src/renderer/src/components/NewThread.vue (2)

139-143: Translate remaining Chinese comments to English (repo rule: English-only comments/logs)

Several in-file comments remain in Chinese; please translate to English to meet repo guidelines and this PR’s objective.

-// 定义偏好模型的类型
+// Define the preferred model type

-    // 如果有现有线程，使用最近线程的模型
+    // If there are existing threads, use the most recent thread's model
-        // 同时匹配 modelId 和 providerId
+        // Match both modelId and providerId

-    // 如果没有现有线程，尝试使用用户上次选择的模型
+    // If there are no existing threads, try the user's last selected model
-        // 验证偏好模型是否还在可用模型列表中
+        // Validate the preferred model is still available

-    // 如果没有偏好模型或偏好模型不可用，使用第一个可用模型
+    // If no preferred model or it is unavailable, fall back to the first available model

-// 监听鼠标悬停
+// Listen for mouse hover

-  // 保存用户的模型偏好设置
+  // Persist the user's model preference

-// 监听 deeplinkCache 变化
+// Watch changes to deeplinkCache

-      // 清理缓存
+      // Clear the cache

-    // 如果打开，立即显示按钮
+    // When opened, show the button immediately
-    // 如果关闭，延迟隐藏按钮，等待动画完成
+    // When closed, hide the button after the animation finishes
-    }, 300) // 300ms是一个常见的动画持续时间，可以根据实际情况调整
+    }, 300) // 300ms is a common animation duration; adjust as needed

-// 初始化时设置showSettingsButton的值与settingsPopoverOpen一致
+// Initialize showSettingsButton to follow settingsPopoverOpen

Also applies to: 207-215, 236-246, 242-246, 264-270, 290-297, 316-321, 325-331, 382-384, 401-409, 412-421

---

439-439: Define a proper settings type for createThread instead of using any

• In src/renderer/src/stores/chat.ts, replace the settings: Partial<CONVERSATION_SETTINGS> parameter with a new interface (e.g. CreateThreadSettings) that extends the conversation config and adds modelId (and other fields like artifacts if required), then update createThread’s signature to use it.
• In src/renderer/src/components/NewThread.vue:439, remove the as any cast and annotate the object literal with the new CreateThreadSettings type.

src/main/presenter/windowPresenter/index.ts (4)

614-618: Harden BrowserWindow: enable contextIsolation and explicitly disable nodeIntegration.

Meets the project’s security guidelines for main/renderer. Keep sandbox as-is if required by preload.

Apply:

       webPreferences: {
         preload: join(__dirname, '../preload/index.mjs'), // Preload script path
-        sandbox: false, // Disable sandbox to allow preload to access Node.js API
-        devTools: is.dev // Enable DevTools in dev mode
+        sandbox: false, // keep if your preload needs Node; otherwise prefer true
+        contextIsolation: true,
+        nodeIntegration: false,
+        webSecurity: true,
+        devTools: is.dev // Enable DevTools in dev mode
       },

---

441-466: Wrap the async setTimeout body in try/catch to avoid unhandled rejections.

The outer try/catch won’t catch errors thrown in the async callback.

Apply:

-    try {
-      setTimeout(async () => {
-        const tabPresenterInstance = presenter.tabPresenter as TabPresenter
-        const tabsData = await tabPresenterInstance.getWindowTabsData(windowId)
-        const activeTab = tabsData.find((tab) => tab.isActive)
-        if (activeTab) {
-          console.log(
-            `Focusing active tab ${activeTab.id} in window ${windowId} (reason: ${reason})`
-          )
-          await tabPresenterInstance.switchTab(activeTab.id)
-          const state = this.windowFocusStates.get(windowId)
-          if (state) {
-            state.lastFocusTime = Date.now()
-            if (reason === 'initial') {
-              state.hasInitialFocus = true
-            }
-            if (reason === 'focus' || reason === 'initial') {
-              state.isNewWindow = false
-            }
-          }
-        }
-      }, 50)
-    } catch (error) {
-      console.error(`Error focusing active tab in window ${windowId}:`, error)
-    }
+    setTimeout(() => {
+      void (async () => {
+        try {
+          const tabPresenterInstance = presenter.tabPresenter as TabPresenter
+          const tabsData = await tabPresenterInstance.getWindowTabsData(windowId)
+          const activeTab = tabsData.find((tab) => tab.isActive)
+          if (activeTab) {
+            console.log(`Focusing active tab ${activeTab.id} in window ${windowId} (reason: ${reason})`)
+            await tabPresenterInstance.switchTab(activeTab.id)
+            const state = this.windowFocusStates.get(windowId)
+            if (state) {
+              state.lastFocusTime = Date.now()
+              if (reason === 'initial') state.hasInitialFocus = true
+              if (reason === 'focus' || reason === 'initial') state.isNewWindow = false
+            }
+          }
+        } catch (error) {
+          console.error(`Error focusing active tab in window ${windowId}:`, error)
+        }
+      })()
+    }, 50)

---

184-193: Handle shell.openPath result/errors.

shell.openPath returns a Promise; current code ignores failures.

Apply:

-      } else {
-        shell.openPath(filePath) // Open with system default app
-      }
+      } else {
+        shell.openPath(filePath).catch((err) => {
+          console.error(`Failed to open file ${filePath}:`, err)
+        })
+      }

---

47-55: Replace sync IPC in preload with async invoke
In src/preload/index.ts (lines 24, 31, 51–52), replace all ipcRenderer.sendSync('get-window-id'|'get-web-contents-id') calls with ipcRenderer.invoke(...) and adjust the surrounding functions to return Promises (e.g. mark them async and await the invoke), including the DOMContentLoaded initialization.

src/shared/types/presenters/thread.presenter.d.ts (4)

15-30: Deduplicate shared types: import/re-export from a single source to avoid drift

These definitions duplicate types exported in src/shared/types/presenters/index.d.ts. Centralize to one source to prevent divergence and ease maintenance.

Apply:

@@
-import {
-  ChatMessage,
-  AssistantMessageBlock,
-  Message,
-  AssistantMessage,
-  UserMessage
-} from '../core/chat'
+import {
+  AssistantMessageBlock,
+  Message,
+  AssistantMessage
+} from '../core/chat'
+import type { CONVERSATION_SETTINGS, MESSAGE_STATUS, MESSAGE_METADATA } from './index'
+export type { CONVERSATION_SETTINGS, MESSAGE_STATUS, MESSAGE_METADATA } from './index'
@@
-export type CONVERSATION_SETTINGS = {
-  systemPrompt: string
-  temperature: number
-  contextLength: number
-  maxTokens: number
-  providerId: string
-  modelId: string
-  artifacts: 0 | 1
-  enabledMcpTools?: string[]
-  thinkingBudget?: number
-  enableSearch?: boolean
-  forcedSearch?: boolean
-  searchStrategy?: 'turbo' | 'max'
-  reasoningEffort?: 'minimal' | 'low' | 'medium' | 'high'
-  verbosity?: 'low' | 'medium' | 'high'
-}
+// moved: re-exported from ./index
@@
-export type MESSAGE_STATUS = 'sent' | 'pending' | 'error'
@@
-export type MESSAGE_METADATA = {
-  totalTokens: number
-  inputTokens: number
-  outputTokens: number
-  generationTime: number
-  firstTokenTime: number
-  tokensPerSecond: number
-  contextUsage: number
-  model?: string
-  provider?: string
-  reasoningStartTime?: number
-  reasoningEndTime?: number
-}
+// moved: re-exported from ./index

Also applies to: 43-58

---

32-41: Narrow flag fields to 0|1 (or boolean) for strict typing consistency

These are currently number; elsewhere artifacts uses 0|1. Align to avoid accidental values.

 export type CONVERSATION = {
   id: string
   title: string
   settings: CONVERSATION_SETTINGS
   createdAt: number
   updatedAt: number
-  is_new?: number
-  artifacts?: number
-  is_pinned?: number
+  is_new?: 0 | 1
+  artifacts?: 0 | 1
+  is_pinned?: 0 | 1
 }

---

137-147: Avoid nullable result for sendMessage; surface errors via rejection

Returning null forces callers to handle two error paths. Prefer Promise rejection with meaningful error info.

-  sendMessage(conversationId: string, content: string, role: MESSAGE_ROLE): Promise<MESSAGE | null>
+  sendMessage(conversationId: string, content: string, role: MESSAGE_ROLE): Promise<MESSAGE>

Also note mismatch with IMessageManager.sendMessage signature below—consider aligning or documenting the abstraction difference.

---

190-199: Relax required params and metadata at send-time in IMessageManager

Root messages may not have a parent; variant is often false by default; full metadata is rarely known at send time.

   sendMessage(
     conversationId: string,
     content: string,
     role: MESSAGE_ROLE,
-    parentId: string,
-    isVariant: boolean,
-    metadata: MESSAGE_METADATA
+    parentId?: string,
+    isVariant?: boolean,
+    metadata?: Partial<MESSAGE_METADATA>
   ): Promise<MESSAGE>

src/renderer/src/App.vue (2)

38-57: Fix theme/font class handling: use html consistently and react to system theme changes

• Currently you add/remove classes on documentElement here but add them on body in onMounted; also system theme changes (isDark) won’t trigger this watch.

Apply:

-// Watch theme and font size changes, update body class directly
+// Watch theme and font size changes; update root (html) classes consistently
 watch(
-  [() => themeStore.themeMode, () => settingsStore.fontSizeClass],
-  ([newTheme, newFontSizeClass], [oldTheme, oldFontSizeClass]) => {
-    let newThemeName = newTheme
-    if (newTheme === 'system') {
-      newThemeName = themeStore.isDark ? 'dark' : 'light'
-    }
-    if (oldTheme) {
-      document.documentElement.classList.remove(oldTheme)
-    }
-    if (oldFontSizeClass) {
-      document.documentElement.classList.remove(oldFontSizeClass)
-    }
-    document.documentElement.classList.add(newThemeName)
-    document.documentElement.classList.add(newFontSizeClass)
-    console.log('newTheme', newThemeName)
-  },
-  { immediate: false } // Initialization is handled in onMounted
+  [() => themeStore.themeMode, () => themeStore.isDark, () => settingsStore.fontSizeClass],
+  ([newThemeMode, isDark, newFontSizeClass], [oldThemeMode, oldIsDark, oldFontSizeClass]) => {
+    const newThemeName = newThemeMode === 'system' ? (isDark ? 'dark' : 'light') : newThemeMode
+    const oldThemeName = oldThemeMode === 'system' ? (oldIsDark ? 'dark' : 'light') : oldThemeMode
+    if (oldThemeName) document.documentElement.classList.remove(oldThemeName)
+    if (oldFontSizeClass) document.documentElement.classList.remove(oldFontSizeClass)
+    document.documentElement.classList.add(newThemeName)
+    document.documentElement.classList.add(newFontSizeClass)
+  },
+  { immediate: false } // Init done in onMounted
 )

---

305-313: Remove SHOW_ERROR listener on unmount to avoid leaks

You add NOTIFICATION_EVENTS.SHOW_ERROR on mount but don’t remove it.

   window.electron.ipcRenderer.removeAllListeners(SHORTCUT_EVENTS.GO_SETTINGS)
+  window.electron.ipcRenderer.removeAllListeners(NOTIFICATION_EVENTS.SHOW_ERROR)
   window.electron.ipcRenderer.removeAllListeners(NOTIFICATION_EVENTS.SYS_NOTIFY_CLICKED)
   window.electron.ipcRenderer.removeAllListeners(NOTIFICATION_EVENTS.DATA_RESET_COMPLETE_DEV)

src/shared/logger.ts (1)

42-52: Console override risks recursion and swallows error logs; preserve stacks and mirror safely.

• With console transport active, console.* -> logger.* -> console.* can recurse.
• Current gating drops console.error when loggingEnabled=false and not dev, violating “don’t swallow errors.”
• console.trace loses stack info.

Mirror to file via log.*, print to screen via saved originalConsole, always persist ERROR/WARN, and include stack for TRACE.

Apply this diff to the console stubs:

   // Replace console methods
   console.log = (...args: unknown[]) => {
-    // Only log when logging is enabled or in development mode
-    if (loggingEnabled || is.dev) {
-      logger.info(...args)
-    }
+    if (is.dev) originalConsole.log(...args)
+    if (loggingEnabled || is.dev) log.info(...args)
   }

   console.error = (...args: unknown[]) => {
-    // Only log when logging is enabled or in development mode
-    if (loggingEnabled || is.dev) {
-      logger.error(...args)
-    }
+    // Never swallow errors
+    originalConsole.error(...args)
+    log.error(...args)
   }

   console.warn = (...args: unknown[]) => {
-    // Only log when logging is enabled or in development mode
-    if (loggingEnabled || is.dev) {
-      logger.warn(...args)
-    }
+    originalConsole.warn(...args)
+    // Keep WARNs by default; drop if you truly want WARN gated by `loggingEnabled`
+    log.warn(...args)
   }

   console.info = (...args: unknown[]) => {
-    // Only log when logging is enabled or in development mode
-    if (loggingEnabled || is.dev) {
-      logger.info(...args)
-    }
+    if (is.dev) originalConsole.info(...args)
+    if (loggingEnabled || is.dev) log.info(...args)
   }

   console.debug = (...args: unknown[]) => {
-    // Only log when logging is enabled or in development mode
-    if (loggingEnabled || is.dev) {
-      logger.debug(...args)
-    }
+    if (is.dev) originalConsole.debug(...args)
+    if (loggingEnabled || is.dev) log.debug(...args)
   }

   console.trace = (...args: unknown[]) => {
-    // Only log when logging is enabled or in development mode
-    if (loggingEnabled || is.dev) {
-      logger.debug(...args)
-    }
+    // Preserve stack in console and file
+    originalConsole.trace(...args)
+    const stack = new Error().stack
+    if (loggingEnabled || is.dev) log.debug('[trace]', ...args, stack)
   }

Notes:

• This relies on applyLogLevels() setting console.level = false to avoid recursion.
• If you truly want WARNs suppressed when disabled, gate log.warn the same as INFO/DEBUG, but keep ERROR unconditional.

Also applies to: 53-96

src/main/presenter/configPresenter/modelDefaultSettings.ts (1)

448-448: Finish translating remaining Chinese comments in this file.

This PR’s goal and project guideline require English-only comments/logs. The following headers are still Chinese; translate to keep consistency.

-  // DeepSeek系列模型配置
+  // DeepSeek series models
-  // Claude系列模型配置
+  // Claude series models
-  // OpenAI GPT系列模型配置
+  // OpenAI GPT series models
-  // Llama系列
+  // Llama series
-  // Mistral系列
+  // Mistral series
-  // Qwen系列
+  // Qwen series
-  // Qwen3系列模型
+  // Qwen3 series models
-  // Yi系列
+  // Yi series
-  // Gemma系列
+  // Gemma series
-  // Phi系列
+  // Phi series
-  // Doubao (字节跳动)模型配置
+  // Doubao (ByteDance) models
-  // MiniMax模型配置
+  // MiniMax models
-  // Zhipu
+  // Zhipu
-  // 智谱AI 推理模型
+  // Zhipu AI reasoning models
-  // 智谱AI 多模态模型
+  // Zhipu AI multimodal models
-  // Fireworks AI模型配置
+  // Fireworks AI models
-  // Moonshot (月之暗面)模型配置
+  // Moonshot (Moonshot AI) models
-  // Azure OpenAI配置
+  // Azure OpenAI configuration

Also applies to: 616-616, 695-695, 1001-1001, 1102-1102, 1148-1148, 1250-1250, 1372-1372, 1396-1396, 1464-1464, 1488-1488, 1534-1534, 1558-1558, 1680-1680, 1714-1714, 1760-1760, 1806-1806, 1863-1863

src/preload/floating-preload.ts (1)

70-83: Set contextIsolation: true in every BrowserWindow’s webPreferences

All new BrowserWindow({ webPreferences: { … } }) calls (e.g. in src/main/presenter/*) must explicitly include contextIsolation: true to enforce isolation and eliminate the insecure window‐attachment fallback.

src/main/lib/watermark.ts (1)

64-91: Escape dynamic SVG text to prevent malformed XML/injection

Texts come from options and are injected directly into SVG. Escape XML entities before embedding.

 const createWatermarkSvg = (width: number, options: WatermarkOptions): string => {
   const { isDark = false, version = '1.0.0', texts = {} } = options
+  const esc = (s: string) =>
+    s.replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/"/g, '&quot;')
@@
-  const brandText = texts.brand || 'DeepChat'
+  const brandText = esc(texts.brand || 'DeepChat')
@@
-  const tipText = texts.tip || 'Generated by DeepChat'
+  const tipText = esc(texts.tip || 'Generated by DeepChat')
@@
-  const timeLineText = `${modelInfo.join(' · ')} · ${timeStr}`
+  const timeLineText = esc(`${modelInfo.join(' · ')} · ${timeStr}`)
@@
-  const versionInfo = 'v' + version
+  const versionInfo = esc('v' + version)

src/main/presenter/configPresenter/aes.ts (1)

62-82: Use AEAD (GCM) instead of CBC; translate decryption errors

CBC without authentication is unsafe (bitflips/Oracle). Switch to aes-256-gcm and include authTag; also translate remaining Chinese text.

-const ALGORITHM = 'aes-256-cbc'
+const ALGORITHM = 'aes-256-gcm'
@@
   static encrypt(
     plainText: string,
     key: Buffer,
-    iv?: Buffer
+    iv?: Buffer
   ): {
-    cipherText: string
-    iv: string
+    cipherText: string
+    iv: string
+    tag: string
   } {
@@
-      const cipher = crypto.createCipheriv(ALGORITHM, key, usedIv)
+      const cipher = crypto.createCipheriv(ALGORITHM, key, usedIv)
@@
-      return {
-        cipherText: encrypted,
-        iv: usedIv.toString('hex')
-      }
+      const tag = cipher.getAuthTag()
+      return { cipherText: encrypted, iv: usedIv.toString('hex'), tag: tag.toString('hex') }
@@
-  static decrypt(cipherText: string, key: Buffer, iv: string): string {
+  static decrypt(cipherText: string, key: Buffer, iv: string, tag: string): string {
     try {
       const ivBuffer = Buffer.from(iv, 'hex')
-      const decipher = crypto.createDecipheriv(ALGORITHM, key, ivBuffer)
+      const decipher = crypto.createDecipheriv(ALGORITHM, key, ivBuffer)
+      decipher.setAuthTag(Buffer.from(tag, 'hex'))
@@
-    } catch (error) {
-      throw new Error(
-        `解密失败: ${error instanceof Error ? error.message : '密文可能被篡改或密钥错误'}`
-      )
-    }
+    } catch (error) {
+      throw new Error(
+        `Decryption failed: ${error instanceof Error ? error.message : 'Ciphertext tampered or wrong key'}`
+      )
+    }

If backward compatibility is required, keep CBC decrypt as legacy fallback behind a version flag.

src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts (1)

96-123: Monkey-patching the OpenAI client is not concurrency-safe

Overwriting this.openai.chat.completions.create can race across simultaneous streams, leaking flags across requests.

Refactor to avoid patching:

• Add a requestParamsTransformer hook in OpenAICompatibleProvider and pass a function here, or
• Create a per-call OpenAI client instance and call super with an injected create method via composition, or
• Extend super.coreStream to accept “extraParams” merged into the single streaming request.

Example (conceptual):

- this.openai.chat.completions.create = ((params, options) => { ... return originalCreate(modifiedParams, options) })
+ const transformer = (params: any) => ({ ...params, ...dashscopeFlags })
+ yield* super.coreStreamWithTransformer(messages, modelId, effectiveModelConfig, temperature, maxTokens, mcpTools, transformer)

src/main/lib/scrollCapture.ts (1)

127-148: JS injection risk via unescaped CSS selectors in executeJavaScript

Selectors are interpolated directly into a JS string; a crafted selector with quotes can break out. Serialize the selector safely.

-      const elementInfo = await this.view.webContents.executeJavaScript(`
-        (function() {
-          const elements = document.querySelectorAll('${selector}')
+      const elementInfo = await this.view.webContents.executeJavaScript(`
+        (function() {
+          const selector = ${JSON.stringify(selector)};
+          const elements = document.querySelectorAll(selector)
           const hiddenElements = []
@@
-        })()
+        })()
       `)

src/main/presenter/llmProviderPresenter/providers/ollamaProvider.ts (3)

161-163: Preserve temperature=0; use nullish coalescing

Using || overrides 0. Use ?? across calls.

-          temperature: temperature || 0.7,
+          temperature: temperature ?? 0.7,

Apply similarly in summaries (Line 229) and generateText (Line 255).

---

300-317: Potential NPE on model.details.family

details or family can be undefined. Guard access.

-    const info = showResponse.model_info
-    const family = model.details.family
-    const context_length = info?.[family + '.context_length'] ?? 4096
-    const embedding_length = info?.[family + '.embedding_length'] ?? 512
+    const info = showResponse.model_info
+    const family = model.details?.family ?? 'default'
+    const context_length = info?.[`${family}.context_length`] ?? 4096
+    const embedding_length = info?.[`${family}.embedding_length`] ?? 512

---

1-1184: Translate or remove Chinese comments/logs in provider files
Multiple provider files (e.g. ollamaProvider.ts, geminiProvider.ts, siliconcloudProvider.ts, etc.) still contain Chinese comments and console messages. Replace all Chinese text in code comments and logs with English equivalents.

src/renderer/src/stores/chat.ts (2)

632-646: Bug: condition always true; should compare against cached.threadId

Both checks use getActiveThreadId() === getActiveThreadId(). Compare to cached.threadId.

-          if (getActiveThreadId() === getActiveThreadId()) {
+          if (getActiveThreadId() === cached.threadId) {
             const mainMsgIndex = getMessages().findIndex((m) => m.id === mainMessage.id)
             if (mainMsgIndex !== -1) {
               getMessages()[mainMsgIndex] = enrichedMainMessage as AssistantMessage | UserMessage
             }
           }
@@
-        if (getActiveThreadId() === getActiveThreadId()) {
+        if (getActiveThreadId() === cached.threadId) {
           const msgIndex = getMessages().findIndex((m) => m.id === msg.eventId)
           if (msgIndex !== -1) {
             getMessages()[msgIndex] = enrichedMessage as AssistantMessage | UserMessage
           }
         }

---

470-476: User-visible string should use i18n key

Replace 'tool call failed' with an i18n key per renderer guidelines.

-                  existingToolCallBlock.tool_call.response =
-                    msg.tool_call_response || 'tool call failed'
+                  existingToolCallBlock.tool_call.response =
+                    msg.tool_call_response || 'common.error.toolCallFailed'

src/main/presenter/configPresenter/providerModelSettings.ts (2)

3091-3107: Prefer maxCompletionTokens when present (GPT‑5 series).

You return maxTokens unconditionally, but comments state GPT‑5 uses maxCompletionTokens. Use the GPT‑5 value when defined to avoid under/over-limiting completions.

Apply:

-      return {
-        maxTokens: config.maxTokens,
+      return {
+        maxTokens: config.maxCompletionTokens ?? config.maxTokens,
         contextLength: config.contextLength,
         temperature: config.temperature, // Keep optional, some models don't support it
         vision: config.vision || false,
         functionCall: config.functionCall || false,
         reasoning: config.reasoning || false,
         type: config.type || ModelType.Chat,
         reasoningEffort: config.reasoningEffort,
         verbosity: config.verbosity,
         maxCompletionTokens: config.maxCompletionTokens,
         thinkingBudget: config.thinkingBudget,
         enableSearch: config.enableSearch ?? false,
         forcedSearch: config.forcedSearch ?? false,
         searchStrategy: config.searchStrategy || 'turbo'
       }

---

3091-3107: Add missing optional fields to ModelConfig
In src/shared/types/presenters/legacy.presenters.d.ts (line 121), extend the ModelConfig interface to include these optional properties—reasoningEffort, verbosity, maxCompletionTokens, thinkingBudget, enableSearch, forcedSearch, and searchStrategy—to match providerModelSettings.ts and prevent implicit any.

src/main/presenter/mcpPresenter/mcpClient.ts (2)

893-905: Populate required PromptListEntry.client for type/UI consumers.

Shared types expect a client field. Provide minimally sane defaults.

-          const validPrompts = promptsArray.map((p) => ({
+          const validPrompts = promptsArray.map((p) => ({
             name: typeof p === 'object' && p !== null && 'name' in p ? String(p.name) : 'unknown',
             description:
               typeof p === 'object' && p !== null && 'description' in p
                 ? String(p.description)
                 : undefined,
             arguments:
               typeof p === 'object' && p !== null && 'arguments' in p ? p.arguments : undefined,
-            files: typeof p === 'object' && p !== null && 'files' in p ? p.files : undefined
+            files: typeof p === 'object' && p !== null && 'files' in p ? p.files : undefined,
+            client: { name: this.serverName, icon: '' }
           })) as PromptListEntry[]

---

999-1006: Populate required ResourceListEntry.client.

Same issue for resources; add client metadata.

-          const validResources = resourcesArray.map((r) => ({
+          const validResources = resourcesArray.map((r) => ({
             uri: typeof r === 'object' && r !== null && 'uri' in r ? String(r.uri) : 'unknown',
-            name: typeof r === 'object' && r !== null && 'name' in r ? String(r.name) : undefined
+            name: typeof r === 'object' && r !== null && 'name' in r ? String(r.name) : undefined,
+            client: { name: this.serverName, icon: '' }
           })) as ResourceListEntry[]

src/main/presenter/configPresenter/mcpConfHelper.ts (3)

692-706: batchImportMcpServers builds the wrong config shape (runtime breakage).

You cast ExtendedMCPServerConfig to MCPServerConfig. MCPServerConfig requires command, descriptions, icons, etc. Missing command will break stdio startup; descriptions key name differs.

Use a proper MCPServerConfig object and map fields:

-        const mcpConfig: ExtendedMCPServerConfig = {
-          name: serverConfig.name,
-          description: serverConfig.description,
-          args: serverConfig.args || [],
-          env: serverConfig.env || {},
-          enabled: serverConfig.enabled ?? enableByDefault,
-          type: (serverConfig.type as MCPServerType) || 'stdio',
-          package: serverConfig.package,
-          version: serverConfig.version || 'latest',
-          source: serverConfig.source as string | undefined,
-          logo_url: serverConfig.logo_url as string | undefined,
-          publisher: serverConfig.publisher as string | undefined,
-          tags: serverConfig.tags as string[] | undefined,
-          view_count: serverConfig.view_count as number | undefined
-        }
-
-        // Add or update the server
-        const success = await this.addMcpServer(serverName, mcpConfig as unknown as MCPServerConfig)
+        const mcpConfig: MCPServerConfig = {
+          command: 'npx',
+          args: [
+            '-y',
+            serverConfig.version
+              ? `${serverConfig.package}@${serverConfig.version}`
+              : serverConfig.package
+          ],
+          env: serverConfig.env || {},
+          descriptions: serverConfig.description, // note: plural key in MCPServerConfig
+          icons: '🔌',
+          autoApprove: ['all'],
+          disable: !(serverConfig.enabled ?? enableByDefault),
+          type: (serverConfig.type as MCPServerType) || 'stdio',
+          source: serverConfig.source as string | undefined,
+          sourceId: serverConfig.package
+        }
+
+        // Add or update the server
+        const success = await this.addMcpServer(serverName, mcpConfig)

---

778-789: findServerByPackage checks a non-existent field.

Stored configs don’t have package; use sourceId (set during import) instead.

-  async findServerByPackage(packageName: string): Promise<string | null> {
+  async findServerByPackage(packageName: string): Promise<string | null> {
     const servers = await this.getMcpServers()
 
     for (const [serverName, config] of Object.entries(servers)) {
-      const extendedConfig = config as unknown as ExtendedMCPServerConfig
-      if (extendedConfig.package === packageName) {
+      if ((config as MCPServerConfig).sourceId === packageName) {
         return serverName
       }
     }
 
     return null
   }

---

338-395: Remove all Chinese comments and strings in presenter/configPresenter and mcpPresenter
Detected Chinese in:

• src/main/presenter/mcpPresenter/toolManager.ts
• src/main/presenter/mcpPresenter/mcpClient.ts
• src/main/presenter/configPresenter/aes.ts
• src/main/presenter/configPresenter/providerModelSettings.ts
• src/main/presenter/configPresenter/modelDefaultSettings.ts
• src/main/presenter/configPresenter/mcpConfHelper.ts
  Replace each Chinese comment, log or user-facing string with an English equivalent.

src/main/presenter/llmProviderPresenter/providers/openAICompatibleProvider.ts (2)

244-291: Multimodal content conversion needs improvement.

The current implementation converts image URLs to text strings (image: ${url}), which loses important information. Consider preserving the image data in a more structured format or documenting why this conversion is necessary.

For assistant messages with multimodal content, instead of converting to plain text, consider:

1. Filtering out image parts if the model doesn't support them
2. Preserving image references in a structured way
3. Adding a comment explaining why this conversion is needed

           // Handle multimodal content arrays
           const textParts: string[] = []
           for (const part of msg.content) {
             if (part.type === 'text' && part.text) {
               textParts.push(part.text)
             }
             if (part.type === 'image_url' && part.image_url?.url) {
+              // Note: Converting image URLs to text for models that don't support multimodal content
+              // This ensures the conversation history remains accessible when switching models
               textParts.push(`image: ${part.image_url.url}`)
             }
           }

---

375-577: Comprehensive image generation implementation with room for improvement.

The image generation handler is well-structured with proper error handling, but there are a few areas that could be improved:

1. Temporary file cleanup could fail silently
2. Hard-coded temporary directory path may not work on all systems
3. Magic strings for model names

-        // 创建临时文件
-        const imagePath = `/tmp/openai_image_${Date.now()}.png`
+        // Create temporary file using OS-appropriate temp directory
+        const tmpDir = app.getPath('temp')
+        const imagePath = path.join(tmpDir, `openai_image_${Date.now()}.png`)
         await new Promise<void>((resolve, reject) => {
           fs.writeFile(imagePath, imageBuffer, (err: Error | null) => {
             if (err) {
               reject(err)
             } else {
               resolve()
             }
           })
         })

         // ... existing code ...

-        // 清理临时文件
+        // Clean up temporary file
         try {
           fs.unlinkSync(imagePath)
         } catch (e) {
           console.error('[handleImgGeneration] Failed to delete temporary file:', e)
+          // Consider scheduling async cleanup or using a cleanup queue
         }

Also consider extracting the model names to constants:

+const HIGH_QUALITY_IMAGE_MODELS = ['gpt-image-1', 'gpt-4o-image', 'gpt-4o-all'] as const

-        if (modelId === 'gpt-image-1' || modelId === 'gpt-4o-image' || modelId === 'gpt-4o-all') {
+        if (HIGH_QUALITY_IMAGE_MODELS.includes(modelId as any)) {

src/main/presenter/mcpPresenter/toolManager.ts (3)

395-407: Fix MCPToolResponse shape mismatch with LLM presenter (permission flow).

LLM presenter reads toolResponse.rawData.requiresPermission, but here you return requiresPermission and permissionRequest at the top level and omit rawData entirely in both the permission and success paths. This will cause runtime errors when a permission is required and also loses the raw tool payload on success.

Apply this diff to standardize the response:

@@
-      if (!hasPermission) {
+      if (!hasPermission) {
         console.warn(
           `Permission required for tool '${originalName}' on server '${toolServerName}'.`
         )
@@
-        return {
-          toolCallId: toolCall.id,
-          content: `components.messageBlockPermissionRequest.description.${permissionType}`,
-          isError: false,
-          requiresPermission: true,
-          permissionRequest: {
-            toolName: originalName,
-            serverName: toolServerName,
-            permissionType,
-            description: `Allow ${originalName} to perform ${permissionType} operations on ${toolServerName}?`
-          }
-        }
+        return {
+          toolCallId: toolCall.id,
+          content: `components.messageBlockPermissionRequest.description.${permissionType}`,
+          isError: false,
+          rawData: {
+            requiresPermission: true,
+            permissionRequest: {
+              toolName: originalName,
+              serverName: toolServerName,
+              permissionType,
+              description: `Allow ${originalName} to perform ${permissionType} operations on ${toolServerName}?`
+            }
+          }
+        }
       }
@@
-      const response: MCPToolResponse = {
+      const response: MCPToolResponse = {
         toolCallId: toolCall.id,
         content: formattedContent,
-        isError: result.isError
+        isError: result.isError,
+        rawData: result
       }

Also applies to: 434-442

---

309-315: Avoid logging full tool-call arguments (risk of PII/secrets).

You log rawArguments and even the full argsString on parse failure. These can contain sensitive data.

Minimize/obfuscate arguments in logs:

-      console.log(`[ToolManager] Calling tool:`, {
-        requestedName: finalName,
-        originalName: finalName,
-        serverName: toolCall.server?.name || 'unknown',
-        rawArguments: argsString
-      })
+      console.log(`[ToolManager] Calling tool:`, {
+        requestedName: finalName,
+        originalName: finalName,
+        serverName: toolCall.server?.name || 'unknown',
+        argsPreview: `${argsString?.slice(0, 200) ?? ''}${argsString && argsString.length > 200 ? '…' : ''}`
+      })
@@
-          console.error('Error parsing tool call arguments even after jsonrepair:', argsString, e)
+          console.error('Error parsing tool call arguments even after jsonrepair:', (e instanceof Error ? e.message : String(e)))

Also applies to: 355-366

---

223-268: Translate remaining Chinese comments to English.

The PR goal and guidelines require English-only logs/comments. These blocks still contain Chinese.

I can push a follow-up patch translating these comments if you want.

Also applies to: 270-303, 455-480, 482-503, 505-584

src/main/presenter/llmProviderPresenter/index.ts (5)

1115-1141: Be resilient to MCPToolResponse shape (permission flow).

After the fix in ToolManager, rawData.requiresPermission will exist, but guard for both shapes to avoid runtime regressions.

-                if (toolResponse.rawData.requiresPermission) {
+                const requiresPermission =
+                  (toolResponse as any)?.rawData?.requiresPermission ??
+                  (toolResponse as any)?.requiresPermission === true
+                if (requiresPermission) {
@@
-                      tool_call_server_name: toolResponse.rawData.permissionRequest?.serverName,
+                      tool_call_server_name:
+                        (toolResponse as any)?.rawData?.permissionRequest?.serverName ??
+                        (toolResponse as any)?.permissionRequest?.serverName,
@@
-                      permission_request: toolResponse.rawData.permissionRequest
+                      permission_request:
+                        (toolResponse as any)?.rawData?.permissionRequest ??
+                        (toolResponse as any)?.permissionRequest

---

1249-1255: Translate user-facing prompt to English.

This string is shown to the model and potentially the UI; keep it English per PR goal.

-                  const userPromptText =
-                    '以上是你刚执行的工具调用及其响应信息，已帮你插入，请仔细阅读工具响应，并继续你的回答。'
+                  const userPromptText =
+                    'The above is the tool call you just executed and its response. It has been inserted for you. Please read the tool response carefully and continue your answer.'

---

1313-1345: Translate tool error insertion to English.

Keep assistant/user-inserted error text fully English.

-                  const formattedErrorText = `编号为 ${toolCall.id} 的工具 ${toolCall.name} 调用执行失败: ${errorMessage}`
+                  const formattedErrorText = `The tool call (id: ${toolCall.id}, name: ${toolCall.name}) failed to execute: ${errorMessage}`
@@
-                  const userPromptText =
-                    '以上是你刚调用的工具及其执行的错误信息，已帮你插入，请根据情况继续回答或重新尝试。'
+                  const userPromptText =
+                    'The above is the tool you just called and its error details. Please continue your answer accordingly or try again.'

---

292-304: Translate remaining Chinese comments to English.

Multiple blocks still use Chinese; align with “Use English for all logs and comments”.

I can submit a quick translation pass over these sections.

Also applies to: 336-356, 358-376, 384-418, 496-540

---

306-334: Await stopAllStreams() in setProviders and update signature to async

• Change setProviders(providers: LLM_PROVIDER[]): void to async setProviders(providers: LLM_PROVIDER[]): Promise<void> and replace this.stopAllStreams() with await this.stopAllStreams().
• Update the ILlmProviderPresenter.setProviders(...) signature in all .d.ts files (src/shared/types/presenters/llmprovider.presenter.d.ts, legacy.presenters.d.ts), adjust all call sites in src/main/presenter/index.ts (around lines 154, 169) to await llmProviderPresenter.setProviders(...), and modify the test at test/main/presenter/llmProviderPresenter.test.ts:515 to await the async call.

src/main/presenter/mcpPresenter/inMemoryServers/filesystem.ts (2)

274-318: Path traversal risk: startsWith() directory check is unsafe. Use path.relative boundary checks.

-    const normalizedRequested = this.normalizePath(absolute)
-
-    // Check if path is within allowed directories
-    const isAllowed = this.allowedDirectories.some((dir) => normalizedRequested.startsWith(dir))
+    const normalizedRequested = this.normalizePath(absolute)
+
+    // Check if path is within allowed directories using path.relative
+    const isAllowed = this.allowedDirectories.some((dir) => {
+      const rel = path.relative(dir, normalizedRequested)
+      return rel === '' || (!rel.startsWith('..') && !path.isAbsolute(rel))
+    })

---

571-589: Bug: duplicate/invalid RegExp flags ("gig"/"gg") and incorrect count when global=false.

-      const modifiedContent = normalizedOriginal.replace(regex, replacement)
-      const matches = Array.from(normalizedOriginal.matchAll(new RegExp(pattern, regexFlags + 'g')))
-      const replacements = matches.length
+      const modifiedContent = normalizedOriginal.replace(regex, replacement)
+      // Always use a counting regex with global flag; preserve case sensitivity
+      const countFlags = caseSensitive ? 'g' : 'gi'
+      const countRegex = new RegExp(pattern, countFlags)
+      const matches = normalizedOriginal.match(countRegex) || []
+      const replacements = global ? matches.length : (matches.length > 0 ? 1 : 0)

src/main/presenter/llmProviderPresenter/providers/doubaoProvider.ts (1)

51-76: Avoid mutating the shared OpenAI client by monkey-patching create

• In DoubaoProvider.coreStream (src/main/presenter/llmProviderPresenter/providers/doubaoProvider.ts:51–76), remove the runtime override of this.openai.chat.completions.create.
• Instead, add a provider-specific hook in OpenAICompatibleProvider (e.g. a buildChatParamsOverride) that merges { thinking: { type: 'enabled' } } into the params for this one request, so you never replace methods on the shared client.

src/main/presenter/mcpPresenter/inMemoryServers/autoPromptingServer.ts (1)

186-189: Unescaped RegExp on placeholder names can break replacement.
If a parameter name contains RegExp metacharacters (e.g., "+", "[]"), replace will misbehave. Use split/join or escape the pattern.

Apply within this block:

-          filledContent = filledContent.replace(new RegExp(`{{${param.name}}}`, 'g'), value)
+          // Avoid RegExp pitfalls by using plain string replacement semantics
+          filledContent = filledContent.split(`{{${param.name}}}`).join(value)

src/main/presenter/llmProviderPresenter/baseProvider.ts (1)

102-121: Use async/await to make errors catchable by the try/catch.
The current then-chain escapes the surrounding try/catch. Convert to await for reliability.

   protected async init() {
     if (this.provider.enable) {
       try {
-        this.isInitialized = true
-        this.fetchModels()
-          .then(() => {
-            return this.autoEnableModelsIfNeeded()
-          })
-          .then(() => {
-            console.info('Provider initialized successfully:', this.provider.name)
-          })
-        // Check if we need to automatically enable all models
+        this.isInitialized = true
+        await this.fetchModels()
+        await this.autoEnableModelsIfNeeded()
+        console.info('Provider initialized successfully:', this.provider.name)
       } catch (error) {
         console.warn('Provider initialization failed:', this.provider.name, error)
       }
     }
   }

src/main/presenter/configPresenter/aes.ts (1)

3-5: Security: consider migrating from AES‑256‑CBC to AES‑256‑GCM (AEAD) in a follow‑up PR

CBC provides confidentiality but not integrity; GCM adds authentication (authTag), preventing malleability. Given this is used in main process config, this is a worthwhile upgrade.

Minimal outline (illustrative):

-const ALGORITHM = 'aes-256-cbc'
+const ALGORITHM = 'aes-256-gcm'
@@
-  static encrypt(plainText: string, key: Buffer, iv?: Buffer): { cipherText: string; iv: string } {
+  static encrypt(plainText: string, key: Buffer, iv?: Buffer): { cipherText: string; iv: string; authTag: string } {
@@
-      const cipher = crypto.createCipheriv(ALGORITHM, key, usedIv)
+      const cipher = crypto.createCipheriv(ALGORITHM, key, usedIv)
@@
-      return {
-        cipherText: encrypted,
-        iv: usedIv.toString('hex')
-      }
+      const authTag = cipher.getAuthTag().toString('hex')
+      return { cipherText: encrypted, iv: usedIv.toString('hex'), authTag }
@@
-  static decrypt(cipherText: string, key: Buffer, iv: string): string {
+  static decrypt(cipherText: string, key: Buffer, iv: string, authTag: string): string {
@@
-      const decipher = crypto.createDecipheriv(ALGORITHM, key, ivBuffer)
+      const decipher = crypto.createDecipheriv(ALGORITHM, key, ivBuffer)
+      decipher.setAuthTag(Buffer.from(authTag, 'hex'))

Note: version as a new API to avoid breaking existing data; add migration path.

Also applies to: 30-55, 62-82