refactor(api): LLM 提供商管理调用适配后端通用 manage 接口

- manage.ts 扩展 manageLlmProvider，端点联合类型纳入 llm/manage
- useLlmProviderDirectory 的目录/模型查询、授权启动/轮询/断开
  全部改走 POST /llm/manage（target+action+params 透传）
- 系统设置页 LLM 测试调用改走 manage，provider 字段转为 target
- 同步更新 composable 与系统设置页测试断言（GET mock 转 POST）

src/api/manage.ts

@@ -9,13 +9,13 @@ export interface ManageRequest {
 }
 
 /**
- * 调用统一管理端点（通知渠道 / 网盘存储）
+ * 调用统一管理端点（通知渠道 / 网盘存储 / LLM 提供商）
  *
  * 端点层不定义任何目标特定的名称与参数，
  * 目标标识、管理动作与表单参数原样透传给后端模块
  */
 function manageTarget<T = Record<string, unknown>>(
-  endpoint: 'notification/manage' | 'storage/manage',
+  endpoint: 'notification/manage' | 'storage/manage' | 'llm/manage',
   request: ManageRequest,
   config?: AxiosRequestConfig,
 ): Promise<T> {
@@ -41,3 +41,13 @@ export function manageStorage<T = Record<string, unknown>>(
 ): Promise<T> {
   return manageTarget<T>('storage/manage', { target: storage, action, params }, config)
 }
+
+/** 对指定 LLM 提供商执行管理动作，返回响应中的业务数据。 */
+export function manageLlmProvider<T = Record<string, unknown>>(
+  provider: string,
+  action: string,
+  params: Record<string, unknown> = {},
+  config?: AxiosRequestConfig,
+): Promise<T> {
+  return manageTarget<T>('llm/manage', { target: provider, action, params }, config)
+}

src/composables/__tests__/useLlmProviderDirectory.spec.ts

@@ -4,12 +4,12 @@ import { defineComponent, nextTick, ref } from 'vue'
 import { beforeEach, describe, expect, it, vi } from 'vitest'
 
 const mocks = vi.hoisted(() => ({
-  apiGet: vi.fn(),
+  apiPost: vi.fn(),
 }))
 
 vi.mock('@/api', () => ({
   default: createDataApiMock({
-    get: (...args: unknown[]) => mocks.apiGet(...args),
+    post: (...args: unknown[]) => mocks.apiPost(...args),
   }),
 }))
 
@@ -33,11 +33,11 @@ function createProvider(id: string, runtime: string) {
 
 describe('useLlmProviderDirectory', () => {
   beforeEach(() => {
-    mocks.apiGet.mockReset()
+    mocks.apiPost.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.apiGet.mockResolvedValue({
+    mocks.apiPost.mockResolvedValue({
       success: true,
       data: [
         createProvider('openai', 'openai_compatible'),
@@ -83,7 +83,7 @@ describe('useLlmProviderDirectory', () => {
 
     expect(wrapper.vm.showApiProtocolField).toBe(false)
 
-    mocks.apiGet.mockResolvedValueOnce({
+    mocks.apiPost.mockResolvedValueOnce({
       success: true,
       data: {
         models: [
@@ -110,7 +110,7 @@ describe('useLlmProviderDirectory', () => {
 
     wrapper.vm.selectProvider('google')
     await nextTick()
-    mocks.apiGet.mockResolvedValueOnce({
+    mocks.apiPost.mockResolvedValueOnce({
       success: true,
       data: {
         models: [

src/composables/useLlmProviderDirectory.ts

@@ -1,5 +1,5 @@
 import { computed, onBeforeUnmount, ref, type Ref } from 'vue'
-import api from '@/api'
+import { manageLlmProvider } from '@/api/manage'
 
 export interface LlmProviderAuthMethod {
   id: string
@@ -248,7 +248,7 @@ export function useLlmProviderDirectory(options: UseLlmProviderDirectoryOptions)
   async function loadProviders(preserveBaseUrl = true) {
     loadingProviders.value = true
     try {
-      const result = await api.get<LlmProvider[]>('llm/providers')
+      const result = await manageLlmProvider<LlmProvider[]>('', 'list_providers')
       providers.value = Array.isArray(result) ? result : []
       if (!selectedProvider.value && providers.value.length > 0) {
         options.provider.value = providers.value[0].id
@@ -266,9 +266,10 @@ export function useLlmProviderDirectory(options: UseLlmProviderDirectoryOptions)
 
     loadingModels.value = true
     try {
-      const payload = await api.get<{ auth_status?: LlmProviderAuthStatus; models?: LlmModel[] }>('llm/models', {
+      const payload = await manageLlmProvider<{ auth_status?: LlmProviderAuthStatus; models?: LlmModel[] }>(
-        params: {
+        normalizeValue(options.provider.value),
-          provider: normalizeValue(options.provider.value),
+        'list_models',
+        {
           api_key: normalizeValue(options.apiKey.value) || undefined,
           base_url: normalizeValue(options.baseUrl.value) || undefined,
           base_url_preset: normalizeValue(options.baseUrlPreset?.value) || undefined,
@@ -276,7 +277,7 @@ export function useLlmProviderDirectory(options: UseLlmProviderDirectoryOptions)
           user_agent: normalizeValue(options.userAgent?.value) || undefined,
           force_refresh: forceRefresh,
         },
-      })
+      )
       models.value = Array.isArray(payload.models) ? payload.models : []
       updateProviderAuthStatus(normalizeValue(options.provider.value), payload.auth_status)
 
@@ -311,9 +312,10 @@ export function useLlmProviderDirectory(options: UseLlmProviderDirectoryOptions)
     authPolling.value = true
     clearPollTimer()
     try {
-      const result = await api.post<Partial<LlmProviderAuthSession>>(
+      const result = await manageLlmProvider<Partial<LlmProviderAuthSession>>(
-        `llm/provider-auth/${authSession.value.session_id}/poll`,
+        normalizeValue(options.provider.value),
-        undefined,
+        'poll_auth',
+        { session_id: authSession.value.session_id },
         { feedback: 'silent' },
       )
 
@@ -347,10 +349,11 @@ export function useLlmProviderDirectory(options: UseLlmProviderDirectoryOptions)
       throw new Error('LLM provider is required')
     }
 
-    const result = await api.post<LlmProviderAuthSession>('llm/provider-auth/start', {
+    const result = await manageLlmProvider<LlmProviderAuthSession>(
-      provider: normalizeValue(options.provider.value),
+      normalizeValue(options.provider.value),
-      method: methodId,
+      'start_auth',
-    })
+      { method: methodId },
+    )
 
     authSession.value = result
     authDialogVisible.value = true
@@ -363,7 +366,7 @@ export function useLlmProviderDirectory(options: UseLlmProviderDirectoryOptions)
   async function disconnectAuth() {
     if (!selectedProvider.value) return false
 
-    await api.delete(`llm/provider-auth/${normalizeValue(options.provider.value)}`)
+    await manageLlmProvider(normalizeValue(options.provider.value), 'disconnect')
 
     await loadProviders()
     return true

src/views/setting/AccountSettingSystem.vue

@@ -2,6 +2,7 @@
 <script lang="ts" setup>
 import { useToast } from 'vue-toastification'
 import api from '@/api'
+import { manageLlmProvider } from '@/api/manage'
 import { useGlobalSettingsStore } from '@/stores'
 import { DownloaderConf, MediaServerConf } from '@/api/types'
 import DownloaderCard from '@/components/cards/DownloaderCard.vue'
@@ -817,7 +818,7 @@ async function testLlmConnection() {
 
   const snapshot = buildLlmSnapshot()
   const snapshotKey = buildLlmSnapshotKey(snapshot)
-  const payload = buildLlmTestPayload(snapshot)
+  const { provider: testProvider, ...testParams } = buildLlmTestPayload(snapshot)
   const requestId = ++llmTestRequestId
   if (llmTestAbortController) llmTestAbortController.abort()
   const abortController = new AbortController()
@@ -825,7 +826,7 @@ async function testLlmConnection() {
 
   testingLlm.value = true
   try {
-    await api.post('llm/test', payload, {
+    await manageLlmProvider(testProvider, 'test', testParams, {
       feedback: 'silent',
       signal: abortController.signal,
     })

src/views/setting/__tests__/AccountSettingSystem.spec.ts

@@ -445,20 +445,23 @@ describe('AccountSettingSystem', () => {
     await fireEvent.click(testLlm)
     await waitFor(() => expect(mocks.toastError).toHaveBeenCalledWith('LLM 调用测试失败：provider rejected'))
     const firstCall = mocks.apiPost.mock.calls[0]
-    expect(firstCall?.[0]).toBe('llm/test')
+    expect(firstCall?.[0]).toBe('llm/manage')
     expect(firstCall?.[1]).toEqual({
-      api_key: 'secret-key',
+      target: 'openai',
-      api_protocol: 'responses',
+      action: 'test',
-      base_url: 'https://llm.example/v1',
+      params: {
-      base_url_preset: 'custom',
+        api_key: 'secret-key',
-      enabled: true,
+        api_protocol: 'responses',
-      model: 'gpt-5',
+        base_url: 'https://llm.example/v1',
-      provider: 'openai',
+        base_url_preset: 'custom',
-      thinking_level: 'high',
+        enabled: true,
-      temperature: 0.7,
+        model: 'gpt-5',
-      use_proxy: false,
+        thinking_level: 'high',
-      user_agent: 'MoviePilot-Test',
+        temperature: 0.7,
-      web_search_mode: 'builtin',
+        use_proxy: false,
+        user_agent: 'MoviePilot-Test',
+        web_search_mode: 'builtin',
+      },
     })
     expect(firstCall?.[2]).toEqual({ signal: expect.any(AbortSignal) })
     expect(testLlm).toBeEnabled()