v2.18.4 (#866)

Co-authored-by: Albert Jiang <aj@mistral.ai>
Co-authored-by: Clément Drouin <clement.drouin@mistral.ai>
Co-authored-by: Laure Hugo <201583486+laure0303@users.noreply.github.com>
Co-authored-by: Mathias Gesbert <mathias.gesbert@mistral.ai>
Co-authored-by: Mert Unsal <mert.unsal@mistral.ai>
Co-authored-by: Michel Thomazo <51709227+michelTho@users.noreply.github.com>
Co-authored-by: Paul VEZIA <166131032+le-codeur-rapide@users.noreply.github.com>
Co-authored-by: Pierre Rossinès <pierre.rossines@mistral.ai>
Co-authored-by: Quentin <quentin.torroba@mistral.ai>
Co-authored-by: Mistral Vibe <vibe@mistral.ai>

tests/agent_loop/e2e/test_e2e_teleport.py

@@ -29,6 +29,7 @@ from vibe.core.teleport.types import (
 # agent manager re-derives config via model_dump(); the default url is unavoidable.
 SESSIONS_BASE_URL = "https://chat.mistral.ai"
 SESSIONS_URL = f"{SESSIONS_BASE_URL}{TELEPORT_SESSIONS_PATH}"
+GITHUB_REMOTE_URL = "https://github.com/owner/repo.git"
 
 
 def _sessions_ok() -> httpx.Response:
@@ -46,27 +47,28 @@ def _sessions_ok() -> httpx.Response:
 
 def _commit(repo: Repo, message: str) -> str:
     (Path(repo.working_dir) / "file.txt").write_text(f"{message}\n")
-    repo.index.add(["file.txt"])
-    repo.index.commit(message)
+    repo.git.add("file.txt")
+    repo.git.commit("-m", message)
     return repo.head.commit.hexsha
 
 
 def _init_repo(workdir: Path) -> Repo:
-    # origin is a local bare repo so fetch/push stay offline and instant; a
-    # separate github-url remote satisfies Teleport's GitHub-only detection.
+    # origin is intentionally not GitHub. The GitHub-looking `hub` remote is
+    # rewritten to a local bare repo so fetch/push stay offline and instant.
     bare = Repo.init(workdir.with_name(f"{workdir.name}_origin.git"), bare=True)
     repo = Repo.init(workdir, initial_branch="work")
     repo.config_writer().set_value("user", "name", "Tester").release()
     repo.config_writer().set_value("user", "email", "t@example.com").release()
     repo.create_remote("origin", str(bare.git_dir))
-    repo.create_remote("hub", "https://github.com/owner/repo.git")
+    repo.git.config(f"url.{bare.git_dir}.insteadOf", GITHUB_REMOTE_URL)
+    repo.create_remote("hub", GITHUB_REMOTE_URL)
     return repo
 
 
 def _repo_with_pushed_branch(workdir: Path) -> Repo:
     repo = _init_repo(workdir)
     _commit(repo, "initial")
-    repo.git.push("origin", "work")
+    repo.git.push("hub", "work")
     return repo
 
 
@@ -150,7 +152,7 @@ async def test_teleport_pushes_then_completes_when_approved(
         TeleportStartingWorkflowEvent,
         TeleportCompleteEvent,
     ]
-    assert repo.remote("origin").refs["work"].commit.hexsha == head
+    assert repo.remote("hub").refs["work"].commit.hexsha == head
 
 
 @pytest.mark.asyncio
@@ -172,7 +174,7 @@ async def test_teleport_fails_when_push_fails(
 ) -> None:
     repo = _repo_with_pushed_branch(tmp_working_directory)
     _commit(repo, "second")
-    repo.git.remote("set-url", "--push", "origin", "/nonexistent/repo.git")
+    repo.git.remote("set-url", "--push", "hub", "/nonexistent/repo.git")
 
     with pytest.raises(TeleportError, match="Failed to push"):
         await _drain(build_e2e_agent_loop(), "ship it", approve=True)

tests/agent_loop/test_agent_auto_compact.py

@@ -40,6 +40,18 @@ def _get_auto_compact_properties(
     return cast(dict[str, object], auto_compact[0]["properties"])
 
 
+def _get_compaction_failed_properties(
+    telemetry_events: list[dict[str, object]],
+) -> dict[str, object]:
+    failed = [
+        event
+        for event in telemetry_events
+        if event.get("event_name") == "vibe.compaction_failed"
+    ]
+    assert len(failed) == 1
+    return cast(dict[str, object], failed[0]["properties"])
+
+
 @pytest.mark.asyncio
 async def test_auto_compact_emits_correct_events(telemetry_events: list[dict]) -> None:
     backend = FakeBackend([
@@ -306,7 +318,9 @@ async def test_compact_without_extra_instructions_has_no_additional_section() ->
 
 
 @pytest.mark.asyncio
-async def test_compact_raises_on_tool_call_when_flag_enabled() -> None:
+async def test_compact_raises_on_tool_call_when_flag_enabled(
+    telemetry_events: list[dict],
+) -> None:
     """With the flag on, a compaction that returns a tool call raises."""
     backend = FakeBackend([
         [
@@ -333,10 +347,13 @@ async def test_compact_raises_on_tool_call_when_flag_enabled() -> None:
     with pytest.raises(CompactionFailedError) as exc_info:
         await agent.compact()
     assert exc_info.value.reason == "tool_call"
+    assert _get_compaction_failed_properties(telemetry_events)["reason"] == "tool_call"
 
 
 @pytest.mark.asyncio
-async def test_compact_raises_on_empty_summary_when_flag_enabled() -> None:
+async def test_compact_raises_on_empty_summary_when_flag_enabled(
+    telemetry_events: list[dict],
+) -> None:
     """With the flag on, a compaction with empty content raises."""
     backend = FakeBackend([[mock_llm_chunk(content="   ")]])
     cfg = build_test_vibe_config(
@@ -350,11 +367,20 @@ async def test_compact_raises_on_empty_summary_when_flag_enabled() -> None:
     with pytest.raises(CompactionFailedError) as exc_info:
         await agent.compact()
     assert exc_info.value.reason == "empty_summary"
+    assert (
+        _get_compaction_failed_properties(telemetry_events)["reason"] == "empty_summary"
+    )
 
 
 @pytest.mark.asyncio
-async def test_compact_falls_back_when_flag_disabled() -> None:
-    """With the flag off (default), empty content uses the legacy fallback."""
+async def test_compact_falls_back_when_flag_disabled(
+    telemetry_events: list[dict],
+) -> None:
+    """With the flag off (default), empty content uses the legacy fallback.
+
+    The compaction failure telemetry event must still be sent even though the
+    flag is off and compact() returns normally.
+    """
     backend = FakeBackend([[mock_llm_chunk(content="")]])
     cfg = build_test_vibe_config(models=make_test_models(auto_compact_threshold=999))
     agent = build_test_agent_loop(config=cfg, backend=backend)
@@ -363,6 +389,9 @@ async def test_compact_falls_back_when_flag_disabled() -> None:
 
     summary = await agent.compact()
     assert summary == "(no summary available)"
+    assert (
+        _get_compaction_failed_properties(telemetry_events)["reason"] == "empty_summary"
+    )
 
 
 @pytest.mark.asyncio

tests/agent_loop/test_agent_backend.py

@@ -17,9 +17,10 @@ from tests.conftest import (
 )
 from tests.mock.utils import mock_llm_chunk
 from tests.stubs.fake_backend import FakeBackend
+from vibe import __version__
 from vibe.core.agents.models import BuiltinAgentName
 from vibe.core.config import ModelConfig, ProviderConfig, VibeConfig
-from vibe.core.telemetry.types import EntrypointMetadata
+from vibe.core.telemetry.types import LaunchContext, TerminalEmulator
 from vibe.core.tools.base import BaseToolConfig, ToolPermission
 from vibe.core.types import (
     Backend,
@@ -32,6 +33,7 @@ from vibe.core.types import (
     StopInfo,
     ToolCall,
 )
+from vibe.core.utils import get_platform_id, get_platform_version
 
 
 def _two_model_vibe_config(active_model: str) -> VibeConfig:
@@ -191,19 +193,20 @@ async def test_passes_parent_session_id_to_backend_after_reset(vibe_config: Vibe
 
 
 @pytest.mark.asyncio
-async def test_passes_entrypoint_metadata_to_backend(vibe_config: VibeConfig):
-    metadata = EntrypointMetadata(
+async def test_passes_launch_context_to_backend(vibe_config: VibeConfig):
+    launch_context = LaunchContext(
         agent_entrypoint="acp",
         agent_version="2.0.0",
         client_name="vibe_ide",
         client_version="0.5.0",
+        terminal_emulator=TerminalEmulator.GHOSTTY,
     )
     backend = FakeBackend([mock_llm_chunk(content="Response")])
     agent = build_test_agent_loop(
         config=vibe_config,
         backend=backend,
         enable_streaming=True,
-        entrypoint_metadata=metadata,
+        launch_context=launch_context,
     )
 
     [_ async for _ in agent.act("Hello")]
@@ -215,6 +218,13 @@ async def test_passes_entrypoint_metadata_to_backend(vibe_config: VibeConfig):
     assert meta["agent_version"] == "2.0.0"
     assert meta["client_name"] == "vibe_ide"
     assert meta["client_version"] == "0.5.0"
+    assert meta["os"] == get_platform_id()
+    if os_version := get_platform_version():
+        assert meta["os_version"] == os_version
+    else:
+        assert "os_version" not in meta
+    assert meta["version"] == __version__
+    assert meta["terminal_emulator"] == TerminalEmulator.GHOSTTY
     assert meta["session_id"] == agent.session_id
     assert "message_id" in meta
     assert meta["call_type"] == "main_call"
@@ -272,11 +282,12 @@ async def test_mcp_sampling_handler_uses_updated_config_when_agent_config_change
 
 @pytest.mark.asyncio
 async def test_mcp_sampling_handler_sends_secondary_call_telemetry_metadata():
-    metadata = EntrypointMetadata(
+    launch_context = LaunchContext(
         agent_entrypoint="acp",
         agent_version="2.0.0",
         client_name="vibe_ide",
         client_version="0.5.0",
+        terminal_emulator=TerminalEmulator.GHOSTTY,
     )
     backend = FakeBackend([
         [mock_llm_chunk(content="Response")],
@@ -285,7 +296,7 @@ async def test_mcp_sampling_handler_sends_secondary_call_telemetry_metadata():
     agent = build_test_agent_loop(
         config=_two_model_vibe_config("devstral-latest"),
         backend=backend,
-        entrypoint_metadata=metadata,
+        launch_context=launch_context,
     )
 
     [_ async for _ in agent.act("Hello")]
@@ -302,6 +313,13 @@ async def test_mcp_sampling_handler_sends_secondary_call_telemetry_metadata():
     assert sampling_metadata["agent_version"] == "2.0.0"
     assert sampling_metadata["client_name"] == "vibe_ide"
     assert sampling_metadata["client_version"] == "0.5.0"
+    assert sampling_metadata["os"] == get_platform_id()
+    if os_version := get_platform_version():
+        assert sampling_metadata["os_version"] == os_version
+    else:
+        assert "os_version" not in sampling_metadata
+    assert sampling_metadata["version"] == __version__
+    assert sampling_metadata["terminal_emulator"] == TerminalEmulator.GHOSTTY
     assert sampling_metadata["session_id"] == agent.session_id
     assert sampling_metadata["parent_session_id"] == "parent-session-456"
     assert sampling_metadata["message_id"] == next(

tests/agent_loop/test_agents.py

@@ -344,6 +344,15 @@ class TestAgentProfileOverrides:
         overrides = BUILTIN_AGENTS[BuiltinAgentName.AUTO_APPROVE].overrides
         assert overrides.get("bypass_tool_permissions") is True
 
+    def test_lean_agent_keeps_tool_permissions_configurable(self) -> None:
+        overrides = BUILTIN_AGENTS[BuiltinAgentName.LEAN].overrides
+        assert "bypass_tool_permissions" not in overrides
+
+    def test_lean_agent_uses_latest_model(self) -> None:
+        overrides = BUILTIN_AGENTS[BuiltinAgentName.LEAN].overrides
+        models = overrides["models"]
+        assert models[0]["name"] == "labs-leanstral-1-5"
+
     def test_plan_agent_restricts_tools(self) -> None:
         overrides = BUILTIN_AGENTS[BuiltinAgentName.PLAN].overrides
         assert "tools" in overrides

tests/agent_loop/test_deferred_init.py

@@ -16,7 +16,7 @@ from tests.stubs.fake_mcp_registry import FakeMCPRegistry
 from vibe.core import agent_loop as agent_loop_module
 from vibe.core.agent_loop import AgentLoop
 from vibe.core.config import MCPStdio
-from vibe.core.telemetry.types import TerminalEmulator
+from vibe.core.telemetry.types import LaunchContext, TerminalEmulator
 from vibe.core.tools.manager import ToolManager
 from vibe.core.tools.mcp import AuthStatus
 from vibe.core.tools.remote import RemoteTool
@@ -371,7 +371,15 @@ class TestStartInitializeExperiments:
 
     @pytest.mark.asyncio
     async def test_refreshes_system_prompt_when_experiments_update(self) -> None:
-        loop = build_test_agent_loop(terminal_emulator=TerminalEmulator.VSCODE)
+        loop = build_test_agent_loop(
+            launch_context=LaunchContext(
+                agent_entrypoint="cli",
+                agent_version="1.0.0",
+                client_name="vibe_cli",
+                client_version="1.0.0",
+                terminal_emulator=TerminalEmulator.VSCODE,
+            )
+        )
         refresh_mock = AsyncMock()
         init_mock = AsyncMock(return_value=True)
 
@@ -387,21 +395,31 @@ class TestStartInitializeExperiments:
         init_mock.assert_awaited_once()
         init_args = init_mock.await_args
         assert init_args is not None
-        assert init_args.kwargs["terminal_emulator"] is TerminalEmulator.VSCODE
+        assert (
+            init_args.kwargs["launch_context"].terminal_emulator
+            is TerminalEmulator.VSCODE
+        )
 
     def test_new_session_telemetry_uses_provided_terminal_emulator(self) -> None:
-        loop = build_test_agent_loop(terminal_emulator=TerminalEmulator.VSCODE)
-        send_new_session = MagicMock()
+        loop = build_test_agent_loop(
+            launch_context=LaunchContext(
+                agent_entrypoint="cli",
+                agent_version="1.0.0",
+                client_name="vibe_cli",
+                client_version="1.0.0",
+                terminal_emulator=TerminalEmulator.VSCODE,
+            )
+        )
+        send_event = MagicMock()
 
         with patch.object(
-            loop.telemetry_client, "send_new_session", new=send_new_session
+            loop.telemetry_client, "send_telemetry_event", new=send_event
         ):
             loop.emit_new_session_telemetry()
 
-        assert (
-            send_new_session.call_args.kwargs["terminal_emulator"]
-            is TerminalEmulator.VSCODE
-        )
+        payload = send_event.call_args.args[1]
+        assert payload["terminal_emulator"] == "vscode"
+        assert type(payload["terminal_emulator"]) is str
 
     @pytest.mark.asyncio
     async def test_does_not_refresh_system_prompt_when_experiments_unchanged(