fix(mothership): surface the real run error for browser-settled workflow tools (#9634)

* fix(mothership): surface the real run error for browser-settled workflow tools

* fix(mothership): clean up run logs and snapshots before the workspace in the test

Author: waleedlatif1

apps/sim/lib/workflows/executor/execution-state.ts

@@ -1,6 +1,6 @@
 import { db } from '@sim/db'
 import { workflowExecutionLogs } from '@sim/db/schema'
-import { isRecordLike } from '@sim/utils/object'
+import { filterUndefined, isRecordLike } from '@sim/utils/object'
 import { and, desc, eq, or, sql } from 'drizzle-orm'
 import { materializeExecutionData, TRACE_STORE_REF_KEY } from '@/lib/logs/execution/trace-store'
 import type { SerializableExecutionState } from '@/executor/execution/types'
@@ -188,6 +188,19 @@ export async function getWorkflowExecutionCopilotToolCallId(
   return typeof toolCallId === 'string' ? toolCallId : undefined
 }
 
+/**
+ * The message an error finalization (`force_failed`) stored as the run's `finalOutput.error`.
+ * Any other run's `finalOutput.error` is the workflow's own data, not a failure message.
+ */
+function readFailureMessage(executionData: Record<string, unknown>): string | undefined {
+  if (typeof executionData.error === 'string') return executionData.error
+  if (executionData.finalizationPath !== 'force_failed') return undefined
+  const { finalOutput } = executionData
+  return isRecordLike(finalOutput) && typeof finalOutput.error === 'string'
+    ? finalOutput.error
+    : undefined
+}
+
 /** Loads a terminal workflow result only when its server-persisted Copilot binding matches. */
 export async function getTrustedWorkflowToolExecution(
   executionId: string,
@@ -232,7 +245,7 @@ export async function getTrustedWorkflowToolExecution(
     ...(Object.hasOwn(executionData, 'finalOutput')
       ? { finalOutput: executionData.finalOutput }
       : {}),
-    ...(typeof executionData.error === 'string' ? { error: executionData.error } : {}),
+    ...filterUndefined({ error: readFailureMessage(executionData) }),
     blockLogs: state.blockLogs,
     provenance,
   }

apps/sim/lib/workflows/executor/trusted-tool-failure.integration.ts

@@ -0,0 +1,154 @@
+/** The failure message a browser-settled Copilot workflow tool reads back, against real PostgreSQL. */
+import { db } from '@sim/db'
+import {
+  user,
+  workflow,
+  workflowExecutionLogs,
+  workflowExecutionSnapshots,
+  workspace,
+} from '@sim/db/schema'
+import { deleteWorkspaceFixture, insertWorkspaceFixture } from '@sim/db/testing/workspace-fixtures'
+import { generateId } from '@sim/utils/id'
+import { eq } from 'drizzle-orm'
+import { afterAll, beforeAll, describe, expect, it } from 'vitest'
+import { LoggingSession } from '@/lib/logs/execution/logging-session'
+import type { TraceSpan, WorkflowState } from '@/lib/logs/types'
+import { getTrustedWorkflowToolExecution } from '@/lib/workflows/executor/execution-state'
+import type { SerializableExecutionState } from '@/executor/execution/types'
+import { emptyResolvedSecretTraceProvenance } from '@/executor/utils/resolved-secret-trace-registry'
+
+const ids = {
+  owner: `trusted-tool-failure-${generateId()}`,
+  workspace: generateId(),
+  workflow: generateId(),
+}
+
+const workflowState: WorkflowState = {
+  blocks: {
+    start: {
+      id: 'start',
+      type: 'starter',
+      name: 'Start',
+      position: { x: 0, y: 0 },
+      subBlocks: {},
+      outputs: {},
+      enabled: true,
+    },
+  },
+  edges: [],
+  loops: {},
+  parallels: {},
+}
+
+function executionState(): SerializableExecutionState {
+  return {
+    blockStates: {},
+    executedBlocks: ['start'],
+    blockLogs: [],
+    decisions: { router: {}, condition: {} },
+    completedLoops: [],
+    activeExecutionPath: [],
+    resolvedSecretTraceProvenance: emptyResolvedSecretTraceProvenance(),
+  }
+}
+
+/** Starts a run bound to a Copilot tool call, the way the browser settlement path launches one. */
+async function startBoundRun() {
+  const executionId = generateId()
+  const toolCallId = generateId()
+  const session = new LoggingSession(ids.workflow, executionId, 'manual', 'trusted-tool-failure')
+  session.setTrustedExecutionCorrelation({
+    executionId,
+    requestId: 'trusted-tool-failure',
+    source: 'workflow',
+    workflowId: ids.workflow,
+    triggerType: 'copilot',
+    copilotToolCallId: toolCallId,
+  })
+  expect(
+    await session.safeStart({ userId: ids.owner, workspaceId: ids.workspace, workflowState })
+  ).toBe(true)
+  return { session, executionId, toolCallId }
+}
+
+function readBack(run: { executionId: string; toolCallId: string }) {
+  return getTrustedWorkflowToolExecution(run.executionId, ids.workflow, run.toolCallId)
+}
+
+beforeAll(async () => {
+  const now = new Date()
+  await db.insert(user).values({
+    id: ids.owner,
+    name: 'Trusted Tool Failure',
+    email: `${ids.owner}@trusted-tool-failure.test`,
+    emailVerified: true,
+    createdAt: now,
+    updatedAt: now,
+  })
+  await insertWorkspaceFixture(db, {
+    id: ids.workspace,
+    name: 'Trusted Tool Failure',
+    ownerId: ids.owner,
+    billedAccountUserId: ids.owner,
+  })
+  await db.insert(workflow).values({
+    id: ids.workflow,
+    userId: ids.owner,
+    workspaceId: ids.workspace,
+    name: 'Trusted Tool Failure',
+    lastSynced: now,
+    createdAt: now,
+    updatedAt: now,
+  })
+}, 30_000)
+
+afterAll(async () => {
+  await db.delete(workflowExecutionLogs).where(eq(workflowExecutionLogs.workflowId, ids.workflow))
+  await db
+    .delete(workflowExecutionSnapshots)
+    .where(eq(workflowExecutionSnapshots.workflowId, ids.workflow))
+  await deleteWorkspaceFixture(db, eq(workspace.id, ids.workspace))
+  await db.delete(user).where(eq(user.id, ids.owner))
+})
+
+describe('trusted workflow tool execution failure message', () => {
+  it('returns the message a run that threw was finalized with', async () => {
+    const run = await startBoundRun()
+
+    await run.session.safeCompleteWithError({
+      error: { message: 'Fetch Orders: Request failed with status 404' },
+      executionState: executionState(),
+    })
+
+    await expect(readBack(run)).resolves.toMatchObject({
+      status: 'failed',
+      contentAvailable: true,
+      error: 'Fetch Orders: Request failed with status 404',
+    })
+  })
+
+  it('returns no message for a run whose failed block left its own output', async () => {
+    const run = await startBoundRun()
+    const now = new Date().toISOString()
+    const failedSpan: TraceSpan = {
+      id: 'span-1',
+      name: 'Check',
+      type: 'function',
+      duration: 1,
+      startTime: now,
+      endTime: now,
+      status: 'error',
+      output: { error: 'boom' },
+    }
+
+    await run.session.safeComplete({
+      finalOutput: { error: 'a field the workflow returned' },
+      traceSpans: [failedSpan],
+      executionState: executionState(),
+    })
+
+    const result = await readBack(run)
+    expect(result).toMatchObject({ status: 'failed', contentAvailable: true })
+    expect(result).not.toHaveProperty('error')
+  })
+})