feat: expand multimodal and knowledge workflows

src/shared/builtin-mcp-servers.ts

@@ -19,8 +19,13 @@ export const builtinMcpServers = [
     id: 'knowledge-base',
     name: '知识库 MCP',
     description:
-      '搜索当前对话明确选择的知识库，并返回可核验的来源与证据引用。',
+      '列出并搜索当前对话明确选择的知识库，返回可核验的来源与证据引用。',
     tools: [
+      {
+        name: 'knowledge_list',
+        description: '列出当前对话已授权的知识库及其说明。',
+        access: 'read'
+      },
       {
         name: 'knowledge_search',
         description: '搜索当前对话已授权的知识库并返回来源引用。',

src/shared/context-contracts.test.ts

@@ -0,0 +1,37 @@
+import { describe, expect, it } from 'vitest'
+import {
+  maximumPastedImageBytes,
+  pastedImageInputSchema
+} from './contracts'
+
+describe('context contracts', () => {
+  it('accepts bounded pasted image bytes in supported formats', () => {
+    expect(
+      pastedImageInputSchema.safeParse({
+        data: Uint8Array.from([0x89, 0x50, 0x4e, 0x47]),
+        mimeType: 'image/png'
+      }).success
+    ).toBe(true)
+  })
+
+  it('rejects empty, oversized, and unsupported pasted images', () => {
+    expect(
+      pastedImageInputSchema.safeParse({
+        data: new Uint8Array(),
+        mimeType: 'image/png'
+      }).success
+    ).toBe(false)
+    expect(
+      pastedImageInputSchema.safeParse({
+        data: new Uint8Array(maximumPastedImageBytes + 1),
+        mimeType: 'image/png'
+      }).success
+    ).toBe(false)
+    expect(
+      pastedImageInputSchema.safeParse({
+        data: Uint8Array.from([1]),
+        mimeType: 'image/gif'
+      }).success
+    ).toBe(false)
+  })
+})

src/shared/contracts.ts

@@ -231,6 +231,7 @@ export const defaultRuntimeSettings = {
   modelName: 'sonnet-5',
   modelProtocol: 'anthropic-messages',
   modelAuthentication: 'api-key',
+  supportsImageInput: false,
   imageGenerationQuality: 'auto',
   opencodeBaseUrl: '',
   opencodeEmbedded: true,
@@ -321,6 +322,7 @@ const modelProfileInputSchema = z
       .regex(/^[\w./:-]+$/, '模型名称包含不支持的字符'),
     protocol: modelProtocolSchema,
     authentication: modelAuthenticationSchema,
+    supportsImageInput: z.boolean().optional(),
     imageGenerationQuality: imageGenerationQualitySchema,
     apiKey: modelApiKeyUpdateSchema
   })
@@ -524,6 +526,7 @@ export type ModelConnectionSettings = {
   modelName: string
   protocol: ModelProtocol
   authentication: ModelAuthentication
+  supportsImageInput?: boolean
   imageGenerationQuality: ImageGenerationQuality
   apiKeyConfigured: boolean
   credentialSource: 'none' | 'encrypted' | 'environment'
@@ -535,6 +538,7 @@ export type RuntimeSettings = {
   modelName: string
   modelProtocol: ModelProtocol
   modelAuthentication: ModelAuthentication
+  supportsImageInput?: boolean
   imageGenerationQuality: ImageGenerationQuality
   opencodeBaseUrl: string
   opencodeEmbedded: boolean
@@ -564,6 +568,23 @@ export type RuntimeSettings = {
 
 export type ContextAttachment = ConversationAttachment
 
+export const maximumPastedImageBytes = 12 * 1024 * 1024
+
+export const pastedImageInputSchema = z
+  .object({
+    data: z
+      .instanceof(Uint8Array)
+      .refine((value) => value.byteLength > 0, '粘贴图片内容为空')
+      .refine(
+        (value) => value.byteLength <= maximumPastedImageBytes,
+        '粘贴图片不能超过 12MB'
+      ),
+    mimeType: z.enum(['image/jpeg', 'image/png', 'image/webp'])
+  })
+  .strict()
+
+export type PastedImageInput = z.infer<typeof pastedImageInputSchema>
+
 export const windowCaptureSourceIdSchema = z
   .string()
   .min(1)
@@ -1169,6 +1190,9 @@ export type DesktopApi = {
   }
   context: {
     selectFiles: () => Promise<ContextAttachment[]>
+    addPastedImage: (
+      input: PastedImageInput
+    ) => Promise<ContextAttachment>
     captureScreen: () => Promise<ContextAttachment>
     listWindows: () => Promise<WindowCaptureOption[]>
     captureWindow: (sourceId: string) => Promise<ContextAttachment>

src/shared/ipc-channels.ts

@@ -112,6 +112,7 @@ export const ipcChannels = {
   capabilitiesDefaultBrowserProfile: 'capabilities:browser-profile:default',
   capabilitiesRemoveBrowserProfile: 'capabilities:browser-profile:remove',
   contextSelectFiles: 'context:select-files',
+  contextAddPastedImage: 'context:add-pasted-image',
   contextCaptureScreen: 'context:capture-screen',
   contextListWindows: 'context:list-windows',
   contextCaptureWindow: 'context:capture-window',