fix(chat): end a turn on a failure when its run failed

The answer shown for a turn whose rows never arrived came from the run's
result, and the run's own verdict was thrown away with it. A flow that
failed before writing anything therefore ended as a successful turn
carrying its error envelope rendered as JSON: Retry was not offered, and
a message queued behind it went straight out into the same failure.

The row now carries what the run said of itself, and a failure is shown
as what it failed with rather than as the envelope around it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

frontend/src/lib/components/flows/conversations/FlowChatManager.svelte.ts

@@ -112,6 +112,14 @@ function toStreamEvent(event: AgentStreamEvent): StreamEvent {
 	}
 }
 
+/** What a run that failed said, as the line a turn ends on. Windmill wraps a failure in an
+ *  error envelope; anything else is shown as it came. */
+function failedRunMessage(result: unknown): string | undefined {
+	const message = (result as { error?: { message?: unknown } })?.error?.message
+	if (typeof message === 'string' && message !== '') return message
+	return extractChatAnswer(result)
+}
+
 function emptyStatus(): TurnStatus {
 	return {
 		isLoading: false,
@@ -1486,11 +1494,13 @@ export class FlowChatManager {
 		const api = this.#chatApi()
 		const signal = turn.signal
 		let flowResult: unknown
+		let flowSucceeded = true
 		while (this.#isCurrent(turn)) {
 			try {
-				const { completed, result } = await api.getCompletedResult(jobId, signal)
+				const { completed, success, result } = await api.getCompletedResult(jobId, signal)
 				if (completed) {
 					flowResult = result
+					flowSucceeded = success !== false
 					break
 				}
 			} catch (error) {
@@ -1507,7 +1517,11 @@ export class FlowChatManager {
 		// answer the row would have carried, and showing it is what keeps a finished turn
 		// from reading as one that produced nothing. A reload replaces it with the row.
 		if (read === 0) {
-			const answer = extractChatAnswer(flowResult)
+			// A run that failed says so in an error rather than an answer, and the row has to
+			// carry that: it is what the transcript reads the turn's outcome off, which is what
+			// offers Retry and what holds a queued message back from running into the same
+			// failure.
+			const answer = flowSucceeded ? extractChatAnswer(flowResult) : failedRunMessage(flowResult)
 			if (typeof answer === 'string' && answer !== '') {
 				this.#rowsById[turn.conversationId] = [
 					...this.#rowsOf(turn.conversationId),
@@ -1519,7 +1533,7 @@ export class FlowChatManager {
 						created_at: new Date().toISOString(),
 						created_seq: 0,
 						job_id: jobId,
-						success: true
+						success: flowSucceeded
 					} as ChatMessage
 				]
 			}

frontend/src/lib/components/flows/conversations/FlowChatManager.test.ts

@@ -27,7 +27,8 @@ const { streamCalls, streamScript, jobCompleted } = vi.hoisted(() => ({
 	jobCompleted: {
 		value: true as boolean | 'throw',
 		gate: undefined as Promise<void> | undefined,
-		result: {} as unknown
+		result: {} as unknown,
+		success: true
 	}
 }))
 
@@ -45,7 +46,11 @@ vi.mock('windmill-chat', async (importOriginal) => {
 		async getCompletedResult() {
 			if (jobCompleted.gate) await jobCompleted.gate
 			if (jobCompleted.value === 'throw') throw new Error('job status unavailable')
-			return { completed: jobCompleted.value, success: true, result: jobCompleted.result }
+			return {
+				completed: jobCompleted.value,
+				success: jobCompleted.success,
+				result: jobCompleted.result
+			}
 		}
 	}
 	return { ...actual, WindmillChatApi: FakeApi }
@@ -748,6 +753,7 @@ describe('a conversation opened while its run is still going', () => {
 			.mockReset()
 			.mockResolvedValue([] as any)
 		jobCompleted.value = true
+		jobCompleted.success = true
 		jobCompleted.result = { windmill_chat_answer: 'the answer the run produced' }
 		const manager = (live = managerWithRows())
 		;(manager as any).initialize(
@@ -765,6 +771,36 @@ describe('a conversation opened while its run is still going', () => {
 		expect(manager.messages.map((m) => m.content)).toContain('the answer the run produced')
 	})
 
+	/**
+	 * A run that fails before writing anything still has to end the turn as a failure: the
+	 * transcript reads a turn's outcome off its last row, and that is what offers Retry and
+	 * what holds a message queued behind the turn back from running into the same failure.
+	 */
+	it('ends on a failure when the run failed and wrote nothing', async () => {
+		vi.mocked(FlowConversationsService.listConversationMessages)
+			.mockReset()
+			.mockResolvedValue([] as any)
+		jobCompleted.value = true
+		jobCompleted.success = false
+		jobCompleted.result = { error: { message: 'the provider refused' } }
+		const manager = (live = managerWithRows())
+		;(manager as any).initialize(
+			vi.fn(async () => 'job-1'),
+			'u/admin/flow',
+			false
+		)
+		manager.operatingWorkspace = () => 'ws'
+		manager.selectedConversationId = 'a'
+		manager.inputMessage = 'ask'
+
+		await manager.sendMessage(undefined, undefined, 'a')
+		await vi.waitFor(() => expect(manager.isConversationBusy('a')).toBe(false))
+
+		const last = manager.messages.at(-1)
+		expect(last?.content).toBe('the provider refused')
+		expect(last?.success).toBe(false)
+	})
+
 	it('leaves no card spinning when Stop ends the turn', async () => {
 		vi.mocked(FlowConversationsService.listConversationMessages)
 			.mockReset()

frontend/src/lib/components/flows/conversations/flowChatViewHost.svelte.ts

@@ -477,16 +477,10 @@ export class FlowChatViewHost implements ChatViewHost {
 			}
 			attachedLocally = true
 		}
-		// The one place a run starts, and so the one place to refuse to start one. Every way
-		// in waits on something first — uploading the attachments, asking a failed job what
-		// it ran with — and the panel can be replaced while any of them is in flight; the
-		// manager it would hand the turn to is then one nobody is reading, pointed at
-		// whichever flow replaced this one.
-		// The one place a run starts, and so the one place to refuse to start one. Every way
-		// in waits on something first — uploading the attachments, asking a failed job what
-		// it ran with — and the panel can be replaced while any of them is in flight; the
-		// manager it would hand the turn to is then one nobody is reading, pointed at
-		// whichever flow replaced this one.
+		// The one place a run starts, and so the one place to refuse to start one: every way
+		// in waits on something first — an upload, a failed job's arguments — and the panel
+		// can be replaced while it waits, leaving the manager it would hand the turn to one
+		// nobody is reading, pointed at whichever flow replaced this one.
 		if (this.#gone) return false
 		const sentInputs = this.#describeSentInputs(args, images, blobs, attachedLocally)