diff --git a/src/api/__tests__/manage.spec.ts b/src/api/__tests__/manage.spec.ts
new file mode 100644
index 00000000..579cb83d
--- /dev/null
+++ b/src/api/__tests__/manage.spec.ts
@@ -0,0 +1,55 @@
+import { manageLlmProvider, manageNotificationChannel, manageStorage } from '@/api/manage'
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const mocks = vi.hoisted(() => ({
+  apiPost: vi.fn(),
+}))
+
+vi.mock('@/api', () => ({
+  default: createDataApiMock({
+    post: (...args: unknown[]) => mocks.apiPost(...args),
+  }),
+}))
+
+describe('manage API adapters', () => {
+  beforeEach(() => {
+    mocks.apiPost.mockReset()
+    mocks.apiPost.mockResolvedValue({ ok: true })
+  })
+
+  it('透传通知渠道的目标、动作、参数和请求配置', async () => {
+    const config = { timeout: 1200 }
+
+    await expect(manageNotificationChannel('telegram', 'refresh', { force: true }, config)).resolves.toEqual({
+      ok: true,
+    })
+
+    expect(mocks.apiPost).toHaveBeenCalledWith(
+      'notification/manage',
+      { target: 'telegram', action: 'refresh', params: { force: true } },
+      config,
+    )
+  })
+
+  it('为网盘存储动作补充空参数对象', async () => {
+    await manageStorage('alist', 'disconnect')
+
+    expect(mocks.apiPost).toHaveBeenCalledWith(
+      'storage/manage',
+      { target: 'alist', action: 'disconnect', params: {} },
+      undefined,
+    )
+  })
+
+  it('透传 LLM Provider 管理动作并返回已解包业务数据', async () => {
+    mocks.apiPost.mockResolvedValueOnce([{ id: 'openai' }])
+
+    await expect(manageLlmProvider('', 'list_providers')).resolves.toEqual([{ id: 'openai' }])
+
+    expect(mocks.apiPost).toHaveBeenCalledWith(
+      'llm/manage',
+      { target: '', action: 'list_providers', params: {} },
+      undefined,
+    )
+  })
+})
diff --git a/src/composables/__tests__/useLlmProviderDirectory.spec.ts b/src/composables/__tests__/useLlmProviderDirectory.spec.ts
index f2e739f9..d1917977 100644
--- a/src/composables/__tests__/useLlmProviderDirectory.spec.ts
+++ b/src/composables/__tests__/useLlmProviderDirectory.spec.ts
@@ -1,23 +1,28 @@
-import { useLlmProviderDirectory } from '@/composables/useLlmProviderDirectory'
-import { mount } from '@vue/test-utils'
-import { defineComponent, nextTick, ref } from 'vue'
-import { beforeEach, describe, expect, it, vi } from 'vitest'
+import {
+  useLlmProviderDirectory,
+  type LlmModel,
+  type LlmProvider,
+  type LlmProviderAuthSession,
+} from '@/composables/useLlmProviderDirectory'
+import { mount, type VueWrapper } from '@vue/test-utils'
+import { defineComponent, h, nextTick, ref } from 'vue'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 
 const mocks = vi.hoisted(() => ({
-  apiPost: vi.fn(),
+  manageLlmProvider: vi.fn(),
 }))
 
-vi.mock('@/api', () => ({
-  default: createDataApiMock({
-    post: (...args: unknown[]) => mocks.apiPost(...args),
-  }),
+vi.mock('@/api/manage', () => ({
+  manageLlmProvider: (...args: unknown[]) => mocks.manageLlmProvider(...args),
 }))
 
-function createProvider(id: string, runtime: string) {
+const wrappers: VueWrapper[] = []
+
+function createProvider(id: string, overrides: Partial<LlmProvider> = {}): LlmProvider {
   return {
     id,
     name: id,
-    runtime,
+    runtime: 'openai_compatible',
     default_base_url: '',
     base_url_presets: [],
     base_url_editable: true,
@@ -28,112 +33,542 @@ function createProvider(id: string, runtime: string) {
     supports_model_refresh: true,
     oauth_methods: [],
     auth_status: { connected: false },
+    ...overrides,
+  }
+}
+
+function createModel(id: string, overrides: Partial<LlmModel> = {}): LlmModel {
+  return {
+    id,
+    name: id,
+    ...overrides,
+  }
+}
+
+function mountDirectory(
+  initial: Partial<{
+    provider: string
+    apiKey: string
+    baseUrl: string
+    baseUrlPreset: string
+    useProxy: boolean
+    userAgent: string
+    apiProtocol: string
+    model: string
+    maxContextTokens: number
+    authConnected: boolean
+  }> = {},
+) {
+  const provider = ref(initial.provider ?? '')
+  const apiKey = ref(initial.apiKey ?? '')
+  const baseUrl = ref(initial.baseUrl ?? '')
+  const baseUrlPreset = ref(initial.baseUrlPreset ?? '')
+  const useProxy = ref(initial.useProxy ?? false)
+  const userAgent = ref(initial.userAgent ?? '')
+  const apiProtocol = ref(initial.apiProtocol ?? 'auto')
+  const model = ref(initial.model ?? '')
+  const maxContextTokens = ref(initial.maxContextTokens ?? 0)
+  const authConnected = ref(initial.authConnected ?? false)
+  let directory!: ReturnType<typeof useLlmProviderDirectory>
+
+  const wrapper = mount(
+    defineComponent({
+      setup() {
+        directory = useLlmProviderDirectory({
+          provider,
+          apiKey,
+          baseUrl,
+          baseUrlPreset,
+          useProxy,
+          userAgent,
+          apiProtocol,
+          model,
+          maxContextTokens,
+          authConnected,
+        })
+        return () => h('div')
+      },
+    }),
+  )
+  wrappers.push(wrapper)
+
+  return {
+    wrapper,
+    directory,
+    provider,
+    apiKey,
+    baseUrl,
+    baseUrlPreset,
+    useProxy,
+    userAgent,
+    apiProtocol,
+    model,
+    maxContextTokens,
+    authConnected,
   }
 }
 
 describe('useLlmProviderDirectory', () => {
   beforeEach(() => {
-    mocks.apiPost.mockReset()
+    mocks.manageLlmProvider.mockReset()
   })
 
-  it('只为 OpenAI 兼容 runtime 或声明 Responses 工具能力的模型显示 API 协议字段', async () => {
-    mocks.apiPost.mockResolvedValue({
-      success: true,
-      data: [
-        createProvider('openai', 'openai_compatible'),
-        createProvider('deepseek', 'deepseek'),
-        createProvider('google', 'google'),
-      ],
+  afterEach(() => {
+    for (const wrapper of wrappers.splice(0)) {
+      if (wrapper.exists()) wrapper.unmount()
+    }
+  })
+
+  it('选择首个 Provider，并回填默认 URL、预设和认证状态', async () => {
+    const provider = createProvider('openai', {
+      default_base_url: 'https://api.example.com/v1',
+      base_url_presets: [{ id: 'official', label: '官方', value: 'https://api.example.com/v1' }],
+      auth_status: { connected: true, label: 'demo@example.com' },
+    })
+    mocks.manageLlmProvider.mockResolvedValue([provider])
+    const state = mountDirectory({ provider: 'missing' })
+
+    await expect(state.directory.loadProviders()).resolves.toEqual([provider])
+
+    expect(mocks.manageLlmProvider).toHaveBeenCalledWith('', 'list_providers')
+    expect(state.provider.value).toBe('openai')
+    expect(state.baseUrl.value).toBe('https://api.example.com/v1')
+    expect(state.baseUrlPreset.value).toBe('official')
+    expect(state.authConnected.value).toBe(true)
+    expect(state.directory.providerItems.value).toEqual([{ title: 'openai', value: 'openai' }])
+    expect(state.directory.loadingProviders.value).toBe(false)
+  })
+
+  it('默认保留已有 URL 和预设，并可显式恢复 Provider 默认值', async () => {
+    const provider = createProvider('openai', {
+      default_base_url: 'https://default.example.com/v1',
+      base_url_presets: [{ id: 'official', label: '官方', value: 'https://default.example.com/v1' }],
+    })
+    mocks.manageLlmProvider.mockResolvedValue([provider])
+    const state = mountDirectory({
+      provider: 'openai',
+      baseUrl: 'https://custom.example.com/v1',
+      baseUrlPreset: 'custom',
     })
 
-    const Harness = defineComponent({
-      setup() {
-        const provider = ref('openai')
-        const model = ref('')
-        const directory = useLlmProviderDirectory({
-          provider,
-          apiKey: ref(''),
-          baseUrl: ref(''),
-          model,
-        })
+    await state.directory.loadProviders()
+    expect(state.baseUrl.value).toBe('https://custom.example.com/v1')
+    expect(state.baseUrlPreset.value).toBe('custom')
 
-        return {
-          loadProviders: directory.loadProviders,
-          selectProvider: (value: string) => {
-            provider.value = value
-          },
-          selectModel: (value: string) => {
-            model.value = value
-          },
-          loadModels: directory.loadModels,
-          showApiProtocolField: directory.showApiProtocolField,
-          supportsBuiltinWebSearch: directory.supportsBuiltinWebSearch,
-        }
+    await state.directory.loadProviders(false)
+    expect(state.baseUrl.value).toBe('https://default.example.com/v1')
+    expect(state.baseUrlPreset.value).toBe('official')
+  })
+
+  it('将非数组 Provider 响应规整为空列表，并在失败后恢复 loading', async () => {
+    const state = mountDirectory()
+    mocks.manageLlmProvider.mockResolvedValueOnce({ providers: [] })
+
+    await expect(state.directory.loadProviders()).resolves.toEqual([])
+    expect(state.directory.providers.value).toEqual([])
+
+    mocks.manageLlmProvider.mockRejectedValueOnce(new Error('provider unavailable'))
+    await expect(state.directory.loadProviders()).rejects.toThrow('provider unavailable')
+    expect(state.directory.loadingProviders.value).toBe(false)
+  })
+
+  it('切换 Provider 时重置从属凭据、模型和协议状态', () => {
+    const state = mountDirectory({
+      provider: 'oauth',
+      apiKey: 'secret',
+      baseUrl: 'https://custom.example.com',
+      baseUrlPreset: 'custom',
+      apiProtocol: 'responses',
+      model: 'old-model',
+      maxContextTokens: 256,
+      authConnected: false,
+    })
+    state.directory.providers.value = [
+      createProvider('oauth', {
+        default_base_url: 'https://oauth.example.com',
+        base_url_presets: [{ id: 'oauth-default', label: '默认', value: 'https://oauth.example.com' }],
+        auth_status: { connected: true },
+      }),
+    ]
+    state.directory.models.value = [createModel('old-model')]
+
+    state.directory.handleProviderSelection()
+
+    expect(state.baseUrl.value).toBe('https://oauth.example.com')
+    expect(state.baseUrlPreset.value).toBe('oauth-default')
+    expect(state.apiKey.value).toBe('')
+    expect(state.apiProtocol.value).toBe('auto')
+    expect(state.model.value).toBe('')
+    expect(state.maxContextTokens.value).toBe(64)
+    expect(state.authConnected.value).toBe(true)
+    expect(state.directory.models.value).toEqual([])
+  })
+
+  it('同步 URL 预设值，并为界面生成带说明的选项', () => {
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [
+      createProvider('openai', {
+        base_url_presets: [{ id: 'proxy', label: '代理', value: 'https://proxy.example.com' }],
+      }),
+    ]
+
+    expect(state.directory.baseUrlPresetItems.value).toEqual([
+      {
+        id: 'proxy',
+        title: 'https://proxy.example.com',
+        value: 'https://proxy.example.com',
+        subtitle: '代理',
       },
-      template: '<div />',
+    ])
+
+    state.directory.setBaseUrlPreset(' proxy ', 'https://proxy.example.com')
+    expect(state.baseUrlPreset.value).toBe('proxy')
+    expect(state.baseUrl.value).toBe('https://proxy.example.com')
+
+    state.directory.setBaseUrlPreset(undefined, '')
+    expect(state.baseUrlPreset.value).toBe('')
+    expect(state.baseUrl.value).toBe('')
+  })
+
+  it('按 Provider 能力、凭据和必填 URL 判断模型刷新资格', () => {
+    const state = mountDirectory({ provider: 'custom' })
+    state.directory.providers.value = [
+      createProvider('custom', {
+        requires_base_url: true,
+        oauth_methods: [{ id: 'device', type: 'device', label: '设备授权' }],
+        supports_api_key: false,
+      }),
+    ]
+
+    expect(state.directory.showBaseUrlField.value).toBe(false)
+    expect(state.directory.showApiKeyField.value).toBe(false)
+    expect(state.directory.hasUsableCredential.value).toBe(false)
+    expect(state.directory.canRefreshModels.value).toBe(false)
+
+    state.apiKey.value = ' key '
+    expect(state.directory.canRefreshModels.value).toBe(false)
+
+    state.baseUrl.value = ' https://custom.example.com '
+    expect(state.directory.canRefreshModels.value).toBe(true)
+
+    state.apiKey.value = ''
+    state.directory.providers.value[0] = {
+      ...state.directory.providers.value[0],
+      auth_status: { connected: true },
+    }
+    expect(state.directory.hasUsableCredential.value).toBe(true)
+    expect(state.directory.canRefreshModels.value).toBe(true)
+
+    state.directory.providers.value[0] = {
+      ...state.directory.providers.value[0],
+      supports_model_refresh: false,
+    }
+    expect(state.directory.canRefreshModels.value).toBe(false)
+  })
+
+  it('没有选中 Provider 时不请求模型', async () => {
+    const state = mountDirectory()
+
+    await expect(state.directory.loadModels(true)).resolves.toEqual([])
+    expect(mocks.manageLlmProvider).not.toHaveBeenCalled()
+    expect(state.directory.loadingModels.value).toBe(false)
+  })
+
+  it('规范化模型刷新 payload，选择首个模型并同步认证与上下文元数据', async () => {
+    const state = mountDirectory({
+      provider: ' openai ',
+      apiKey: ' secret ',
+      baseUrl: ' https://api.example.com/v1 ',
+      baseUrlPreset: ' official ',
+      useProxy: false,
+      userAgent: ' MoviePilot-Test ',
+      maxContextTokens: 64,
+    })
+    state.directory.providers.value = [createProvider('openai')]
+    mocks.manageLlmProvider.mockResolvedValue({
+      auth_status: { connected: true, label: 'demo@example.com' },
+      models: [createModel('gpt-test', { context_tokens_k: 128 })],
     })
 
-    const wrapper = mount(Harness)
-    await wrapper.vm.loadProviders()
+    await expect(state.directory.loadModels(true)).resolves.toEqual([
+      createModel('gpt-test', { context_tokens_k: 128 }),
+    ])
 
-    expect(wrapper.vm.showApiProtocolField).toBe(true)
+    expect(mocks.manageLlmProvider).toHaveBeenCalledWith('openai', 'list_models', {
+      api_key: 'secret',
+      base_url: 'https://api.example.com/v1',
+      base_url_preset: 'official',
+      use_proxy: false,
+      user_agent: 'MoviePilot-Test',
+      force_refresh: true,
+    })
+    expect(state.model.value).toBe('gpt-test')
+    expect(state.maxContextTokens.value).toBe(128)
+    expect(state.authConnected.value).toBe(true)
+    expect(state.directory.selectedModel.value?.id).toBe('gpt-test')
+    expect(state.directory.loadingModels.value).toBe(false)
+  })
 
-    wrapper.vm.selectProvider('deepseek')
-    await nextTick()
+  it('保留仍存在的模型选择，并仅从匹配模型回填上下文', async () => {
+    const state = mountDirectory({ provider: 'openai', model: 'model-b', maxContextTokens: 32 })
+    state.directory.providers.value = [createProvider('openai')]
+    mocks.manageLlmProvider.mockResolvedValue({
+      models: [createModel('model-a', { context_tokens_k: 64 }), createModel('model-b', { context_tokens_k: 256 })],
+    })
 
-    expect(wrapper.vm.showApiProtocolField).toBe(false)
+    await state.directory.loadModels()
 
-    mocks.apiPost.mockResolvedValueOnce({
-      success: true,
-      data: {
-        models: [
+    expect(state.model.value).toBe('model-b')
+    expect(state.maxContextTokens.value).toBe(256)
+    expect(state.directory.applyModelMetadata('missing')).toBeNull()
+  })
+
+  it('将非数组模型响应规整为空列表，并在失败后恢复 loading', async () => {
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [createProvider('openai')]
+    mocks.manageLlmProvider.mockResolvedValueOnce({ models: { id: 'invalid' } })
+
+    await expect(state.directory.loadModels()).resolves.toEqual([])
+    expect(state.directory.models.value).toEqual([])
+
+    mocks.manageLlmProvider.mockRejectedValueOnce(new Error('model unavailable'))
+    await expect(state.directory.loadModels()).rejects.toThrow('model unavailable')
+    expect(state.directory.loadingModels.value).toBe(false)
+  })
+
+  it('只为 OpenAI 兼容 runtime 或要求 Responses 的工具模型显示协议字段', async () => {
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [
+      createProvider('openai'),
+      createProvider('deepseek', { runtime: 'deepseek' }),
+      createProvider('google', { runtime: 'google' }),
+    ]
+
+    expect(state.directory.showApiProtocolField.value).toBe(true)
+
+    state.provider.value = 'deepseek'
+    state.directory.models.value = [
+      createModel('deepseek-v4-flash', {
+        server_tools: [
           {
-            id: 'deepseek-v4-flash',
-            name: 'deepseek-v4-flash',
-            server_tools: [
-              {
-                id: 'web_search',
-                required_api_protocol: 'responses',
-                client_adapter: 'openai_responses',
-              },
-            ],
+            id: 'web_search',
+            required_api_protocol: 'responses',
+            client_adapter: 'openai_responses',
           },
         ],
-      },
+      }),
+    ]
+    state.model.value = 'deepseek-v4-flash'
+    await nextTick()
+    expect(state.directory.supportsBuiltinWebSearch.value).toBe(true)
+    expect(state.directory.showApiProtocolField.value).toBe(true)
+
+    state.provider.value = 'google'
+    state.directory.models.value = [
+      createModel('gemini-test', {
+        server_tools: [{ id: 'web_search', required_api_protocol: 'native', client_adapter: 'google_native' }],
+      }),
+    ]
+    state.model.value = 'gemini-test'
+    await nextTick()
+    expect(state.directory.supportsBuiltinWebSearch.value).toBe(true)
+    expect(state.directory.showApiProtocolField.value).toBe(false)
+  })
+
+  it('没有 Provider 时拒绝开始 OAuth 授权', async () => {
+    const state = mountDirectory()
+
+    await expect(state.directory.startAuth('device')).rejects.toThrow('LLM provider is required')
+    expect(mocks.manageLlmProvider).not.toHaveBeenCalled()
+  })
+
+  it('开始 OAuth 授权后打开页面、显示弹窗并安排首次轮询', async () => {
+    vi.useFakeTimers()
+    const open = vi.spyOn(window, 'open').mockReturnValue(null)
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [createProvider('openai')]
+    const session: LlmProviderAuthSession = {
+      session_id: 'session-1',
+      provider_id: 'openai',
+      flow_type: 'device',
+      status: 'pending',
+      authorize_url: 'https://auth.example.com/device',
+    }
+    mocks.manageLlmProvider.mockResolvedValue(session)
+
+    await expect(state.directory.startAuth('device')).resolves.toEqual(session)
+
+    expect(mocks.manageLlmProvider).toHaveBeenCalledWith('openai', 'start_auth', { method: 'device' })
+    expect(open).toHaveBeenCalledWith(
+      'https://auth.example.com/device',
+      '_blank',
+      'noopener,noreferrer,width=960,height=780',
+    )
+    expect(state.directory.authDialogVisible.value).toBe(true)
+    expect(state.directory.authPopupBlocked.value).toBe(true)
+    expect(vi.getTimerCount()).toBe(1)
+  })
+
+  it('优先打开 authorize URL，并回退到 verification URL', () => {
+    const open = vi.spyOn(window, 'open').mockReturnValue({} as Window)
+    const state = mountDirectory({ provider: 'openai' })
+
+    state.directory.openAuthPage()
+    expect(open).not.toHaveBeenCalled()
+
+    state.directory.authSession.value = {
+      session_id: 'session-1',
+      provider_id: 'openai',
+      flow_type: 'device',
+      status: 'pending',
+      verification_url: 'https://auth.example.com/verify',
+    }
+    state.directory.openAuthPage()
+
+    expect(open).toHaveBeenCalledWith(
+      'https://auth.example.com/verify',
+      '_blank',
+      'noopener,noreferrer,width=960,height=780',
+    )
+    expect(state.directory.authPopupBlocked.value).toBe(false)
+  })
+
+  it('待授权时合并会话状态，并按服务端间隔安排下一次静默轮询', async () => {
+    vi.useFakeTimers()
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.authSession.value = {
+      session_id: 'session-1',
+      provider_id: 'openai',
+      flow_type: 'device',
+      status: 'pending',
+      interval_seconds: 5,
+    }
+    mocks.manageLlmProvider.mockResolvedValue({ status: 'pending', interval_seconds: 0.5, message: 'waiting' })
+
+    await expect(state.directory.pollAuthSession()).resolves.toMatchObject({
+      status: 'pending',
+      interval_seconds: 0.5,
+      message: 'waiting',
     })
-    await wrapper.vm.loadModels()
-    wrapper.vm.selectModel('deepseek-v4-flash')
-    await nextTick()
 
-    expect(wrapper.vm.supportsBuiltinWebSearch).toBe(true)
-    expect(wrapper.vm.showApiProtocolField).toBe(true)
+    expect(mocks.manageLlmProvider).toHaveBeenCalledWith(
+      'openai',
+      'poll_auth',
+      { session_id: 'session-1' },
+      { feedback: 'silent' },
+    )
+    expect(state.directory.authPolling.value).toBe(false)
+    expect(vi.getTimerCount()).toBe(1)
 
-    wrapper.vm.selectProvider('google')
-    await nextTick()
-    mocks.apiPost.mockResolvedValueOnce({
-      success: true,
-      data: {
-        models: [
-          {
-            id: 'gemini-3.6-flash-preview',
-            name: 'gemini-3.6-flash-preview',
-            server_tools: [
-              {
-                id: 'web_search',
-                required_api_protocol: 'native',
-                client_adapter: 'google_native',
-              },
-            ],
-          },
-        ],
-      },
+    await vi.advanceTimersByTimeAsync(999)
+    expect(mocks.manageLlmProvider).toHaveBeenCalledTimes(1)
+
+    await vi.advanceTimersByTimeAsync(1)
+    expect(mocks.manageLlmProvider).toHaveBeenNthCalledWith(
+      2,
+      'openai',
+      'poll_auth',
+      { session_id: 'session-1' },
+      { feedback: 'silent' },
+    )
+  })
+
+  it('授权成功后刷新 Provider 与模型，并将模型请求标记为强制刷新', async () => {
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [createProvider('openai')]
+    state.directory.authSession.value = {
+      session_id: 'session-1',
+      provider_id: 'openai',
+      flow_type: 'device',
+      status: 'pending',
+    }
+    mocks.manageLlmProvider.mockImplementation((_provider: string, action: string) => {
+      if (action === 'poll_auth') return Promise.resolve({ status: 'authorized' })
+      if (action === 'list_providers') {
+        return Promise.resolve([createProvider('openai', { auth_status: { connected: true } })])
+      }
+      if (action === 'list_models') return Promise.resolve({ models: [createModel('gpt-test')] })
+      return Promise.reject(new Error(`unexpected action: ${action}`))
     })
-    await wrapper.vm.loadModels()
-    wrapper.vm.selectModel('gemini-3.6-flash-preview')
-    await nextTick()
 
-    expect(wrapper.vm.supportsBuiltinWebSearch).toBe(true)
-    expect(wrapper.vm.showApiProtocolField).toBe(false)
-    wrapper.unmount()
+    await expect(state.directory.pollAuthSession()).resolves.toMatchObject({ status: 'authorized' })
+
+    expect(mocks.manageLlmProvider).toHaveBeenNthCalledWith(
+      1,
+      'openai',
+      'poll_auth',
+      { session_id: 'session-1' },
+      { feedback: 'silent' },
+    )
+    expect(mocks.manageLlmProvider).toHaveBeenNthCalledWith(2, '', 'list_providers')
+    expect(mocks.manageLlmProvider).toHaveBeenNthCalledWith(3, 'openai', 'list_models', {
+      api_key: undefined,
+      base_url: undefined,
+      base_url_preset: undefined,
+      use_proxy: false,
+      user_agent: undefined,
+      force_refresh: true,
+    })
+    expect(state.authConnected.value).toBe(true)
+    expect(state.model.value).toBe('gpt-test')
+    expect(state.directory.authPolling.value).toBe(false)
+  })
+
+  it('轮询失败后恢复 polling 状态且不安排下一次请求', async () => {
+    vi.useFakeTimers()
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.authSession.value = {
+      session_id: 'session-1',
+      provider_id: 'openai',
+      flow_type: 'device',
+      status: 'pending',
+    }
+    mocks.manageLlmProvider.mockRejectedValue(new Error('poll failed'))
+
+    await expect(state.directory.pollAuthSession()).rejects.toThrow('poll failed')
+    expect(state.directory.authPolling.value).toBe(false)
+    expect(vi.getTimerCount()).toBe(0)
+  })
+
+  it('断开认证后刷新 Provider，并在没有 Provider 时直接返回', async () => {
+    const emptyState = mountDirectory()
+    await expect(emptyState.directory.disconnectAuth()).resolves.toBe(false)
+
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [createProvider('openai', { auth_status: { connected: true } })]
+    mocks.manageLlmProvider.mockImplementation((_provider: string, action: string) => {
+      if (action === 'disconnect') return Promise.resolve(undefined)
+      if (action === 'list_providers') return Promise.resolve([createProvider('openai')])
+      return Promise.reject(new Error(`unexpected action: ${action}`))
+    })
+
+    await expect(state.directory.disconnectAuth()).resolves.toBe(true)
+    expect(mocks.manageLlmProvider).toHaveBeenNthCalledWith(1, 'openai', 'disconnect')
+    expect(mocks.manageLlmProvider).toHaveBeenNthCalledWith(2, '', 'list_providers')
+    expect(state.authConnected.value).toBe(false)
+  })
+
+  it('关闭授权弹窗和组件卸载都会清理轮询 timer', async () => {
+    vi.useFakeTimers()
+    const state = mountDirectory({ provider: 'openai' })
+    state.directory.providers.value = [createProvider('openai')]
+    mocks.manageLlmProvider.mockResolvedValue({
+      session_id: 'session-1',
+      provider_id: 'openai',
+      flow_type: 'device',
+      status: 'pending',
+    })
+
+    await state.directory.startAuth('device')
+    expect(vi.getTimerCount()).toBe(1)
+
+    state.directory.closeAuthDialog()
+    expect(state.directory.authDialogVisible.value).toBe(false)
+    expect(vi.getTimerCount()).toBe(0)
+
+    await state.directory.startAuth('device')
+    expect(vi.getTimerCount()).toBe(1)
+    state.wrapper.unmount()
+    expect(vi.getTimerCount()).toBe(0)
   })
 })
diff --git a/vite.config.ts b/vite.config.ts
index e8eb90f4..c13bc24c 100644
--- a/vite.config.ts
+++ b/vite.config.ts
@@ -304,6 +304,7 @@ export default defineConfig(({ command, mode, isPreview }) => ({
     unstubGlobals: true,
     coverage: {
       include: [
+        'src/api/manage.ts',
         'src/utils/recommendSources.ts',
         'src/utils/permission.ts',
         'src/utils/pluginSidebarNav.ts',
@@ -334,6 +335,7 @@ export default defineConfig(({ command, mode, isPreview }) => ({
         'src/views/subscribe/SubscribePopularView.vue',
         'src/views/subscribe/SubscribeShareView.vue',
         'src/composables/useMediaSubscribe.ts',
+        'src/composables/useLlmProviderDirectory.ts',
         'src/composables/useTorrentFilter.ts',
         'src/components/cards/SubscribeCard.vue',
         'src/components/cards/UserCard.vue',
@@ -411,6 +413,18 @@ export default defineConfig(({ command, mode, isPreview }) => ({
         functions: 85,
         lines: 85,
         statements: 85,
+        'src/api/manage.ts': {
+          branches: 75,
+          functions: 80,
+          lines: 80,
+          statements: 80,
+        },
+        'src/composables/useLlmProviderDirectory.ts': {
+          branches: 85,
+          functions: 90,
+          lines: 90,
+          statements: 90,
+        },
         'src/components/cards/SubscribeCard.vue': {
           branches: 75,
           functions: 80,