feat(bot): timeout kick, SSE restart, turbo config and repaired tests

bot/README.md

@@ -55,7 +55,7 @@ cp config.json.example config.json
 | Параметр | Описание | По умолчанию |
 |----------|----------|-------------|
 | `vk_token` | VK API токен (получить [тут](https://vk.com/dev)) | — |
-| `opencode_url` | URL opencode serve | `http://127.0.0.1:4096` |
+| `opencode_url` | URL opencode serve | `http://127.0.0.1:4098` |
 | `session_file` | Файл хранения сессий | `sessions.json` |
 | `vk_api_version` | Версия VK API | `5.200` |
 | `longpoll_wait` | Время ожидания longpoll (сек) | `25` |
@@ -65,6 +65,7 @@ cp config.json.example config.json
 | `opencode_bin_path` | Путь к бинарнику opencode | — |
 | `llama_server_path` | Путь к llama-server | `llama-server` |
 | `llama_server_host` | URL llama-server | `http://localhost:8081` |
+| `llama_server_env` | Переменные окружения для llama-server (например, `{"TURBOPREFILL": "1"}`) | `{}` |
 | `models` | Словарь моделей и параметров запуска | — |
 | `default_model` | Алиас модели по умолчанию | — |
 | `mcp_servers` | (опционально) MCP серверы для opencode | — |

bot/config-turbo.json

@@ -0,0 +1,61 @@
+{
+  "vk_token": "vk1.a.Iff3QxRdMeE9iG6HPbTYSMyj6-4Uv-euwpd6sM1EUWtL9Tf3ZH8zRW5G7nHqmYPu--JFdvBtepc4Yj3AtQyDuQW4OuNxQGU-CwDxYezHK2-kzKi4_h3RORGeriMbKMDyBCJwwDjgsD8ssYd5rHrt0MYh1JB9rnDjiPtu1Xlr6Gr5yqnt8fOBFINHJCsAjUjvax8HT0_IW0VRbC5DYwnYFw",
+  "opencode_url": "http://127.0.0.1:4098",
+  "session_file": "sessions.json",
+  "vk_api_version": "5.200",
+  "longpoll_wait": 25,
+  "thinking_peer_id": 2000000003,
+  "llama_server_path": "/home/grishberg/llama.cpp.turbo/build/bin/llama-server",
+  "llama_server_host": "http://127.0.0.1:8081",
+  "llama_server_env": {
+    "TURBOPREFILL": "1"
+  },
+  "opencode_config_path": "/home/grishberg/.config/opencode2/config.json",
+  "models": {
+    "qwen3.8-27b": {
+      "model": "qwen3.8-27b",
+      "args": "-m /home/grishberg/models/Qwen3.8-27B-UD-Q8_K_XL.gguf --port 8081 --host 0.0.0.0 --ctx-size 200000  -fit off -sm tensor --tensor-split 1,1 -fa on --spec-type draft-mtp --spec-draft-n-max 3 --no-mmap --cache-type-k q8_0 --cache-type-v q8_0 --mmproj /home/grishberg/models/Qwen3.8-27B-mmproj-F16.gguf --no-mmproj-offload --repeat-penalty 1.1 --temp 1.0 --top-p 0.95 --top-k 20 --min-p 0.0 --presence-penalty 0.0"
+    },
+    "qwen3.6-27b-vision": {
+      "model": "qwen3.6-27b-vision",
+      "args": "-m /home/grishberg/models/Qwen3.6-27B-UD-Q8_K_XL.gguf --port 8081 --host 0.0.0.0 --ctx-size 200000  -fit off -sm tensor --tensor-split 1,1 -fa on --spec-type draft-mtp --spec-draft-n-max 3 --no-mmap --cache-type-k q8_0 --cache-type-v q8_0 --mmproj /home/grishberg/models/Qwen3.6-27B-mmproj-F16.gguf --no-mmproj-offload --repeat-penalty 1.1"
+    }
+  },
+  "default_model": "qwen3.8-27b",
+  "allowed_folders": [
+    "/tmp",
+    "/home/grishberg/projects"
+  ],
+  "reasoning_timeout_seconds": 900,
+  "server_type": "llama",
+  "permission": {
+    "external_directory": {
+      "/tmp/**": "allow",
+      "*": "ask"
+    }
+  },
+  "peer_id": 2000000002,
+  "opencode_bin_path": "/home/grishberg/projects/ts/opencode/packages/opencode/dist/opencode-linux-x64/bin/opencode",
+  "opencode-app-name": "opencode2",
+  "shutdown": "/home/grishberg/shutdown-pc.sh",
+  "mcp_servers": {
+    "ya-disk-uploader": {
+      "type": "local",
+      "command": [
+        "/home/grishberg/projects/go/ya-disk-uploader/ya-disk-uploader",
+        "mcp"
+      ],
+      "enabled": true
+    },
+    "send-files": {
+      "type": "local",
+      "command": [
+        "/home/grishberg/projects/go/mcp-vk-files/mcp-vk-files",
+        "mcp"
+      ],
+      "enabled": true
+    }
+  },
+  "voice_auto_transcribe_enabled": true,
+  "whisper_api_url": "http://127.0.0.1:8013"
+}

bot/config.py

@@ -55,6 +55,7 @@ def _resolve_config_path(name: str) -> Path:
     "models": [],
     "default_model": "qwen3.5-122b",
     "allowed_folders": ["/tmp"],
+    "reasoning_timeout_seconds": 900,
 }
 
 
@@ -126,6 +127,7 @@ def load_config(config_path: str = "config.json") -> dict:
 ALLOWED_FOLDERS = CONFIG.get("allowed_folders", ["/tmp"])
 SHUTDOWN_SCRIPT = CONFIG.get("shutdown")
 RESTART_SCRIPT = CONFIG.get("restart")
+REASONING_TIMEOUT_SECONDS = CONFIG.get("reasoning_timeout_seconds", 900)
 
 if not VK_TOKEN:
     raise ValueError("VK_TOKEN is required in config file")
@@ -210,6 +212,7 @@ def switch_config(config_name: str) -> bool:
     current_module.ALLOWED_FOLDERS = new_config.get("allowed_folders", ["/tmp"])
     current_module.SHUTDOWN_SCRIPT = new_config.get("shutdown")
     current_module.RESTART_SCRIPT = new_config.get("restart")
+    current_module.REASONING_TIMEOUT_SECONDS = new_config.get("reasoning_timeout_seconds", 900)
     current_module.OPENCODE_BIN = Path(new_config["opencode_bin_path"])
     if not current_module.OPENCODE_BIN.is_absolute():
         current_module.OPENCODE_BIN = (SCRIPT_DIR / current_module.OPENCODE_BIN).resolve()

bot/llama_server.py

@@ -99,6 +99,12 @@ async def restart_llama_server(
         env = os.environ.copy()
         env.pop("TMUX", None)
 
+        # Дополнительные переменные окружения для llama-server из конфига
+        # (например, {"TURBOPREFILL": "1"} в config-turbo.json)
+        import config
+        for key, value in (config.CONFIG.get("llama_server_env") or {}).items():
+            env[str(key)] = str(value)
+
         log_path = f"/tmp/llama-server-{alias or 'unknown'}.log"
         logger.info(f"Starting llama server, logging to {log_path}")
 
@@ -196,7 +202,7 @@ async def test_llama_server_speed(complete_url: str, model_name: str = None) ->
         async with aiohttp.ClientSession() as session:
             payload = {
                 "prompt": "Test",
-                "n_predict": 10,
+                "n_predict": 100,
                 "stream": False,
                 "temperature": 0.7,
             }

bot/sse_listener.py

@@ -4,6 +4,7 @@
 """
 import asyncio
 import json
+from pathlib import Path
 from typing import Awaitable, Callable, Optional
 
 import aiohttp
@@ -18,8 +19,9 @@
 class SSEEventListener:
     """Асинхронный SSE клиент для OpenCode events."""
 
-    def __init__(self, base_url: str):
+    def __init__(self, base_url: str, directory: Optional[str] = None):
         self.base_url = base_url.rstrip("/")
+        self.directory = directory
         self._task: Optional[asyncio.Task] = None
         self.running = False
         self.processed_events: set[str] = set()
@@ -45,8 +47,14 @@ async def stop(self):
             except (asyncio.CancelledError, Exception):
                 pass
 
-    async def _run(self):
+    def _build_url(self) -> str:
         url = f"{self.base_url}/event"
+        if self.directory:
+            url += f"?directory={self.directory}"
+        return url
+
+    async def _run(self):
+        url = self._build_url()
         while self.running:
             try:
                 async with aiohttp.ClientSession(timeout=SSE_TIMEOUT) as session:

bot/tests/test_auto_grant.py

@@ -12,41 +12,36 @@ class TestAutoGrantPermissionResponse:
     @pytest.mark.asyncio
     async def test_auto_grant_uses_always_not_once(self, mock_vk_client, mock_opencode_client,
                                                     mock_opencode_process, temp_config_file):
-        """
-        BUG: При авто-одобрении используется "once" вместо "always".
-        "once" означает разовое одобрение, но пользователь всё равно видит запрос.
-        "always" означает постоянное одобрение.
-        """
+        """При включённом grant_mode авто-одобрение использует "always"."""
         from vk_longpoll import VKLongPoll
         from session_manager import SessionManager
-        
+
         # Создаём SessionManager и VKLongPoll
         session_mgr = SessionManager(temp_config_file)
         longpoll = VKLongPoll(mock_vk_client, session_mgr, mock_opencode_process)
         longpoll.opencode_client = mock_opencode_client
         session_id = "test-session-auto-grant"
-        longpoll.user_session[999] = session_id
-        longpoll.seen_permissions[session_id] = set()
+        longpoll.session_to_user[session_id] = 999
         # Включаем авто-одобрение через session_mgr
         longpoll.session_mgr.set_grant_mode(session_id, True)
-
-        # Создаём разрешение
-        permission = {
+
+        # SSE-событие permission.asked
+        permission_event = {
+            "sessionID": session_id,
             "id": "perm-auto-123",
-            "session_id": session_id,
             "permission": "write_file",
             "metadata": {"filepath": "/tmp/test.txt"}
         }
-        
+
         # Вызываем обработку разрешения
-        await longpoll._process_permission(permission, session_id, 999)
+        await longpoll._on_permission("permission.asked", permission_event)
 
         # Проверяем, что ответ был отправлен
         mock_opencode_client.send_permission_response.assert_called_once()
 
-        # BUG: Проверка, что используется "always" не "once"
+        # Ответ уходит третьим позиционным аргументом: (session_id, perm_id, response)
         call_args = mock_opencode_client.send_permission_response.call_args
-        response_type = call_args[0][2]  # Третий аргумент: response
+        response_type = call_args[0][2]
 
         assert response_type == "always", \
             f"Авто-одобрение должно использовать 'always', а не '{response_type}'"
@@ -62,19 +57,18 @@ async def test_auto_grant_does_not_send_prompt(self, mock_vk_client, mock_openco
         longpoll = VKLongPoll(mock_vk_client, session_mgr, mock_opencode_process)
         longpoll.opencode_client = mock_opencode_client
         session_id = "test-session-no-prompt"
-        longpoll.user_session[999] = session_id
-        longpoll.seen_permissions[session_id] = set()
+        longpoll.session_to_user[session_id] = 999
         # Включаем авто-одобрение через session_mgr
         longpoll.session_mgr.set_grant_mode(session_id, True)
-
-        permission = {
+
+        permission_event = {
+            "sessionID": session_id,
             "id": "perm-no-prompt",
-            "session_id": session_id,
             "permission": "write_file",
             "metadata": {"filepath": "/tmp/test.txt"}
         }
-        
-        await longpoll._process_permission(permission, session_id, 999)
+
+        await longpoll._on_permission("permission.asked", permission_event)
 
         # mock_vk_client.send_message не должен вызываться для permission
         send_message_calls = [
@@ -95,19 +89,18 @@ async def test_normal_mode_sends_permission_prompt(self, mock_vk_client, mock_op
         longpoll = VKLongPoll(mock_vk_client, session_mgr, mock_opencode_process)
         longpoll.opencode_client = mock_opencode_client
         session_id = "test-session-normal-mode"
-        longpoll.user_session[999] = session_id
-        longpoll.seen_permissions[session_id] = set()
+        longpoll.session_to_user[session_id] = 999
         # Выключаем авто-одобрение через session_mgr
         longpoll.session_mgr.set_grant_mode(session_id, False)
-
-        permission = {
+
+        permission_event = {
+            "sessionID": session_id,
             "id": "perm-normal-456",
-            "session_id": session_id,
             "permission": "write_file",
             "metadata": {"filepath": "/tmp/test.txt"}
         }
-        
-        await longpoll._process_permission(permission, session_id, 999)
+
+        await longpoll._on_permission("permission.asked", permission_event)
 
         # Должно быть отправлено сообщение пользователю
         mock_vk_client.send_message.assert_called_once()

bot/tests/test_config.py

@@ -6,7 +6,7 @@
 ROOT_DIR = Path(__file__).parent.parent
 sys.path.insert(0, str(ROOT_DIR))
 
-from opencode_vk_gateway import load_config
+from config import load_config
 
 
 class TestLoadConfig:
@@ -60,7 +60,7 @@ def test_load_config_file_not_found(self, tmp_path):
         config = load_config(str(config_file))
 
         assert config["vk_token"] == "token"
-        assert config["opencode_url"] == "http://127.0.0.1:4096"
+        assert config["opencode_url"] == "http://127.0.0.1:4098"
         assert config["session_file"] == "sessions.json"
 
     def test_load_config_invalid_json(self, tmp_path):