v2.7.4 (#579)

Co-authored-by: Clément Sirieix <clement.sirieix@mistral.ai>
Co-authored-by: Kim-Adeline Miguel <kimadeline.miguel@mistral.ai>
Co-authored-by: Lucas Marandat <31749711+lucasmrdt@users.noreply.github.com>
Co-authored-by: Michel Thomazo <51709227+michelTho@users.noreply.github.com>
Co-authored-by: Paul Cacheux <paul.cacheux@mistral.ai>
Co-authored-by: Peter Evers <pevers90@gmail.com>
Co-authored-by: Pierre Rossinès <pierre.rossines@mistral.ai>
Co-authored-by: Pierre Rossinès <pierre.rossines@protonmail.com>
Co-authored-by: Quentin <quentin.torroba@mistral.ai>
Co-authored-by: Simon Van de Kerckhove <simon.vandekerckhove@mistral.ai>
Co-authored-by: Val <102326092+vdeva@users.noreply.github.com>
Co-authored-by: Vincent G <10739306+VinceOPS@users.noreply.github.com>
Co-authored-by: Mistral Vibe <vibe@mistral.ai>

tests/tools/test_ask_user_question.py

@@ -18,7 +18,7 @@ from vibe.core.types import ToolCallEvent, ToolResultEvent
 @pytest.fixture
 def tool():
     config = AskUserQuestionConfig()
-    return AskUserQuestion(config=config, state=BaseToolState())
+    return AskUserQuestion(config_getter=lambda: config, state=BaseToolState())
 
 
 @pytest.fixture

tests/tools/test_bash.py

@@ -12,7 +12,7 @@ from vibe.core.tools.permissions import PermissionContext
 def bash(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = BashToolConfig()
-    return Bash(config=config, state=BaseToolState())
+    return Bash(config_getter=lambda: config, state=BaseToolState())
 
 
 @pytest.mark.asyncio
@@ -39,7 +39,7 @@ async def test_fails_cat_command_with_missing_file(bash):
 async def test_uses_effective_workdir(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = BashToolConfig()
-    bash_tool = Bash(config=config, state=BaseToolState())
+    bash_tool = Bash(config_getter=lambda: config, state=BaseToolState())
 
     result = await collect_result(bash_tool.run(BashArgs(command="pwd")))
 
@@ -57,7 +57,7 @@ async def test_handles_timeout(bash):
 @pytest.mark.asyncio
 async def test_truncates_output_to_max_bytes(bash):
     config = BashToolConfig(max_output_bytes=5)
-    bash_tool = Bash(config=config, state=BaseToolState())
+    bash_tool = Bash(config_getter=lambda: config, state=BaseToolState())
 
     result = await collect_result(
         bash_tool.run(BashArgs(command="printf 'abcdefghij'"))
@@ -78,7 +78,7 @@ async def test_decodes_non_utf8_bytes(bash):
 
 
 def test_find_not_in_default_allowlist():
-    bash_tool = Bash(config=BashToolConfig(), state=BaseToolState())
+    bash_tool = Bash(config_getter=lambda: BashToolConfig(), state=BaseToolState())
     # find -exec runs arbitrary commands; must not be allowlisted by default
     permission = bash_tool.resolve_permission(BashArgs(command="find . -exec id \\;"))
     assert (
@@ -89,7 +89,7 @@ def test_find_not_in_default_allowlist():
 
 def test_resolve_permission():
     config = BashToolConfig(allowlist=["echo", "pwd"], denylist=["rm"])
-    bash_tool = Bash(config=config, state=BaseToolState())
+    bash_tool = Bash(config_getter=lambda: config, state=BaseToolState())
 
     allowlisted = bash_tool.resolve_permission(BashArgs(command="echo hi"))
     denylisted = bash_tool.resolve_permission(BashArgs(command="rm -rf /tmp"))
@@ -111,7 +111,7 @@ class TestResolvePermissionWindowsSyntax:
 
     def _make_bash(self, **kwargs) -> Bash:
         config = BashToolConfig(**kwargs)
-        return Bash(config=config, state=BaseToolState())
+        return Bash(config_getter=lambda: config, state=BaseToolState())
 
     def test_dir_with_windows_flags_allowlisted(self):
         bash_tool = self._make_bash(allowlist=["dir"])
@@ -215,7 +215,7 @@ class TestDenylistWordBoundary:
 
     def _make_bash(self, **kwargs) -> Bash:
         config = BashToolConfig(**kwargs)
-        return Bash(config=config, state=BaseToolState())
+        return Bash(config_getter=lambda: config, state=BaseToolState())
 
     def test_vi_blocks_vi_exact(self):
         bash_tool = self._make_bash(denylist=["vi"])

tests/tools/test_exit_plan_mode.py

@@ -56,7 +56,9 @@ def _default_profile() -> AgentProfile:
 
 @pytest.fixture
 def tool() -> ExitPlanMode:
-    return ExitPlanMode(config=ExitPlanModeConfig(), state=BaseToolState())
+    return ExitPlanMode(
+        config_getter=lambda: ExitPlanModeConfig(), state=BaseToolState()
+    )
 
 
 @pytest.fixture

tests/tools/test_granular_permissions.py

@@ -46,7 +46,7 @@ class TestBashGranularPermissions:
 
     def _bash(self, **kwargs):
         config = BashToolConfig(**kwargs)
-        return Bash(config=config, state=BaseToolState())
+        return Bash(config_getter=lambda: config, state=BaseToolState())
 
     def test_allowlisted_command_always(self):
         bash = self._bash()
@@ -264,7 +264,7 @@ class TestReadFileGranularPermissions:
 
     def _read_file(self, **kwargs):
         config = ReadFileToolConfig(**kwargs)
-        return ReadFile(config=config, state=ReadFileState())
+        return ReadFile(config_getter=lambda: config, state=ReadFileState())
 
     def test_in_workdir_normal_file_returns_none(self):
         (self.workdir / "test.py").touch()
@@ -346,7 +346,7 @@ class TestWriteFileGranularPermissions:
 
     def _write_file(self):
         config = WriteFileConfig()
-        return WriteFile(config=config, state=BaseToolState())
+        return WriteFile(config_getter=lambda: config, state=BaseToolState())
 
     def test_in_workdir_returns_none(self):
         tool = self._write_file()
@@ -379,7 +379,7 @@ class TestSearchReplaceGranularPermissions:
 
     def test_outside_workdir_returns_permission_context(self):
         config = SearchReplaceConfig()
-        tool = SearchReplace(config=config, state=BaseToolState())
+        tool = SearchReplace(config_getter=lambda: config, state=BaseToolState())
         result = tool.resolve_permission(
             SearchReplaceArgs(file_path="/tmp/file.py", content="x")
         )
@@ -395,7 +395,7 @@ class TestGrepGranularPermissions:
 
     def _grep(self):
         config = GrepToolConfig()
-        return Grep(config=config, state=BaseToolState())
+        return Grep(config_getter=lambda: config, state=BaseToolState())
 
     def test_in_workdir_normal_path_returns_none(self):
         tool = self._grep()
@@ -442,7 +442,7 @@ class TestApprovalFlowSimulation:
                 session_pattern="mkdir *",
             )
         ]
-        bash = Bash(config=BashToolConfig(), state=BaseToolState())
+        bash = Bash(config_getter=lambda: BashToolConfig(), state=BaseToolState())
         result = bash.resolve_permission(BashArgs(command="mkdir another_dir"))
         assert isinstance(result, PermissionContext)
         uncovered = [
@@ -460,7 +460,7 @@ class TestApprovalFlowSimulation:
                 session_pattern="mkdir *",
             )
         ]
-        bash = Bash(config=BashToolConfig(), state=BaseToolState())
+        bash = Bash(config_getter=lambda: BashToolConfig(), state=BaseToolState())
         result = bash.resolve_permission(BashArgs(command="npm install"))
         assert isinstance(result, PermissionContext)
         uncovered = [
@@ -472,7 +472,7 @@ class TestApprovalFlowSimulation:
         assert uncovered[0].session_pattern == "npm install *"
 
     def test_outside_dir_approved_covers_subsequent(self):
-        bash = Bash(config=BashToolConfig(), state=BaseToolState())
+        bash = Bash(config_getter=lambda: BashToolConfig(), state=BaseToolState())
         result = bash.resolve_permission(BashArgs(command="mkdir /tmp/newdir"))
         assert isinstance(result, PermissionContext)
         outside_rps = [
@@ -509,7 +509,7 @@ class TestApprovalFlowSimulation:
                 session_pattern="rm *",
             )
         ]
-        bash = Bash(config=BashToolConfig(), state=BaseToolState())
+        bash = Bash(config_getter=lambda: BashToolConfig(), state=BaseToolState())
         result = bash.resolve_permission(BashArgs(command="rm -rf /tmp/something"))
         assert isinstance(result, PermissionContext)
         cmd_perms = [
@@ -529,7 +529,7 @@ class TestApprovalFlowSimulation:
                 session_pattern="sudo apt install foo",
             )
         ]
-        bash = Bash(config=BashToolConfig(), state=BaseToolState())
+        bash = Bash(config_getter=lambda: BashToolConfig(), state=BaseToolState())
         result = bash.resolve_permission(BashArgs(command="sudo apt install bar"))
         assert isinstance(result, PermissionContext)
         cmd_perms = [
@@ -575,7 +575,7 @@ class TestApprovalFlowSimulation:
 
 class TestWebFetchPermissions:
     def _make_webfetch(self) -> WebFetch:
-        return WebFetch(config=WebFetchConfig(), state=BaseToolState())
+        return WebFetch(config_getter=lambda: WebFetchConfig(), state=BaseToolState())
 
     def test_returns_url_pattern_with_domain(self):
         wf = self._make_webfetch()
@@ -671,7 +671,7 @@ class TestWebFetchPermissions:
 
     def test_config_permission_always_honored(self):
         wf = WebFetch(
-            config=WebFetchConfig(permission=ToolPermission.ALWAYS),
+            config_getter=lambda: WebFetchConfig(permission=ToolPermission.ALWAYS),
             state=BaseToolState(),
         )
         result = wf.resolve_permission(WebFetchArgs(url="https://example.com"))
@@ -680,7 +680,7 @@ class TestWebFetchPermissions:
 
     def test_config_permission_never_honored(self):
         wf = WebFetch(
-            config=WebFetchConfig(permission=ToolPermission.NEVER),
+            config_getter=lambda: WebFetchConfig(permission=ToolPermission.NEVER),
             state=BaseToolState(),
         )
         result = wf.resolve_permission(WebFetchArgs(url="https://example.com"))
@@ -689,7 +689,8 @@ class TestWebFetchPermissions:
 
     def test_config_permission_ask_falls_through_to_domain(self):
         wf = WebFetch(
-            config=WebFetchConfig(permission=ToolPermission.ASK), state=BaseToolState()
+            config_getter=lambda: WebFetchConfig(permission=ToolPermission.ASK),
+            state=BaseToolState(),
         )
         result = wf.resolve_permission(WebFetchArgs(url="https://example.com"))
         assert isinstance(result, PermissionContext)

tests/tools/test_grep.py

@@ -13,7 +13,7 @@ from vibe.core.tools.builtins.grep import Grep, GrepArgs, GrepBackend, GrepToolC
 def grep(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = GrepToolConfig()
-    return Grep(config=config, state=BaseToolState())
+    return Grep(config_getter=lambda: config, state=BaseToolState())
 
 
 @pytest.fixture
@@ -28,7 +28,7 @@ def grep_gnu_only(tmp_path, monkeypatch):
 
     monkeypatch.setattr("shutil.which", mock_which)
     config = GrepToolConfig()
-    return Grep(config=config, state=BaseToolState())
+    return Grep(config_getter=lambda: config, state=BaseToolState())
 
 
 def test_detects_ripgrep_when_available(grep):
@@ -137,7 +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=BaseToolState())
+    grep_tool = Grep(config_getter=lambda: 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")))
@@ -189,7 +189,7 @@ async def test_ignores_comments_in_vibeignore(grep, tmp_path):
 async def test_uses_effective_workdir(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
     config = GrepToolConfig()
-    grep_tool = Grep(config=config, state=BaseToolState())
+    grep_tool = Grep(config_getter=lambda: 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_invoke_context.py

@@ -36,7 +36,7 @@ class SimpleTool(BaseTool[SimpleArgs, SimpleResult, BaseToolConfig, BaseToolStat
 
 @pytest.fixture
 def simple_tool() -> SimpleTool:
-    return SimpleTool(config=BaseToolConfig(), state=BaseToolState())
+    return SimpleTool(config_getter=lambda: BaseToolConfig(), state=BaseToolState())
 
 
 class TestInvokeContext:

tests/tools/test_skill.py

@@ -54,7 +54,7 @@ def _make_ctx(skill_manager: SkillManager | None = None) -> InvokeContext:
 
 @pytest.fixture
 def skill_tool() -> Skill:
-    return Skill(config=SkillToolConfig(), state=BaseToolState())
+    return Skill(config_getter=lambda: SkillToolConfig(), state=BaseToolState())
 
 
 class TestSkillRun:

tests/tools/test_task.py

@@ -16,7 +16,7 @@ from vibe.core.types import AssistantEvent, LLMMessage, Role
 
 @pytest.fixture
 def task_tool() -> Task:
-    return Task(config=TaskToolConfig(), state=BaseToolState())
+    return Task(config_getter=lambda: TaskToolConfig(), state=BaseToolState())
 
 
 class TestTaskArgs:
@@ -91,7 +91,7 @@ class TestTaskToolResolvePermission:
 
     def test_denylist_takes_precedence(self) -> None:
         config = TaskToolConfig(allowlist=["explore"], denylist=["explore"])
-        tool = Task(config=config, state=BaseToolState())
+        tool = Task(config_getter=lambda: config, state=BaseToolState())
         args = TaskArgs(task="do something", agent="explore")
         result = tool.resolve_permission(args)
         assert isinstance(result, PermissionContext)
@@ -99,7 +99,7 @@ class TestTaskToolResolvePermission:
 
     def test_glob_pattern_in_allowlist(self) -> None:
         config = TaskToolConfig(allowlist=["exp*"])
-        tool = Task(config=config, state=BaseToolState())
+        tool = Task(config_getter=lambda: config, state=BaseToolState())
         args = TaskArgs(task="do something", agent="explore")
         result = tool.resolve_permission(args)
         assert isinstance(result, PermissionContext)
@@ -107,7 +107,7 @@ class TestTaskToolResolvePermission:
 
     def test_glob_pattern_in_denylist(self) -> None:
         config = TaskToolConfig(denylist=["danger*"])
-        tool = Task(config=config, state=BaseToolState())
+        tool = Task(config_getter=lambda: config, state=BaseToolState())
         args = TaskArgs(task="do something", agent="dangerous_agent")
         result = tool.resolve_permission(args)
         assert isinstance(result, PermissionContext)
@@ -115,7 +115,7 @@ class TestTaskToolResolvePermission:
 
     def test_empty_lists_returns_none(self) -> None:
         config = TaskToolConfig(allowlist=[], denylist=[])
-        tool = Task(config=config, state=BaseToolState())
+        tool = Task(config_getter=lambda: config, state=BaseToolState())
         args = TaskArgs(task="do something", agent="explore")
         result = tool.resolve_permission(args)
         assert result is None

tests/tools/test_ui_bash_execution.py

@@ -5,9 +5,13 @@ import time
 import pytest
 from textual.widgets import Static
 
+from tests.conftest import build_test_agent_loop, build_test_vibe_app
+from tests.mock.utils import mock_llm_chunk
+from tests.stubs.fake_backend import FakeBackend
 from vibe.cli.textual_ui.app import VibeApp
 from vibe.cli.textual_ui.widgets.chat_input.container import ChatInputContainer
 from vibe.cli.textual_ui.widgets.messages import BashOutputMessage, ErrorMessage
+from vibe.core.types import Role
 
 
 async def _wait_for_bash_output_message(
@@ -118,3 +122,38 @@ async def test_ui_handles_non_utf8_stderr(vibe_app: VibeApp) -> None:
         output_widget = message.query_one(".bash-output", Static)
         assert str(output_widget.render()) == "<EFBFBD><EFBFBD>"
         assert_no_command_error(vibe_app)
+
+
+@pytest.mark.asyncio
+async def test_ui_sends_manual_command_output_to_next_agent_turn() -> None:
+    backend = FakeBackend(mock_llm_chunk(content="I saw it."))
+    vibe_app = build_test_vibe_app(agent_loop=build_test_agent_loop(backend=backend))
+
+    async with vibe_app.run_test() as pilot:
+        chat_input = vibe_app.query_one(ChatInputContainer)
+        chat_input.value = "!echo hello"
+
+        await pilot.press("enter")
+        await _wait_for_bash_output_message(vibe_app, pilot)
+
+        injected_message = vibe_app.agent_loop.messages[-1]
+        assert injected_message.role == Role.user
+        assert injected_message.injected is True
+        assert injected_message.content is not None
+        assert "Manual `!` command result from the user." in injected_message.content
+        assert "Command: `echo hello`" in injected_message.content
+        assert "Exit code: 0" in injected_message.content
+        assert "Stdout:\n```text\nhello\n```" in injected_message.content
+
+        chat_input.value = "what did the command print?"
+        await pilot.press("enter")
+        await pilot.app.workers.wait_for_complete()
+
+        assert len(backend.requests_messages) == 1
+        user_messages = [
+            msg for msg in backend.requests_messages[0] if msg.role == Role.user
+        ]
+        assert len(user_messages) >= 2
+        assert user_messages[-2].content == injected_message.content
+        assert user_messages[-2].injected is True
+        assert user_messages[-1].content == "what did the command print?"

tests/tools/test_webfetch.py

@@ -12,13 +12,13 @@ from vibe.core.tools.builtins.webfetch import WebFetch, WebFetchArgs, WebFetchCo
 @pytest.fixture
 def webfetch():
     config = WebFetchConfig()
-    return WebFetch(config=config, state=BaseToolState())
+    return WebFetch(config_getter=lambda: config, state=BaseToolState())
 
 
 @pytest.fixture
 def webfetch_small():
     config = WebFetchConfig(max_content_bytes=100)
-    return WebFetch(config=config, state=BaseToolState())
+    return WebFetch(config_getter=lambda: config, state=BaseToolState())
 
 
 @pytest.mark.asyncio
@@ -32,6 +32,7 @@ async def test_bare_domain_gets_https(webfetch):
     result = await collect_result(webfetch.run(WebFetchArgs(url="example.com")))
     assert result.url == "https://example.com"
     assert result.content == "ok"
+    assert result.was_truncated is False
 
 
 @pytest.mark.asyncio
@@ -167,6 +168,7 @@ async def test_truncates_to_max_bytes_with_disclaimer(webfetch_small):
     )
     assert result.content.startswith("a" * 100)
     assert "[Content truncated due to size limit]" in result.content
+    assert result.was_truncated is True
 
 
 @pytest.mark.asyncio
@@ -189,6 +191,7 @@ async def test_truncates_html_with_disclaimer(webfetch_small):
     assert "## first title" in result.content
     assert "## second title" not in result.content
     assert "[Content truncated due to size limit]" in result.content
+    assert result.was_truncated is True
 
 
 @pytest.mark.asyncio

tests/tools/test_websearch.py

@@ -38,7 +38,7 @@ def _make_response(
 def websearch(monkeypatch):
     monkeypatch.setenv("MISTRAL_API_KEY", "test-key")
     config = WebSearchConfig()
-    return WebSearch(config=config, state=BaseToolState())
+    return WebSearch(config_getter=lambda: config, state=BaseToolState())
 
 
 def test_parse_text_chunks(websearch):
@@ -104,7 +104,7 @@ def test_parse_skips_non_message_entries(websearch):
 async def test_run_missing_api_key(monkeypatch):
     monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
     config = WebSearchConfig()
-    ws = WebSearch(config=config, state=BaseToolState())
+    ws = WebSearch(config_getter=lambda: config, state=BaseToolState())
     with pytest.raises(ToolError, match="MISTRAL_API_KEY"):
         await collect_result(ws.run(WebSearchArgs(query="test")))