feat(config): watch runtime configuration changes (#5026)

tests/agent/test_max_messages_config.py

@@ -98,6 +98,7 @@ class TestMaxMessagesInit:
 
         initial = loop.runtime_resolver.runtime
         assert replay_max_messages_for_context(initial.context_window_tokens) == 327
+        loop.runtime_resolver.invalidate()
         refreshed = loop.llm_runtime()
         assert replay_max_messages_for_context(refreshed.context_window_tokens) == FILE_MAX_MESSAGES
 

tests/agent/test_model_runtime_resolver.py

@@ -177,12 +177,59 @@ def test_resolver_refreshes_provider_generation_for_next_default_turn() -> None:
     admitted = resolver.current()
 
     provider.generation = GenerationSettings(temperature=0.8, max_tokens=512)
-    refreshed = resolver.current(refresh=True)
+    refreshed = resolver.admit()
 
     assert admitted.generation == GenerationSettings(0.2, 2048, None)
     assert refreshed.generation == GenerationSettings(0.8, 512, None)
 
 
+def test_resolver_admission_reloads_config_only_after_invalidation() -> None:
+    initial = _runtime()
+    refreshed_provider = _provider()
+    load_count = 0
+
+    def load_snapshot() -> ProviderSnapshot:
+        nonlocal load_count
+        load_count += 1
+        return ProviderSnapshot(
+            provider=refreshed_provider,
+            model="refreshed-model",
+            context_window_tokens=20_000,
+            signature=("refreshed-model", "auto"),
+        )
+
+    resolver = ModelRuntimeResolver(initial, provider_snapshot_loader=load_snapshot)
+
+    assert resolver.admit() is initial
+    assert load_count == 0
+
+    resolver.invalidate()
+    refreshed = resolver.admit()
+
+    assert refreshed.provider is refreshed_provider
+    assert refreshed.model == "refreshed-model"
+    assert resolver.admit() is refreshed
+    assert load_count == 1
+
+
+def test_current_refresh_forces_config_reload() -> None:
+    initial = _runtime()
+    load_snapshot = MagicMock(
+        return_value=ProviderSnapshot(
+            provider=_provider(),
+            model="refreshed-model",
+            context_window_tokens=20_000,
+            signature=("refreshed-model", "auto"),
+        )
+    )
+    resolver = ModelRuntimeResolver(initial, provider_snapshot_loader=load_snapshot)
+
+    refreshed = resolver.current(refresh=True)
+
+    assert refreshed.model == "refreshed-model"
+    load_snapshot.assert_called_once_with()
+
+
 def test_selected_preset_generation_does_not_fall_back_to_provider_defaults() -> None:
     provider = _provider(temperature=0.1, max_tokens=1024)
     resolver = ModelRuntimeResolver(
@@ -198,7 +245,7 @@ def test_selected_preset_generation_does_not_fall_back_to_provider_defaults() ->
     selected = resolver.select_preset("creative")
 
     provider.generation = GenerationSettings(temperature=0.9, max_tokens=64)
-    refreshed = resolver.current(refresh=True)
+    refreshed = resolver.admit()
 
     assert refreshed is selected
     assert refreshed.generation == GenerationSettings(0.7, 4096, None)

tests/agent/test_runtime_refresh.py

@@ -2,6 +2,8 @@ from pathlib import Path
 from types import SimpleNamespace
 from unittest.mock import MagicMock
 
+import pytest
+
 from nanobot.agent.loop import AgentLoop
 from nanobot.bus.queue import MessageBus
 from nanobot.config.loader import save_config
@@ -34,6 +36,7 @@ def test_provider_refresh_updates_only_runtime_resolver(tmp_path: Path) -> None:
             signature=("new-model",),
         ),
     )
+    loop.runtime_resolver.invalidate()
 
     runtime = loop.llm_runtime()
 
@@ -90,6 +93,7 @@ def test_llm_runtime_refreshes_provider_snapshot(tmp_path: Path) -> None:
             signature=("new-model",),
         ),
     )
+    loop.runtime_resolver.invalidate()
 
     runtime = loop.llm_runtime()
 
@@ -99,6 +103,24 @@ def test_llm_runtime_refreshes_provider_snapshot(tmp_path: Path) -> None:
     assert not hasattr(loop.runner, "provider")
 
 
+def test_llm_runtime_surfaces_invalidated_config_errors(tmp_path: Path) -> None:
+    def fail_refresh() -> ProviderSnapshot:
+        raise ValueError("invalid config")
+
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=_provider("old-model"),
+        workspace=tmp_path,
+        model="old-model",
+        context_window_tokens=1000,
+        provider_snapshot_loader=fail_refresh,
+    )
+    loop.runtime_resolver.invalidate()
+
+    with pytest.raises(ValueError, match="invalid config"):
+        loop.llm_runtime()
+
+
 def test_same_snapshot_default_clears_preset_and_publishes_update(tmp_path: Path) -> None:
     base_provider = _provider("base-model")
     fast_provider = _provider("fast-model")
@@ -122,6 +144,7 @@ def test_same_snapshot_default_clears_preset_and_publishes_update(tmp_path: Path
         preset_snapshot_loader=lambda _name: fast_snapshot,
         runtime_model_publisher=lambda model, preset: published.append((model, preset)),
     )
+    loop.runtime_resolver.invalidate()
 
     runtime = loop.llm_runtime()
 
@@ -173,6 +196,7 @@ def test_settings_context_window_refreshes_runtime_state(
     loop = AgentLoop.from_config(config, provider_snapshot_loader=loader)
 
     payload = update_agent_settings({"context_window_tokens": ["262144"]})
+    loop.runtime_resolver.invalidate()
     loop.llm_runtime()
 
     assert payload["requires_restart"] is False

tests/agent/test_self_model_preset.py

@@ -175,6 +175,7 @@ def test_active_model_preset_survives_unchanged_config_refresh(tmp_path) -> None
     )
 
     loop.set_model_preset("fast")
+    loop.runtime_resolver.invalidate()
     loop.llm_runtime()
 
     assert loop.model_preset == "fast"
@@ -211,6 +212,7 @@ def test_config_model_refresh_clears_active_model_preset(tmp_path) -> None:
     )
 
     loop.set_model_preset("fast")
+    loop.runtime_resolver.invalidate()
     loop.llm_runtime()
 
     assert loop.model_preset is None

tests/cli/test_commands.py

@@ -2737,12 +2737,14 @@ def test_gateway_local_trigger_queue_submits_agent_turns(
             self.context = _FakeContext()
             self.sessions = kwargs["session_manager"]
             self.submit_local_trigger_turn = AsyncMock()
+            self.runtime_resolver = MagicMock()
             seen["agent"] = self
 
         def _schedule_background(self, _coro) -> None:
             return None
 
         async def run(self) -> None:
+            self.runtime_resolver.invalidate.assert_called_once_with()
             await asyncio.Event().wait()
 
         async def close_mcp(self) -> None:
@@ -2971,6 +2973,7 @@ def test_gateway_health_endpoint_binds_and_serves_expected_responses(
             self.model = "test-model"
             self.provider = object()
             self.sessions = _FakeSessionManager()
+            self.runtime_resolver = MagicMock()
 
         def llm_runtime(self) -> None:
             return None
@@ -3164,6 +3167,7 @@ def test_gateway_shutdown_lets_agent_task_own_mcp_cleanup(
             self.model = "test-model"
             self.provider = object()
             self.sessions = _FakeSessionManager()
+            self.runtime_resolver = MagicMock()
 
         def llm_runtime(self) -> None:
             return None
@@ -3262,6 +3266,7 @@ def test_gateway_shutdown_event_exits_forever_runtime_tasks(
             self.model = "test-model"
             self.provider = object()
             self.sessions = _FakeSessionManager()
+            self.runtime_resolver = MagicMock()
 
         def llm_runtime(self) -> None:
             return None

tests/config/test_watcher.py

@@ -0,0 +1,60 @@
+import asyncio
+from contextlib import suppress
+from pathlib import Path
+
+import pytest
+from watchfiles import Change
+
+import nanobot.config.watcher as config_watcher
+
+
+@pytest.mark.asyncio
+async def test_watch_config_file_filters_directory_events(
+    tmp_path: Path,
+    monkeypatch,
+) -> None:
+    config_path = tmp_path / "config.json"
+    other_path = tmp_path / "other.json"
+    seen: dict[str, object] = {}
+
+    async def fake_awatch(*paths, **kwargs):
+        seen["paths"] = paths
+        seen["recursive"] = kwargs["recursive"]
+        watch_filter = kwargs["watch_filter"]
+        assert watch_filter(Change.modified, str(config_path)) is True
+        assert watch_filter(Change.modified, str(other_path)) is False
+        yield {(Change.modified, str(config_path))}
+
+    monkeypatch.setattr(config_watcher, "awatch", fake_awatch)
+    changes: list[None] = []
+
+    await config_watcher.watch_config_file(config_path, lambda: changes.append(None))
+
+    assert seen == {"paths": (tmp_path,), "recursive": False}
+    assert changes == [None]
+
+
+@pytest.mark.asyncio
+async def test_watch_config_file_observes_atomic_replace(tmp_path: Path) -> None:
+    config_path = tmp_path / "config.json"
+    config_path.write_text("{}", encoding="utf-8")
+    changed = asyncio.Event()
+    task = asyncio.create_task(
+        config_watcher.watch_config_file(config_path, changed.set)
+    )
+
+    try:
+        for attempt in range(10):
+            replacement = tmp_path / "config.tmp"
+            replacement.write_text(f'{{"attempt": {attempt}}}', encoding="utf-8")
+            replacement.replace(config_path)
+            try:
+                await asyncio.wait_for(changed.wait(), timeout=0.2)
+                break
+            except TimeoutError:
+                continue
+        assert changed.is_set()
+    finally:
+        task.cancel()
+        with suppress(asyncio.CancelledError):
+            await task