terminal-r3" x="439.2" y="312.8" textLength="183" clip-path="url(#terminal-line-12)">devstral-latest</text><text class="terminal-r1" x="1220" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
+</text><text class="terminal-r1" x="0" y="337.2" textLength="134.2" clip-path="url(#terminal-line-13)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="337.2" textLength="414.8" clip-path="url(#terminal-line-13)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-r1" x="1220" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
+</text><text class="terminal-r1" x="0" y="361.6" textLength="134.2" clip-path="url(#terminal-line-14)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="361.6" textLength="61" clip-path="url(#terminal-line-14)">Type&#160;</text><text class="terminal-r3" x="231.8" y="361.6" textLength="61" clip-path="url(#terminal-line-14)">/help</text><text class="terminal-r1" x="292.8" y="361.6" textLength="256.2" clip-path="url(#terminal-line-14)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1220" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
+</text><text class="terminal-r1" x="1220" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
+</text><text class="terminal-r4" x="24.4" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">⎣</text><text class="terminal-r1" x="48.8" y="410.4" textLength="256.2" clip-path="url(#terminal-line-16)">Proxy&#160;setup&#160;opened...</text><text class="terminal-r1" x="1220" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
+</text><text class="terminal-r1" x="1220" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
+</text><text class="terminal-r1" x="1220" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
+</text><text class="terminal-r4" x="0" y="483.6" textLength="1220" clip-path="url(#terminal-line-19)">┌──────────────────────────────────────────────────────────────────────────────────────────────────┐</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="508" textLength="231.8" clip-path="url(#terminal-line-20)">Proxy&#160;Configuration</text><text class="terminal-r4" x="1207.8" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="532.4" textLength="122" clip-path="url(#terminal-line-21)">HTTP_PROXY</text><text class="terminal-r4" x="1207.8" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">▎</text><text class="terminal-r6" x="48.8" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">P</text><text class="terminal-r7" x="61" y="556.8" textLength="317.2" clip-path="url(#terminal-line-22)">roxy&#160;URL&#160;for&#160;HTTP&#160;requests</text><text class="terminal-r4" x="1207.8" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="581.2" textLength="134.2" clip-path="url(#terminal-line-23)">HTTPS_PROXY</text><text class="terminal-r4" x="1207.8" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">▎</text><text class="terminal-r7" x="48.8" y="605.6" textLength="341.6" clip-path="url(#terminal-line-24)">Proxy&#160;URL&#160;for&#160;HTTPS&#160;requests</text><text class="terminal-r4" x="1207.8" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="630" textLength="109.8" clip-path="url(#terminal-line-25)">ALL_PROXY</text><text class="terminal-r4" x="1207.8" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">│</text><text class="terminal-r1" x="1220" 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-r4" x="24.4" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">▎</text><text class="terminal-r7" x="48.8" y="654.4" textLength="451.4" clip-path="url(#terminal-line-26)">Proxy&#160;URL&#160;for&#160;all&#160;requests&#160;(fallback)</text><text class="terminal-r4" x="1207.8" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r1" x="1220" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
+</text><text class="terminal-r4" x="0" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">│</text><text class="terminal-r5" x="24.4" y="678.8" textLength="97.6" clip-path="url(#terminal-line-27)">NO_PROXY</text><text class="terminal-r4" x="1207.8" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">│</text><text class="terminal-r1" x="1220" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r4" x="0" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r4" x="24.4" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">▎</text><text class="terminal-r7" x="48.8" y="703.2" textLength="549" clip-path="url(#terminal-line-28)">Comma-separated&#160;list&#160;of&#160;hosts&#160;to&#160;bypass&#160;proxy</text><text class="terminal-r4" x="1207.8" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r1" x="1220" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
+</text><text class="terminal-r4" x="0" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">│</text><text class="terminal-r5" x="24.4" y="727.6" textLength="158.6" clip-path="url(#terminal-line-29)">SSL_CERT_FILE</text><text class="terminal-r4" x="1207.8" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">│</text><text class="terminal-r1" x="1220" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
+</text><text class="terminal-r4" x="0" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r4" x="24.4" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">▎</text><text class="terminal-r7" x="48.8" y="752" textLength="427" clip-path="url(#terminal-line-30)">Path&#160;to&#160;custom&#160;SSL&#160;certificate&#160;file</text><text class="terminal-r4" x="1207.8" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="776.4" textLength="146.4" clip-path="url(#terminal-line-31)">SSL_CERT_DIR</text><text class="terminal-r4" x="1207.8" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">▎</text><text class="terminal-r7" x="48.8" y="800.8" textLength="549" clip-path="url(#terminal-line-32)">Path&#160;to&#160;directory&#160;containing&#160;SSL&#160;certificates</text><text class="terminal-r4" x="1207.8" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="825.2" textLength="512.4" clip-path="url(#terminal-line-33)">↑↓&#160;navigate&#160;&#160;Enter&#160;save&#160;&amp;&#160;exit&#160;&#160;ESC&#160;cancel</text><text class="terminal-r4" x="1207.8" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r1" x="1220" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">
+</text><text class="terminal-r4" x="0" y="849.6" textLength="1220" clip-path="url(#terminal-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1220" 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="1012.6" 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_proxy_setup/test_snapshot_proxy_setup_initial_with_values.svg

@@ -33,12 +33,11 @@
     }
 
     .terminal-r1 { fill: #c5c8c6 }
-.terminal-r2 { fill: #4b4e55 }
+.terminal-r2 { fill: #ff8205;font-weight: bold }
 .terminal-r3 { fill: #68a0b3 }
-.terminal-r4 { fill: #292929 }
-.terminal-r5 { fill: #9a9b99 }
-.terminal-r6 { fill: #608ab1;font-weight: bold }
-.terminal-r7 { fill: #868887 }
+.terminal-r4 { fill: #9a9b99 }
+.terminal-r5 { fill: #608ab1;font-weight: bold }
+.terminal-r6 { fill: #868887 }
     </style>
 
     <defs>
@@ -160,44 +159,44 @@
             </g>
         
     <g transform="translate(9, 41)" clip-path="url(#terminal-clip-terminal)">
-    <rect fill="#4b4e55" x="1207.8" y="25.9" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="1207.8" y="50.3" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="1207.8" y="74.7" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#608ab1" x="48.8" y="269.9" width="256.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="305" y="269.9" width="12.2" height="24.65" shape-rendering="crispEdges"/>
+    <rect fill="#608ab1" x="48.8" y="538.3" width="256.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#4b4e55" x="305" y="538.3" width="12.2" height="24.65" shape-rendering="crispEdges"/>
     <g class="terminal-matrix">
-    <text class="terminal-r1" x="0" y="20" textLength="134.2" clip-path="url(#terminal-line-0)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="20" textLength="414.8" clip-path="url(#terminal-line-0)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-r2" x="1207.8" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">▂</text><text class="terminal-r1" x="1220" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
-</text><text class="terminal-r1" x="0" y="44.4" textLength="134.2" clip-path="url(#terminal-line-1)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="44.4" textLength="61" clip-path="url(#terminal-line-1)">Type&#160;</text><text class="terminal-r3" x="231.8" y="44.4" textLength="61" clip-path="url(#terminal-line-1)">/help</text><text class="terminal-r1" x="292.8" y="44.4" textLength="256.2" clip-path="url(#terminal-line-1)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1220" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+    <text class="terminal-r1" x="1220" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r1" x="1220" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
 </text><text class="terminal-r1" x="1220" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
-</text><text class="terminal-r5" x="24.4" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">⎣</text><text class="terminal-r1" x="48.8" y="93.2" textLength="256.2" clip-path="url(#terminal-line-3)">Proxy&#160;setup&#160;opened...</text><text class="terminal-r1" x="1220" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r1" x="1220" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
 </text><text class="terminal-r1" x="1220" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
 </text><text class="terminal-r1" x="1220" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
-</text><text class="terminal-r5" x="0" y="166.4" textLength="1220" clip-path="url(#terminal-line-6)">┌──────────────────────────────────────────────────────────────────────────────────────────────────┐</text><text class="terminal-r1" x="1220" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
-</text><text class="terminal-r5" x="0" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">│</text><text class="terminal-r6" x="24.4" y="190.8" textLength="231.8" clip-path="url(#terminal-line-7)">Proxy&#160;Configuration</text><text class="terminal-r5" x="1207.8" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">│</text><text class="terminal-r1" x="1220" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
-</text><text class="terminal-r5" x="0" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">│</text><text class="terminal-r5" x="1207.8" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">│</text><text class="terminal-r1" x="1220" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
-</text><text class="terminal-r5" x="0" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">│</text><text class="terminal-r6" x="24.4" y="239.6" textLength="122" clip-path="url(#terminal-line-9)">HTTP_PROXY</text><text class="terminal-r7" x="170.8" y="239.6" textLength="329.4" clip-path="url(#terminal-line-9)">Proxy&#160;URL&#160;for&#160;HTTP&#160;requests</text><text class="terminal-r5" x="1207.8" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">│</text><text class="terminal-r1" x="1220" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
-</text><text class="terminal-r5" x="0" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">│</text><text class="terminal-r5" x="1207.8" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">│</text><text class="terminal-r1" x="1220" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
-</text><text class="terminal-r5" x="0" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">│</text><text class="terminal-r5" x="24.4" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">▎</text><text class="terminal-r1" x="48.8" y="288.4" textLength="256.2" clip-path="url(#terminal-line-11)">http://old-proxy:8080</text><text class="terminal-r5" x="1207.8" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">│</text><text class="terminal-r1" x="1220" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
-</text><text class="terminal-r5" x="0" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">│</text><text class="terminal-r5" x="1207.8" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">│</text><text class="terminal-r1" x="1220" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
-</text><text class="terminal-r5" x="0" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">│</text><text class="terminal-r6" x="24.4" y="337.2" textLength="134.2" clip-path="url(#terminal-line-13)">HTTPS_PROXY</text><text class="terminal-r7" x="183" y="337.2" textLength="341.6" clip-path="url(#terminal-line-13)">Proxy&#160;URL&#160;for&#160;HTTPS&#160;requests</text><text class="terminal-r5" x="1207.8" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">│</text><text class="terminal-r1" x="1220" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
-</text><text class="terminal-r5" x="0" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">│</text><text class="terminal-r5" x="1207.8" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">│</text><text class="terminal-r1" x="1220" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
-</text><text class="terminal-r5" x="0" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">│</text><text class="terminal-r5" x="24.4" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">▎</text><text class="terminal-r1" x="48.8" y="386" textLength="1146.8" clip-path="url(#terminal-line-15)">https://old-proxy:8443&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;</text><text class="terminal-r5" x="1207.8" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">│</text><text class="terminal-r1" x="1220" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
-</text><text class="terminal-r5" x="0" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">│</text><text class="terminal-r5" x="1207.8" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">│</text><text class="terminal-r1" x="1220" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
-</text><text class="terminal-r5" 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="109.8" clip-path="url(#terminal-line-17)">ALL_PROXY</text><text class="terminal-r7" x="158.6" y="434.8" textLength="451.4" clip-path="url(#terminal-line-17)">Proxy&#160;URL&#160;for&#160;all&#160;requests&#160;(fallback)</text><text class="terminal-r5" x="1207.8" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">│</text><text class="terminal-r1" x="1220" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
-</text><text class="terminal-r5" x="0" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">│</text><text class="terminal-r5" x="1207.8" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">│</text><text class="terminal-r1" x="1220" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
-</text><text class="terminal-r5" x="0" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">│</text><text class="terminal-r5" x="24.4" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">▎</text><text class="terminal-r7" x="48.8" y="483.6" textLength="85.4" clip-path="url(#terminal-line-19)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">│</text><text class="terminal-r1" x="1220" y="483.6" textLength="12.2" clip-path="url(#terminal-line-19)">
-</text><text class="terminal-r5" x="0" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r5" x="1207.8" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r1" x="1220" 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="97.6" clip-path="url(#terminal-line-21)">NO_PROXY</text><text class="terminal-r7" x="146.4" y="532.4" textLength="549" clip-path="url(#terminal-line-21)">Comma-separated&#160;list&#160;of&#160;hosts&#160;to&#160;bypass&#160;proxy</text><text class="terminal-r5" x="1207.8" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">│</text><text class="terminal-r1" x="1220" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">
-</text><text class="terminal-r5" x="0" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r5" x="1207.8" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r1" x="1220" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">
-</text><text class="terminal-r5" x="0" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r5" x="24.4" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">▎</text><text class="terminal-r7" x="48.8" y="581.2" textLength="85.4" clip-path="url(#terminal-line-23)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r1" x="1220" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">
-</text><text class="terminal-r5" x="0" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r5" x="1207.8" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r1" x="1220" 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="158.6" clip-path="url(#terminal-line-25)">SSL_CERT_FILE</text><text class="terminal-r7" x="207.4" y="630" textLength="427" clip-path="url(#terminal-line-25)">Path&#160;to&#160;custom&#160;SSL&#160;certificate&#160;file</text><text class="terminal-r5" x="1207.8" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">│</text><text class="terminal-r1" x="1220" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">
-</text><text class="terminal-r5" x="0" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r5" x="1207.8" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">▎</text><text class="terminal-r7" x="48.8" y="678.8" textLength="85.4" clip-path="url(#terminal-line-27)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">│</text><text class="terminal-r1" x="1220" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
-</text><text class="terminal-r5" x="0" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r5" x="1207.8" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r1" x="1220" 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-r6" x="24.4" y="727.6" textLength="146.4" clip-path="url(#terminal-line-29)">SSL_CERT_DIR</text><text class="terminal-r7" x="195.2" y="727.6" textLength="549" clip-path="url(#terminal-line-29)">Path&#160;to&#160;directory&#160;containing&#160;SSL&#160;certificates</text><text class="terminal-r5" x="1207.8" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">│</text><text class="terminal-r1" x="1220" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
-</text><text class="terminal-r5" x="0" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r5" x="1207.8" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">▎</text><text class="terminal-r7" x="48.8" y="776.4" textLength="85.4" clip-path="url(#terminal-line-31)">NOT&#160;SET</text><text class="terminal-r5" x="1207.8" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r1" x="1220" 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="1207.8" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="825.2" textLength="512.4" clip-path="url(#terminal-line-33)">↑↓&#160;navigate&#160;&#160;Enter&#160;save&#160;&amp;&#160;exit&#160;&#160;ESC&#160;cancel</text><text class="terminal-r5" x="1207.8" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r1" x="1220" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">
-</text><text class="terminal-r5" x="0" y="849.6" textLength="1220" clip-path="url(#terminal-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1220" 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="1012.6" y="874" textLength="207.4" clip-path="url(#terminal-line-35)">0%&#160;of&#160;200k&#160;tokens</text>
+</text><text class="terminal-r1" x="1220" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="1220" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="1220" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text><text class="terminal-r1" x="1220" y="239.6" textLength="12.2" clip-path="url(#terminal-line-9)">
+</text><text class="terminal-r1" x="1220" y="264" textLength="12.2" clip-path="url(#terminal-line-10)">
+</text><text class="terminal-r1" x="1220" y="288.4" textLength="12.2" clip-path="url(#terminal-line-11)">
+</text><text class="terminal-r1" x="0" y="312.8" textLength="134.2" clip-path="url(#terminal-line-12)">&#160;&#160;⡠⣒⠄&#160;&#160;⡔⢄⠔⡄</text><text class="terminal-r2" x="170.8" y="312.8" textLength="146.4" clip-path="url(#terminal-line-12)">Mistral&#160;Vibe</text><text class="terminal-r1" x="317.2" y="312.8" textLength="122" clip-path="url(#terminal-line-12)">&#160;v0.0.0&#160;·&#160;</text><text class="terminal-r3" x="439.2" y="312.8" textLength="183" clip-path="url(#terminal-line-12)">devstral-latest</text><text class="terminal-r1" x="1220" y="312.8" textLength="12.2" clip-path="url(#terminal-line-12)">
+</text><text class="terminal-r1" x="0" y="337.2" textLength="134.2" clip-path="url(#terminal-line-13)">&#160;⢸⠸⣀⡔⢉⠱⣃⡢⣂⡣</text><text class="terminal-r1" x="170.8" y="337.2" textLength="414.8" clip-path="url(#terminal-line-13)">1&#160;model&#160;·&#160;0&#160;MCP&#160;servers&#160;·&#160;0&#160;skills</text><text class="terminal-r1" x="1220" y="337.2" textLength="12.2" clip-path="url(#terminal-line-13)">
+</text><text class="terminal-r1" x="0" y="361.6" textLength="134.2" clip-path="url(#terminal-line-14)">&#160;&#160;⠉⠒⠣⠤⠵⠤⠬⠮⠆</text><text class="terminal-r1" x="170.8" y="361.6" textLength="61" clip-path="url(#terminal-line-14)">Type&#160;</text><text class="terminal-r3" x="231.8" y="361.6" textLength="61" clip-path="url(#terminal-line-14)">/help</text><text class="terminal-r1" x="292.8" y="361.6" textLength="256.2" clip-path="url(#terminal-line-14)">&#160;for&#160;more&#160;information</text><text class="terminal-r1" x="1220" y="361.6" textLength="12.2" clip-path="url(#terminal-line-14)">
+</text><text class="terminal-r1" x="1220" y="386" textLength="12.2" clip-path="url(#terminal-line-15)">
+</text><text class="terminal-r4" x="24.4" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">⎣</text><text class="terminal-r1" x="48.8" y="410.4" textLength="256.2" clip-path="url(#terminal-line-16)">Proxy&#160;setup&#160;opened...</text><text class="terminal-r1" x="1220" y="410.4" textLength="12.2" clip-path="url(#terminal-line-16)">
+</text><text class="terminal-r1" x="1220" y="434.8" textLength="12.2" clip-path="url(#terminal-line-17)">
+</text><text class="terminal-r1" x="1220" y="459.2" textLength="12.2" clip-path="url(#terminal-line-18)">
+</text><text class="terminal-r4" x="0" y="483.6" textLength="1220" clip-path="url(#terminal-line-19)">┌──────────────────────────────────────────────────────────────────────────────────────────────────┐</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="508" textLength="231.8" clip-path="url(#terminal-line-20)">Proxy&#160;Configuration</text><text class="terminal-r4" x="1207.8" y="508" textLength="12.2" clip-path="url(#terminal-line-20)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="532.4" textLength="122" clip-path="url(#terminal-line-21)">HTTP_PROXY</text><text class="terminal-r4" x="1207.8" y="532.4" textLength="12.2" clip-path="url(#terminal-line-21)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">▎</text><text class="terminal-r1" x="48.8" y="556.8" textLength="256.2" clip-path="url(#terminal-line-22)">http://old-proxy:8080</text><text class="terminal-r4" x="1207.8" y="556.8" textLength="12.2" clip-path="url(#terminal-line-22)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="581.2" textLength="134.2" clip-path="url(#terminal-line-23)">HTTPS_PROXY</text><text class="terminal-r4" x="1207.8" y="581.2" textLength="12.2" clip-path="url(#terminal-line-23)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">▎</text><text class="terminal-r1" x="48.8" y="605.6" textLength="1146.8" clip-path="url(#terminal-line-24)">https://old-proxy:8443&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;&#160;</text><text class="terminal-r4" x="1207.8" y="605.6" textLength="12.2" clip-path="url(#terminal-line-24)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="630" textLength="109.8" clip-path="url(#terminal-line-25)">ALL_PROXY</text><text class="terminal-r4" x="1207.8" y="630" textLength="12.2" clip-path="url(#terminal-line-25)">│</text><text class="terminal-r1" x="1220" 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-r4" x="24.4" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">▎</text><text class="terminal-r6" x="48.8" y="654.4" textLength="451.4" clip-path="url(#terminal-line-26)">Proxy&#160;URL&#160;for&#160;all&#160;requests&#160;(fallback)</text><text class="terminal-r4" x="1207.8" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">│</text><text class="terminal-r1" x="1220" y="654.4" textLength="12.2" clip-path="url(#terminal-line-26)">
+</text><text class="terminal-r4" x="0" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">│</text><text class="terminal-r5" x="24.4" y="678.8" textLength="97.6" clip-path="url(#terminal-line-27)">NO_PROXY</text><text class="terminal-r4" x="1207.8" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">│</text><text class="terminal-r1" x="1220" y="678.8" textLength="12.2" clip-path="url(#terminal-line-27)">
+</text><text class="terminal-r4" x="0" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r4" x="24.4" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">▎</text><text class="terminal-r6" x="48.8" y="703.2" textLength="549" clip-path="url(#terminal-line-28)">Comma-separated&#160;list&#160;of&#160;hosts&#160;to&#160;bypass&#160;proxy</text><text class="terminal-r4" x="1207.8" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">│</text><text class="terminal-r1" x="1220" y="703.2" textLength="12.2" clip-path="url(#terminal-line-28)">
+</text><text class="terminal-r4" x="0" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">│</text><text class="terminal-r5" x="24.4" y="727.6" textLength="158.6" clip-path="url(#terminal-line-29)">SSL_CERT_FILE</text><text class="terminal-r4" x="1207.8" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">│</text><text class="terminal-r1" x="1220" y="727.6" textLength="12.2" clip-path="url(#terminal-line-29)">
+</text><text class="terminal-r4" x="0" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r4" x="24.4" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">▎</text><text class="terminal-r6" x="48.8" y="752" textLength="427" clip-path="url(#terminal-line-30)">Path&#160;to&#160;custom&#160;SSL&#160;certificate&#160;file</text><text class="terminal-r4" x="1207.8" y="752" textLength="12.2" clip-path="url(#terminal-line-30)">│</text><text class="terminal-r1" x="1220" 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-r5" x="24.4" y="776.4" textLength="146.4" clip-path="url(#terminal-line-31)">SSL_CERT_DIR</text><text class="terminal-r4" x="1207.8" y="776.4" textLength="12.2" clip-path="url(#terminal-line-31)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">▎</text><text class="terminal-r6" x="48.8" y="800.8" textLength="549" clip-path="url(#terminal-line-32)">Path&#160;to&#160;directory&#160;containing&#160;SSL&#160;certificates</text><text class="terminal-r4" x="1207.8" y="800.8" textLength="12.2" clip-path="url(#terminal-line-32)">│</text><text class="terminal-r1" x="1220" 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="24.4" y="825.2" textLength="512.4" clip-path="url(#terminal-line-33)">↑↓&#160;navigate&#160;&#160;Enter&#160;save&#160;&amp;&#160;exit&#160;&#160;ESC&#160;cancel</text><text class="terminal-r4" x="1207.8" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">│</text><text class="terminal-r1" x="1220" y="825.2" textLength="12.2" clip-path="url(#terminal-line-33)">
+</text><text class="terminal-r4" x="0" y="849.6" textLength="1220" clip-path="url(#terminal-line-34)">└──────────────────────────────────────────────────────────────────────────────────────────────────┘</text><text class="terminal-r1" x="1220" 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="1012.6" 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_streaming_tool_call/test_snapshot_tool_call_partial.svg

@@ -0,0 +1,94 @@
+<svg class="rich-terminal" viewBox="0 0 994 294.0" 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: #d9d9d9 }
+    </style>
+
+    <defs>
+    <clipPath id="terminal-clip-terminal">
+      <rect x="0" y="0" width="975.0" height="243.0" />
+    </clipPath>
+    <clipPath id="terminal-line-0">
+    <rect x="0" y="1.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-1">
+    <rect x="0" y="25.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-2">
+    <rect x="0" y="50.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-3">
+    <rect x="0" y="74.7" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-4">
+    <rect x="0" y="99.1" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-5">
+    <rect x="0" y="123.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-6">
+    <rect x="0" y="147.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-7">
+    <rect x="0" y="172.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-8">
+    <rect x="0" y="196.7" width="976" height="24.65"/>
+            </clipPath>
+    </defs>
+
+    <rect fill="#292929" stroke="rgba(255,255,255,0.35)" stroke-width="1" x="1" y="1" width="992" height="292" rx="8"/><text class="terminal-title" fill="#c5c8c6" text-anchor="middle" x="496" y="27">ToolCallStreamingUpdateTest</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)">
+    <rect fill="#121212" x="0" y="1.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="25.9" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="12.2" y="25.9" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="24.4" y="25.9" width="109.8" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="134.2" y="25.9" width="841.8" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="50.3" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="74.7" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="99.1" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="123.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="147.9" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="172.3" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="196.7" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="221.1" width="976" height="24.65" shape-rendering="crispEdges"/>
+    <g class="terminal-matrix">
+    <text class="terminal-r1" x="976" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r2" x="0" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">■</text><text class="terminal-r2" x="24.4" y="44.4" textLength="109.8" clip-path="url(#terminal-line-1)">read_file</text><text class="terminal-r1" x="976" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+</text><text class="terminal-r1" x="976" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
+</text><text class="terminal-r1" x="976" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r1" x="976" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
+</text><text class="terminal-r1" x="976" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
+</text><text class="terminal-r1" x="976" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="976" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="976" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text>
+    </g>
+    </g>
+</svg>

tests/snapshots/__snapshots__/test_ui_snapshot_streaming_tool_call/test_snapshot_tool_call_updated.svg

@@ -0,0 +1,94 @@
+<svg class="rich-terminal" viewBox="0 0 994 294.0" 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: #d9d9d9 }
+    </style>
+
+    <defs>
+    <clipPath id="terminal-clip-terminal">
+      <rect x="0" y="0" width="975.0" height="243.0" />
+    </clipPath>
+    <clipPath id="terminal-line-0">
+    <rect x="0" y="1.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-1">
+    <rect x="0" y="25.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-2">
+    <rect x="0" y="50.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-3">
+    <rect x="0" y="74.7" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-4">
+    <rect x="0" y="99.1" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-5">
+    <rect x="0" y="123.5" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-6">
+    <rect x="0" y="147.9" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-7">
+    <rect x="0" y="172.3" width="976" height="24.65"/>
+            </clipPath>
+<clipPath id="terminal-line-8">
+    <rect x="0" y="196.7" width="976" height="24.65"/>
+            </clipPath>
+    </defs>
+
+    <rect fill="#292929" stroke="rgba(255,255,255,0.35)" stroke-width="1" x="1" y="1" width="992" height="292" rx="8"/><text class="terminal-title" fill="#c5c8c6" text-anchor="middle" x="496" y="27">ToolCallStreamingUpdateTest</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)">
+    <rect fill="#121212" x="0" y="1.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="25.9" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="12.2" y="25.9" width="12.2" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="24.4" y="25.9" width="292.8" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="317.2" y="25.9" width="658.8" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="50.3" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="74.7" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="99.1" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="123.5" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="147.9" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="172.3" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="196.7" width="976" height="24.65" shape-rendering="crispEdges"/><rect fill="#121212" x="0" y="221.1" width="976" height="24.65" shape-rendering="crispEdges"/>
+    <g class="terminal-matrix">
+    <text class="terminal-r1" x="976" y="20" textLength="12.2" clip-path="url(#terminal-line-0)">
+</text><text class="terminal-r2" x="0" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">■</text><text class="terminal-r2" x="24.4" y="44.4" textLength="292.8" clip-path="url(#terminal-line-1)">Reading&#160;/test/example.py</text><text class="terminal-r1" x="976" y="44.4" textLength="12.2" clip-path="url(#terminal-line-1)">
+</text><text class="terminal-r1" x="976" y="68.8" textLength="12.2" clip-path="url(#terminal-line-2)">
+</text><text class="terminal-r1" x="976" y="93.2" textLength="12.2" clip-path="url(#terminal-line-3)">
+</text><text class="terminal-r1" x="976" y="117.6" textLength="12.2" clip-path="url(#terminal-line-4)">
+</text><text class="terminal-r1" x="976" y="142" textLength="12.2" clip-path="url(#terminal-line-5)">
+</text><text class="terminal-r1" x="976" y="166.4" textLength="12.2" clip-path="url(#terminal-line-6)">
+</text><text class="terminal-r1" x="976" y="190.8" textLength="12.2" clip-path="url(#terminal-line-7)">
+</text><text class="terminal-r1" x="976" y="215.2" textLength="12.2" clip-path="url(#terminal-line-8)">
+</text>
+    </g>
+    </g>
+</svg>

tests/snapshots/test_ui_snapshot_empty_assistant_before_reasoning.py

@@ -0,0 +1,58 @@
+"""Snapshot tests for empty assistant message removed when reasoning starts (e.g. Opus)."""
+
+from __future__ import annotations
+
+from textual.pilot import Pilot
+
+from tests.conftest import build_test_agent_loop
+from tests.mock.utils import mock_llm_chunk
+from tests.snapshots.base_snapshot_test_app import BaseSnapshotTestApp, default_config
+from tests.snapshots.snap_compare import SnapCompare
+from tests.stubs.fake_backend import FakeBackend
+
+
+class SnapshotTestAppEmptyAssistantThenReasoning(BaseSnapshotTestApp):
+    """Backend stream: first chunk is assistant content only (empty), then reasoning.
+
+    Ensures the empty assistant bubble is removed when the first reasoning chunk
+    arrives, so the UI does not show a blank assistant message above the thinking block.
+    """
+
+    def __init__(self) -> None:
+        config = default_config()
+        fake_backend = FakeBackend(
+            chunks=[
+                mock_llm_chunk(content="\n\n"),
+                mock_llm_chunk(
+                    content="", reasoning_content="Let me think about this..."
+                ),
+                mock_llm_chunk(
+                    content="", reasoning_content=" Considering the options."
+                ),
+                mock_llm_chunk(content="The answer is 42."),
+            ]
+        )
+        super().__init__(config=config)
+        self.agent_loop = build_test_agent_loop(
+            config=config,
+            agent_name=self._current_agent_name,
+            enable_streaming=True,
+            backend=fake_backend,
+        )
+
+
+def test_snapshot_empty_assistant_removed_when_reasoning_starts(
+    snap_compare: SnapCompare,
+) -> None:
+    """Empty assistant message is removed when reasoning starts; no blank bubble above thinking."""
+
+    async def run_before(pilot: Pilot) -> None:
+        await pilot.press(*"What is the answer?")
+        await pilot.press("enter")
+        await pilot.pause(0.5)
+
+    assert snap_compare(
+        "test_ui_snapshot_empty_assistant_before_reasoning.py:SnapshotTestAppEmptyAssistantThenReasoning",
+        terminal_size=(120, 36),
+        run_before=run_before,
+    )

tests/snapshots/test_ui_snapshot_modes.py

@@ -24,6 +24,7 @@ def test_snapshot_cycle_to_plan_mode(snap_compare: SnapCompare) -> None:
     async def run_before(pilot: Pilot) -> None:
         await pilot.pause(0.1)
         await pilot.press("shift+tab")  # default -> plan
+        await pilot.app.workers.wait_for_complete()
         await pilot.pause(0.1)
 
     assert snap_compare(
@@ -40,6 +41,7 @@ def test_snapshot_cycle_to_accept_edits_mode(snap_compare: SnapCompare) -> None:
         await pilot.pause(0.1)
         await pilot.press("shift+tab")  # default -> plan
         await pilot.press("shift+tab")  # plan -> accept edits
+        await pilot.app.workers.wait_for_complete()
         await pilot.pause(0.1)
 
     assert snap_compare(
@@ -57,6 +59,7 @@ def test_snapshot_cycle_to_auto_approve_mode(snap_compare: SnapCompare) -> None:
         await pilot.press("shift+tab")  # default -> plan
         await pilot.press("shift+tab")  # plan -> accept edits
         await pilot.press("shift+tab")  # accept edits -> auto approve
+        await pilot.app.workers.wait_for_complete()
         await pilot.pause(0.1)
 
     assert snap_compare(
@@ -75,6 +78,7 @@ def test_snapshot_cycle_wraps_to_default(snap_compare: SnapCompare) -> None:
         await pilot.press("shift+tab")  # plan -> accept edits
         await pilot.press("shift+tab")  # accept edits -> auto approve
         await pilot.press("shift+tab")  # auto approve -> default (wrap)
+        await pilot.app.workers.wait_for_complete()
         await pilot.pause(0.1)
 
     assert snap_compare(

tests/snapshots/test_ui_snapshot_parallel_tool_calls.py

@@ -0,0 +1,114 @@
+from __future__ import annotations
+
+from typing import cast
+
+from textual.app import App, ComposeResult
+from textual.containers import VerticalScroll
+from textual.pilot import Pilot
+from textual.widget import Widget
+
+from tests.snapshots.snap_compare import SnapCompare
+from vibe.cli.textual_ui.handlers.event_handler import EventHandler
+from vibe.cli.textual_ui.widgets.tools import ToolCallMessage
+from vibe.core.tools.builtins.read_file import ReadFile, ReadFileArgs, ReadFileResult
+from vibe.core.types import ToolCallEvent, ToolResultEvent
+
+
+class ParallelToolCallsApp(App):
+    CSS_PATH = "../../vibe/cli/textual_ui/app.tcss"
+
+    def __init__(self) -> None:
+        super().__init__()
+        self._scroll: VerticalScroll | None = None
+        self._handler: EventHandler | None = None
+
+    def compose(self) -> ComposeResult:
+        self._scroll = VerticalScroll(id="messages")
+        yield self._scroll
+
+    def on_mount(self) -> None:
+        async def mount_callback(
+            widget: Widget, *, after: Widget | None = None
+        ) -> None:
+            if self._scroll is None:
+                return
+            if after is not None and after.parent is self._scroll:
+                await self._scroll.mount(widget, after=after)
+            else:
+                await self._scroll.mount(widget)
+
+        self._handler = EventHandler(
+            mount_callback=mount_callback, get_tools_collapsed=lambda: False
+        )
+
+    async def emit_all_tool_calls(self) -> None:
+        if self._handler is None:
+            return
+        for i in range(3):
+            await self._handler.handle_event(
+                ToolCallEvent(
+                    tool_call_id=f"tc_{i}",
+                    tool_call_index=i,
+                    tool_name="read_file",
+                    tool_class=ReadFile,
+                    args=ReadFileArgs(path=f"/src/file_{i}.py"),
+                )
+            )
+
+    def freeze_spinners(self) -> None:
+        for widget in self.query(ToolCallMessage):
+            widget._is_spinning = False
+            if widget._spinner_timer:
+                widget._spinner_timer.stop()
+                widget._spinner_timer = None
+            widget._spinner.reset()
+            if widget._indicator_widget:
+                widget._indicator_widget.update(widget._spinner.current_frame())
+
+    async def resolve_all_results(self) -> None:
+        if self._handler is None:
+            return
+        for i in range(3):
+            await self._handler.handle_event(
+                ToolResultEvent(
+                    tool_name="read_file",
+                    tool_class=ReadFile,
+                    result=ReadFileResult(
+                        path=f"/src/file_{i}.py",
+                        content=f"# content of file_{i}.py",
+                        lines_read=1,
+                        was_truncated=False,
+                    ),
+                    tool_call_id=f"tc_{i}",
+                )
+            )
+
+
+def test_snapshot_parallel_tool_calls_pending(snap_compare: SnapCompare) -> None:
+    async def run_before(pilot: Pilot) -> None:
+        app = cast(ParallelToolCallsApp, pilot.app)
+        await app.emit_all_tool_calls()
+        await pilot.pause(0.3)
+        app.freeze_spinners()
+        await pilot.pause(0.1)
+
+    assert snap_compare(
+        "test_ui_snapshot_parallel_tool_calls.py:ParallelToolCallsApp",
+        terminal_size=(80, 15),
+        run_before=run_before,
+    )
+
+
+def test_snapshot_parallel_tool_calls_resolved(snap_compare: SnapCompare) -> None:
+    async def run_before(pilot: Pilot) -> None:
+        app = cast(ParallelToolCallsApp, pilot.app)
+        await app.emit_all_tool_calls()
+        await pilot.pause(0.3)
+        await app.resolve_all_results()
+        await pilot.pause(0.3)
+
+    assert snap_compare(
+        "test_ui_snapshot_parallel_tool_calls.py:ParallelToolCallsApp",
+        terminal_size=(80, 20),
+        run_before=run_before,
+    )

tests/snapshots/test_ui_snapshot_streaming_tool_call.py

@@ -0,0 +1,75 @@
+from __future__ import annotations
+
+from typing import cast
+from unittest.mock import patch
+
+from textual.app import App, ComposeResult
+from textual.containers import VerticalScroll
+from textual.pilot import Pilot
+
+from tests.snapshots.snap_compare import SnapCompare
+from vibe.cli.textual_ui.widgets.status_message import StatusMessage
+from vibe.cli.textual_ui.widgets.tools import ToolCallMessage
+from vibe.core.tools.builtins.read_file import ReadFile, ReadFileArgs
+from vibe.core.types import ToolCallEvent
+
+
+class ToolCallStreamingUpdateTest(App):
+    CSS_PATH = "../../vibe/cli/textual_ui/app.tcss"
+
+    def __init__(self) -> None:
+        super().__init__()
+        self._widget: ToolCallMessage | None = None
+
+    def compose(self) -> ComposeResult:
+        partial_event = ToolCallEvent(
+            tool_call_id="tc_streaming",
+            tool_call_index=0,
+            tool_name="read_file",
+            tool_class=ReadFile,
+            args=None,
+        )
+        self._widget = ToolCallMessage(partial_event)
+
+        with VerticalScroll():
+            yield self._widget
+
+    def update_with_full_event(self) -> None:
+        if self._widget is None:
+            return
+
+        full_event = ToolCallEvent(
+            tool_call_id="tc_streaming",
+            tool_call_index=0,
+            tool_name="read_file",
+            tool_class=ReadFile,
+            args=ReadFileArgs(path="/test/example.py"),
+        )
+        self._widget.update_event(full_event)
+
+
+def test_snapshot_tool_call_partial(snap_compare: SnapCompare) -> None:
+    async def run_before(pilot: Pilot) -> None:
+        await pilot.pause(0.1)
+
+    with patch.object(StatusMessage, "start_spinner_timer"):
+        assert snap_compare(
+            "test_ui_snapshot_streaming_tool_call.py:ToolCallStreamingUpdateTest",
+            terminal_size=(80, 10),
+            run_before=run_before,
+        )
+
+
+def test_snapshot_tool_call_updated(snap_compare: SnapCompare) -> None:
+    async def run_before(pilot: Pilot) -> None:
+        await pilot.pause(0.1)
+        app = cast(ToolCallStreamingUpdateTest, pilot.app)
+        app.update_with_full_event()
+        await pilot.pause(0.1)
+
+    with patch.object(StatusMessage, "start_spinner_timer"):
+        assert snap_compare(
+            "test_ui_snapshot_streaming_tool_call.py:ToolCallStreamingUpdateTest",
+            terminal_size=(80, 10),
+            run_before=run_before,
+        )

tests/stubs/fake_backend.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from collections.abc import AsyncGenerator, Callable, Iterable
+from collections.abc import AsyncGenerator, Callable, Iterable, Sequence
 from typing import cast
 
 from tests.mock.utils import mock_llm_chunk
@@ -21,7 +21,7 @@ class FakeBackend:
         | Iterable[Iterable[LLMChunk]]
         | None = None,
         *,
-        token_counter: Callable[[list[LLMMessage]], int] | None = None,
+        token_counter: Callable[[Sequence[LLMMessage]], int] | None = None,
         exception_to_raise: Exception | None = None,
     ) -> None:
         """Fake backend that will output the given chunks in the order they are given.
@@ -35,6 +35,7 @@ class FakeBackend:
         """
         self._requests_messages: list[list[LLMMessage]] = []
         self._requests_extra_headers: list[dict[str, str] | None] = []
+        self._requests_metadata: list[dict[str, str] | None] = []
         self._count_tokens_calls: list[list[LLMMessage]] = []
         self._token_counter = token_counter or self._default_token_counter
         self._exception_to_raise = exception_to_raise
@@ -65,8 +66,12 @@ class FakeBackend:
     def requests_extra_headers(self) -> list[dict[str, str] | None]:
         return self._requests_extra_headers
 
+    @property
+    def requests_metadata(self) -> list[dict[str, str] | None]:
+        return self._requests_metadata
+
     @staticmethod
-    def _default_token_counter(messages: list[LLMMessage]) -> int:
+    def _default_token_counter(messages: Sequence[LLMMessage]) -> int:
         return 1
 
     async def __aenter__(self):
@@ -85,12 +90,14 @@ class FakeBackend:
         tool_choice,
         extra_headers,
         max_tokens,
+        metadata=None,
     ) -> LLMChunk:
         if self._exception_to_raise:
             raise self._exception_to_raise
 
         self._requests_messages.append(messages)
         self._requests_extra_headers.append(extra_headers)
+        self._requests_metadata.append(metadata)
 
         if self._streams:
             stream = self._streams.pop(0)
@@ -111,12 +118,14 @@ class FakeBackend:
         tool_choice,
         extra_headers,
         max_tokens,
+        metadata=None,
     ) -> AsyncGenerator[LLMChunk]:
         if self._exception_to_raise:
             raise self._exception_to_raise
 
         self._requests_messages.append(messages)
         self._requests_extra_headers.append(extra_headers)
+        self._requests_metadata.append(metadata)
 
         if self._streams:
             stream = list(self._streams.pop(0))
@@ -134,6 +143,7 @@ class FakeBackend:
         tools,
         tool_choice=None,
         extra_headers,
+        metadata=None,
     ) -> int:
         self._count_tokens_calls.append(list(messages))
         return self._token_counter(messages)

tests/test_agent_auto_compact.py

@@ -16,22 +16,13 @@ from vibe.core.types import (
 
 
 @pytest.mark.asyncio
-async def test_auto_compact_triggers_and_batches_observer(
-    telemetry_events: list[dict],
-) -> None:
-    observed: list[tuple[Role, str | None]] = []
-
-    def observer(msg: LLMMessage) -> None:
-        observed.append((msg.role, msg.content))
-
+async def test_auto_compact_emits_correct_events(telemetry_events: list[dict]) -> None:
     backend = FakeBackend([
         [mock_llm_chunk(content="<summary>")],
         [mock_llm_chunk(content="<final>")],
     ])
     cfg = build_test_vibe_config(auto_compact_threshold=1)
-    agent = build_test_agent_loop(
-        config=cfg, message_observer=observer, backend=backend
-    )
+    agent = build_test_agent_loop(config=cfg, backend=backend)
     agent.stats.context_tokens = 2
 
     events = [ev async for ev in agent.act("Hello")]
@@ -50,14 +41,83 @@ async def test_auto_compact_triggers_and_batches_observer(
     assert end.new_context_tokens >= 1
     assert final.content == "<final>"
 
-    roles = [r for r, _ in observed]
-    assert roles == [Role.system, Role.user, Role.assistant]
-    assert observed[1][1] is not None and "<summary>" in observed[1][1]
-    assert observed[2][1] == "<final>"
-
     auto_compact = [
         e
         for e in telemetry_events
-        if e.get("event_name") == "vibe/auto_compact_triggered"
+        if e.get("event_name") == "vibe.auto_compact_triggered"
     ]
     assert len(auto_compact) == 1
+
+
+@pytest.mark.asyncio
+async def test_auto_compact_observer_sees_user_msg_not_summary() -> None:
+    """Observer sees the original user message and final response.
+
+    Compact internals (summary request, LLM summary) are invisible
+    to the observer because they happen inside silent() / reset().
+    """
+    observed: list[tuple[Role, str | None]] = []
+
+    def observer(msg: LLMMessage) -> None:
+        observed.append((msg.role, msg.content))
+
+    backend = FakeBackend([
+        [mock_llm_chunk(content="<summary>")],
+        [mock_llm_chunk(content="<final>")],
+    ])
+    cfg = build_test_vibe_config(auto_compact_threshold=1)
+    agent = build_test_agent_loop(
+        config=cfg, message_observer=observer, backend=backend
+    )
+    agent.stats.context_tokens = 2
+
+    [_ async for _ in agent.act("Hello")]
+
+    roles = [r for r, _ in observed]
+    assert roles == [Role.system, Role.user, Role.assistant]
+    assert observed[1][1] == "Hello"
+    assert observed[2][1] == "<final>"
+
+
+@pytest.mark.asyncio
+async def test_auto_compact_observer_does_not_see_summary_request() -> None:
+    """The compact summary request and LLM response must not leak to observer."""
+    observed: list[tuple[Role, str | None]] = []
+
+    def observer(msg: LLMMessage) -> None:
+        observed.append((msg.role, msg.content))
+
+    backend = FakeBackend([
+        [mock_llm_chunk(content="<summary>")],
+        [mock_llm_chunk(content="<final>")],
+    ])
+    cfg = build_test_vibe_config(auto_compact_threshold=1)
+    agent = build_test_agent_loop(
+        config=cfg, message_observer=observer, backend=backend
+    )
+    agent.stats.context_tokens = 2
+
+    [_ async for _ in agent.act("Hello")]
+
+    contents = [c for _, c in observed]
+    assert "<summary>" not in contents
+    assert all("compact" not in (c or "").lower() for c in contents)
+
+
+@pytest.mark.asyncio
+async def test_compact_replaces_messages_with_summary() -> None:
+    """After compact, messages list contains only system + summary."""
+    backend = FakeBackend([
+        [mock_llm_chunk(content="<summary>")],
+        [mock_llm_chunk(content="<final>")],
+    ])
+    cfg = build_test_vibe_config(auto_compact_threshold=1)
+    agent = build_test_agent_loop(config=cfg, backend=backend)
+    agent.stats.context_tokens = 2
+
+    [_ async for _ in agent.act("Hello")]
+
+    # After compact + final response: system, summary, final
+    assert agent.messages[0].role == Role.system
+    assert agent.messages[-1].role == Role.assistant
+    assert agent.messages[-1].content == "<final>"

tests/test_agent_backend.py

@@ -1,11 +1,54 @@
 from __future__ import annotations
 
+from unittest.mock import MagicMock
+
+from mcp.types import (
+    CreateMessageRequestParams,
+    CreateMessageResult,
+    SamplingMessage,
+    TextContent,
+)
 import pytest
 
-from tests.conftest import build_test_agent_loop
+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.config import VibeConfig
+from vibe.core.config import Backend, ModelConfig, ProviderConfig, VibeConfig
+from vibe.core.types import EntrypointMetadata
+
+
+def _two_model_vibe_config(active_model: str) -> VibeConfig:
+    """VibeConfig with two models so we can switch active_model."""
+    models = [
+        ModelConfig(
+            name="mistral-vibe-cli-latest", provider="mistral", alias="devstral-latest"
+        ),
+        ModelConfig(
+            name="devstral-small-latest", provider="mistral", alias="devstral-small"
+        ),
+    ]
+    providers = [
+        ProviderConfig(
+            name="mistral",
+            api_base="https://api.mistral.ai/v1",
+            api_key_env_var="MISTRAL_API_KEY",
+            backend=Backend.MISTRAL,
+        )
+    ]
+    return build_test_vibe_config(
+        active_model=active_model, models=models, providers=providers
+    )
+
+
+def _make_sampling_params() -> CreateMessageRequestParams:
+    return CreateMessageRequestParams(
+        messages=[
+            SamplingMessage(role="user", content=TextContent(type="text", text="Hi"))
+        ],
+        systemPrompt=None,
+        temperature=None,
+        maxTokens=100,
+    )
 
 
 @pytest.mark.asyncio
@@ -66,3 +109,79 @@ async def test_updates_tokens_stats_based_on_backend_response_streaming(
     [_ async for _ in agent.act("Hello")]
 
     assert agent.stats.context_tokens == 275
+
+
+@pytest.mark.asyncio
+async def test_passes_entrypoint_metadata_to_backend(vibe_config: VibeConfig):
+    metadata = EntrypointMetadata(
+        agent_entrypoint="acp",
+        agent_version="2.0.0",
+        client_name="vibe_ide",
+        client_version="0.5.0",
+    )
+    backend = FakeBackend([mock_llm_chunk(content="Response")])
+    agent = build_test_agent_loop(
+        config=vibe_config,
+        backend=backend,
+        enable_streaming=True,
+        entrypoint_metadata=metadata,
+    )
+
+    [_ async for _ in agent.act("Hello")]
+
+    assert len(backend.requests_metadata) > 0
+    assert backend.requests_metadata[0] == {
+        "agent_entrypoint": "acp",
+        "agent_version": "2.0.0",
+        "client_name": "vibe_ide",
+        "client_version": "0.5.0",
+    }
+
+
+@pytest.mark.asyncio
+async def test_mcp_sampling_handler_uses_updated_backend_when_agent_backend_changes():
+    """AgentLoop's MCP sampling handler uses current backend when backend is reassigned."""
+    backend1 = FakeBackend([mock_llm_chunk(content="from-backend-1")])
+    backend2 = FakeBackend([mock_llm_chunk(content="from-backend-2")])
+    config = _two_model_vibe_config("devstral-latest")
+    agent = build_test_agent_loop(config=config, backend=backend1)
+    handler = agent._sampling_handler
+    params = _make_sampling_params()
+    context = MagicMock()
+
+    result1 = await handler(context, params)
+    assert isinstance(result1, CreateMessageResult)
+    assert result1.content.type == "text"
+    assert result1.content.text == "from-backend-1"
+    assert len(backend1.requests_messages) == 1
+    assert len(backend2.requests_messages) == 0
+
+    agent.backend = backend2
+    result2 = await handler(context, params)
+    assert isinstance(result2, CreateMessageResult)
+    assert result2.content.type == "text"
+    assert result2.content.text == "from-backend-2"
+    assert len(backend1.requests_messages) == 1
+    assert len(backend2.requests_messages) == 1
+
+
+@pytest.mark.asyncio
+async def test_mcp_sampling_handler_uses_updated_config_when_agent_config_changes():
+    chunk = mock_llm_chunk(content="ok")
+    backend = FakeBackend([chunk])
+    config1 = _two_model_vibe_config("devstral-latest")
+    config2 = _two_model_vibe_config("devstral-small")
+    agent = build_test_agent_loop(config=config1, backend=backend)
+    handler = agent._sampling_handler
+    params = _make_sampling_params()
+    context = MagicMock()
+
+    result1 = await handler(context, params)
+    assert isinstance(result1, CreateMessageResult)
+    assert result1.model == "mistral-vibe-cli-latest"
+
+    agent._base_config = config2
+    agent.agent_manager.invalidate_config()
+    result2 = await handler(context, params)
+    assert isinstance(result2, CreateMessageResult)
+    assert result2.model == "devstral-small-latest"

tests/test_agent_observer_streaming.py

@@ -1,9 +1,10 @@
 from __future__ import annotations
 
+from collections import Counter
 from collections.abc import Callable
 from http import HTTPStatus
 from typing import cast
-from unittest.mock import AsyncMock
+from unittest.mock import AsyncMock, Mock
 
 import httpx
 import pytest
@@ -180,6 +181,27 @@ async def test_act_streams_chunks_in_order() -> None:
     assert agent.messages[-1].content == "Hello from Vibe! More and end"
 
 
+@pytest.mark.asyncio
+async def test_act_streaming_does_not_cleanup_tmp_files_directly() -> None:
+    backend = FakeBackend([
+        mock_llm_chunk(content="Hello"),
+        mock_llm_chunk(content=" from"),
+        mock_llm_chunk(content=" Vibe"),
+    ])
+    agent = build_test_agent_loop(
+        config=make_config(), backend=backend, enable_streaming=True
+    )
+    agent.session_logger.save_interaction = AsyncMock(return_value=None)
+    cleanup_spy = Mock()
+    agent.session_logger.maybe_cleanup_tmp_files = cleanup_spy
+
+    events = [event async for event in agent.act("Stream, please.")]
+
+    assistant_events = [event for event in events if isinstance(event, AssistantEvent)]
+    assert len(assistant_events) == 3
+    assert cleanup_spy.call_count == 0
+
+
 @pytest.mark.asyncio
 async def test_act_handles_streaming_with_tool_call_events_in_sequence() -> None:
     todo_tool_call = ToolCall(
@@ -210,6 +232,7 @@ async def test_act_handles_streaming_with_tool_call_events_in_sequence() -> None
         UserMessageEvent,
         AssistantEvent,
         ToolCallEvent,
+        ToolCallEvent,
         ToolResultEvent,
         AssistantEvent,
     ]
@@ -217,12 +240,17 @@ async def test_act_handles_streaming_with_tool_call_events_in_sequence() -> None
     assert isinstance(events[1], AssistantEvent)
     assert events[1].content == "Checking your todos."
     assert isinstance(events[2], ToolCallEvent)
+    assert events[2].args is None  # streaming event
+    assert events[2].tool_call_id == "tc_stream"
     assert events[2].tool_name == "todo"
-    assert isinstance(events[3], ToolResultEvent)
-    assert events[3].error is None
-    assert events[3].skipped is False
-    assert isinstance(events[4], AssistantEvent)
-    assert events[4].content == "Done reviewing todos."
+    assert isinstance(events[3], ToolCallEvent)
+    assert events[3].args is not None
+    assert events[3].tool_name == "todo"
+    assert isinstance(events[4], ToolResultEvent)
+    assert events[4].error is None
+    assert events[4].skipped is False
+    assert isinstance(events[5], AssistantEvent)
+    assert events[5].content == "Done reviewing todos."
     assert agent.messages[-1].content == "Done reviewing todos."
 
 
@@ -250,21 +278,27 @@ async def test_act_handles_tool_call_chunk_with_content() -> None:
 
     events = [event async for event in agent.act("Check todos with content.")]
 
-    assert [type(event) for event in events] == [
-        UserMessageEvent,
-        AssistantEvent,
-        AssistantEvent,
-        AssistantEvent,
-        ToolCallEvent,
-        ToolResultEvent,
-    ]
-    assert isinstance(events[0], UserMessageEvent)
-    assert isinstance(events[1], AssistantEvent)
-    assert isinstance(events[2], AssistantEvent)
-    assert isinstance(events[3], AssistantEvent)
-    assert events[1].content == "Preparing "
-    assert events[2].content == "todo request"
-    assert events[3].content == " complete"
+    event_types = [type(e) for e in events]
+    assert Counter(event_types) == Counter({
+        UserMessageEvent: 1,
+        AssistantEvent: 3,
+        ToolCallEvent: 2,
+        ToolResultEvent: 1,
+    })
+
+    tool_call_events = [e for e in events if isinstance(e, ToolCallEvent)]
+    assert len(tool_call_events) == 2
+    assert any(
+        tc.args is None and tc.tool_call_id == "tc_content" for tc in tool_call_events
+    )
+    assert any(tc.args is not None for tc in tool_call_events)
+
+    assistant_events = [e for e in events if isinstance(e, AssistantEvent)]
+    assistant_contents = {e.content for e in assistant_events}
+    assert "Preparing " in assistant_contents
+    assert "todo request" in assistant_contents
+    assert " complete" in assistant_contents
+
     assert any(
         m.role == Role.assistant and m.content == "Preparing todo request complete"
         for m in agent.messages
@@ -304,17 +338,21 @@ async def test_act_merges_streamed_tool_call_arguments() -> None:
         UserMessageEvent,
         AssistantEvent,
         ToolCallEvent,
+        ToolCallEvent,
         ToolResultEvent,
     ]
     assert isinstance(events[0], UserMessageEvent)
-    call_event = events[2]
+    assert isinstance(events[2], ToolCallEvent)
+    assert events[2].args is None  # streaming event
+    assert events[2].tool_call_id == "tc_merge"
+    call_event = events[3]
     assert isinstance(call_event, ToolCallEvent)
     assert call_event.tool_call_id == "tc_merge"
     call_args = cast(TodoArgs, call_event.args)
     assert call_args.action == "read"
-    assert isinstance(events[3], ToolResultEvent)
-    assert events[3].error is None
-    assert events[3].skipped is False
+    assert isinstance(events[4], ToolResultEvent)
+    assert events[4].error is None
+    assert events[4].skipped is False
     assistant_with_calls = next(
         m for m in agent.messages if m.role == Role.assistant and m.tool_calls
     )
@@ -371,6 +409,7 @@ async def test_act_handles_user_cancellation_during_streaming() -> None:
     assert [type(event) for event in events] == [
         UserMessageEvent,
         AssistantEvent,
+        ToolCallEvent,
         AssistantEvent,
         ToolCallEvent,
         ToolResultEvent,

tests/test_agent_stats.py

@@ -531,7 +531,7 @@ class TestAutoCompactIntegration:
 
         roles = [r for r, _ in observed]
         assert roles == [Role.system, Role.user, Role.assistant]
-        assert observed[1][1] is not None and "<summary>" in observed[1][1]
+        assert observed[1][1] == "Hello"
 
 
 class TestClearHistoryFullReset:
@@ -623,6 +623,37 @@ class TestClearHistoryFullReset:
         assert agent.session_id == agent.session_logger.session_id
 
 
+class TestClearHistoryObserverBugfix:
+    @pytest.mark.asyncio
+    async def test_clear_history_observer_sees_new_messages(
+        self, observer_capture
+    ) -> None:
+        """Bug fix: clear_history previously left a stale index, so new messages
+        appended after clearing were never observed.
+        """
+        observed, observer = observer_capture
+        backend = FakeBackend([
+            [mock_llm_chunk(content="First")],
+            [mock_llm_chunk(content="Second")],
+        ])
+        agent = build_test_agent_loop(
+            config=make_config(), message_observer=observer, backend=backend
+        )
+
+        async for _ in agent.act("Hello"):
+            pass
+
+        await agent.clear_history()
+        observed.clear()
+
+        async for _ in agent.act("After clear"):
+            pass
+
+        roles = [msg.role for msg in observed]
+        assert Role.user in roles
+        assert Role.assistant in roles
+
+
 class TestStatsEdgeCases:
     @pytest.mark.asyncio
     async def test_session_cost_approximation_on_model_change(

tests/test_agent_tool_call.py

@@ -113,7 +113,7 @@ async def test_single_tool_call_executes_under_auto_approve(
     assert "total_count" in (tool_msgs[-1].content or "")
 
     tool_finished = [
-        e for e in telemetry_events if e.get("event_name") == "vibe/tool_call_finished"
+        e for e in telemetry_events if e.get("event_name") == "vibe.tool_call_finished"
     ]
     assert len(tool_finished) == 1
     assert tool_finished[0]["properties"]["tool_name"] == "todo"
@@ -158,7 +158,7 @@ async def test_tool_call_requires_approval_if_not_auto_approved(
     assert agent_loop.stats.tool_calls_succeeded == 0
 
     tool_finished = [
-        e for e in telemetry_events if e.get("event_name") == "vibe/tool_call_finished"
+        e for e in telemetry_events if e.get("event_name") == "vibe.tool_call_finished"
     ]
     assert len(tool_finished) == 1
     assert tool_finished[0]["properties"]["approval_type"] == "ask"
@@ -198,7 +198,7 @@ async def test_tool_call_approved_by_callback(telemetry_events: list[dict]) -> N
     assert agent_loop.stats.tool_calls_succeeded == 1
 
     tool_finished = [
-        e for e in telemetry_events if e.get("event_name") == "vibe/tool_call_finished"
+        e for e in telemetry_events if e.get("event_name") == "vibe.tool_call_finished"
     ]
     assert len(tool_finished) == 1
     assert tool_finished[0]["properties"]["approval_type"] == "ask"
@@ -243,7 +243,7 @@ async def test_tool_call_rejected_when_auto_approve_disabled_and_rejected_by_cal
     assert agent_loop.stats.tool_calls_succeeded == 0
 
     tool_finished = [
-        e for e in telemetry_events if e.get("event_name") == "vibe/tool_call_finished"
+        e for e in telemetry_events if e.get("event_name") == "vibe.tool_call_finished"
     ]
     assert len(tool_finished) == 1
     assert tool_finished[0]["properties"]["approval_type"] == "ask"
@@ -287,7 +287,7 @@ async def test_tool_call_skipped_when_permission_is_never(
     assert agent_loop.stats.tool_calls_succeeded == 0
 
     tool_finished = [
-        e for e in telemetry_events if e.get("event_name") == "vibe/tool_call_finished"
+        e for e in telemetry_events if e.get("event_name") == "vibe.tool_call_finished"
     ]
     assert len(tool_finished) == 1
     assert tool_finished[0]["properties"]["approval_type"] == "never"
@@ -492,14 +492,14 @@ async def test_fill_missing_tool_responses_inserts_placeholders() -> None:
     assistant_msg = LLMMessage(
         role=Role.assistant, content="Calling tools...", tool_calls=tool_calls_messages
     )
-    agent_loop.messages = [
+    agent_loop.messages.reset([
         agent_loop.messages[0],
         assistant_msg,
         # only one tool responded: the second is missing
         LLMMessage(
             role=Role.tool, tool_call_id="tc1", name="todo", content="Retrieved 0 todos"
         ),
-    ]
+    ])
 
     await act_and_collect_events(agent_loop, "Proceed")
 
@@ -524,7 +524,7 @@ async def test_ensure_assistant_after_tool_appends_understood() -> None:
     tool_msg = LLMMessage(
         role=Role.tool, tool_call_id="tc_z", name="todo", content="Done"
     )
-    agent_loop.messages = [agent_loop.messages[0], tool_msg]
+    agent_loop.messages.reset([agent_loop.messages[0], tool_msg])
 
     await act_and_collect_events(agent_loop, "Next")
 

tests/test_agents.py

@@ -126,8 +126,9 @@ class TestAgentSafety:
 
 
 class TestAgentProfile:
-    def test_all_builtin_agents_exist(self) -> None:
-        assert set(BUILTIN_AGENTS.keys()) == set(BuiltinAgentName)
+    def test_all_builtin_agents_have_valid_names(self) -> None:
+        acp_only = {BuiltinAgentName.CHAT}
+        assert set(BUILTIN_AGENTS.keys()) == set(BuiltinAgentName) - acp_only
 
     def test_display_name_property(self) -> None:
         assert BUILTIN_AGENTS[BuiltinAgentName.DEFAULT].display_name == "Default"

tests/test_cli_programmatic_preload.py

@@ -79,7 +79,7 @@ def test_run_programmatic_preload_streaming_is_batched(
         ]
 
         new_session = [
-            e for e in telemetry_events if e.get("event_name") == "vibe/new_session"
+            e for e in telemetry_events if e.get("event_name") == "vibe.new_session"
         ]
         assert len(new_session) == 1
         assert new_session[0]["properties"]["entrypoint"] == "programmatic"

tests/test_middleware.py

@@ -3,44 +3,67 @@ from __future__ import annotations
 import pytest
 
 from tests.conftest import build_test_agent_loop, build_test_vibe_config
-from vibe.core.agents.models import BUILTIN_AGENTS, AgentProfile, BuiltinAgentName
+from vibe.core.agents.models import BUILTIN_AGENTS, CHAT, AgentProfile, BuiltinAgentName
 from vibe.core.config import VibeConfig
 from vibe.core.middleware import (
+    CHAT_AGENT_EXIT,
+    CHAT_AGENT_REMINDER,
     PLAN_AGENT_EXIT,
     PLAN_AGENT_REMINDER,
     ConversationContext,
     MiddlewareAction,
     MiddlewarePipeline,
-    PlanAgentMiddleware,
+    ReadOnlyAgentMiddleware,
     ResetReason,
 )
-from vibe.core.types import AgentStats
+from vibe.core.types import AgentStats, MessageList
+
+REMINDER = "test reminder"
+EXIT_MSG = "test exit"
+TARGET_AGENT = BuiltinAgentName.PLAN
+
+
+def _build_middleware(
+    profile_getter,
+    agent_name: str = TARGET_AGENT,
+    reminder: str = REMINDER,
+    exit_message: str = EXIT_MSG,
+) -> ReadOnlyAgentMiddleware:
+    return ReadOnlyAgentMiddleware(profile_getter, agent_name, reminder, exit_message)
 
 
 @pytest.fixture
 def ctx(vibe_config: VibeConfig) -> ConversationContext:
-    return ConversationContext(messages=[], stats=AgentStats(), config=vibe_config)
+    return ConversationContext(
+        messages=MessageList(), stats=AgentStats(), config=vibe_config
+    )
 
 
-class TestPlanAgentMiddleware:
+class TestReadOnlyAgentMiddleware:
     @pytest.mark.asyncio
-    async def test_injects_reminder_when_plan_agent_active(
+    async def test_injects_reminder_when_target_agent_active(
         self, ctx: ConversationContext
     ) -> None:
-        middleware = PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
+        middleware = _build_middleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
 
         result = await middleware.before_turn(ctx)
 
         assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_REMINDER
+        assert result.message == REMINDER
 
     @pytest.mark.asyncio
-    async def test_does_not_inject_when_default_agent(
-        self, ctx: ConversationContext
+    @pytest.mark.parametrize(
+        "agent_name",
+        [
+            BuiltinAgentName.DEFAULT,
+            BuiltinAgentName.AUTO_APPROVE,
+            BuiltinAgentName.ACCEPT_EDITS,
+        ],
+    )
+    async def test_does_not_inject_when_non_target_agent(
+        self, ctx: ConversationContext, agent_name: str
     ) -> None:
-        middleware = PlanAgentMiddleware(
-            lambda: BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-        )
+        middleware = _build_middleware(lambda: BUILTIN_AGENTS[agent_name])
 
         result = await middleware.before_turn(ctx)
 
@@ -48,89 +71,56 @@ class TestPlanAgentMiddleware:
         assert result.message is None
 
     @pytest.mark.asyncio
-    async def test_does_not_inject_when_auto_approve_agent(
-        self, ctx: ConversationContext
-    ) -> None:
-        middleware = PlanAgentMiddleware(
-            lambda: BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
-        )
-
-        result = await middleware.before_turn(ctx)
-
-        assert result.action == MiddlewareAction.CONTINUE
-        assert result.message is None
-
-    @pytest.mark.asyncio
-    async def test_does_not_inject_when_accept_edits_agent(
-        self, ctx: ConversationContext
-    ) -> None:
-        middleware = PlanAgentMiddleware(
-            lambda: BUILTIN_AGENTS[BuiltinAgentName.ACCEPT_EDITS]
-        )
-
-        result = await middleware.before_turn(ctx)
-
-        assert result.action == MiddlewareAction.CONTINUE
-        assert result.message is None
-
-    @pytest.mark.asyncio
-    async def test_injects_reminder_only_once_while_in_plan_mode(
-        self, ctx: ConversationContext
-    ) -> None:
-        middleware = PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
+    async def test_injects_reminder_only_once(self, ctx: ConversationContext) -> None:
+        middleware = _build_middleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
 
         result1 = await middleware.before_turn(ctx)
         assert result1.action == MiddlewareAction.INJECT_MESSAGE
-        assert result1.message == PLAN_AGENT_REMINDER
+        assert result1.message == REMINDER
 
         result2 = await middleware.before_turn(ctx)
         assert result2.action == MiddlewareAction.CONTINUE
         assert result2.message is None
 
     @pytest.mark.asyncio
-    async def test_injects_exit_message_when_leaving_plan_mode(
+    async def test_injects_exit_message_when_leaving(
         self, ctx: ConversationContext
     ) -> None:
         current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
+        middleware = _build_middleware(lambda: current_profile)
 
-        # Enter plan mode
         await middleware.before_turn(ctx)
 
-        # Leave plan mode
         current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
         result = await middleware.before_turn(ctx)
         assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_EXIT
+        assert result.message == EXIT_MSG
 
     @pytest.mark.asyncio
-    async def test_reinjects_reminder_when_reentering_plan_mode(
+    async def test_reinjects_reminder_on_reentry(
         self, ctx: ConversationContext
     ) -> None:
         current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
+        middleware = _build_middleware(lambda: current_profile)
 
-        # Enter plan mode - should inject reminder
         result1 = await middleware.before_turn(ctx)
         assert result1.action == MiddlewareAction.INJECT_MESSAGE
-        assert result1.message == PLAN_AGENT_REMINDER
+        assert result1.message == REMINDER
 
-        # Leave plan mode - should inject exit message
         current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
         result2 = await middleware.before_turn(ctx)
         assert result2.action == MiddlewareAction.INJECT_MESSAGE
-        assert result2.message == PLAN_AGENT_EXIT
+        assert result2.message == EXIT_MSG
 
-        # Re-enter plan mode - should inject reminder again
         current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
         result3 = await middleware.before_turn(ctx)
         assert result3.action == MiddlewareAction.INJECT_MESSAGE
-        assert result3.message == PLAN_AGENT_REMINDER
+        assert result3.message == REMINDER
 
     @pytest.mark.asyncio
     async def test_custom_reminder(self, ctx: ConversationContext) -> None:
-        custom_reminder = "Custom plan agent reminder"
-        middleware = PlanAgentMiddleware(
+        custom_reminder = "Custom reminder"
+        middleware = _build_middleware(
             lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN], reminder=custom_reminder
         )
 
@@ -138,246 +128,216 @@ class TestPlanAgentMiddleware:
 
         assert result.message == custom_reminder
 
-    @pytest.mark.asyncio
-    async def test_reset_clears_state(self, ctx: ConversationContext) -> None:
-        middleware = PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
-        await middleware.before_turn(ctx)  # Enter and inject
-
-        middleware.reset()
-
-        # Should inject again after reset
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-
-    @pytest.mark.asyncio
-    async def test_exit_message_fires_only_once(self, ctx: ConversationContext) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        # Enter plan mode
-        await middleware.before_turn(ctx)
-
-        # Leave plan mode - first call should inject exit
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_EXIT
-
-        # Subsequent calls in default mode should be CONTINUE
-        result2 = await middleware.before_turn(ctx)
-        assert result2.action == MiddlewareAction.CONTINUE
-        assert result2.message is None
-
-    @pytest.mark.asyncio
-    async def test_multiple_turns_in_plan_mode_after_entry(
-        self, ctx: ConversationContext
-    ) -> None:
-        middleware = PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
-
-        # First turn: inject reminder
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-
-        # Several more turns in plan mode: all should be CONTINUE
-        for _ in range(5):
-            result = await middleware.before_turn(ctx)
-            assert result.action == MiddlewareAction.CONTINUE
-            assert result.message is None
-
-    @pytest.mark.asyncio
-    async def test_multiple_turns_in_default_mode_after_exit(
-        self, ctx: ConversationContext
-    ) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        await middleware.before_turn(ctx)  # enter plan
-
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-        await middleware.before_turn(ctx)  # exit plan (fires exit message)
-
-        # Several more turns in default mode: all should be CONTINUE
-        for _ in range(5):
-            result = await middleware.before_turn(ctx)
-            assert result.action == MiddlewareAction.CONTINUE
-            assert result.message is None
-
-    @pytest.mark.asyncio
-    async def test_rapid_toggling_plan_default_multiple_cycles(
-        self, ctx: ConversationContext
-    ) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        for _ in range(3):
-            # Enter plan
-            current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-            result = await middleware.before_turn(ctx)
-            assert result.action == MiddlewareAction.INJECT_MESSAGE
-            assert result.message == PLAN_AGENT_REMINDER
-
-            # Leave plan
-            current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-            result = await middleware.before_turn(ctx)
-            assert result.action == MiddlewareAction.INJECT_MESSAGE
-            assert result.message == PLAN_AGENT_EXIT
-
-    @pytest.mark.asyncio
-    async def test_exit_to_non_default_agent(self, ctx: ConversationContext) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        await middleware.before_turn(ctx)  # enter plan
-
-        # Switch to auto_approve (not default)
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_EXIT
-
-    @pytest.mark.asyncio
-    async def test_exit_to_accept_edits_agent(self, ctx: ConversationContext) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        await middleware.before_turn(ctx)  # enter plan
-
-        # Switch to accept_edits
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.ACCEPT_EDITS]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_EXIT
-
-    @pytest.mark.asyncio
-    async def test_switching_between_non_plan_agents(
-        self, ctx: ConversationContext
-    ) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.CONTINUE
-
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.CONTINUE
-
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.ACCEPT_EDITS]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.CONTINUE
-
-    @pytest.mark.asyncio
-    async def test_non_plan_to_plan_entry(self, ctx: ConversationContext) -> None:
-        """Starting in a non-plan agent then entering plan should inject reminder."""
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.CONTINUE
-
-        # Now switch to plan
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_REMINDER
-
-    @pytest.mark.asyncio
-    async def test_reset_while_in_default_after_exiting_plan(
-        self, ctx: ConversationContext
-    ) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        await middleware.before_turn(ctx)  # enter plan
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-        await middleware.before_turn(ctx)  # exit plan
-
-        middleware.reset()
-
-        # Still in default mode - should CONTINUE (no phantom exit message)
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.CONTINUE
-
-    @pytest.mark.asyncio
-    async def test_reset_while_in_default_then_reenter_plan(
-        self, ctx: ConversationContext
-    ) -> None:
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
-
-        await middleware.before_turn(ctx)  # enter plan
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
-        await middleware.before_turn(ctx)  # exit plan
-
-        middleware.reset()
-
-        # Re-enter plan after reset
-        current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_REMINDER
-
-    @pytest.mark.asyncio
-    async def test_reset_with_compact_reason(self, ctx: ConversationContext) -> None:
-        middleware = PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
-        await middleware.before_turn(ctx)  # enter and inject
-
-        middleware.reset(ResetReason.COMPACT)
-
-        # Should reinject reminder after compact reset
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_REMINDER
-
     @pytest.mark.asyncio
     async def test_custom_exit_message(self, ctx: ConversationContext) -> None:
         custom_exit = "Custom exit message"
         current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(
+        middleware = _build_middleware(
             lambda: current_profile, exit_message=custom_exit
         )
 
-        await middleware.before_turn(ctx)  # enter plan
+        await middleware.before_turn(ctx)
 
         current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
         result = await middleware.before_turn(ctx)
         assert result.message == custom_exit
 
     @pytest.mark.asyncio
-    async def test_plan_entry_then_immediate_exit_same_not_possible(
-        self, ctx: ConversationContext
-    ) -> None:
-        """Even if profile changes between two calls, each call sees one transition."""
-        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
-        middleware = PlanAgentMiddleware(lambda: current_profile)
+    async def test_reset_clears_state(self, ctx: ConversationContext) -> None:
+        middleware = _build_middleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
+        await middleware.before_turn(ctx)
+
+        middleware.reset()
 
-        # First call: entry
         result = await middleware.before_turn(ctx)
         assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_REMINDER
 
-        # Second call (still plan): no injection
-        result = await middleware.before_turn(ctx)
-        assert result.action == MiddlewareAction.CONTINUE
+    @pytest.mark.asyncio
+    async def test_exit_message_fires_only_once(self, ctx: ConversationContext) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        await middleware.before_turn(ctx)
 
-        # Third call (switched to default): exit
         current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
         result = await middleware.before_turn(ctx)
         assert result.action == MiddlewareAction.INJECT_MESSAGE
-        assert result.message == PLAN_AGENT_EXIT
+        assert result.message == EXIT_MSG
+
+        result2 = await middleware.before_turn(ctx)
+        assert result2.action == MiddlewareAction.CONTINUE
+        assert result2.message is None
+
+    @pytest.mark.asyncio
+    async def test_multiple_turns_after_entry(self, ctx: ConversationContext) -> None:
+        middleware = _build_middleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+
+        for _ in range(5):
+            result = await middleware.before_turn(ctx)
+            assert result.action == MiddlewareAction.CONTINUE
+            assert result.message is None
+
+    @pytest.mark.asyncio
+    async def test_multiple_turns_after_exit(self, ctx: ConversationContext) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        await middleware.before_turn(ctx)
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
+        await middleware.before_turn(ctx)
+
+        for _ in range(5):
+            result = await middleware.before_turn(ctx)
+            assert result.action == MiddlewareAction.CONTINUE
+            assert result.message is None
+
+    @pytest.mark.asyncio
+    async def test_rapid_toggling_multiple_cycles(
+        self, ctx: ConversationContext
+    ) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        for _ in range(3):
+            current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+            result = await middleware.before_turn(ctx)
+            assert result.action == MiddlewareAction.INJECT_MESSAGE
+            assert result.message == REMINDER
+
+            current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
+            result = await middleware.before_turn(ctx)
+            assert result.action == MiddlewareAction.INJECT_MESSAGE
+            assert result.message == EXIT_MSG
+
+    @pytest.mark.asyncio
+    async def test_exit_to_non_default_agent(self, ctx: ConversationContext) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        await middleware.before_turn(ctx)
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert result.message == EXIT_MSG
+
+    @pytest.mark.asyncio
+    async def test_switching_between_non_target_agents(
+        self, ctx: ConversationContext
+    ) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
+        middleware = _build_middleware(lambda: current_profile)
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.CONTINUE
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.CONTINUE
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.ACCEPT_EDITS]
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.CONTINUE
+
+    @pytest.mark.asyncio
+    async def test_non_target_to_target_entry(self, ctx: ConversationContext) -> None:
+        """Starting in a non-target agent then entering target should inject reminder."""
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE]
+        middleware = _build_middleware(lambda: current_profile)
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.CONTINUE
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert result.message == REMINDER
+
+    @pytest.mark.asyncio
+    async def test_reset_while_inactive_after_exit(
+        self, ctx: ConversationContext
+    ) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        await middleware.before_turn(ctx)
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
+        await middleware.before_turn(ctx)
+
+        middleware.reset()
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.CONTINUE
+
+    @pytest.mark.asyncio
+    async def test_reset_while_inactive_then_reenter(
+        self, ctx: ConversationContext
+    ) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        await middleware.before_turn(ctx)
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
+        await middleware.before_turn(ctx)
+
+        middleware.reset()
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert result.message == REMINDER
+
+    @pytest.mark.asyncio
+    async def test_reset_with_compact_reason(self, ctx: ConversationContext) -> None:
+        middleware = _build_middleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN])
+        await middleware.before_turn(ctx)
+
+        middleware.reset(ResetReason.COMPACT)
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert result.message == REMINDER
+
+    @pytest.mark.asyncio
+    async def test_entry_then_continuation_then_exit_then_continuation(
+        self, ctx: ConversationContext
+    ) -> None:
+        """Each call sees one transition at a time."""
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        middleware = _build_middleware(lambda: current_profile)
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert result.message == REMINDER
+
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.CONTINUE
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.DEFAULT]
+        result = await middleware.before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert result.message == EXIT_MSG
 
-        # Fourth call (still default): no injection
         result = await middleware.before_turn(ctx)
         assert result.action == MiddlewareAction.CONTINUE
 
 
-class TestMiddlewarePipelineWithPlanAgent:
+class TestMiddlewarePipelineWithReadOnlyAgent:
     @pytest.mark.asyncio
-    async def test_pipeline_includes_plan_agent_injection(
-        self, ctx: ConversationContext
-    ) -> None:
+    async def test_pipeline_includes_injection(self, ctx: ConversationContext) -> None:
         pipeline = MiddlewarePipeline()
-        pipeline.add(PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN]))
+        pipeline.add(
+            ReadOnlyAgentMiddleware(
+                lambda: BUILTIN_AGENTS[BuiltinAgentName.PLAN],
+                BuiltinAgentName.PLAN,
+                PLAN_AGENT_REMINDER,
+                PLAN_AGENT_EXIT,
+            )
+        )
 
         result = await pipeline.run_before_turn(ctx)
 
@@ -385,20 +345,73 @@ class TestMiddlewarePipelineWithPlanAgent:
         assert PLAN_AGENT_REMINDER in (result.message or "")
 
     @pytest.mark.asyncio
-    async def test_pipeline_skips_injection_when_not_plan_agent(
+    async def test_pipeline_skips_injection_when_not_target_agent(
         self, ctx: ConversationContext
     ) -> None:
         pipeline = MiddlewarePipeline()
         pipeline.add(
-            PlanAgentMiddleware(lambda: BUILTIN_AGENTS[BuiltinAgentName.DEFAULT])
+            ReadOnlyAgentMiddleware(
+                lambda: BUILTIN_AGENTS[BuiltinAgentName.DEFAULT],
+                BuiltinAgentName.PLAN,
+                PLAN_AGENT_REMINDER,
+                PLAN_AGENT_EXIT,
+            )
         )
 
         result = await pipeline.run_before_turn(ctx)
 
         assert result.action == MiddlewareAction.CONTINUE
 
+    @pytest.mark.asyncio
+    async def test_direct_plan_to_chat_transition_delivers_both_messages(
+        self, ctx: ConversationContext
+    ) -> None:
+        current_profile: AgentProfile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        pipeline = MiddlewarePipeline()
+        pipeline.add(
+            ReadOnlyAgentMiddleware(
+                lambda: current_profile,
+                BuiltinAgentName.PLAN,
+                PLAN_AGENT_REMINDER,
+                PLAN_AGENT_EXIT,
+            )
+        )
+        pipeline.add(
+            ReadOnlyAgentMiddleware(
+                lambda: current_profile,
+                BuiltinAgentName.CHAT,
+                CHAT_AGENT_REMINDER,
+                CHAT_AGENT_EXIT,
+            )
+        )
 
-class TestPlanAgentMiddlewareIntegration:
+        result = await pipeline.run_before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert PLAN_AGENT_REMINDER in (result.message or "")
+
+        current_profile = CHAT
+        result = await pipeline.run_before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert PLAN_AGENT_EXIT in (result.message or "")
+        assert CHAT_AGENT_REMINDER in (result.message or "")
+
+        current_profile = BUILTIN_AGENTS[BuiltinAgentName.PLAN]
+        result = await pipeline.run_before_turn(ctx)
+        assert result.action == MiddlewareAction.INJECT_MESSAGE
+        assert CHAT_AGENT_EXIT in (result.message or "")
+        assert PLAN_AGENT_REMINDER in (result.message or "")
+
+
+def _find_plan_middleware(agent) -> ReadOnlyAgentMiddleware:
+    return next(
+        mw
+        for mw in agent.middleware_pipeline.middlewares
+        if isinstance(mw, ReadOnlyAgentMiddleware)
+        and mw._agent_name == BuiltinAgentName.PLAN
+    )
+
+
+class TestReadOnlyAgentMiddlewareIntegration:
     @pytest.mark.asyncio
     async def test_switch_agent_preserves_middleware_state_for_exit_message(
         self,
@@ -414,11 +427,7 @@ class TestPlanAgentMiddlewareIntegration:
         )
         agent = build_test_agent_loop(config=config, agent_name=BuiltinAgentName.PLAN)
 
-        plan_middleware = next(
-            mw
-            for mw in agent.middleware_pipeline.middlewares
-            if isinstance(mw, PlanAgentMiddleware)
-        )
+        plan_middleware = _find_plan_middleware(agent)
 
         ctx = ConversationContext(
             messages=agent.messages, stats=agent.stats, config=agent.config
@@ -429,11 +438,7 @@ class TestPlanAgentMiddlewareIntegration:
 
         await agent.switch_agent(BuiltinAgentName.DEFAULT)
 
-        plan_middleware_after = next(
-            mw
-            for mw in agent.middleware_pipeline.middlewares
-            if isinstance(mw, PlanAgentMiddleware)
-        )
+        plan_middleware_after = _find_plan_middleware(agent)
         assert plan_middleware is plan_middleware_after
 
         ctx = ConversationContext(
@@ -456,11 +461,7 @@ class TestPlanAgentMiddlewareIntegration:
         )
         agent = build_test_agent_loop(config=config, agent_name=BuiltinAgentName.PLAN)
 
-        plan_middleware = next(
-            mw
-            for mw in agent.middleware_pipeline.middlewares
-            if isinstance(mw, PlanAgentMiddleware)
-        )
+        plan_middleware = _find_plan_middleware(agent)
 
         ctx = ConversationContext(
             messages=agent.messages, stats=agent.stats, config=agent.config
@@ -497,11 +498,7 @@ class TestPlanAgentMiddlewareIntegration:
         )
         agent = build_test_agent_loop(config=config, agent_name=BuiltinAgentName.PLAN)
 
-        plan_middleware = next(
-            mw
-            for mw in agent.middleware_pipeline.middlewares
-            if isinstance(mw, PlanAgentMiddleware)
-        )
+        plan_middleware = _find_plan_middleware(agent)
 
         ctx = ConversationContext(
             messages=agent.messages, stats=agent.stats, config=agent.config
@@ -532,11 +529,7 @@ class TestPlanAgentMiddlewareIntegration:
             config=config, agent_name=BuiltinAgentName.DEFAULT
         )
 
-        plan_middleware = next(
-            mw
-            for mw in agent.middleware_pipeline.middlewares
-            if isinstance(mw, PlanAgentMiddleware)
-        )
+        plan_middleware = _find_plan_middleware(agent)
 
         ctx = ConversationContext(
             messages=agent.messages, stats=agent.stats, config=agent.config
@@ -566,11 +559,7 @@ class TestPlanAgentMiddlewareIntegration:
         )
         agent = build_test_agent_loop(config=config, agent_name=BuiltinAgentName.PLAN)
 
-        plan_middleware = next(
-            mw
-            for mw in agent.middleware_pipeline.middlewares
-            if isinstance(mw, PlanAgentMiddleware)
-        )
+        plan_middleware = _find_plan_middleware(agent)
 
         def _ctx():
             return ConversationContext(

tests/tools/test_bash.py

@@ -76,14 +76,14 @@ async def test_decodes_non_utf8_bytes(bash):
     assert result.stderr == ""
 
 
-def test_check_allowlist_denylist():
+def test_resolve_permission():
     config = BashToolConfig(allowlist=["echo", "pwd"], denylist=["rm"])
     bash_tool = Bash(config=config, state=BaseToolState())
 
-    allowlisted = bash_tool.check_allowlist_denylist(BashArgs(command="echo hi"))
-    denylisted = bash_tool.check_allowlist_denylist(BashArgs(command="rm -rf /tmp"))
-    mixed = bash_tool.check_allowlist_denylist(BashArgs(command="pwd && whoami"))
-    empty = bash_tool.check_allowlist_denylist(BashArgs(command=""))
+    allowlisted = bash_tool.resolve_permission(BashArgs(command="echo hi"))
+    denylisted = bash_tool.resolve_permission(BashArgs(command="rm -rf /tmp"))
+    mixed = bash_tool.resolve_permission(BashArgs(command="pwd && whoami"))
+    empty = bash_tool.resolve_permission(BashArgs(command=""))
 
     assert allowlisted is ToolPermission.ALWAYS
     assert denylisted is ToolPermission.NEVER

tests/tools/test_grep.py

@@ -5,21 +5,15 @@ import shutil
 import pytest
 
 from tests.mock.utils import collect_result
-from vibe.core.tools.base import ToolError
-from vibe.core.tools.builtins.grep import (
-    Grep,
-    GrepArgs,
-    GrepBackend,
-    GrepState,
-    GrepToolConfig,
-)
+from vibe.core.tools.base import BaseToolState, ToolError
+from vibe.core.tools.builtins.grep import Grep, GrepArgs, GrepBackend, GrepToolConfig
 
 
 @pytest.fixture
 def grep(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = GrepToolConfig()
-    return Grep(config=config, state=GrepState())
+    return Grep(config=config, state=BaseToolState())
 
 
 @pytest.fixture
@@ -34,7 +28,7 @@ def grep_gnu_only(tmp_path, monkeypatch):
 
     monkeypatch.setattr("shutil.which", mock_which)
     config = GrepToolConfig()
-    return Grep(config=config, state=GrepState())
+    return Grep(config=config, state=BaseToolState())
 
 
 def test_detects_ripgrep_when_available(grep):
@@ -143,7 +137,7 @@ async def test_truncates_to_max_matches(grep, tmp_path):
 async def test_truncates_to_max_output_bytes(grep, tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = GrepToolConfig(max_output_bytes=100)
-    grep_tool = Grep(config=config, state=GrepState())
+    grep_tool = Grep(config=config, state=BaseToolState())
     (tmp_path / "test.py").write_text("\n".join("x" * 100 for _ in range(10)))
 
     result = await collect_result(grep_tool.run(GrepArgs(pattern="x")))
@@ -191,22 +185,11 @@ async def test_ignores_comments_in_vibeignore(grep, tmp_path):
     assert result.match_count >= 1
 
 
-@pytest.mark.asyncio
-async def test_tracks_search_history(grep, tmp_path):
-    (tmp_path / "test.py").write_text("content\n")
-
-    await collect_result(grep.run(GrepArgs(pattern="first")))
-    await collect_result(grep.run(GrepArgs(pattern="second")))
-    await collect_result(grep.run(GrepArgs(pattern="third")))
-
-    assert grep.state.search_history == ["first", "second", "third"]
-
-
 @pytest.mark.asyncio
 async def test_uses_effective_workdir(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = GrepToolConfig()
-    grep_tool = Grep(config=config, state=GrepState())
+    grep_tool = Grep(config=config, state=BaseToolState())
     (tmp_path / "test.py").write_text("match\n")
 
     result = await collect_result(grep_tool.run(GrepArgs(pattern="match", path=".")))

tests/tools/test_manager_get_tool_config.py

@@ -227,6 +227,70 @@ class FileTool(BaseTool[FileToolArgs, FileToolResult, BaseToolConfig, BaseToolSt
         assert "file_tool" in tools
 
 
+class TestToolRuntimeAvailability:
+    """Tests for is_available() filtering in ToolManager."""
+
+    def test_unavailable_tool_excluded_from_available_tools(
+        self, tmp_path: Path, monkeypatch
+    ):
+        """Tools where is_available() returns False should be excluded."""
+        import sys
+
+        tool_dir = tmp_path / "tools"
+        tool_dir.mkdir()
+        (tool_dir / "conditional_tool.py").write_text("""
+import os
+from vibe.core.tools.base import BaseTool, BaseToolConfig, BaseToolState
+from pydantic import BaseModel
+
+class ConditionalToolArgs(BaseModel):
+    pass
+
+class ConditionalToolResult(BaseModel):
+    pass
+
+class ConditionalTool(BaseTool[ConditionalToolArgs, ConditionalToolResult, BaseToolConfig, BaseToolState]):
+    description = "Tool that requires TEST_VAR"
+
+    @classmethod
+    def is_available(cls) -> bool:
+        return bool(os.getenv("TEST_VAR"))
+
+    async def run(self, args, ctx=None):
+        yield ConditionalToolResult()
+""")
+
+        to_remove = [k for k in sys.modules if "conditional_tool" in k]
+        for k in to_remove:
+            del sys.modules[k]
+
+        monkeypatch.delenv("TEST_VAR", raising=False)
+        vibe_config = build_test_vibe_config(
+            system_prompt_id="tests",
+            include_project_context=False,
+            tool_paths=[tool_dir],
+        )
+        manager = ToolManager(lambda: vibe_config)
+        assert "conditional_tool" not in manager.available_tools
+
+        to_remove = [k for k in sys.modules if "conditional_tool" in k]
+        for k in to_remove:
+            del sys.modules[k]
+
+        monkeypatch.setenv("TEST_VAR", "1")
+        manager2 = ToolManager(lambda: vibe_config)
+        assert "conditional_tool" in manager2.available_tools
+
+    def test_default_is_available_returns_true(self):
+        """Tools without is_available() override should be available."""
+        vibe_config = build_test_vibe_config(
+            system_prompt_id="tests", include_project_context=False
+        )
+        manager = ToolManager(lambda: vibe_config)
+
+        assert "bash" in manager.available_tools
+
+
 class TestToolManagerModuleReuse:
     """Tests for module reuse across ToolManager instances.
 

tests/tools/test_mcp.py

@@ -11,6 +11,7 @@ import pytest
 
 from vibe.core.config import MCPHttp, MCPStdio, MCPStreamableHttp
 from vibe.core.tools.mcp import (
+    MCPRegistry,
     MCPToolResult,
     RemoteTool,
     _mcp_stderr_capture,
@@ -372,3 +373,181 @@ class TestMCPConfigModels:
 
         # Trailing special chars become underscores which are then stripped
         assert config.name == "my_server"
+
+
+class TestMCPRegistry:
+    def _make_http_server(
+        self, name: str, url: str = "http://localhost:8080"
+    ) -> MCPHttp:
+        return MCPHttp(name=name, transport="http", url=url)
+
+    def _make_stdio_server(self, name: str, command: str = "python -m srv") -> MCPStdio:
+        return MCPStdio(name=name, transport="stdio", command=command)
+
+    def test_server_key_is_stable(self):
+        srv = self._make_http_server("s1")
+        registry = MCPRegistry()
+
+        assert registry._server_key(srv) == registry._server_key(srv)
+
+    def test_different_configs_produce_different_keys(self):
+        registry = MCPRegistry()
+        s1 = self._make_http_server("s1", url="http://a:1")
+        s2 = self._make_http_server("s2", url="http://b:2")
+
+        assert registry._server_key(s1) != registry._server_key(s2)
+
+    def test_get_tools_caches_discovery(self):
+        registry = MCPRegistry()
+        srv = self._make_http_server("cached")
+        remote = RemoteTool(name="tool_a", description="A tool")
+        proxy = create_mcp_http_proxy_tool_class(
+            url="http://localhost:8080", remote=remote, alias="cached"
+        )
+
+        key = registry._server_key(srv)
+        registry._cache[key] = {proxy.get_name(): proxy}
+
+        tools = registry.get_tools([srv])
+        assert "cached_tool_a" in tools
+        assert tools["cached_tool_a"] is proxy
+
+    def test_get_tools_returns_empty_for_no_servers(self):
+        registry = MCPRegistry()
+
+        assert registry.get_tools([]) == {}
+
+    def test_clear_drops_cache(self):
+        registry = MCPRegistry()
+        srv = self._make_http_server("s")
+        proxy = create_mcp_http_proxy_tool_class(
+            url="http://localhost:8080", remote=RemoteTool(name="t"), alias="s"
+        )
+        key = registry._server_key(srv)
+        registry._cache[key] = {proxy.get_name(): proxy}
+
+        registry.clear()
+
+        assert len(registry._cache) == 0
+
+    def test_cache_survives_multiple_get_tools_calls(self):
+        registry = MCPRegistry()
+        srv = self._make_http_server("stable")
+        remote = RemoteTool(name="t1")
+        proxy = create_mcp_http_proxy_tool_class(
+            url="http://localhost:8080", remote=remote, alias="stable"
+        )
+
+        key = registry._server_key(srv)
+        registry._cache[key] = {proxy.get_name(): proxy}
+
+        first = registry.get_tools([srv])
+        second = registry.get_tools([srv])
+
+        assert first == second
+        assert first["stable_t1"] is second["stable_t1"]
+
+    def test_disjoint_server_lists_across_agents(self):
+        registry = MCPRegistry()
+
+        srv_x = self._make_http_server("x", url="http://x:1")
+        srv_y = self._make_http_server("y", url="http://y:2")
+
+        proxy_x = create_mcp_http_proxy_tool_class(
+            url="http://x:1", remote=RemoteTool(name="tx"), alias="x"
+        )
+        proxy_y = create_mcp_http_proxy_tool_class(
+            url="http://y:2", remote=RemoteTool(name="ty"), alias="y"
+        )
+
+        registry._cache[registry._server_key(srv_x)] = {proxy_x.get_name(): proxy_x}
+        registry._cache[registry._server_key(srv_y)] = {proxy_y.get_name(): proxy_y}
+
+        agent_a_tools = registry.get_tools([srv_x])
+        agent_b_tools = registry.get_tools([srv_y])
+
+        assert "x_tx" in agent_a_tools
+        assert "y_ty" not in agent_a_tools
+        assert "y_ty" in agent_b_tools
+        assert "x_tx" not in agent_b_tools
+
+    @pytest.mark.asyncio
+    async def test_discover_http_success(self):
+        registry = MCPRegistry()
+        srv = self._make_http_server("demo", url="http://demo:9090")
+        remote = RemoteTool(name="hello", description="Hi")
+
+        with patch(
+            "vibe.core.tools.mcp.registry.list_tools_http", return_value=[remote]
+        ):
+            tools = await registry._discover_http(srv)
+
+        assert tools is not None
+        assert len(tools) == 1
+        name = next(iter(tools))
+        assert name == "demo_hello"
+
+    @pytest.mark.asyncio
+    async def test_discover_http_failure_returns_none(self):
+        registry = MCPRegistry()
+        srv = self._make_http_server("fail", url="http://fail:1")
+
+        with patch(
+            "vibe.core.tools.mcp.registry.list_tools_http",
+            side_effect=ConnectionError("down"),
+        ):
+            tools = await registry._discover_http(srv)
+
+        assert tools is None
+
+    @pytest.mark.asyncio
+    async def test_discover_stdio_success(self):
+        registry = MCPRegistry()
+        srv = self._make_stdio_server("local", command="python -m local_srv")
+        remote = RemoteTool(name="run", description="Run it")
+
+        with patch(
+            "vibe.core.tools.mcp.registry.list_tools_stdio", return_value=[remote]
+        ):
+            tools = await registry._discover_stdio(srv)
+
+        assert tools is not None
+        assert len(tools) == 1
+        name = next(iter(tools))
+        assert name == "local_run"
+
+    @pytest.mark.asyncio
+    async def test_discover_stdio_failure_returns_none(self):
+        registry = MCPRegistry()
+        srv = self._make_stdio_server("broken")
+
+        with patch(
+            "vibe.core.tools.mcp.registry.list_tools_stdio",
+            side_effect=OSError("no binary"),
+        ):
+            tools = await registry._discover_stdio(srv)
+
+        assert tools is None
+
+    def test_get_tools_discovers_only_uncached(self):
+        registry = MCPRegistry()
+
+        cached_srv = self._make_http_server("cached", url="http://c:1")
+        new_srv = self._make_http_server("new", url="http://n:2")
+
+        cached_proxy = create_mcp_http_proxy_tool_class(
+            url="http://c:1", remote=RemoteTool(name="ct"), alias="cached"
+        )
+        registry._cache[registry._server_key(cached_srv)] = {
+            cached_proxy.get_name(): cached_proxy
+        }
+
+        new_remote = RemoteTool(name="nt")
+        with patch(
+            "vibe.core.tools.mcp.registry.list_tools_http", return_value=[new_remote]
+        ):
+            tools = registry.get_tools([cached_srv, new_srv])
+
+        assert "cached_ct" in tools
+        assert "new_nt" in tools
+        assert len(registry._cache) == 2

tests/tools/test_mcp_sampling.py

@@ -0,0 +1,171 @@
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+from mcp.types import (
+    CreateMessageRequestParams,
+    CreateMessageResult,
+    ErrorData,
+    SamplingMessage,
+    TextContent,
+)
+import pytest
+
+from tests.mock.utils import mock_llm_chunk
+from tests.stubs.fake_backend import FakeBackend
+from vibe.core.tools.mcp_sampling import (
+    MCPSamplingHandler,
+    _extract_text_content,
+    _map_sampling_messages,
+)
+from vibe.core.types import LLMMessage, Role
+
+
+def _make_config(model_name: str = "test-model") -> MagicMock:
+    config = MagicMock()
+    model = MagicMock()
+    model.name = model_name
+    model.temperature = 0.7
+    config.get_active_model.return_value = model
+    return config
+
+
+def _make_params(
+    messages: list[SamplingMessage] | None = None,
+    system_prompt: str | None = None,
+    temperature: float | None = None,
+    max_tokens: int = 100,
+) -> CreateMessageRequestParams:
+    if messages is None:
+        messages = [
+            SamplingMessage(role="user", content=TextContent(type="text", text="Hello"))
+        ]
+    return CreateMessageRequestParams(
+        messages=messages,
+        systemPrompt=system_prompt,
+        temperature=temperature,
+        maxTokens=max_tokens,
+    )
+
+
+class TestExtractTextContent:
+    def test_single_text_block(self) -> None:
+        block = TextContent(type="text", text="hello")
+        assert _extract_text_content(block) == "hello"
+
+    def test_list_of_text_blocks(self) -> None:
+        blocks = [
+            TextContent(type="text", text="a"),
+            TextContent(type="text", text="b"),
+        ]
+        assert _extract_text_content(blocks) == "a\nb"
+
+    def test_unsupported_single_block(self) -> None:
+        block = MagicMock(type="image", text=None)
+        assert _extract_text_content(block) == ""
+
+    def test_mixed_blocks_skips_non_text(self) -> None:
+        blocks = [TextContent(type="text", text="keep"), MagicMock(type="image")]
+        assert _extract_text_content(blocks) == "keep"
+
+
+class TestMapSamplingMessages:
+    def test_maps_user_message(self) -> None:
+        msgs = [
+            SamplingMessage(role="user", content=TextContent(type="text", text="hi"))
+        ]
+        result = _map_sampling_messages(msgs)
+        assert len(result) == 1
+        assert result[0].role == Role.user
+        assert result[0].content == "hi"
+
+    def test_maps_assistant_message(self) -> None:
+        msgs = [
+            SamplingMessage(
+                role="assistant", content=TextContent(type="text", text="hello")
+            )
+        ]
+        result = _map_sampling_messages(msgs)
+        assert result[0].role == Role.assistant
+        assert result[0].content == "hello"
+
+    def test_maps_multiple_messages(self) -> None:
+        msgs = [
+            SamplingMessage(role="user", content=TextContent(type="text", text="q")),
+            SamplingMessage(
+                role="assistant", content=TextContent(type="text", text="a")
+            ),
+        ]
+        result = _map_sampling_messages(msgs)
+        assert len(result) == 2
+        assert result[0].role == Role.user
+        assert result[1].role == Role.assistant
+
+
+class TestMCPSamplingHandler:
+    @pytest.mark.asyncio
+    async def test_basic_text_response(self) -> None:
+        chunk = mock_llm_chunk(content="LLM says hi")
+        backend = FakeBackend(chunk)
+        config = _make_config("my-model")
+        handler = MCPSamplingHandler(
+            backend_getter=lambda: backend, config_getter=lambda: config
+        )
+
+        result = await handler(MagicMock(), _make_params())
+
+        assert not isinstance(result, Exception)
+        assert isinstance(result, CreateMessageResult)
+        assert result.role == "assistant"
+        assert result.content.type == "text"
+        assert result.content.text == "LLM says hi"
+        assert result.model == "my-model"
+        assert result.stopReason == "endTurn"
+
+    @pytest.mark.asyncio
+    async def test_system_prompt_prepended(self) -> None:
+        chunk = mock_llm_chunk(content="ok")
+        backend = FakeBackend(chunk)
+        config = _make_config()
+        handler = MCPSamplingHandler(
+            backend_getter=lambda: backend, config_getter=lambda: config
+        )
+
+        await handler(MagicMock(), _make_params(system_prompt="Be helpful"))
+
+        sent_messages: list[LLMMessage] = backend.requests_messages[0]
+        assert sent_messages[0].role == Role.system
+        assert sent_messages[0].content == "Be helpful"
+        assert sent_messages[1].role == Role.user
+
+    @pytest.mark.asyncio
+    async def test_calls_backend_with_messages(self) -> None:
+        """Verify the handler forwards messages to the backend."""
+        chunk = mock_llm_chunk(content="ok")
+        backend = FakeBackend(chunk)
+        config = _make_config()
+        handler = MCPSamplingHandler(
+            backend_getter=lambda: backend, config_getter=lambda: config
+        )
+
+        await handler(MagicMock(), _make_params())
+
+        assert len(backend.requests_messages) == 1
+        sent = backend.requests_messages[0]
+        assert len(sent) == 1
+        assert sent[0].role == Role.user
+        assert sent[0].content == "Hello"
+
+    @pytest.mark.asyncio
+    async def test_returns_error_on_backend_failure(self) -> None:
+        backend = FakeBackend(exception_to_raise=RuntimeError("boom"))
+        config = _make_config()
+        handler = MCPSamplingHandler(
+            backend_getter=lambda: backend, config_getter=lambda: config
+        )
+
+        result = await handler(MagicMock(), _make_params())
+
+        assert isinstance(result, ErrorData)
+        assert result.code == -1
+        assert "boom" in result.message

tests/tools/test_task.py

@@ -8,7 +8,7 @@ from tests.conftest import build_test_vibe_config
 from tests.mock.utils import collect_result
 from vibe.core.agents.manager import AgentManager
 from vibe.core.agents.models import BUILTIN_AGENTS, AgentType
-from vibe.core.tools.base import BaseToolState, InvokeContext, ToolError
+from vibe.core.tools.base import BaseToolState, InvokeContext, ToolError, ToolPermission
 from vibe.core.tools.builtins.task import Task, TaskArgs, TaskResult, TaskToolConfig
 from vibe.core.types import AssistantEvent, LLMMessage, Role
 
@@ -76,6 +76,50 @@ class TestTaskToolValidation:
         assert agent.agent_type == AgentType.SUBAGENT
 
 
+class TestTaskToolResolvePermission:
+    def test_explore_allowed_by_default(self, task_tool: Task) -> None:
+        args = TaskArgs(task="do something", agent="explore")
+        result = task_tool.resolve_permission(args)
+        assert result == ToolPermission.ALWAYS
+
+    def test_unknown_agent_returns_none(self, task_tool: Task) -> None:
+        args = TaskArgs(task="do something", agent="custom_agent")
+        result = task_tool.resolve_permission(args)
+        assert result is None
+
+    def test_denylist_takes_precedence(self) -> None:
+        config = TaskToolConfig(allowlist=["explore"], denylist=["explore"])
+        tool = Task(config=config, state=BaseToolState())
+        args = TaskArgs(task="do something", agent="explore")
+        result = tool.resolve_permission(args)
+        assert result == ToolPermission.NEVER
+
+    def test_glob_pattern_in_allowlist(self) -> None:
+        config = TaskToolConfig(allowlist=["exp*"])
+        tool = Task(config=config, state=BaseToolState())
+        args = TaskArgs(task="do something", agent="explore")
+        result = tool.resolve_permission(args)
+        assert result == ToolPermission.ALWAYS
+
+    def test_glob_pattern_in_denylist(self) -> None:
+        config = TaskToolConfig(denylist=["danger*"])
+        tool = Task(config=config, state=BaseToolState())
+        args = TaskArgs(task="do something", agent="dangerous_agent")
+        result = tool.resolve_permission(args)
+        assert result == ToolPermission.NEVER
+
+    def test_empty_lists_returns_none(self) -> None:
+        config = TaskToolConfig(allowlist=[], denylist=[])
+        tool = Task(config=config, state=BaseToolState())
+        args = TaskArgs(task="do something", agent="explore")
+        result = tool.resolve_permission(args)
+        assert result is None
+
+    def test_default_config_has_explore_in_allowlist(self) -> None:
+        config = TaskToolConfig()
+        assert "explore" in config.allowlist
+
+
 class TestTaskToolExecution:
     @pytest.fixture
     def ctx(self) -> InvokeContext:

tests/tools/test_webfetch.py

@@ -0,0 +1,253 @@
+from __future__ import annotations
+
+import httpx
+import pytest
+import respx
+
+from tests.mock.utils import collect_result
+from vibe.core.tools.base import BaseToolState, ToolError
+from vibe.core.tools.builtins.webfetch import WebFetch, WebFetchArgs, WebFetchConfig
+
+
+@pytest.fixture
+def webfetch():
+    config = WebFetchConfig()
+    return WebFetch(config=config, state=BaseToolState())
+
+
+@pytest.fixture
+def webfetch_small():
+    config = WebFetchConfig(max_content_bytes=100)
+    return WebFetch(config=config, state=BaseToolState())
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_bare_domain_gets_https(webfetch):
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text="ok", headers={"Content-Type": "text/plain"}
+        )
+    )
+    result = await collect_result(webfetch.run(WebFetchArgs(url="example.com")))
+    assert result.url == "https://example.com"
+    assert result.content == "ok"
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_http_url_stays_http(webfetch):
+    respx.get("http://example.com").mock(
+        return_value=httpx.Response(
+            200, text="ok", headers={"Content-Type": "text/plain"}
+        )
+    )
+    result = await collect_result(webfetch.run(WebFetchArgs(url="http://example.com")))
+    assert result.url == "http://example.com"
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_https_url_stays_https(webfetch):
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text="ok", headers={"Content-Type": "text/plain"}
+        )
+    )
+    result = await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+    assert result.url == "https://example.com"
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_protocol_relative_url_normalized(webfetch):
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text="ok", headers={"Content-Type": "text/plain"}
+        )
+    )
+    result = await collect_result(webfetch.run(WebFetchArgs(url="//example.com")))
+    assert result.url == "https://example.com"
+    assert result.content == "ok"
+
+
+@pytest.mark.asyncio
+async def test_ftp_scheme_rejected(webfetch):
+    with pytest.raises(ToolError, match="Invalid URL scheme: ftp"):
+        await collect_result(webfetch.run(WebFetchArgs(url="ftp://example.com")))
+
+
+@pytest.mark.asyncio
+async def test_empty_url_rejected(webfetch):
+    with pytest.raises(ToolError, match="URL cannot be empty"):
+        await collect_result(webfetch.run(WebFetchArgs(url="   ")))
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_html_converted_to_markdown(webfetch):
+    html = "<html><body><h1>Title</h1><p>Hello world</p></body></html>"
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text=html, headers={"Content-Type": "text/html; charset=utf-8"}
+        )
+    )
+    result = await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+    assert "# Title" in result.content
+    assert "Hello world" in result.content
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_plain_text_unchanged(webfetch):
+    respx.get("https://example.com/file.txt").mock(
+        return_value=httpx.Response(
+            200, text="just text", headers={"Content-Type": "text/plain"}
+        )
+    )
+    result = await collect_result(
+        webfetch.run(WebFetchArgs(url="https://example.com/file.txt"))
+    )
+    assert result.content == "just text"
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_scripts_stripped_from_markdown(webfetch):
+    html = "<html><body><script>alert('xss')</script><style>.x{}</style><p>Clean</p></body></html>"
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text=html, headers={"Content-Type": "text/html"}
+        )
+    )
+    result = await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+    assert "alert" not in result.content
+    assert ".x{}" not in result.content
+    assert "Clean" in result.content
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_cloudflare_retry_on_challenge(webfetch):
+    route = respx.get("https://example.com")
+    route.side_effect = [
+        httpx.Response(403, headers={"cf-mitigated": "challenge"}),
+        httpx.Response(200, text="success", headers={"Content-Type": "text/plain"}),
+    ]
+    result = await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+    assert result.content == "success"
+    assert route.call_count == 2
+
+    second_request = route.calls[1].request
+    assert second_request.headers["User-Agent"] == "vibe-cli"
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_regular_403_not_retried(webfetch):
+    route = respx.get("https://example.com").mock(
+        return_value=httpx.Response(403, headers={"Content-Type": "text/plain"})
+    )
+    with pytest.raises(ToolError, match="HTTP error 403"):
+        await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+    assert route.call_count == 1
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_truncates_to_max_bytes_with_disclaimer(webfetch_small):
+    body = "a" * 200
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text=body, headers={"Content-Type": "text/plain"}
+        )
+    )
+    result = await collect_result(
+        webfetch_small.run(WebFetchArgs(url="https://example.com"))
+    )
+    assert result.content.startswith("a" * 100)
+    assert "[Content truncated due to size limit]" in result.content
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_truncates_html_with_disclaimer(webfetch_small):
+    html = (
+        "<html><body><h2>first title</h2>"
+        + "x" * 200
+        + "<h2>second title</h2></body></html>"
+    )
+    respx.get("https://example.com").mock(
+        return_value=httpx.Response(
+            200, text=html, headers={"Content-Type": "text/html"}
+        )
+    )
+    result = await collect_result(
+        webfetch_small.run(WebFetchArgs(url="https://example.com"))
+    )
+
+    assert "## first title" in result.content
+    assert "## second title" not in result.content
+    assert "[Content truncated due to size limit]" in result.content
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_http_404_raises_tool_error(webfetch):
+    respx.get("https://example.com").mock(return_value=httpx.Response(404))
+    with pytest.raises(ToolError, match="HTTP error 404"):
+        await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_http_500_raises_tool_error(webfetch):
+    respx.get("https://example.com").mock(return_value=httpx.Response(500))
+    with pytest.raises(ToolError, match="HTTP error 500"):
+        await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_timeout_raises_tool_error(webfetch):
+    respx.get("https://example.com").mock(side_effect=httpx.ReadTimeout("timed out"))
+    with pytest.raises(ToolError, match="Request timed out"):
+        await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+
+
+@pytest.mark.asyncio
+@respx.mock
+async def test_network_error_raises_tool_error(webfetch):
+    respx.get("https://example.com").mock(
+        side_effect=httpx.ConnectError("connection refused")
+    )
+    with pytest.raises(ToolError, match="Failed to fetch URL"):
+        await collect_result(webfetch.run(WebFetchArgs(url="https://example.com")))
+
+
+@pytest.mark.asyncio
+async def test_negative_timeout_rejected(webfetch):
+    with pytest.raises(ToolError, match="Timeout must be a positive number"):
+        await collect_result(
+            webfetch.run(WebFetchArgs(url="https://example.com", timeout=-1))
+        )
+
+
+@pytest.mark.asyncio
+async def test_zero_timeout_rejected(webfetch):
+    with pytest.raises(ToolError, match="Timeout must be a positive number"):
+        await collect_result(
+            webfetch.run(WebFetchArgs(url="https://example.com", timeout=0))
+        )
+
+
+@pytest.mark.asyncio
+async def test_over_max_timeout_rejected(webfetch):
+    with pytest.raises(ToolError, match="Timeout cannot exceed"):
+        await collect_result(
+            webfetch.run(WebFetchArgs(url="https://example.com", timeout=999))
+        )
+
+
+def test_get_status_text():
+    assert WebFetch.get_status_text() == "Fetching URL"

tests/tools/test_websearch.py

@@ -0,0 +1,165 @@
+from __future__ import annotations
+
+from unittest.mock import AsyncMock, patch
+
+import mistralai
+import pytest
+
+from tests.mock.utils import collect_result
+from vibe.core.tools.base import BaseToolState, ToolError
+from vibe.core.tools.builtins.websearch import WebSearch, WebSearchArgs, WebSearchConfig
+
+
+def _make_response(
+    content: list | None = None, outputs: list | None = None
+) -> mistralai.ConversationResponse:
+    if outputs is None:
+        outputs = [mistralai.MessageOutputEntry(content=content or [])]
+    return mistralai.ConversationResponse(
+        conversation_id="test",
+        outputs=outputs,
+        usage=mistralai.ConversationUsageInfo(
+            prompt_tokens=10, completion_tokens=20, total_tokens=30
+        ),
+    )
+
+
+@pytest.fixture
+def websearch(monkeypatch):
+    monkeypatch.setenv("MISTRAL_API_KEY", "test-key")
+    config = WebSearchConfig()
+    return WebSearch(config=config, state=BaseToolState())
+
+
+def test_parse_text_chunks(websearch):
+    response = _make_response(
+        content=[mistralai.TextChunk(text="Hello "), mistralai.TextChunk(text="world")]
+    )
+    result = websearch._parse_response(response)
+    assert result.answer == "Hello world"
+    assert result.sources == []
+
+
+def test_parse_sources_deduped(websearch):
+    response = _make_response(
+        content=[
+            mistralai.TextChunk(text="Answer"),
+            mistralai.ToolReferenceChunk(
+                tool="web_search", title="Site A", url="https://a.com"
+            ),
+            mistralai.ToolReferenceChunk(
+                tool="web_search", title="Site A duplicate", url="https://a.com"
+            ),
+            mistralai.ToolReferenceChunk(
+                tool="web_search", title="Site B", url="https://b.com"
+            ),
+        ]
+    )
+    result = websearch._parse_response(response)
+    assert result.answer == "Answer"
+    assert len(result.sources) == 2
+    assert result.sources[0].url == "https://a.com"
+    assert result.sources[0].title == "Site A"
+    assert result.sources[1].url == "https://b.com"
+
+
+def test_parse_skips_source_without_url(websearch):
+    response = _make_response(
+        content=[
+            mistralai.TextChunk(text="Answer"),
+            mistralai.ToolReferenceChunk(tool="web_search", title="No URL"),
+        ]
+    )
+    result = websearch._parse_response(response)
+    assert result.sources == []
+
+
+def test_parse_empty_text_raises(websearch):
+    response = _make_response(content=[])
+    with pytest.raises(ToolError, match="No text in agent response"):
+        websearch._parse_response(response)
+
+
+def test_parse_whitespace_only_raises(websearch):
+    response = _make_response(content=[mistralai.TextChunk(text="   ")])
+    with pytest.raises(ToolError, match="No text in agent response"):
+        websearch._parse_response(response)
+
+
+def test_parse_skips_non_message_entries(websearch):
+    response = _make_response(
+        outputs=[
+            mistralai.MessageOutputEntry(content=[mistralai.TextChunk(text="Answer")])
+        ]
+    )
+    result = websearch._parse_response(response)
+    assert result.answer == "Answer"
+
+
+@pytest.mark.asyncio
+async def test_run_missing_api_key(monkeypatch):
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+    config = WebSearchConfig()
+    ws = WebSearch(config=config, state=BaseToolState())
+    with pytest.raises(ToolError, match="MISTRAL_API_KEY"):
+        await collect_result(ws.run(WebSearchArgs(query="test")))
+
+
+@pytest.mark.asyncio
+async def test_run_returns_parsed_result(websearch):
+    response = _make_response(
+        content=[
+            mistralai.TextChunk(text="The answer"),
+            mistralai.ToolReferenceChunk(
+                tool="web_search", title="Source", url="https://example.com"
+            ),
+        ]
+    )
+
+    mock_start = AsyncMock(return_value=response)
+    with patch.object(mistralai.Mistral, "beta", create=True) as mock_beta:
+        mock_beta.conversations.start_async = mock_start
+        with patch.object(mistralai.Mistral, "__aenter__", return_value=None):
+            with patch.object(mistralai.Mistral, "__aexit__", return_value=None):
+                result = await collect_result(
+                    websearch.run(WebSearchArgs(query="test query"))
+                )
+
+    assert result.answer == "The answer"
+    assert len(result.sources) == 1
+    assert result.sources[0].url == "https://example.com"
+
+
+@pytest.mark.asyncio
+async def test_run_sdk_error_wrapped(websearch):
+    from unittest.mock import Mock
+
+    import httpx
+
+    mock_response = Mock(spec=httpx.Response)
+    mock_response.status_code = 500
+    mock_response.text = "error"
+    mock_response.headers = httpx.Headers({"content-type": "application/json"})
+
+    with patch.object(mistralai.Mistral, "beta", create=True) as mock_beta:
+        mock_beta.conversations.start_async = AsyncMock(
+            side_effect=mistralai.SDKError("API failed", mock_response)
+        )
+        with patch.object(mistralai.Mistral, "__aenter__", return_value=None):
+            with patch.object(mistralai.Mistral, "__aexit__", return_value=None):
+                with pytest.raises(ToolError, match="Mistral API error"):
+                    await collect_result(websearch.run(WebSearchArgs(query="test")))
+
+
+def test_is_available_with_key(monkeypatch):
+    monkeypatch.setenv("MISTRAL_API_KEY", "key")
+    assert WebSearch.is_available() is True
+
+
+def test_is_available_without_key(monkeypatch):
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+    assert WebSearch.is_available() is False
+
+
+def test_get_status_text():
+    assert WebSearch.get_status_text() == "Searching the web"