refactor: 移除 LiteLLM proxy 架构，后端直连 Provider API

- 移除 backend/scripts/build_litellm_proxy_config.py
- 简化 LiteLLMService，移除 run_completion_with_cost 方法
- AgentScopeRunner 改为从 LlmFactory 获取 api_base 和 api_key
- 部署配置移除 litellm/litellm-config-job 服务
- Flutter 新增 AuthBootScreen 引导页
- Android 添加通知权限 (POST_NOTIFICATIONS, RECEIVE_BOOT_COMPLETED, SCHEDULE_EXACT_ALARM)
- 优化 LocalNotificationService 调度失败 fallback
- 更新 manifest.json (version 3)

.env.example

@@ -16,11 +16,6 @@ SOCIAL_WEB__HOST=0.0.0.0
 SOCIAL_WEB__PORT=5775
 SOCIAL_WEB__WORKERS=2
 
-############
-# LiteLLM Proxy 网关配置
-############
-SOCIAL_LITELLM__PORT=3875
-
 ############
 # Redis 配置
 ############

apps/android/app/src/main/AndroidManifest.xml

@@ -1,5 +1,8 @@
 <manifest xmlns:android="http://schemas.android.com/apk/res/android">
     <uses-permission android:name="android.permission.INTERNET" />
+    <uses-permission android:name="android.permission.POST_NOTIFICATIONS" />
+    <uses-permission android:name="android.permission.RECEIVE_BOOT_COMPLETED" />
+    <uses-permission android:name="android.permission.SCHEDULE_EXACT_ALARM" />
     <uses-permission android:name="android.permission.RECORD_AUDIO" />
     <uses-permission android:name="android.permission.CAMERA" />
     <uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE" android:maxSdkVersion="32" />

apps/lib/core/notifications/local_notification_service.dart

@@ -76,16 +76,29 @@ class LocalNotificationService {
       ),
     );
 
-    await _plugin.zonedSchedule(
-      notificationId,
-      event.title,
-      _buildReminderBody(event, reminderMinutes),
-      scheduledAt,
-      details,
-      androidScheduleMode: AndroidScheduleMode.exactAllowWhileIdle,
-      uiLocalNotificationDateInterpretation:
-          UILocalNotificationDateInterpretation.absoluteTime,
-    );
+    try {
+      await _plugin.zonedSchedule(
+        notificationId,
+        event.title,
+        _buildReminderBody(event, reminderMinutes),
+        scheduledAt,
+        details,
+        androidScheduleMode: AndroidScheduleMode.exactAllowWhileIdle,
+        uiLocalNotificationDateInterpretation:
+            UILocalNotificationDateInterpretation.absoluteTime,
+      );
+    } catch (_) {
+      await _plugin.zonedSchedule(
+        notificationId,
+        event.title,
+        _buildReminderBody(event, reminderMinutes),
+        scheduledAt,
+        details,
+        androidScheduleMode: AndroidScheduleMode.inexactAllowWhileIdle,
+        uiLocalNotificationDateInterpretation:
+            UILocalNotificationDateInterpretation.absoluteTime,
+      );
+    }
   }
 
   Future<void> cancelEventReminder(String eventId) async {

apps/lib/core/router/app_router.dart

@@ -3,6 +3,7 @@ import '../../features/auth/presentation/bloc/auth_bloc.dart';
 import '../../features/auth/presentation/bloc/auth_state.dart';
 import 'go_router_refresh_stream.dart';
 import '../../features/auth/ui/screens/login_screen.dart';
+import '../../features/auth/ui/screens/auth_boot_screen.dart';
 import '../../features/auth/ui/screens/register_screen.dart';
 import '../../features/auth/ui/screens/register_verification_screen.dart';
 import '../../features/auth/ui/screens/reset_password_screen.dart';
@@ -43,11 +44,14 @@ final _protectedRoutes = [
 
 GoRouter createAppRouter(AuthBloc authBloc) {
   return GoRouter(
-    initialLocation: '/',
+    initialLocation: '/boot',
     refreshListenable: GoRouterRefreshStream(authBloc.stream),
     redirect: (context, state) {
       final authState = authBloc.state;
       final isAuthenticated = authState is AuthAuthenticated;
+      final isAuthChecking =
+          authState is AuthInitial || authState is AuthLoading;
+      final isBootRoute = state.matchedLocation == '/boot';
       final isAuthRoute =
           state.matchedLocation == '/' ||
           state.matchedLocation.startsWith('/login') ||
@@ -56,6 +60,12 @@ GoRouter createAppRouter(AuthBloc authBloc) {
         (route) => state.matchedLocation.startsWith(route),
       );
 
+      if (isAuthChecking && !isBootRoute) {
+        return '/boot';
+      }
+      if (!isAuthChecking && isBootRoute) {
+        return isAuthenticated ? '/home' : '/';
+      }
       if (!isAuthenticated && isProtected) {
         return '/';
       }
@@ -65,6 +75,10 @@ GoRouter createAppRouter(AuthBloc authBloc) {
       return null;
     },
     routes: [
+      GoRoute(
+        path: '/boot',
+        builder: (context, state) => const AuthBootScreen(),
+      ),
       GoRoute(path: '/', builder: (context, state) => const LoginScreen()),
       GoRoute(
         path: '/calendar/events/:id',

apps/lib/features/auth/ui/screens/auth_boot_screen.dart

@@ -0,0 +1,26 @@
+import 'package:flutter/material.dart';
+
+import '../../../../core/theme/design_tokens.dart';
+import '../../../../shared/widgets/app_loading_indicator.dart';
+
+class AuthBootScreen extends StatelessWidget {
+  const AuthBootScreen({super.key});
+
+  @override
+  Widget build(BuildContext context) {
+    return const Scaffold(
+      backgroundColor: AppColors.authBackgroundTop,
+      body: SafeArea(
+        child: Center(
+          child: AppLoadingIndicator(
+            variant: AppLoadingVariant.surface,
+            size: 28,
+            strokeWidth: 2.5,
+            color: AppColors.authPrimaryButton,
+            trackColor: AppColors.authPrimaryButtonDisabled,
+          ),
+        ),
+      ),
+    );
+  }
+}

apps/lib/features/calendar/ui/widgets/create_event_sheet.dart

@@ -638,7 +638,11 @@ class _CreateEventSheetState extends State<CreateEventSheet>
       try {
         final notificationService = sl<LocalNotificationService>();
         await notificationService.upsertEventReminder(saved);
-      } catch (_) {}
+      } catch (_) {
+        if (mounted) {
+          Toast.show(context, '提醒创建失败，请检查通知权限', type: ToastType.warning);
+        }
+      }
 
       widget.onSaved?.call();
       if (mounted) {

apps/pubspec.yaml

@@ -1,7 +1,7 @@
 name: social_app
 description: "Social App - A Flutter mobile application"
 publish_to: 'none'
-version: 0.1.0+2
+version: 0.1.0+3
 environment:
   sdk: ^3.10.7
 

backend/scripts/build_litellm_proxy_config.py

@@ -1,90 +0,0 @@
-from __future__ import annotations
-
-import argparse
-from pathlib import Path
-from typing import Any
-
-import yaml
-
-from core.config.initial.init_data import load_llm_catalog
-
-
-def _provider_key_env_name(factory_name: str) -> str:
-    normalized = factory_name.strip().upper()
-    if normalized == "VOLCENGINE":
-        normalized = "ARK"
-    return f"SOCIAL_LLM__PROVIDER_KEYS__{normalized}"
-
-
-def build_proxy_config() -> dict[str, Any]:
-    catalog = load_llm_catalog()
-
-    factories = catalog.get("factories", [])
-    llms = catalog.get("llms", [])
-    if not isinstance(factories, list) or not isinstance(llms, list):
-        raise ValueError("invalid llm catalog format")
-
-    factory_url_map: dict[str, str] = {}
-    for factory in factories:
-        if not isinstance(factory, dict):
-            continue
-        name = str(factory.get("name", "")).strip().lower()
-        request_url = str(factory.get("request_url", "")).strip()
-        if name and request_url:
-            factory_url_map[name] = request_url
-
-    model_list: list[dict[str, Any]] = []
-    for llm in llms:
-        if not isinstance(llm, dict):
-            continue
-        model_code = str(llm.get("model_code", "")).strip()
-        factory_name = str(llm.get("factory_name", "")).strip()
-        litellm_model = str(llm.get("litellm_model", "")).strip()
-        if not model_code or not factory_name or not litellm_model:
-            continue
-
-        api_base = factory_url_map.get(factory_name.lower())
-        if not api_base:
-            raise ValueError(
-                f"factory request_url missing for model {model_code}: {factory_name}"
-            )
-
-        env_key_name = _provider_key_env_name(factory_name)
-        provider_model = (
-            litellm_model.split("/", 1)[1] if "/" in litellm_model else litellm_model
-        )
-
-        model_list.append(
-            {
-                "model_name": model_code,
-                "litellm_params": {
-                    "model": f"openai/{provider_model}",
-                    "api_base": api_base,
-                    "api_key": f"os.environ/{env_key_name}",
-                },
-            }
-        )
-
-    if not model_list:
-        raise ValueError("no models found in llm catalog")
-
-    return {"model_list": model_list}
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description="Build LiteLLM proxy config")
-    parser.add_argument("--output", required=True, help="Output YAML file path")
-    args = parser.parse_args()
-
-    output_path = Path(args.output).resolve()
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-
-    config = build_proxy_config()
-    with output_path.open("w", encoding="utf-8") as file:
-        yaml.safe_dump(config, file, sort_keys=False, allow_unicode=False)
-
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())

backend/src/core/agentscope/runtime/runner.py

@@ -21,9 +21,11 @@ from core.agentscope.utils import (
     finalize_json_response,
     patch_agentscope_json_repair_compat,
 )
+from core.config.settings import config
 from core.db.session import AsyncSessionLocal
 from core.logging import get_logger
 from models.llm import Llm
+from models.llm_factory import LlmFactory
 from models.system_agents import SystemAgents
 from schemas.agent.runtime_models import (
     RouterAgentOutput,
@@ -50,6 +52,8 @@ logger = get_logger("core.agentscope.runtime.runner")
 class SystemAgentRuntimeConfig:
     agent_type: AgentType
     model_code: str
+    api_base_url: str
+    api_key: str
     llm_config: SystemAgentLLMConfig
 
 
@@ -63,7 +67,7 @@ class StageExecutionResult:
 class AgentScopeRunner:
     def __init__(self, *, litellm_service: LiteLLMService | None = None) -> None:
         patch_agentscope_json_repair_compat()
-        self._litellm_service = litellm_service or LiteLLMService()
+        self._litellm_service: LiteLLMService = litellm_service or LiteLLMService()
 
     async def execute(
         self,
@@ -221,23 +225,42 @@ class AgentScopeRunner:
         agent_type: AgentType,
     ) -> SystemAgentRuntimeConfig:
         stmt = (
-            select(SystemAgents, Llm)
+            select(SystemAgents, Llm, LlmFactory)
             .join(Llm, SystemAgents.llm_id == Llm.id)
+            .join(LlmFactory, Llm.factory_id == LlmFactory.id)
             .where(SystemAgents.agent_type == agent_type.value)
         )
         row = (await session.execute(stmt)).one_or_none()
         if row is None:
             raise RuntimeError(f"system agent config not found: {agent_type.value}")
-        system_agent, llm = row
+        system_agent, llm, factory = row
         status = str(system_agent.status).strip().lower()
         if status != "active":
             raise RuntimeError(f"system agent is not active: {agent_type.value}")
         return SystemAgentRuntimeConfig(
             agent_type=agent_type,
             model_code=llm.model_code,
+            api_base_url=factory.request_url,
+            api_key=self._resolve_provider_api_key(factory_name=factory.name),
             llm_config=SystemAgentLLMConfig.model_validate(system_agent.config or {}),
         )
 
+    @staticmethod
+    def _resolve_provider_api_key(*, factory_name: str) -> str:
+        normalized_factory_name = factory_name.strip().upper()
+        if normalized_factory_name == "VOLCENGINE":
+            normalized_factory_name = "ARK"
+
+        provider_keys = {
+            str(key).strip().upper(): str(value).strip()
+            for key, value in config.llm.provider_keys.items()
+            if str(value).strip()
+        }
+        api_key = provider_keys.get(normalized_factory_name, "")
+        if not api_key:
+            raise RuntimeError(f"provider api key missing for factory: {factory_name}")
+        return api_key
+
     async def _run_router_stage(
         self,
         *,
@@ -363,9 +386,9 @@ class AgentScopeRunner:
 
         model = OpenAIChatModel(
             model_name=stage_config.model_code,
-            api_key=self._litellm_service.proxy_api_key,
+            api_key=stage_config.api_key,
             stream=False,
-            client_kwargs={"base_url": self._litellm_service.proxy_base_url},
+            client_kwargs={"base_url": stage_config.api_base_url},
             generate_kwargs=generate_kwargs,
         )
         return TrackingChatModel(model)

backend/src/services/litellm/__init__.py

@@ -1,9 +1,5 @@
 from __future__ import annotations
 
-from services.litellm.service import (
-    LiteLLMResponseWithCost,
-    LiteLLMService,
-    LiteLLMUsage,
-)
+from services.litellm.service import LiteLLMService
 
-__all__ = ["LiteLLMService", "LiteLLMUsage", "LiteLLMResponseWithCost"]
+__all__ = ["LiteLLMService"]

backend/src/services/litellm/service.py

@@ -1,11 +1,8 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
-from typing import Any, Callable
+from typing import Any
 
-from litellm import completion
-
-from core.config.settings import config
 from core.config.initial.init_data import load_llm_catalog
 
 
@@ -17,34 +14,10 @@ class PricingTier:
     cache_hit_cost_per_token: float
 
 
-@dataclass(frozen=True)
-class LiteLLMUsage:
-    prompt_tokens: int
-    completion_tokens: int
-    total_tokens: int
-    cached_prompt_tokens: int
-    cost: float
-
-
-@dataclass(frozen=True)
-class LiteLLMResponseWithCost:
-    response: dict[str, Any]
-    usage: LiteLLMUsage
-
-
 class LiteLLMService:
-    proxy_base_url: str
-    proxy_api_key: str
     _pricing_by_model: dict[str, tuple[PricingTier, ...]]
 
-    def __init__(
-        self,
-        *,
-        proxy_base_url: str | None = None,
-        proxy_api_key: str | None = None,
-    ) -> None:
-        self.proxy_base_url = proxy_base_url or config.litellm.base_url
-        self.proxy_api_key = proxy_api_key or config.litellm.api_key
+    def __init__(self) -> None:
         self._pricing_by_model = self._build_pricing_map()
 
     @staticmethod
@@ -142,78 +115,3 @@ class LiteLLMService:
             "cost": cost,
             "latencyMs": latency_ms,
         }
-
-    def run_completion_with_cost(
-        self,
-        *,
-        model: str,
-        messages: list[dict[str, Any]],
-        temperature: float | None = None,
-        max_tokens: int | None = None,
-        timeout: float | None = None,
-        response_format: dict[str, Any] | None = None,
-        completion_fn: Callable[..., dict[str, Any]] | None = None,
-    ) -> LiteLLMResponseWithCost:
-        caller = completion_fn or completion
-        request_model = model if model.startswith("openai/") else f"openai/{model}"
-
-        request_kwargs: dict[str, Any] = {
-            "model": request_model,
-            "api_key": self.proxy_api_key,
-            "api_base": self.proxy_base_url,
-            "messages": messages,
-            "temperature": temperature,
-            "max_tokens": max_tokens,
-            "timeout": timeout,
-            "stream": False,
-        }
-        if response_format is not None:
-            request_kwargs["response_format"] = response_format
-
-        response_any = caller(**request_kwargs)
-        response = self._normalize_response(response_any)
-
-        usage_raw = response.get("usage")
-        if not isinstance(usage_raw, dict):
-            raise ValueError("missing usage in response")
-
-        prompt_tokens = int(usage_raw.get("prompt_tokens", 0) or 0)
-        completion_tokens = int(usage_raw.get("completion_tokens", 0) or 0)
-        total_tokens = int(
-            usage_raw.get("total_tokens", prompt_tokens + completion_tokens) or 0
-        )
-        cached_prompt_tokens = 0
-        prompt_tokens_details = usage_raw.get("prompt_tokens_details")
-        if isinstance(prompt_tokens_details, dict):
-            cached_prompt_tokens = int(
-                prompt_tokens_details.get("cached_tokens", 0) or 0
-            )
-
-        resolved_model = str(response.get("model", model)).strip()
-        cost = self.calculate_cost(
-            model=resolved_model,
-            prompt_tokens=prompt_tokens,
-            completion_tokens=completion_tokens,
-            cached_prompt_tokens=cached_prompt_tokens,
-        )
-        return LiteLLMResponseWithCost(
-            response=response,
-            usage=LiteLLMUsage(
-                prompt_tokens=prompt_tokens,
-                completion_tokens=completion_tokens,
-                total_tokens=total_tokens,
-                cached_prompt_tokens=cached_prompt_tokens,
-                cost=cost,
-            ),
-        )
-
-    @staticmethod
-    def _normalize_response(response_any: Any) -> dict[str, Any]:
-        if isinstance(response_any, dict):
-            return response_any
-        model_dump = getattr(response_any, "model_dump", None)
-        if callable(model_dump):
-            dumped = model_dump()
-            if isinstance(dumped, dict):
-                return dumped
-        raise ValueError("litellm response is not serializable")

backend/tests/unit/core/agentscope/runtime/test_runner.py

@@ -133,6 +133,8 @@ async def test_execute_uses_router_ui_mode_to_select_worker_output_model(
             model_code="qwen3.5-flash"
             if kwargs["agent_type"] == AgentType.ROUTER
             else "deepseek-chat",
+            api_base_url="https://example.com/v1",
+            api_key="sk-test",
             llm_config=SystemAgentLLMConfig(
                 temperature=0.1, max_tokens=256, timeout_seconds=30
             ),
@@ -233,6 +235,8 @@ async def test_execute_passes_runtime_client_time_to_router_and_worker(
         return SystemAgentRuntimeConfig(
             agent_type=kwargs["agent_type"],
             model_code="model-a",
+            api_base_url="https://example.com/v1",
+            api_key="sk-test",
             llm_config=SystemAgentLLMConfig(
                 temperature=0.1, max_tokens=256, timeout_seconds=30
             ),
@@ -296,3 +300,29 @@ async def test_execute_passes_runtime_client_time_to_router_and_worker(
 
     assert captured["router_timezone"] == "America/Los_Angeles"
     assert captured["worker_timezone"] == "America/Los_Angeles"
+
+
+def test_resolve_provider_api_key_maps_volcengine_to_ark(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        "core.agentscope.runtime.runner.config.llm.provider_keys",
+        {"ARK": "ark-key", "DASHSCOPE": "dash-key"},
+    )
+
+    assert (
+        AgentScopeRunner._resolve_provider_api_key(factory_name="volcengine")
+        == "ark-key"
+    )
+
+
+def test_resolve_provider_api_key_raises_when_missing(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        "core.agentscope.runtime.runner.config.llm.provider_keys",
+        {"DASHSCOPE": "dash-key"},
+    )
+
+    with pytest.raises(RuntimeError, match="provider api key missing"):
+        AgentScopeRunner._resolve_provider_api_key(factory_name="deepseek")

backend/tests/unit/services/test_litellm_service.py

@@ -31,37 +31,6 @@ def test_calculate_cost_uses_second_qwen_tier() -> None:
     assert cost == pytest.approx(0.1856)
 
 
-def test_run_completion_extracts_usage_and_cost() -> None:
-    service = LiteLLMService()
-    captured: dict[str, object] = {}
-
-    def _fake_completion(**kwargs: object) -> dict[str, object]:
-        captured.update(kwargs)
-        return {
-            "model": "dashscope/qwen3.5-flash",
-            "usage": {
-                "prompt_tokens": 2000,
-                "completion_tokens": 100,
-                "total_tokens": 2100,
-                "prompt_tokens_details": {"cached_tokens": 500},
-            },
-            "choices": [{"message": {"content": "ok"}}],
-        }
-
-    result = service.run_completion_with_cost(
-        model="dashscope/qwen3.5-flash",
-        messages=[{"role": "user", "content": "hello"}],
-        response_format={"type": "json_object"},
-        completion_fn=_fake_completion,
-    )
-
-    assert result.usage.prompt_tokens == 2000
-    assert result.usage.completion_tokens == 100
-    assert result.usage.total_tokens == 2100
-    assert result.usage.cost == pytest.approx(0.00051)
-    assert captured["response_format"] == {"type": "json_object"}
-
-
 def test_build_usage_metadata_calculates_cost_from_usage_summary() -> None:
     service = LiteLLMService()
 

deploy/.env.prod.example

@@ -16,13 +16,6 @@ SOCIAL_WEB__HOST=0.0.0.0
 SOCIAL_WEB__PORT=5775
 SOCIAL_WEB__WORKERS=2
 
-############
-# LiteLLM Proxy 网关配置
-############
-# 可选：覆盖官方 LiteLLM 镜像（默认使用 compose 内置 digest）
-# SOCIAL_LITELLM_IMAGE=ghcr.io/berriai/litellm@sha256:b959a1816fa454a14d2842242d0fa1cd0d39f96fc94d3a1f4e1de4e48e2398c6
-SOCIAL_LITELLM__PORT=3875
-
 ############
 # Redis 配置
 ############

deploy/README.md

@@ -2,7 +2,7 @@
 
 本目录是单机 `docker compose` 的生产交付包，架构为：
 
-- 应用层：`litellm + web + worker-critical + worker-default + worker-bulk + init-job`
+- 应用层：`web + worker-critical + worker-default + worker-bulk + init-job`
 - 中间件：`redis`
 - 数据与认证：云 Supabase（通过环境变量访问）
 - 反向代理：由服务器侧 nginx 托管（不在本目录编排）
@@ -19,7 +19,7 @@
 - `deploy/.env.prod.example` 仅作为模板，真实密钥请在服务器上填写到 `deploy/.env.prod`，不要提交仓库。
 - Redis 密码必填；为空时容器会启动失败。
 - 后端镜像默认使用非 root 用户运行。
-- 容器间通信仅走 Docker 内网（`redis`、`litellm` 服务名）。
+- 容器间通信仅走 Docker 内网（`redis` 服务名）。
 
 ## 目录结构
 
@@ -75,14 +75,13 @@ cp deploy/.env.prod.example deploy/.env.prod
 
 说明：
 
-- 容器内通信统一使用 Docker 内网：`SOCIAL_REDIS__HOST=redis`、`SOCIAL_LITELLM__HOST=litellm`。
-- `SOCIAL_WEB__HOST`/`SOCIAL_LITELLM__BIND_HOST` 是容器内监听地址，生产建议保持 `0.0.0.0`。
+- 容器内通信统一使用 Docker 内网：`SOCIAL_REDIS__HOST=redis`。
+- `SOCIAL_WEB__HOST` 是容器内监听地址，生产建议保持 `0.0.0.0`。
 
 ### 2) 启动常驻服务
 
 ```bash
-docker compose --env-file deploy/.env.prod -f deploy/docker-compose.prod.yml up -d redis litellm-config-job
-docker compose --env-file deploy/.env.prod -f deploy/docker-compose.prod.yml up -d litellm web worker-critical worker-default worker-bulk
+docker compose --env-file deploy/.env.prod -f deploy/docker-compose.prod.yml up -d redis web worker-critical worker-default worker-bulk
 ```
 
 ### 3) 执行一次性 bootstrap
@@ -125,5 +124,4 @@ docker compose --env-file deploy/.env.prod -f deploy/docker-compose.prod.yml up
 
 ## 已知约束
 
-- LiteLLM 配置由 `litellm-config-job` 一次性生成到共享 volume（`litellm_config`）。若更新了 LLM 目录或 Provider Key，需重新执行 `up -d litellm-config-job` 后重启 `litellm`。
 - `init-job` 为一次性任务，不长期驻留。

deploy/docker-compose.prod.yml

@@ -23,50 +23,6 @@ services:
             timeout: 3s
             retries: 10
 
-    litellm-config-job:
-        image: ${SOCIAL_BACKEND_IMAGE:-social-app-backend:prod}
-        container_name: social-prod-litellm-config-job
-        restart: "no"
-        env_file:
-            - ./.env.prod
-        environment:
-            - PYTHONPATH=/app/backend/src
-            - PYTHONDONTWRITEBYTECODE=1
-        command: >
-            .venv/bin/python backend/scripts/build_litellm_proxy_config.py --output /config/litellm-proxy-config.yaml
-        volumes:
-            - litellm_config:/config
-        depends_on:
-            redis:
-                condition: service_healthy
-
-    litellm:
-        image: ghcr.io/berriai/litellm@sha256:b959a1816fa454a14d2842242d0fa1cd0d39f96fc94d3a1f4e1de4e48e2398c6
-        container_name: social-prod-litellm
-        restart: unless-stopped
-        env_file:
-            - ./.env.prod
-        command: >
-            --config /config/litellm-proxy-config.yaml --host ${SOCIAL_LITELLM__BIND_HOST:-0.0.0.0} --port ${SOCIAL_LITELLM__PORT:-3875}
-        volumes:
-            - litellm_config:/config:ro
-        depends_on:
-            redis:
-                condition: service_healthy
-            litellm-config-job:
-                condition: service_completed_successfully
-        healthcheck:
-            test:
-                [
-                    "CMD",
-                    "python",
-                    "-c",
-                    "import os,sys,urllib.request;port=os.getenv('SOCIAL_LITELLM__PORT','3875');u=f'http://127.0.0.1:{port}/health';sys.exit(0 if urllib.request.urlopen(u, timeout=3).getcode() < 500 else 1)",
-                ]
-            interval: 15s
-            timeout: 5s
-            retries: 10
-
     web:
         image: ${SOCIAL_BACKEND_IMAGE:-social-app-backend:prod}
         container_name: social-prod-web
@@ -80,8 +36,6 @@ services:
             - SOCIAL_RUNTIME__ENVIRONMENT=${SOCIAL_RUNTIME__ENVIRONMENT:-prod}
             - SOCIAL_REDIS__HOST=redis
             - SOCIAL_REDIS__PORT=6379
-            - SOCIAL_LITELLM__HOST=litellm
-            - SOCIAL_LITELLM__PORT=${SOCIAL_LITELLM__PORT:-3875}
         command: >
             sh -c '.venv/bin/uvicorn app:app --host ${SOCIAL_WEB__HOST:-0.0.0.0} --port ${SOCIAL_WEB__PORT:-5775} --workers ${SOCIAL_WEB__WORKERS:-2} --log-level $(printf "%s" "${SOCIAL_RUNTIME__LOG_LEVEL:-info}" | tr "[:upper:]" "[:lower:]")'
         ports:
@@ -89,8 +43,6 @@ services:
         depends_on:
             redis:
                 condition: service_healthy
-            litellm:
-                condition: service_healthy
         volumes:
             - ../logs:/app/logs
             - ./static/releases:/app/deploy/static/releases:ro
@@ -119,15 +71,11 @@ services:
             - SOCIAL_RUNTIME__ENVIRONMENT=${SOCIAL_RUNTIME__ENVIRONMENT:-prod}
             - SOCIAL_REDIS__HOST=redis
             - SOCIAL_REDIS__PORT=6379
-            - SOCIAL_LITELLM__HOST=litellm
-            - SOCIAL_LITELLM__PORT=${SOCIAL_LITELLM__PORT:-3875}
         command: >
             sh -c '.venv/bin/taskiq worker core.taskiq.app:critical_broker core.agentscope.runtime.tasks --workers ${SOCIAL_WORKER__GROUPS__CRITICAL__CONCURRENCY:-2}'
         depends_on:
             redis:
                 condition: service_healthy
-            litellm:
-                condition: service_healthy
         volumes:
             - ../logs:/app/logs
             - ./static/releases:/app/deploy/static/releases:ro
@@ -145,15 +93,11 @@ services:
             - SOCIAL_RUNTIME__ENVIRONMENT=${SOCIAL_RUNTIME__ENVIRONMENT:-prod}
             - SOCIAL_REDIS__HOST=redis
             - SOCIAL_REDIS__PORT=6379
-            - SOCIAL_LITELLM__HOST=litellm
-            - SOCIAL_LITELLM__PORT=${SOCIAL_LITELLM__PORT:-3875}
         command: >
             sh -c '.venv/bin/taskiq worker core.taskiq.app:default_broker core.agentscope.runtime.tasks --workers ${SOCIAL_WORKER__GROUPS__DEFAULT__CONCURRENCY:-2}'
         depends_on:
             redis:
                 condition: service_healthy
-            litellm:
-                condition: service_healthy
         volumes:
             - ../logs:/app/logs
             - ./static/releases:/app/deploy/static/releases:ro
@@ -171,15 +115,11 @@ services:
             - SOCIAL_RUNTIME__ENVIRONMENT=${SOCIAL_RUNTIME__ENVIRONMENT:-prod}
             - SOCIAL_REDIS__HOST=redis
             - SOCIAL_REDIS__PORT=6379
-            - SOCIAL_LITELLM__HOST=litellm
-            - SOCIAL_LITELLM__PORT=${SOCIAL_LITELLM__PORT:-3875}
         command: >
             sh -c '.venv/bin/taskiq worker core.taskiq.app:bulk_broker core.agentscope.runtime.tasks --workers ${SOCIAL_WORKER__GROUPS__BULK__CONCURRENCY:-1}'
         depends_on:
             redis:
                 condition: service_healthy
-            litellm:
-                condition: service_healthy
         volumes:
             - ../logs:/app/logs
             - ./static/releases:/app/deploy/static/releases:ro
@@ -197,14 +137,10 @@ services:
             - SOCIAL_RUNTIME__ENVIRONMENT=${SOCIAL_RUNTIME__ENVIRONMENT:-prod}
             - SOCIAL_REDIS__HOST=redis
             - SOCIAL_REDIS__PORT=6379
-            - SOCIAL_LITELLM__HOST=litellm
-            - SOCIAL_LITELLM__PORT=${SOCIAL_LITELLM__PORT:-3875}
         command: .venv/bin/python -m core.runtime.cli bootstrap
         depends_on:
             redis:
                 condition: service_healthy
-            litellm:
-                condition: service_healthy
         volumes:
             - ../logs:/app/logs
             - ./static/releases:/app/deploy/static/releases:ro
@@ -213,4 +149,3 @@ services:
 
 volumes:
     redis_data:
-    litellm_config:

deploy/static/releases/manifest.json

@@ -4,23 +4,12 @@
       "platform": "android",
       "channel": "release",
       "version_name": "0.1.0",
-      "version_code": 1,
-      "min_supported_version_code": 1,
-      "file_name": "social-app-android-v0.1.0+1-release.apk",
-      "release_notes": "\u95ee\u9898\u4fee\u590d\u548c\u4f53\u9a8c\u4f18\u5316",
-      "file_size": 21371504,
-      "sha256": "6cf53601f36e0037b6de909ea3567d1e18a1bcec1164e1b70d88c1802eafd44b"
-    },
-    {
-      "platform": "android",
-      "channel": "release",
-      "version_name": "0.1.0",
-      "version_code": 2,
-      "min_supported_version_code": 2,
-      "file_name": "social-app-android-v0.1.0+2-release.apk",
-      "release_notes": "\u95ee\u9898\u4fee\u590d\u548c\u4f53\u9a8c\u4f18\u5316",
-      "file_size": 21371504,
-      "sha256": "8f769bda3ba5414dfd5712ac026d8a13663990b7e83a4e92b8e85caf9945d5eb"
+      "version_code": 3,
+      "min_supported_version_code": 3,
+      "file_name": "social-app-android-v0.1.0+3-release.apk",
+      "release_notes": null,
+      "file_size": 21371568,
+      "sha256": "34691f96004b3dc3b2070d84ae0e7f0d2943f6c9978160eb78550081bc72a74a"
     }
   ]
 }

docs/plans/2026-03-16-account-and-chat-surface-language-implementation.md

@@ -1,86 +0,0 @@
-# 账户页与主页工具渲染统一表面语言 Implementation Plan
-
-> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
-
-**Goal:** 统一 `我的账户` 页面与主页 `ui_schema` 渲染的视觉语言，修正 STEP 协议映射并将等待文案从“正在思考”切换为协议阶段语义。
-
-**Architecture:** 设置域继续复用 `AccountSurfaceScaffold + AccountSectionCard`；聊天域在 `UiSchemaRenderer` 中重做 surface 层级与按钮/徽章/KV 呈现，保持协议字段不变。将 stage 映射改为 `router/worker` 协议值，在 UI 显示为“意图识别中/任务执行中/任务处理中”。
-
-**Tech Stack:** Flutter, flutter_bloc, design tokens (`AppColors/AppSpacing/AppRadius`), AG-UI SSE protocol mapping
-
----
-
-### Task 1: 先补失败测试（Stage 文案与 UI Schema 呈现）
-
-**Files:**
-- Create: `apps/test/features/chat/presentation/agent_stage_mapping_test.dart`
-- Modify: `apps/test/features/chat/ui_schema_renderer_test.dart`
-
-**Step 1: 写失败测试**
-- 新增 stage 映射测试，断言 `router -> intentLabel`、`worker -> executionLabel`、未知 -> processingLabel
-- 扩展 `ui_schema_renderer_test.dart`，断言 card surface、按钮文案、badge 仍可渲染
-
-**Step 2: 跑测试确认失败**
-
-Run: `flutter test test/features/chat/presentation/agent_stage_mapping_test.dart test/features/chat/ui_schema_renderer_test.dart`
-Expected: FAIL（映射函数/新视觉结构尚未实现）
-
-**Step 3: 最小实现使测试通过**
-- 新增 stage 映射文件并接入 chat/home
-- 重构 `UiSchemaRenderer` 的 surface 风格并保持现有 schema 兼容
-
-**Step 4: 再跑测试确认通过**
-
-Run: `flutter test test/features/chat/presentation/agent_stage_mapping_test.dart test/features/chat/ui_schema_renderer_test.dart`
-Expected: PASS
-
-### Task 2: 重构我的账户页面为统一表面语言
-
-**Files:**
-- Modify: `apps/lib/features/settings/ui/screens/account_screen.dart`
-
-**Step 1: 写失败测试（轻量）**
-- 可选新增页面结构测试，断言标题/分组/退出按钮存在
-
-**Step 2: 运行测试确认失败（若新增测试）**
-
-Run: `flutter test test/features/settings/ui/screens/...`
-Expected: FAIL
-
-**Step 3: 最小实现**
-- 用 `AccountSurfaceScaffold` 承载页面
-- 用 `AccountSectionCard` 承载账户菜单与安全操作分组
-- 退出登录按钮改为统一 token 风格
-
-**Step 4: 测试通过**
-
-Run: `flutter test test/features/settings/ui/screens/...`
-Expected: PASS
-
-### Task 3: 协议对齐与主页等待文案修正
-
-**Files:**
-- Modify: `apps/lib/features/chat/presentation/bloc/chat_bloc.dart`
-- Modify: `apps/lib/features/home/ui/screens/home_screen.dart`
-- Create: `apps/lib/features/chat/presentation/bloc/agent_stage.dart`
-
-**Step 1: 保持协议语义**
-- 将 step 映射改为文档约定：`router`、`worker`
-- 显示文案：`意图识别中`、`任务执行中`、`任务处理中`
-
-**Step 2: 运行相关测试**
-
-Run: `flutter test test/features/chat/presentation/agent_stage_mapping_test.dart test/features/home/ui/widgets/home_screen_layout_test.dart`
-Expected: PASS
-
-### Task 4: 全量验证
-
-**Step 1: 静态检查**
-
-Run: `flutter analyze`
-Expected: 无新增错误
-
-**Step 2: 回归测试**
-
-Run: `flutter test`
-Expected: PASS

docs/plans/2026-03-16-calendar-timezone-unification.md

@@ -1,285 +0,0 @@
-# Calendar Timezone Unification Implementation Plan
-
-> **For agentic workers:** REQUIRED: Use superpowers:subagent-driven-development (if subagents available) or superpowers:executing-plans to implement this plan. Steps use checkbox (`- [ ]`) syntax for tracking.
-
-**Goal:** Eliminate calendar time mismatches by enforcing one end-to-end timezone policy across App input, Agent runtime context, tool execution, and UTC database storage.
-
-**Architecture:** Keep database schema unchanged (`start_at/end_at TIMESTAMPTZ + timezone`) and enforce strict runtime normalization. Device timezone is injected from `RunAgentInput.forwardedProps`, resolved into a single `effective_timezone`, then written explicitly into tool arguments and persisted as event timezone while timestamps are stored in UTC. Calendar read responses include deterministic event-timezone-rendered values so frontend rendering is stable and no implicit `toLocal()` conversion remains.
-
-**Tech Stack:** FastAPI, Pydantic v2, AgentScope runtime/tooling, Flutter (Dart), PostgreSQL TIMESTAMPTZ, pytest, Flutter test.
-
----
-
-## Chunk 1: Protocol and Backend Runtime Normalization
-
-### Task 1: Freeze protocol and timezone precedence contract
-
-**Files:**
-- Modify: `docs/protocols/agent/run-agent-input.md`
-- Create: `docs/protocols/calendar/timezone-policy.md`
-
-- [ ] **Step 1: Write protocol delta checklist in docs first**
-
-Document the exact policy:
-- `event_timezone > device_timezone > profile.timezone > UTC`
-- `event_timezone` must be present in final tool call
-- `start_at/end_at` must be timezone-aware
-- DB stores UTC timestamps and IANA timezone string
-
-- [ ] **Step 2: Update RunAgentInput protocol with forwardedProps contract**
-
-Add canonical payload example:
-
-```json
-{
-  "forwardedProps": {
-    "client_time": {
-      "device_timezone": "America/Los_Angeles",
-      "client_now_iso": "2026-03-16T09:12:33-07:00",
-      "client_epoch_ms": 1773658353000
-    }
-  }
-}
-```
-
-- [ ] **Step 3: Add calendar timezone policy protocol doc**
-
-Include:
-- accepted datetime formats
-- explicit error codes
-- write/read response semantics
-- DST handling rule
-
-- [ ] **Step 4: Verify docs consistency**
-
-Run: `cd backend && uv run python -m pytest tests/unit/core/agentscope/test_system_prompt.py -q`
-Expected: PASS (no protocol-breaking prompt assumptions)
-
-
-### Task 2: Parse forwarded device time and compute effective timezone
-
-**Files:**
-- Modify: `backend/src/core/agentscope/schemas/agui_input.py`
-- Modify: `backend/src/core/agentscope/runtime/runner.py`
-- Modify: `backend/src/core/agentscope/prompts/system_prompt.py`
-- Test: `backend/tests/unit/core/agentscope/test_system_prompt.py`
-
-- [ ] **Step 1: Write failing tests for effective timezone resolution**
-
-Add tests covering:
-- forwarded `device_timezone` present -> selected
-- missing forwarded timezone -> fallback profile timezone
-- invalid forwarded timezone -> fallback profile timezone
-
-- [ ] **Step 2: Run tests to confirm RED**
-
-Run: `cd backend && uv run pytest tests/unit/core/agentscope/test_system_prompt.py -k timezone -v`
-Expected: FAIL on new assertions
-
-- [ ] **Step 3: Implement minimal runtime context extraction**
-
-Implement a typed helper in runner path to read:
-- `run_input.forwarded_props.client_time.device_timezone`
-- `client_now_iso`
-- `client_epoch_ms`
-
-Compute `effective_timezone` using fixed precedence and pass it into `build_system_prompt(...)`.
-
-- [ ] **Step 4: Inject effective_timezone into ENV section**
-
-Update `build_system_prompt` env payload to include:
-- `timezone_profile`
-- `timezone_device`
-- `timezone_effective`
-
-Update guidance sentence to resolve ambiguous time with `timezone_effective`.
-
-- [ ] **Step 5: Run tests to confirm GREEN**
-
-Run: `cd backend && uv run pytest tests/unit/core/agentscope/test_system_prompt.py -v`
-Expected: PASS
-
-
-### Task 3: Remove timezone ambiguity and hidden fallbacks from calendar write
-
-**Files:**
-- Modify: `backend/src/core/agentscope/tools/utils/calendar_domain.py`
-- Modify: `backend/src/core/agentscope/tools/custom/calendar.py`
-- Modify: `backend/src/v1/schedule_items/schemas.py`
-- Modify: `backend/src/v1/schedule_items/service.py`
-- Test: `backend/tests/unit/core/agentscope/test_calendar_tools.py`
-- Test: `backend/tests/unit/v1/schedule_items/test_schemas.py`
-- Test: `backend/tests/unit/v1/schedule_items/test_service.py`
-
-- [ ] **Step 1: Write failing tests for forbidden naive datetime and required timezone**
-
-Add tests for:
-- naive `start_at` rejected
-- missing `event_timezone` rejected in tool path
-- parse failure does not fallback to `now + 1h`
-
-- [ ] **Step 2: Run tests to confirm RED**
-
-Run: `cd backend && uv run pytest tests/unit/core/agentscope/test_calendar_tools.py tests/unit/v1/schedule_items/test_schemas.py -v`
-Expected: FAIL on new constraints
-
-- [ ] **Step 3: Implement strict parsing and normalization**
-
-Implementation requirements:
-- `parse_iso_datetime` rejects naive input
-- remove default `Asia/Shanghai` in tool
-- remove fallback auto-generated start time
-- validate IANA timezone and normalize `start_at/end_at` to UTC before persistence
-
-- [ ] **Step 4: Enforce service-level invariants**
-
-Service invariant set:
-- timezone non-empty and valid IANA
-- `end_at is None or end_at >= start_at`
-
-- [ ] **Step 5: Run backend tests**
-
-Run: `cd backend && uv run pytest tests/unit/core/agentscope/test_calendar_tools.py tests/unit/v1/schedule_items/test_schemas.py tests/unit/v1/schedule_items/test_service.py tests/integration/test_schedule_items_routes.py -v`
-Expected: PASS
-
-
-### Task 4: Keep DB schema, add non-breaking constraint migration only
-
-**Files:**
-- Create: `backend/alembic/versions/20260316_000x_schedule_items_time_constraints.py`
-- Test: `backend/tests/integration/test_schedule_items_routes.py`
-
-- [ ] **Step 1: Write migration test expectation first**
-
-Add/extend integration assertion for invalid `end_at < start_at` returning 422.
-
-- [ ] **Step 2: Run integration test to confirm RED**
-
-Run: `cd backend && uv run pytest tests/integration/test_schedule_items_routes.py -k end_at -v`
-Expected: FAIL
-
-- [ ] **Step 3: Implement migration with CHECK only (no new columns)**
-
-Migration includes:
-- `CHECK (end_at IS NULL OR end_at >= start_at)`
-
-- [ ] **Step 4: Run migration + integration test**
-
-Run: `cd backend && uv run alembic upgrade head && uv run pytest tests/integration/test_schedule_items_routes.py -v`
-Expected: PASS
-
----
-
-## Chunk 2: Frontend Deterministic Display and Agent Input Wiring
-
-### Task 5: Wire device timezone into RunAgentInput forwardedProps
-
-**Files:**
-- Modify: `apps/lib/features/chat/data/services/ag_ui_service.dart`
-- Modify: `apps/lib/features/chat/data/models/ag_ui_event.dart` (only if serialization helper is needed)
-- Test: `apps/test/features/chat/ag_ui_event_test.dart`
-
-- [ ] **Step 1: Write failing test for forwarded client_time payload**
-
-Assert outgoing run request contains:
-- `forwardedProps.client_time.device_timezone`
-- `client_now_iso`
-- `client_epoch_ms`
-
-- [ ] **Step 2: Run test to confirm RED**
-
-Run: `cd apps && flutter test test/features/chat/ag_ui_event_test.dart`
-Expected: FAIL
-
-- [ ] **Step 3: Implement payload injection in one place**
-
-Add a single helper to build client time context and attach it to run input requests.
-
-- [ ] **Step 4: Run test to confirm GREEN**
-
-Run: `cd apps && flutter test test/features/chat/ag_ui_event_test.dart`
-Expected: PASS
-
-
-### Task 6: Remove implicit local-time rendering and render by event timezone
-
-**Files:**
-- Modify: `apps/lib/features/calendar/data/models/schedule_item_model.dart`
-- Modify: `apps/lib/features/calendar/ui/screens/calendar_event_detail_screen.dart`
-- Modify: `apps/lib/features/calendar/ui/screens/calendar_dayweek_screen.dart`
-- Modify: `apps/lib/features/calendar/ui/screens/calendar_month_screen.dart`
-- Modify: `apps/lib/features/messages/ui/widgets/calendar_message_card.dart`
-- Modify: `apps/lib/features/calendar/ui/widgets/create_event_sheet.dart`
-- Test: `apps/test/features/calendar/ui/calendar_time_utils_test.dart`
-- Test: `apps/test/features/calendar/ui/create_event_sheet_time_align_test.dart`
-
-- [ ] **Step 1: Write failing tests for timezone-specific rendering**
-
-Cover cases:
-- same UTC event shows different local clock time under different `event.timezone`
-- list/day/week/month are consistent for one event
-- create sheet sends explicit timezone in payload
-
-- [ ] **Step 2: Run tests to confirm RED**
-
-Run: `cd apps && flutter test test/features/calendar/ui/calendar_time_utils_test.dart test/features/calendar/ui/create_event_sheet_time_align_test.dart`
-Expected: FAIL
-
-- [ ] **Step 3: Implement deterministic time conversion utility**
-
-Implement one utility used by all calendar UI surfaces:
-- input: UTC datetime + IANA timezone
-- output: event-local datetime
-
-Replace direct `.toLocal()` usage in calendar model/view with this utility.
-
-- [ ] **Step 4: Enforce explicit timezone on create/update payload**
-
-Create/update must always include `timezone` field from selected event timezone.
-
-- [ ] **Step 5: Run Flutter tests**
-
-Run: `cd apps && flutter test test/features/calendar/ui/calendar_time_utils_test.dart test/features/calendar/ui/create_event_sheet_time_align_test.dart`
-Expected: PASS
-
-
-### Task 7: End-to-end verification matrix and release checklist
-
-**Files:**
-- Modify: `docs/plans/timezone-e2e-checklist.md`
-- Test: `backend/tests/integration/test_schedule_items_routes.py`
-- Test: `apps/test/features/calendar/ui/create_event_sheet_time_align_test.dart`
-
-- [ ] **Step 1: Add reproducible matrix**
-
-Matrix axes:
-- device timezone: `America/Los_Angeles`, `Asia/Shanghai`
-- profile timezone: `Asia/Shanghai`, `Europe/Paris`
-- explicit event timezone: `Asia/Tokyo`
-
-- [ ] **Step 2: Run backend + frontend verification commands**
-
-Run:
-- `cd backend && uv run pytest tests/unit/core/agentscope/test_system_prompt.py tests/unit/core/agentscope/test_calendar_tools.py tests/integration/test_schedule_items_routes.py -v`
-- `cd apps && flutter test test/features/chat/ag_ui_event_test.dart test/features/calendar/ui/calendar_time_utils_test.dart test/features/calendar/ui/create_event_sheet_time_align_test.dart`
-
-Expected: all PASS
-
-- [ ] **Step 3: Manual scenario check**
-
-Manual script:
-1. device timezone set to Los Angeles
-2. profile timezone set to Shanghai
-3. ask agent create "明天上午9点开会"
-4. verify assistant text, tool card, DB UTC value, and calendar detail all align to chosen event timezone semantics
-
-- [ ] **Step 4: Capture release notes**
-
-Record:
-- removed hidden timezone defaults
-- deterministic precedence
-- no schema expansion
-
----
-
-Plan complete and saved to `docs/superpowers/plans/2026-03-16-calendar-timezone-unification.md`. Ready to execute?

docs/plans/2026-03-16-settings-account-subpages-design.md

@@ -1,99 +0,0 @@
-# 设置-账户子页面视觉重构设计（编辑资料 / 修改密码）
-
-## 背景与目标
-
-当前 `编辑资料` 与 `修改密码` 子页面存在明显的“文档页/默认表单堆叠”观感，未满足 `apps/rules/visual_design_language.md` 要求的“高级、柔和、分层、助手产品感”。
-
-本次设计目标：
-
-- 延续项目现有蓝灰体系，不改变品牌语气
-- 建立清晰的表面层级：背景层 -> 主内容层 -> 次级分组层 -> 交互强调层
-- 将 `修改密码` 对齐 `忘记密码` 的设计语言（分段、状态提示、密码输入体验），但保持“设置域”语义
-- 严格使用 `AppColors` / `AppSpacing` / `AppRadius`，移除新增硬编码视觉值
-
-## 约束来源
-
-- `apps/AGENTS.md`
-- `apps/rules/visual_design_language.md`
-- 现有共享组件体系（`AppButton`、`AppBanner`、`FixedLengthCodeInput`、Toast 系统）
-
-## 方案选择
-
-已确认采用方案 B：设置域统一壳层。
-
-方案要点：
-
-- 在 settings feature 内建立可复用的页面壳与分组卡片语义
-- 两个子页面共享同一结构语言与间距节奏
-- `修改密码` 采用步骤型结构（Step 1 验证邮箱 / Step 2 设置新密码）
-
-## 信息架构
-
-### 编辑资料
-
-- 页面头部：返回 + 标题 + 简短辅助说明
-- 主内容卡：
-  - 资料概览组（头像占位、当前账号信息）
-  - 基础信息组（用户名输入）
-  - 个人简介组（多行输入 + 字数）
-- 底部强调操作区：`保存修改` 按钮（仅在有变更且可提交时可用）
-
-### 修改密码（步骤型）
-
-- 页面头部：返回 + 标题 + 简短辅助说明
-- Step 1：邮箱验证与验证码发送
-  - 展示当前邮箱
-  - 发送/重发验证码 + 倒计时
-- Step 2：输入验证码并设置新密码
-  - 验证码输入
-  - 新密码、确认密码
-  - 状态提示（使用 `AppBanner`）
-- 底部强调操作区：`确认修改`
-
-## 视觉与交互规则
-
-- 背景：`AppColors.surfaceSecondary`
-- 主卡：`AppColors.white` + `AppColors.borderSecondary` + 圆角 `AppRadius.lg/xl`
-- 次级分组：`AppColors.surfaceTertiary/surfaceInfoLight`（按语义使用）+ `AppColors.borderTertiary`
-- 输入聚焦态：`AppColors.blue500`
-- 间距节奏：
-  - 页面外边距：`AppSpacing.xl`
-  - 组内：`AppSpacing.sm/lg`
-  - 组间：`AppSpacing.xxl`
-- 动效策略：仅保留柔和状态反馈（按钮 loading/disabled、步骤区显隐、输入 focus），不添加噱头动画
-
-## 组件与复用策略
-
-- 优先复用：
-  - `AppButton`
-  - `FixedLengthCodeInput`
-  - `AppBanner`
-  - Toast 系统
-  - `PasswordField`（与忘记密码页统一密码输入体验）
-- 在 settings 域新增可复用 UI：
-  - `account_surface_scaffold.dart`
-  - `account_section_card.dart`
-
-## 影响文件
-
-- 修改：`apps/lib/features/settings/ui/screens/edit_profile_screen.dart`
-- 修改：`apps/lib/features/settings/ui/screens/change_password_screen.dart`
-- 新增：`apps/lib/features/settings/ui/widgets/account_surface_scaffold.dart`
-- 新增：`apps/lib/features/settings/ui/widgets/account_section_card.dart`
-- 可能补充：`apps/lib/core/theme/design_tokens.dart`（仅缺失 token 时）
-
-## 验收标准
-
-- 两个页面具备一致的“柔和卡片分层”结构
-- 修改密码页面与忘记密码页面在设计语言上可感知一致
-- 不出现“纯白文档页/后台表单页”观感
-- 不新增视觉硬编码，遵守 token 体系
-- 核心交互逻辑不回归（验证码、倒计时、提交、错误提示）
-
-## 验证计划
-
-- `flutter analyze`
-- `flutter test`（至少覆盖 auth/reset-password 相关与 settings 相关用例）
-- 手工验证：
-  - 编辑资料：无改动禁用、改动后可保存、成功后返回
-  - 修改密码：发送验证码、倒计时、错误提示、成功返回

docs/plans/2026-03-16-settings-account-subpages-implementation.md

@@ -1,151 +0,0 @@
-# 设置账户子页面视觉重构 Implementation Plan
-
-> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
-
-**Goal:** 重构设置页中的编辑资料与修改密码子页面，使其符合视觉设计语言并统一为柔和卡片分层风格。
-
-**Architecture:** 在 settings feature 内新增复用型页面壳与分组卡组件，统一两页的表面层级、间距节奏与交互强调区；修改密码页面借鉴忘记密码页面的步骤分段与状态提示语义，但保留设置域信息架构与文案。保持现有业务逻辑不变，优先做样式与结构重构。
-
-**Tech Stack:** Flutter, go_router, flutter_bloc, formz, design tokens (`AppColors/AppSpacing/AppRadius`), shared widgets (`AppButton`, `AppBanner`, `FixedLengthCodeInput`, Toast)
-
----
-
-### Task 1: 创建设置域复用 UI 壳层组件
-
-**Files:**
-- Create: `apps/lib/features/settings/ui/widgets/account_surface_scaffold.dart`
-- Create: `apps/lib/features/settings/ui/widgets/account_section_card.dart`
-
-**Step 1: 写一个失败的基础渲染测试（可选，若项目已有 widget test 基础）**
-
-```dart
-testWidgets('AccountSectionCard renders title and child', (tester) async {
-  await tester.pumpWidget(...);
-  expect(find.text('标题'), findsOneWidget);
-});
-```
-
-**Step 2: 运行测试确认失败（若执行 Step 1）**
-
-Run: `flutter test apps/test/... -r expanded`
-Expected: FAIL（新组件不存在）
-
-**Step 3: 实现最小组件**
-
-- `AccountSurfaceScaffold`：统一背景、头部、滚动主内容、可选底部强调区
-- `AccountSectionCard`：统一分组卡片容器（标题、描述、child）
-
-**Step 4: 运行测试确认通过（若执行 Step 1）**
-
-Run: `flutter test apps/test/... -r expanded`
-Expected: PASS
-
-**Step 5: Commit**
-
-```bash
-git add apps/lib/features/settings/ui/widgets/account_surface_scaffold.dart apps/lib/features/settings/ui/widgets/account_section_card.dart
-git commit -m "feat: add reusable account surface widgets"
-```
-
-### Task 2: 重构编辑资料页面为柔和卡片分层
-
-**Files:**
-- Modify: `apps/lib/features/settings/ui/screens/edit_profile_screen.dart`
-
-**Step 1: 写失败测试（仅在当前测试基建允许时）**
-
-```dart
-testWidgets('Edit profile shows grouped sections and disabled save initially', (tester) async {
-  // verify section titles and save button disabled by default
-});
-```
-
-**Step 2: 运行测试确认失败（若执行 Step 1）**
-
-Run: `flutter test apps/test/... -r expanded`
-Expected: FAIL（旧结构不满足）
-
-**Step 3: 实现最小重构**
-
-- 接入 `AccountSurfaceScaffold`
-- 将头像区、用户名区、简介区改为分组卡语义
-- 统一输入风格（token 化）
-- 底部强调操作区承载 `保存修改`
-
-**Step 4: 运行测试确认通过（若执行 Step 1）**
-
-Run: `flutter test apps/test/... -r expanded`
-Expected: PASS
-
-**Step 5: Commit**
-
-```bash
-git add apps/lib/features/settings/ui/screens/edit_profile_screen.dart
-git commit -m "feat: redesign edit profile screen with layered surfaces"
-```
-
-### Task 3: 重构修改密码页面为步骤型结构
-
-**Files:**
-- Modify: `apps/lib/features/settings/ui/screens/change_password_screen.dart`
-- Reference: `apps/lib/features/auth/ui/screens/reset_password_screen.dart`
-
-**Step 1: 写失败测试（仅在当前测试基建允许时）**
-
-```dart
-testWidgets('Change password renders step sections and submit state', (tester) async {
-  // verify step titles and CTA disabled before code sent
-});
-```
-
-**Step 2: 运行测试确认失败（若执行 Step 1）**
-
-Run: `flutter test apps/test/... -r expanded`
-Expected: FAIL（步骤结构不存在）
-
-**Step 3: 实现最小重构**
-
-- 接入 `AccountSurfaceScaffold` + `AccountSectionCard`
-- 按 Step 1/Step 2 分段展示（邮箱验证 -> 验证码与新密码）
-- 引入 `AppBanner` 做状态提示
-- 保持验证码、倒计时、提交逻辑不变
-
-**Step 4: 运行测试确认通过（若执行 Step 1）**
-
-Run: `flutter test apps/test/... -r expanded`
-Expected: PASS
-
-**Step 5: Commit**
-
-```bash
-git add apps/lib/features/settings/ui/screens/change_password_screen.dart
-git commit -m "feat: redesign change password flow with step-based sections"
-```
-
-### Task 4: 验证与文档同步
-
-**Files:**
-- Modify (if needed): `docs/plans/2026-03-16-settings-account-subpages-design.md`
-
-**Step 1: 运行静态检查**
-
-Run: `flutter analyze`
-Expected: PASS
-
-**Step 2: 运行回归测试**
-
-Run: `flutter test`
-Expected: PASS
-
-**Step 3: 手工验收**
-
-- 编辑资料：无变更禁用、有变更可保存
-- 修改密码：发码、倒计时、错误提示、成功返回
-- 视觉层级：背景/主卡/分组/强调区清晰
-
-**Step 4: Commit**
-
-```bash
-git add docs/plans/2026-03-16-settings-account-subpages-design.md
-git commit -m "docs: finalize account subpages redesign validation notes"
-```

docs/plans/2026-03-17-automation-memory-design.md

@@ -0,0 +1,227 @@
+# 自动化记忆任务设计方案（v1）
+
+## 1. 背景与目标
+
+本方案用于落地后端自动化记忆任务：
+
+- 提供每日/每周自动执行能力；
+- 自动提取用户近期对话中的可沉淀记忆并写入 `memories`；
+- 支持“忘记”操作；
+- 复用现有 Agent 运行链路，避免并行维护两套运行时；
+- 保证可审计、可追溯，同时对用户界面隐藏自动输入提示词。
+
+## 2. 范围与非目标
+
+### 2.1 范围
+
+- `automation_jobs` 增加 `config`（JSONB，强约束）；
+- 调度器执行 `automation_jobs`，按用户本地时区计算执行时间；
+- 自动任务复用现有 `router -> agent` 运行策略；
+- 自动输入落库但对用户不可见；
+- 工具集按任务配置分发，仅允许白名单工具；
+- 正式启用 `memory_prompt`，将 `memories` 注入系统提示词。
+
+### 2.2 非目标
+
+- 不开放复杂 DSL 编排；
+- 不开放细粒度策略配置（如 `execution_profile/history_window_days/memory_policy`）；
+- 不引入第二套独立 Agent 运行框架。
+
+## 3. 关键决策
+
+1. **调度与执行分离**
+   - `run_at/next_run_at/timezone` 负责“何时执行”；
+   - `config` 负责“如何执行”（提示词与工具集）。
+
+2. **配置最小化**
+   - `automation_jobs.config` 仅保留：
+     - `prompt: string`
+     - `tools: string[]`
+
+3. **自动输入可审计但用户不可见**
+   - 自动任务提示词作为“用户消息”写入 `messages`；
+   - 在 metadata 加可见性标记并在用户历史接口中默认过滤；
+   - 审计路径可查询完整消息链路。
+
+4. **时区策略**
+   - 调度语义采用“用户本地时区 10:00”；
+   - 存储执行时间统一为 UTC 的 `next_run_at`；
+   - 调度器仅按 UTC 扫描，降低运维复杂度。
+
+## 4. 数据模型设计
+
+## 4.1 `automation_jobs` 调整
+
+- 保留字段：`id`, `owner_id`, `title`, `schedule_type`, `run_at`, `next_run_at`, `timezone`, `last_run_at`, `status` 等；
+- 新增字段：`config JSONB NOT NULL`；
+- 迁移策略：
+  - 将历史 `prompt` 迁移到 `config.prompt`；
+  - 切换业务代码后删除旧 `prompt` 字段（可分两步灰度）。
+
+### 4.2 `config` 强约束 schema
+
+```json
+{
+  "prompt": "提取最近对话中的稳定记忆并写入",
+  "tools": ["memory_write", "memory_forget"]
+}
+```
+
+约束要求：
+
+- `prompt`：必填、非空、长度受限（建议 <= 2000）；
+- `tools`：必填数组，元素为注册工具名；
+- `extra="forbid"`，拒绝未知字段；
+- 执行前二次校验：`tools` 必须与后端 allowlist 取交集，禁止越权。
+
+### 4.3 `messages.metadata` 扩展
+
+新增建议字段：
+
+- `hidden_from_user: bool`
+- `origin: "chat" | "automation"`
+
+用途：
+
+- `hidden_from_user=true` 的消息在用户历史默认不可见；
+- 审计查询可读取全量。
+
+## 5. 运行时架构
+
+## 5.1 复用现有 Agent 链路
+
+自动任务不新建运行时，沿用现有 `router -> orchestrator -> worker` 流程：
+
+- 输入：来自 `automation_jobs.config.prompt` 的合成“用户输入”；
+- 上下文：默认加载“今天 + 昨天”历史；
+- 工具：仅启用 `config.tools` 且通过 allowlist 校验。
+
+## 5.2 会话与消息策略
+
+- 自动任务写入独立 `session_type=automation` 会话，并关联 `job_id`；
+- 普通对话维持 `session_type=chat`；
+- 自动输入消息：`role=user` 但 `hidden_from_user=true`；
+- Assistant 输出消息：默认可见。
+
+## 6. 调度器设计
+
+## 6.1 扫描与触发
+
+- 周期扫描条件：
+  - `status='active'`
+  - `next_run_at <= now_utc`
+- 命中后入队执行命令（复用现有任务队列）。
+
+## 6.2 幂等与并发控制
+
+- 使用任务槽位去重键：`job_id + scheduled_slot`；
+- 使用行级抢占或原子更新防止并发重复执行；
+- 失败可重试并记录错误上下文。
+
+## 6.3 下次执行时间计算
+
+- 每次执行完成后按 `schedule_type + timezone` 计算下一次触发时间；
+- 保存到 `next_run_at`（UTC）；
+- `last_run_at` 记录实际执行时间。
+
+## 7. 默认任务创建
+
+用户创建后自动创建一条默认 daily 记忆任务：
+
+- `schedule_type = 'daily'`
+- 本地时区目标时间：10:00
+- `timezone`：优先用户设置时区，缺省 `Asia/Shanghai`
+- `config.prompt`：内置记忆提取提示词模板
+- `config.tools = ["memory_write", "memory_forget"]`
+
+## 8. 记忆工具设计
+
+## 8.1 `memory_write`
+
+- 用途：写入/更新用户记忆；
+- 输入：受 schema 限制（标题、内容、来源等）；
+- 安全：owner 作用域强制绑定，不信任模型输入的 owner 信息。
+
+## 8.2 `memory_forget`
+
+- 用途：失效或删除用户记忆；
+- 输入：`memory_id` 或受限条件；
+- 安全：仅允许当前 owner 范围内操作。
+
+## 8.3 工具授权
+
+- 任务声明工具集来自 `config.tools`；
+- 运行时执行 `declared_tools ∩ allowlist`；
+- 非白名单工具直接拒绝并记录审计日志。
+
+## 9. `memory_prompt` 启用策略
+
+- 从数据库读取当前用户可用 `memories`；
+- 组装为系统提示词 memory section；
+- 设置条数与长度上限，避免 token 膨胀；
+- 自动任务与普通对话统一使用该注入机制。
+
+## 10. 前端展示策略
+
+- 历史消息接口默认过滤 `hidden_from_user=true`；
+- 用户仅看到 assistant 输出（以及可选系统摘要，不包含原始自动 prompt）；
+- 前端无需大量工具分支 if/else，可统一按 metadata 标记处理。
+
+## 11. 安全与审计
+
+- 关键要求：
+  - 禁止越权工具调用；
+  - 禁止跨用户 memory 读写；
+  - 日志不输出敏感数据和完整私密上下文。
+- 审计能力：
+  - 自动输入、工具调用、输出全链路可追踪；
+  - 用户界面只显示允许可见内容。
+
+## 12. 迁移与发布步骤
+
+1. 新增 `config` 字段与 schema 代码；
+2. 数据迁移：旧 `prompt -> config.prompt`；
+3. 运行时切换到 `config.prompt/config.tools`；
+4. 灰度验证后移除旧 `prompt`；
+5. 上线调度器与默认任务创建；
+6. 启用 `memory_prompt` 并完成联调。
+
+## 13. 测试与验收
+
+### 13.1 单元测试
+
+- `config` schema 校验（非法字段、非法工具、空 prompt）；
+- `next_run_at` 计算（跨时区、边界时间）；
+- 隐藏消息过滤逻辑；
+- `memory_write/memory_forget` owner 边界。
+
+### 13.2 集成测试
+
+- 用户创建触发默认 daily 任务创建；
+- 调度触发 -> Agent 执行 -> memory 写入 -> `next_run_at` 更新；
+- 自动输入在审计可见、在用户历史不可见。
+
+### 13.3 验收标准
+
+- 每日 10:00（用户本地时区）稳定执行；
+- 自动化链路有完整审计；
+- 工具授权边界有效；
+- 前端历史展示符合“隐藏自动输入、展示输出”的预期。
+
+## 14. 实施顺序建议
+
+1. 先更新协议文档（`docs/protocols`）明确契约；
+2. 数据库与 schema 改造（`automation_jobs.config`）；
+3. 工具集与授权边界实现；
+4. 调度器与默认任务创建；
+5. `memory_prompt` 注入与端到端验证；
+6. 前端过滤逻辑收口。
+
+---
+
+本方案是最小可行版本（MVP）设计，重点保障：
+
+- 复用现有架构；
+- 低配置复杂度；
+- 可审计与可维护并存；
+- 为后续扩展保留演进空间。

docs/todo/2026-03-17-asr.md

@@ -0,0 +1 @@
+- 语音识别计费

docs/todo/2026-03-17-worker-token-latency-optimization.md

@@ -0,0 +1,41 @@
+# Worker Token/Latency Optimization TODO
+
+Date: 2026-03-17
+Owner: backend runtime
+Status: pending
+
+## Background
+
+- Router cost/latency is acceptable.
+- Worker stage (deepseek-chat) has significantly higher input tokens and latency.
+- Current optimization work is deferred due to prioritization.
+
+## Observations (from `public.messages`)
+
+- Worker avg input tokens are much higher than router (about 12k+ vs 3k).
+- Worker avg latency is much higher than router (about 41s vs 4s).
+- Worker cost dominates total cost.
+
+## Root Cause Hypothesis
+
+- Worker ReAct path repeatedly includes full tool schemas per model call.
+- `calendar_write` tool schema is large and contributes major prompt overhead.
+- Finalize JSON step performs an additional model call after ReAct.
+
+## Deferred Optimization Items
+
+1. Tool schema slimming for calendar write path.
+   - Split `calendar_write` into focused tools (`calendar_create`, `calendar_update`, `calendar_delete`).
+   - Reduce redundant/verbose field descriptions where possible.
+2. Dynamic tool set exposure by routed intent.
+   - Only expose tools needed for current task.
+3. Evaluate finalize overhead.
+   - Verify whether finalize call can be reduced or replaced in specific flows.
+4. Add before/after benchmark script.
+   - Compare worker `input_tokens`, `latency_ms`, and `cost` for the same scripted multi-turn scenario.
+
+## Acceptance Metrics (target)
+
+- Reduce worker input tokens by >= 30% in multi-turn calendar CRUD scenario.
+- Reduce worker p95 latency by >= 25%.
+- Keep functional behavior unchanged for agent runs.

infra/docker/docker-compose.yml

@@ -19,50 +19,6 @@ services:
             timeout: 3s
             retries: 5
 
-    litellm-config-job:
-        build:
-            context: ../..
-            dockerfile: backend/Dockerfile
-        image: social-local-backend
-        container_name: social-local-litellm-config-job
-        restart: "no"
-        env_file:
-            - ../../.env
-        environment:
-            - PYTHONPATH=/app/backend/src
-        command: >
-            uv run python backend/scripts/build_litellm_proxy_config.py --output /config/litellm-proxy-config.yaml
-        volumes:
-            - litellm_config:/config
-        depends_on:
-            redis:
-                condition: service_healthy
-
-    litellm:
-        image: ghcr.io/berriai/litellm@sha256:b959a1816fa454a14d2842242d0fa1cd0d39f96fc94d3a1f4e1de4e48e2398c6
-        container_name: social-local-litellm
-        restart: unless-stopped
-        env_file:
-            - ../../.env
-        ports:
-            - "${SOCIAL_LITELLM__PORT:-3875}:${SOCIAL_LITELLM__PORT:-3875}"
-        volumes:
-            - litellm_config:/config:ro
-        command:
-            [
-                "--config",
-                "/config/litellm-proxy-config.yaml",
-                "--host",
-                "0.0.0.0",
-                "--port",
-                "${SOCIAL_LITELLM__PORT:-3875}",
-            ]
-        depends_on:
-            redis:
-                condition: service_healthy
-            litellm-config-job:
-                condition: service_completed_successfully
-
     init-job:
         build:
             context: ../..
@@ -91,4 +47,3 @@ services:
 
 volumes:
     redis_data:
-    litellm_config:

infra/scripts/app.sh

@@ -116,7 +116,7 @@ kill_listening_processes() {
 start() {
   echo "=== App Up ==="
   echo "This script starts local web + worker processes in tmux."
-  echo "LiteLLM/Redis should be managed separately as long-running docker services."
+  echo "Redis should be managed separately as a long-running docker service."
   echo "NOTE: Bootstrap (migrate + init-data) must be run separately."
   echo ""
 
@@ -184,7 +184,7 @@ ${SOCIAL_WEB__WORKERS:-2} --log-level ${UVICORN_LOG_LEVEL}"
 
 stop() {
   echo "=== App Down ==="
-  echo "Stopping tmux app processes (docker redis/litellm are not managed here)."
+  echo "Stopping tmux app processes (docker redis is not managed here)."
   load_env_if_exists
   WEB_PORT="${SOCIAL_WEB__PORT:-5775}"