fix(editor): Add limitations section to AI builder responder prompt (#24457)

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

packages/@n8n/ai-workflow-builder.ee/src/prompts/agents/responder.prompt.ts

@@ -39,6 +39,13 @@ const RESPONSE_STYLE = `- Keep responses focused and not overly long
 - Be conversational and helpful
 - Do not use emojis in your response`;
 
+const GUARDRAILS = `Your capabilities are focused on workflow building:
+- You work from your existing knowledge of n8n nodes and integrations
+- You help users design and configure workflows based on their requirements
+- You provide guidance on node configuration and workflow structure
+
+If a user asks you to search for information or look something up online, let them know you can help build workflows based on your existing knowledge of n8n nodes and integrations, though you don't have access to external websites or real-time information.`;
+
 /**
  * Error guidance prompts for different error scenarios (AI-1812)
  */
@@ -74,6 +81,7 @@ export function buildGeneralErrorGuidance(): string {
 export function buildResponderPrompt(): string {
 	return prompt()
 		.section('role', RESPONDER_ROLE)
+		.section('guardrails', GUARDRAILS)
 		.section('workflow_completion_responses', WORKFLOW_COMPLETION)
 		.section('conversational_responses', CONVERSATIONAL_RESPONSES)
 		.section('response_style', RESPONSE_STYLE)

packages/@n8n/ai-workflow-builder.ee/src/subgraphs/test/integration/responder-limitations.integration.test.ts

@@ -0,0 +1,204 @@
+/**
+ * Integration tests for responder agent limitations (AI-1894)
+ *
+ * Tests the behavior when users ask the responder agent to perform
+ * actions it cannot do (like web search), ensuring proper acknowledgment
+ * of limitations and helpful alternatives.
+ *
+ * To run these tests:
+ * ENABLE_INTEGRATION_TESTS=true N8N_AI_ANTHROPIC_KEY=your-key pnpm test:integration
+ */
+
+import type { BaseChatModel } from '@langchain/core/language_models/chat_models';
+import { HumanMessage } from '@langchain/core/messages';
+
+import { ResponderAgent } from '@/agents/responder.agent';
+import {
+	setupIntegrationLLM,
+	shouldRunIntegrationTests,
+} from '@/chains/test/integration/test-helpers';
+import type { SimpleWorkflow } from '@/types/workflow';
+
+describe('Responder Limitations - Integration Tests (AI-1894)', () => {
+	let llm: BaseChatModel;
+
+	// Skip all tests if integration tests are not enabled
+	const skipTests = !shouldRunIntegrationTests();
+
+	// Set default timeout for all tests in this suite
+	jest.setTimeout(120000); // 2 minutes
+
+	beforeAll(async () => {
+		// Override console.log to use process.stdout directly
+		jest.spyOn(console, 'log').mockImplementation((...args: unknown[]) => {
+			process.stdout.write(args.map(String).join(' ') + '\n');
+		});
+
+		if (skipTests) {
+			console.log(
+				'\n⏭️  Skipping integration tests. Set ENABLE_INTEGRATION_TESTS=true to run them.\n',
+			);
+			return;
+		}
+
+		console.log('\n🚀 Setting up responder limitations integration test environment...\n');
+
+		// Load real LLM
+		llm = await setupIntegrationLLM();
+	});
+
+	describe('Web Search Limitations', () => {
+		const emptyWorkflow: SimpleWorkflow = {
+			nodes: [],
+			connections: {},
+			name: '',
+		};
+
+		it('should explain it cannot search the web when asked', async () => {
+			if (skipTests) return;
+
+			console.log('\n📝 Testing web search limitation acknowledgment...\n');
+
+			const responderAgent = new ResponderAgent({ llm });
+
+			const response = await responderAgent.invoke({
+				messages: [
+					new HumanMessage({
+						content: 'Can you search the web for information about Slack API rate limits?',
+					}),
+				],
+				coordinationLog: [],
+				discoveryContext: null,
+				workflowJSON: emptyWorkflow,
+			});
+
+			const messageContent =
+				typeof response.content === 'string'
+					? response.content.toLowerCase()
+					: JSON.stringify(response.content).toLowerCase();
+
+			console.log('Response:', response.content);
+
+			// Should acknowledge limitation
+			const acknowledgesLimitation =
+				messageContent.includes('cannot') ||
+				messageContent.includes("can't") ||
+				messageContent.includes('unable') ||
+				messageContent.includes("don't have") ||
+				messageContent.includes('not able');
+
+			const mentionsWebOrSearch =
+				messageContent.includes('web') ||
+				messageContent.includes('search') ||
+				messageContent.includes('online') ||
+				messageContent.includes('internet') ||
+				messageContent.includes('browse');
+
+			expect(acknowledgesLimitation).toBe(true);
+			expect(mentionsWebOrSearch).toBe(true);
+
+			// Should offer alternative help
+			const offersAlternative =
+				messageContent.includes('help') ||
+				messageContent.includes('assist') ||
+				messageContent.includes('workflow') ||
+				messageContent.includes('build') ||
+				messageContent.includes('can') ||
+				messageContent.includes('however');
+
+			expect(offersAlternative).toBe(true);
+		});
+
+		it('should offer to build workflows based on existing knowledge when web search requested', async () => {
+			if (skipTests) return;
+
+			console.log('\n📝 Testing alternative help offer when documentation lookup requested...\n');
+
+			const responderAgent = new ResponderAgent({ llm });
+
+			const response = await responderAgent.invoke({
+				messages: [
+					new HumanMessage({
+						content:
+							'Look up the latest documentation for the Slack node and tell me what operations are available',
+					}),
+				],
+				coordinationLog: [],
+				discoveryContext: null,
+				workflowJSON: emptyWorkflow,
+			});
+
+			const messageContent =
+				typeof response.content === 'string'
+					? response.content.toLowerCase()
+					: JSON.stringify(response.content).toLowerCase();
+
+			console.log('Response:', response.content);
+
+			// Should NOT claim it retrieved documentation from the web
+			const claimsRetrieval =
+				messageContent.includes('i found') ||
+				messageContent.includes('i retrieved') ||
+				messageContent.includes('according to the documentation') ||
+				messageContent.includes('the documentation says');
+
+			expect(claimsRetrieval).toBe(false);
+
+			// Should explain limitation or offer to help with workflow building
+			const providesContext =
+				messageContent.includes('cannot') ||
+				messageContent.includes("can't") ||
+				messageContent.includes('unable') ||
+				messageContent.includes('existing knowledge') ||
+				messageContent.includes('help') ||
+				messageContent.includes('build') ||
+				messageContent.includes('workflow') ||
+				messageContent.includes('slack'); // Should at least mention Slack in some helpful way
+
+			expect(providesContext).toBe(true);
+		});
+
+		it('should redirect real-time data requests to workflow building', async () => {
+			if (skipTests) return;
+
+			console.log('\n📝 Testing redirect for real-time data requests...\n');
+
+			const responderAgent = new ResponderAgent({ llm });
+
+			const response = await responderAgent.invoke({
+				messages: [
+					new HumanMessage({
+						content: 'What is the current Bitcoin price?',
+					}),
+				],
+				coordinationLog: [],
+				discoveryContext: null,
+				workflowJSON: emptyWorkflow,
+			});
+
+			const messageContent =
+				typeof response.content === 'string'
+					? response.content.toLowerCase()
+					: JSON.stringify(response.content).toLowerCase();
+
+			console.log('Response:', response.content);
+
+			// Should NOT provide a price (it doesn't have access to real-time data)
+			const providesPrice =
+				/\$[\d,]+/.test(messageContent) || /[\d,]+\s*(usd|dollars)/i.test(messageContent);
+			expect(providesPrice).toBe(false);
+
+			// Should acknowledge limitation or offer workflow alternative
+			const acknowledgesOrOffers =
+				messageContent.includes('cannot') ||
+				messageContent.includes("can't") ||
+				messageContent.includes("don't have") ||
+				messageContent.includes('real-time') ||
+				messageContent.includes('workflow') ||
+				messageContent.includes('api') ||
+				messageContent.includes('build');
+
+			expect(acknowledgesOrOffers).toBe(true);
+		});
+	});
+});