fix: truncate userInputMessage to first 500k characters (#1327)

* fix: truncate userInputMessage to 500k

Author: leigaol

server/aws-lsp-codewhisperer/src/language-server/agenticChat/agenticChatController.test.ts

@@ -962,19 +962,20 @@ describe('AgenticChatController', () => {
             assert.strictEqual(typedChatResult.data?.body, errorMsg)
         })
 
-        it('does not make backend request when input is too long ', async function () {
-            const input = 'X'.repeat(generateAssistantResponseInputLimit + 1)
-            const chatResult = await chatController.onChatPrompt(
-                { tabId: mockTabId, prompt: { prompt: input } },
-                mockCancellationToken
+        it('truncate input to 500k character ', async function () {
+            const input = 'X'.repeat(generateAssistantResponseInputLimit + 10)
+            generateAssistantResponseStub.restore()
+            generateAssistantResponseStub = sinon.stub(CodeWhispererStreaming.prototype, 'generateAssistantResponse')
+            generateAssistantResponseStub.callsFake(() => {})
+            await chatController.onChatPrompt({ tabId: mockTabId, prompt: { prompt: input } }, mockCancellationToken)
+            assert.ok(generateAssistantResponseStub.called)
+            const calledRequestInput: GenerateAssistantResponseCommandInput =
+                generateAssistantResponseStub.firstCall.firstArg
+            assert.deepStrictEqual(
+                calledRequestInput.conversationState?.currentMessage?.userInputMessage?.content?.length,
+                generateAssistantResponseInputLimit
             )
-
-            const typedChatResult = chatResult as ResponseError<ChatResult>
-            assert.ok(typedChatResult.message.includes('too long'))
-            assert.ok(typedChatResult.data?.body?.includes('too long'))
-            assert.ok(generateAssistantResponseStub.notCalled)
         })
-
         it('shows generic errorMsg on internal errors', async function () {
             const chatResult = await chatController.onChatPrompt(
                 { tabId: mockTabId, prompt: { prompt: 'Hello' } },
@@ -1072,26 +1073,6 @@ describe('AgenticChatController', () => {
             assert.strictEqual(typedChatResult.message, 'invalid state')
         })
 
-        it('returns a user-friendly message when input is too long', async () => {
-            generateAssistantResponseStub.restore()
-            generateAssistantResponseStub = sinon.stub(CodeWhispererStreaming.prototype, 'generateAssistantResponse')
-            generateAssistantResponseStub.callsFake(() => {
-                const error = new Error('Input is too long')
-                throw error
-            })
-
-            const chatResult = await chatController.onChatPrompt(
-                { tabId: mockTabId, prompt: { prompt: 'Hello with large context' } },
-                mockCancellationToken
-            )
-
-            const typedChatResult = chatResult as ResponseError<ChatResult>
-            assert.strictEqual(
-                typedChatResult.data?.body,
-                'Too much context loaded. I have cleared the conversation history. Please retry your request with smaller input.'
-            )
-        })
-
         describe('#extractDocumentContext', () => {
             const typescriptDocument = TextDocument.create('file:///test.ts', 'typescript', 1, 'test')
             let extractDocumentContextStub: sinon.SinonStub

server/aws-lsp-codewhisperer/src/language-server/agenticChat/agenticChatController.ts

@@ -576,7 +576,7 @@ export class AgenticChatController implements ChatHandlers {
             await chatResultStream.writeResultBlock({ ...loadingMessage, messageId: loadingMessageId })
 
             // Phase 3: Request Execution
-            this.#validateRequest(currentRequestInput)
+            this.#truncateRequest(currentRequestInput)
             const response = await session.generateAssistantResponse(currentRequestInput)
             this.#features.logging.info(
                 `generateAssistantResponse ResponseMetadata: ${loggingUtils.formatObj(response.$metadata)}`
@@ -695,17 +695,21 @@ export class AgenticChatController implements ChatHandlers {
     }
 
     /**
-     * performs pre-validation of request before sending to backend service.
+     * performs truncation of request before sending to backend service.
      * @param request
      */
-    #validateRequest(request: GenerateAssistantResponseCommandInput) {
+    #truncateRequest(request: GenerateAssistantResponseCommandInput) {
         // Note: these logs are very noisy, but contain information redacted on the backend.
         this.#debug(`generateAssistantResponse Request: ${JSON.stringify(request, undefined, 2)}`)
+        if (!request?.conversationState?.currentMessage?.userInputMessage) {
+            return
+        }
         const message = request.conversationState?.currentMessage?.userInputMessage?.content
         if (message && message.length > generateAssistantResponseInputLimit) {
-            throw new AgenticChatError(
-                `Message is too long with ${message.length} characters, max is ${generateAssistantResponseInputLimit}`,
-                'PromptCharacterLimit'
+            this.#debug(`Truncating userInputMessage to ${generateAssistantResponseInputLimit} characters}`)
+            request.conversationState.currentMessage.userInputMessage.content = message.substring(
+                0,
+                generateAssistantResponseInputLimit
             )
         }
     }