fix(agent): release completed task groups

nanobot/agent/loop.py

@@ -852,6 +852,22 @@ class AgentLoop:
             metadata={**metadata, "render_as": "text"},
         )
 
+    def _track_active_task(self, key: str, task: asyncio.Task[Any]) -> None:
+        """Track active session work until its task group becomes empty."""
+        tasks = self._active_tasks.setdefault(key, set())
+        tasks.add(task)
+        task.add_done_callback(partial(self._active_task_done, key, tasks))
+
+    def _active_task_done(
+        self,
+        key: str,
+        tasks: set[asyncio.Task[Any]],
+        task: asyncio.Task[Any],
+    ) -> None:
+        tasks.discard(task)
+        if not tasks and self._active_tasks.get(key) is tasks:
+            self._active_tasks.pop(key, None)
+
     async def _cancel_active_tasks(self, key: str) -> int:
         """Cancel and await all active work for *key*.
 
@@ -1350,12 +1366,7 @@ class AgentLoop:
                 # Compute the effective session key before dispatching
                 # This ensures /stop command can find tasks correctly when unified session is enabled
                 task = asyncio.create_task(self._dispatch(msg))
-                active_tasks: set[asyncio.Task[Any]] = self._active_tasks.setdefault(
-                    effective_key,
-                    set(),
-                )
-                active_tasks.add(task)
-                task.add_done_callback(active_tasks.discard)
+                self._track_active_task(effective_key, task)
         finally:
             await self.aclose()
 

tests/agent/test_task_cancel.py

@@ -42,6 +42,65 @@ def _make_loop(*, tools_config=None):
     return loop, bus
 
 
+class TestActiveTaskTracking:
+    @pytest.mark.asyncio
+    async def test_completed_task_removes_empty_session_group(self):
+        loop, _bus = _make_loop()
+        release = asyncio.Event()
+        task = asyncio.create_task(release.wait())
+
+        loop._track_active_task("test:c1", task)
+        release.set()
+        await task
+        await asyncio.sleep(0)
+
+        assert "test:c1" not in loop._active_tasks
+
+    @pytest.mark.asyncio
+    async def test_session_group_remains_until_last_task_completes(self):
+        loop, _bus = _make_loop()
+        releases = [asyncio.Event(), asyncio.Event()]
+        tasks = [asyncio.create_task(release.wait()) for release in releases]
+        for task in tasks:
+            loop._track_active_task("test:c1", task)
+
+        releases[0].set()
+        await tasks[0]
+        await asyncio.sleep(0)
+
+        assert loop._active_tasks["test:c1"] == {tasks[1]}
+
+        releases[1].set()
+        await tasks[1]
+        await asyncio.sleep(0)
+
+        assert "test:c1" not in loop._active_tasks
+
+    @pytest.mark.asyncio
+    async def test_old_callback_preserves_replacement_session_group(self):
+        loop, _bus = _make_loop()
+        old_release = asyncio.Event()
+        new_release = asyncio.Event()
+        old_task = asyncio.create_task(old_release.wait())
+        new_task = asyncio.create_task(new_release.wait())
+
+        loop._track_active_task("test:c1", old_task)
+        loop._active_tasks.pop("test:c1")
+        loop._track_active_task("test:c1", new_task)
+
+        old_release.set()
+        await old_task
+        await asyncio.sleep(0)
+
+        assert loop._active_tasks["test:c1"] == {new_task}
+
+        new_release.set()
+        await new_task
+        await asyncio.sleep(0)
+
+        assert "test:c1" not in loop._active_tasks
+
+
 class TestHandleStop:
     @pytest.mark.asyncio
     async def test_stop_no_active_task(self):