fix(ai-builder): Handle data table name conflict gracefully instead of looping (no-changelog)

When the builder creates a workflow involving a data table, it creates
the table in run #1 then tries to create it again in run #2 (rebuild
phase). The DataTableNameConflictError causes the agent to retry in a
loop until the build times out.

Fix: catch the conflict error and return `denied: true` with a reason
guiding the agent to use the existing table. The agent already handles
denied responses gracefully.

Also adds family-assistant eval test case (Telegram + AI agent +
calendar + shopping list + web search) which exercises this code path.

Before: BUILD FAILED after 175s of looping
After: BUILD SUCCESS in ~3 min, 3/3 scenarios passing

Ref: https://linear.app/n8n/issue/AI-2358

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: JoseBra

packages/@n8n/instance-ai/evaluations/data/workflows/family-assistant.json

@@ -0,0 +1,26 @@
+{
+	"prompt": "I want to build a personal family assistant that can handle our calendars, maintain a shopping list and search the web. I want to interact with it via telegram. Configure all nodes as completely as possible and don't ask me for credentials, I'll set them up later.",
+	"complexity": "complex",
+	"tags": ["build", "telegram", "ai-agent", "google-calendar", "web-search"],
+	"triggerType": "webhook",
+	"scenarios": [
+		{
+			"name": "add-shopping-item",
+			"description": "User asks to add an item to the shopping list via Telegram",
+			"dataSetup": "The Telegram trigger receives a message: 'Add milk and eggs to the shopping list'. The AI agent processes the request and adds the items. All tool calls return success responses.",
+			"successCriteria": "The workflow executes without errors. The AI agent receives the Telegram message and identifies it as a shopping list request. The items are added to the shopping list. A confirmation message is sent back via Telegram."
+		},
+		{
+			"name": "check-calendar",
+			"description": "User asks about upcoming calendar events",
+			"dataSetup": "The Telegram trigger receives a message: 'What do I have on my calendar this week?'. The Google Calendar API returns 2 events: 'Team standup' on Monday at 10am and 'Dentist appointment' on Wednesday at 3pm. The AI agent processes and responds.",
+			"successCriteria": "The workflow executes without errors. The AI agent receives the Telegram message, queries the calendar, and sends back a summary of the upcoming events via Telegram."
+		},
+		{
+			"name": "web-search",
+			"description": "User asks to search the web for something",
+			"dataSetup": "The Telegram trigger receives a message: 'Search for the best Italian restaurants near me'. The web search returns a few results. The AI agent summarizes and responds.",
+			"successCriteria": "The workflow executes without errors. The AI agent receives the Telegram message, performs a web search, and sends back a summary of the results via Telegram."
+		}
+	]
+}

packages/@n8n/instance-ai/src/tools/data-tables/create-data-table.tool.ts

@@ -28,6 +28,20 @@ export const createDataTableResumeSchema = z.object({
 	approved: z.boolean(),
 });
 
+/**
+ * Check if an error (or its cause chain) is a DataTableNameConflictError.
+ * The error class lives in packages/cli so we can't import it directly —
+ * instead we match on the class name through the cause chain.
+ */
+function isNameConflictError(error: unknown): boolean {
+	let current: unknown = error;
+	while (current instanceof Error) {
+		if (current.constructor.name === 'DataTableNameConflictError') return true;
+		current = (current as Error & { cause?: unknown }).cause;
+	}
+	return false;
+}
+
 export function createCreateDataTableTool(context: InstanceAiContext) {
 	return createTool({
 		id: 'create-data-table',
@@ -91,10 +105,26 @@ export function createCreateDataTableTool(context: InstanceAiContext) {
 			}
 
 			// State 3: Approved or always_allow — execute
-			const table = await context.dataTableService.create(input.name, input.columns, {
-				projectId: input.projectId,
-			});
-			return { table };
+			try {
+				const table = await context.dataTableService.create(input.name, input.columns, {
+					projectId: input.projectId,
+				});
+				return { table };
+			} catch (error) {
+				// If table already exists, guide the agent to use the existing one
+				// rather than throwing — which would cause the agent to retry in a loop
+				if (isNameConflictError(error)) {
+					const tables = await context.dataTableService.list({ projectId: input.projectId });
+					const existing = tables.find((t) => t.name === input.name);
+					if (existing) {
+						return {
+							denied: true,
+							reason: `Table "${input.name}" already exists. Use list-data-tables to find it and get-data-table-schema to check its columns.`,
+						};
+					}
+				}
+				throw error;
+			}
 		},
 	});
 }