Improve agent reliability and token efficiency

tests/test_yandex_ai.py

@@ -16,6 +16,7 @@ class FakeCompletion:
         self.configuration = None
         self.messages = None
         self.timeout = None
+        self.result = [SimpleNamespace(text="  готово  ")]
 
     def configure(self, **kwargs):
         self.configuration = kwargs
@@ -24,7 +25,7 @@ class FakeCompletion:
     async def run(self, messages, timeout):
         self.messages = messages
         self.timeout = timeout
-        return [SimpleNamespace(text="  готово  ")]
+        return self.result
 
 
 class FakeChatCompletions:
@@ -131,6 +132,38 @@ class YandexAIClientTests(unittest.IsolatedAsyncioTestCase):
             ["gpt://folder-id/qwen3.6-35b-a3b/latest"],
         )
 
+    async def test_chat_logs_reported_token_usage(self) -> None:
+        class FakeResult(list):
+            pass
+
+        completion = FakeCompletion()
+        completion.result = FakeResult(
+            [SimpleNamespace(text="готово")]
+        )
+        completion.result.usage = SimpleNamespace(
+            prompt_tokens=120,
+            completion_tokens=8,
+            total_tokens=128,
+        )
+        completions = FakeChatCompletions(completion, self._list_models)
+        sdk = SimpleNamespace(
+            chat=SimpleNamespace(completions=completions),
+        )
+        client = YandexAIClient(folder_id="folder-id", sdk=sdk)
+
+        with self.assertLogs(
+            "assistant_bot.yandex_ai",
+            level="INFO",
+        ) as captured:
+            await client.chat(
+                "yandexgpt",
+                [{"role": "user", "content": "Привет"}],
+            )
+
+        self.assertIn("input_tokens=120", captured.output[0])
+        self.assertIn("output_tokens=8", captured.output[0])
+        self.assertIn("total_tokens=128", captured.output[0])
+
     async def test_list_models_returns_sorted_uris(self) -> None:
         sdk = SimpleNamespace(
             chat=SimpleNamespace(