diff --git a/apps/api/src/handlers/tasks/automation-work-items/__tests__/launch.test.ts b/apps/api/src/handlers/tasks/automation-work-items/__tests__/launch.test.ts
index 30fb4bec04..ffb803b29a 100644
--- a/apps/api/src/handlers/tasks/automation-work-items/__tests__/launch.test.ts
+++ b/apps/api/src/handlers/tasks/automation-work-items/__tests__/launch.test.ts
@@ -288,33 +288,39 @@ describe('launchActWorkItems', () => {
     });
 
     expect(result).toEqual({ launchedCount: 1, failedCount: 0 });
+    expect(mockEnqueueTask.mock.calls[0]?.[0]).toMatchObject({
+      title: 'Fix parser nil access',
+    });
     const enqueuePayload = mockEnqueueTask.mock.calls[0]?.[0].task
       .payload as Record<string, unknown>;
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.description).toBe(
+      'Fix parser nil access\n\nNil access is driving a production Sentry issue.',
+    );
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining('keep progress visible in the web task'),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining('without posting status updates back to Slack'),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining('<untrusted_content_policy/>'),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         '<untrusted_external_content source="automation_work_item_brief">Nil access is driving a production Sentry issue.</untrusted_external_content>',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Investigation context:\n<untrusted_external_content source="automation_investigation_context">$sentry-triage\nIssue: SENTRY-123</untrusted_external_content>',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Execution guidance from the scan run (apply only within the scope of this work item):\n<automation_execution_guidance>Reproduce the nil access, fix it, add regression coverage, and open a PR.</automation_execution_guidance>',
       ),
     );
-    expect(enqueuePayload.visibleInTranscript).toBe(false);
+    expect(enqueuePayload.visibleInTranscript).toBe(true);
     expect(enqueuePayload).not.toHaveProperty('channel');
     expect(enqueuePayload).not.toHaveProperty('slackChannel');
     expect(enqueuePayload).not.toHaveProperty('thread_ts');
@@ -382,121 +388,121 @@ describe('launchActWorkItems', () => {
     expect(result).toEqual({ launchedCount: 1, failedCount: 0 });
     const enqueuePayload = mockEnqueueTask.mock.calls[0]?.[0].task
       .payload as Record<string, unknown>;
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining('$update-dependencies'),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Do not send Slack progress updates, elapsed-time updates, validation-started updates, or partial findings',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Keep intermediate status in the web task and todo list only',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'do not send a Slack-visible opening acknowledgement',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Default to finishing silently. Post to Slack only when there is something a human should see now',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Routine success, no-change results, and no-op or deferred outcomes that require no human action should not produce a message',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'If you report, keep the first Slack-visible closeout as one self-contained message instead of a separate opener plus a result',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'make it fully standalone and do not assume readers have seen any earlier scan, audit, or research task',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Write that closeout like a helpful coworker summarizing completed work',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Lead with a direct plain-language sentence that names the object of the work, says what you reviewed, why it mattered, and what changed or how far it got',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'On first mention, spell out the object before shorthand or identifiers',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Avoid thread-local references like "this", "that follow-up", "the issue", "the investigation", or "the risk"',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Keep the whole message to at most two short paragraphs',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'use one human-readable reference from that context so the reader knows what prompted the work',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'make the label describe the object instead of showing an unexplained code',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'if there is a clean URL, render it as a named inline link such as `[Sentry issue ROOMOTE-WORKER-381](...)` or `[alert #123](...)`, not a bare URL or a bare-ID label',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'keep a stable plain-text identifier such as `GHSA-123`, `SENTRY-123`, or `owner/repo#123` if that is the clearest reference',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'make the delivery state explicit; if the work stopped at a draft PR, say you opened a draft PR instead of saying it shipped',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Link the draft PR number if the URL is available; otherwise keep the PR identifier plain text',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'I reviewed [alert #275](...) and opened [draft PR #4783](...) to address a high and two medium `undici` vulnerabilities',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'do not recite file paths, code identifiers, or step-by-step verification in Slack',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Do not append a verification or validation sentence',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'For blocker, no-op, deferred, or input-needed outcomes that are worth reporting under the rule above, keep the same single-message shape but report only the outcome-relevant details',
       ),
     );
     expect(enqueuePayload.automationWorkItemId).toBe(workItem.id);
-    expect(enqueuePayload.visibleInTranscript).toBe(false);
+    expect(enqueuePayload.visibleInTranscript).toBe(true);
     expect(enqueuePayload.channel).toBe('C456');
     expect(enqueuePayload.slackChannel).toBe('C456');
     expect(enqueuePayload).not.toHaveProperty('thread_ts');
@@ -524,73 +530,73 @@ describe('launchActWorkItems', () => {
     expect(result).toEqual({ launchedCount: 1, failedCount: 0 });
     const enqueuePayload = mockEnqueueTask.mock.calls[0]?.[0].task
       .payload as Record<string, unknown>;
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'do not send a Slack-visible opening acknowledgement',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'If you report, keep the first Slack-visible closeout as one self-contained message instead of a separate opener plus a result',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Write that closeout like a helpful coworker summarizing completed work',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Pretend the Slack reader only sees that one closeout message and none of the hidden automation or environment context behind it',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Do not frame the message as a follow-up on a hidden scan, audit, evaluator, research task, or spawned environment task',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Prefer result-first wording like "I reviewed [alert #275](...) and opened [draft PR #4783](...) to address ..." or "I reviewed the Sentry issue SENTRY-123 and opened the resulting draft PR to address ..."',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'If you mention a prior PR, alert, issue, task, workflow run, or environment identifier, say in the same sentence what it is and what about it was under review',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Keep the whole message to at most two short paragraphs',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'explain what the automation investigated and the concrete outcome, but do not recite file paths, code identifiers, or step-by-step verification in Slack',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'use one human-readable reference from that context so the reader knows what prompted the work',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'make the label describe the object instead of showing an unexplained code',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'keep a stable plain-text identifier such as `GHSA-123`, `SENTRY-123`, or `owner/repo#123` if that is the clearest reference',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Prefer wording like "I reviewed [alert #275](...) and opened [draft PR #4783](...)" or "I reviewed the Sentry issue SENTRY-123 and opened the resulting draft PR ..." over "That shipped in draft PR #4783."',
       ),
     );
     expect(enqueuePayload.automationWorkItemId).toBe(workItem.id);
-    expect(enqueuePayload.visibleInTranscript).toBe(false);
+    expect(enqueuePayload.visibleInTranscript).toBe(true);
     expect(enqueuePayload.channel).toBe('C456');
     expect(enqueuePayload.slackChannel).toBe('C456');
     expect(updateSets).toEqual([
@@ -620,18 +626,18 @@ describe('launchActWorkItems', () => {
     expect(enqueuePayload.thread_ts).toBe('1781300000.000100');
     expect(enqueuePayload.slackThreadTs).toBe('1781300000.000100');
     expect(enqueuePayload.channel).toBe('C456');
-    expect(enqueuePayload.visibleInTranscript).toBe(false);
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.visibleInTranscript).toBe(true);
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Reply in the existing Slack investigation thread',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'still assume readers do not know about the hidden research task or spawned environment task behind this run',
       ),
     );
-    expect(enqueuePayload.description).not.toEqual(
+    expect(enqueuePayload.agentPromptText).not.toEqual(
       expect.stringContaining('may create a new top-level thread'),
     );
     expect(mockFinalizeAutomationLaunch).toHaveBeenCalledWith(
@@ -744,12 +750,12 @@ describe('launchActWorkItems', () => {
     expect(enqueuePayload.automationWorkItemId).toBe(workItem.id);
     expect(enqueuePayload).not.toHaveProperty('slackChannel');
     expect(enqueuePayload).not.toHaveProperty('channel');
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Do not send Telegram progress updates, elapsed-time updates, validation-started updates, or partial findings',
       ),
     );
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'do not send a Telegram-visible opening acknowledgement',
       ),
@@ -785,7 +791,7 @@ describe('launchActWorkItems', () => {
       communicationMessageId: 'discord-message-1',
       discordTaskThread: true,
     });
-    expect(enqueuePayload.description).toEqual(
+    expect(enqueuePayload.agentPromptText).toEqual(
       expect.stringContaining(
         'Do not send Discord progress updates, elapsed-time updates, validation-started updates, or partial findings',
       ),
diff --git a/apps/api/src/handlers/tasks/automation-work-items/launch.ts b/apps/api/src/handlers/tasks/automation-work-items/launch.ts
index 56a0d8e353..ac444791ff 100644
--- a/apps/api/src/handlers/tasks/automation-work-items/launch.ts
+++ b/apps/api/src/handlers/tasks/automation-work-items/launch.ts
@@ -7,7 +7,11 @@ import {
   escapeTaskContextText,
 } from '@roomote/cloud-agents/server';
 import { finalizeAutomationLaunch } from '@roomote/sdk/server/automation-post-launch-finalization';
-import { TaskPayloadKind, type BackgroundAutomationKey } from '@roomote/types';
+import {
+  TaskPayloadKind,
+  buildAutomationWorkItemDisplayPrompt,
+  type BackgroundAutomationKey,
+} from '@roomote/types';
 import {
   and,
   claimWorkItem,
@@ -247,8 +251,27 @@ export async function launchActWorkItems(params: {
         });
       }
 
+      const agentPromptText = buildExecutionTaskPrompt(workItem, {
+        executionTaskBootstrap: params.executionTaskBootstrap,
+        lateBoundChatReplies: workItemChatTarget !== null,
+        hasChatThread:
+          (workItemChatTarget?.provider === 'slack' &&
+            Boolean(workItemChatTarget.threadTs)) ||
+          (workItemChatTarget?.provider === 'discord' &&
+            Boolean(workItemChatTarget.threadId)),
+        chatSurface:
+          workItemChatTarget?.provider === 'telegram'
+            ? 'Telegram'
+            : workItemChatTarget?.provider === 'teams'
+              ? 'Teams'
+              : workItemChatTarget?.provider === 'discord'
+                ? 'Discord'
+                : 'Slack',
+      });
+
       await enqueueTask(
         {
+          title: workItem.title,
           task: {
             type: TaskPayloadKind.StandardTask,
             payload: {
@@ -257,23 +280,8 @@ export async function launchActWorkItems(params: {
                 ? { environmentId: workItem.targetEnvironmentId }
                 : {}),
               selectedRepositories: [workItem.targetRepositoryFullName],
-              description: buildExecutionTaskPrompt(workItem, {
-                executionTaskBootstrap: params.executionTaskBootstrap,
-                lateBoundChatReplies: workItemChatTarget !== null,
-                hasChatThread:
-                  (workItemChatTarget?.provider === 'slack' &&
-                    Boolean(workItemChatTarget.threadTs)) ||
-                  (workItemChatTarget?.provider === 'discord' &&
-                    Boolean(workItemChatTarget.threadId)),
-                chatSurface:
-                  workItemChatTarget?.provider === 'telegram'
-                    ? 'Telegram'
-                    : workItemChatTarget?.provider === 'teams'
-                      ? 'Teams'
-                      : workItemChatTarget?.provider === 'discord'
-                        ? 'Discord'
-                        : 'Slack',
-              }),
+              description: buildAutomationWorkItemDisplayPrompt(workItem),
+              agentPromptText,
               ...(workItemChatTarget?.provider === 'slack'
                 ? {
                     automationWorkItemId: workItem.id,
@@ -319,7 +327,7 @@ export async function launchActWorkItems(params: {
                     discordTaskThread: Boolean(workItemChatTarget.threadId),
                   }
                 : {}),
-              visibleInTranscript: false,
+              visibleInTranscript: true,
             },
           },
           initiator: { kind: 'automation', key: params.automationKey },
diff --git a/apps/web/src/app/(sandbox)/task/[taskId]/hooks/__tests__/use-task-session.client.test.ts b/apps/web/src/app/(sandbox)/task/[taskId]/hooks/__tests__/use-task-session.client.test.ts
index d2b62be2b5..15d8479d99 100644
--- a/apps/web/src/app/(sandbox)/task/[taskId]/hooks/__tests__/use-task-session.client.test.ts
+++ b/apps/web/src/app/(sandbox)/task/[taskId]/hooks/__tests__/use-task-session.client.test.ts
@@ -63,6 +63,26 @@ describe('getTaskRunPromptText', () => {
 });
 
 describe('getTaskRunVisiblePrompt', () => {
+  it('projects linked work-item fields for legacy automation execution payloads', () => {
+    expect(
+      getTaskRunVisiblePrompt(
+        buildTaskRun({
+          repo: 'Roomote/example-app',
+          description:
+            '$implement-changes\n\nAutomation work item: Fix the failing route\n<untrusted_external_content>Useful brief</untrusted_external_content>\n\nInternal execution policy',
+          visibleInTranscript: false,
+        } satisfies TaskPayload<typeof TaskPayloadKind.StandardTask>),
+        {
+          title: 'Fix the failing route',
+          brief: 'A production route fails for signed-in users.',
+        },
+      ),
+    ).toEqual({
+      text: 'Fix the failing route\n\nA production route fails for signed-in users.',
+      visibleInTranscript: true,
+    });
+  });
+
   it('strips Slack thread context and reply targets from the visible prompt text', () => {
     expect(
       getTaskRunVisiblePrompt(
diff --git a/apps/web/src/lib/task-run-visible-prompt.ts b/apps/web/src/lib/task-run-visible-prompt.ts
index 33563a397c..b1728dad6c 100644
--- a/apps/web/src/lib/task-run-visible-prompt.ts
+++ b/apps/web/src/lib/task-run-visible-prompt.ts
@@ -1,4 +1,5 @@
 import {
+  buildAutomationWorkItemDisplayPrompt,
   isSystemInjectedAcpPromptText,
   normalizeTranscriptUserText,
   ACP_ENVELOPE_EVENT_TYPES,
@@ -133,7 +134,15 @@ function getTaskRunPromptImages(
  */
 export function getTaskRunVisiblePrompt(
   taskRun: Pick<TaskRunDetail, 'payload'> | null | undefined,
+  automationWorkItem?: { title: string; brief?: string | null } | null,
 ): TaskRunVisiblePrompt | null {
+  if (automationWorkItem) {
+    return {
+      text: buildAutomationWorkItemDisplayPrompt(automationWorkItem),
+      visibleInTranscript: true,
+    };
+  }
+
   const text = getTaskRunPromptText(taskRun);
 
   const visibleText = normalizeTranscriptUserText(
diff --git a/apps/web/src/trpc/commands/sandbox-session/index.ts b/apps/web/src/trpc/commands/sandbox-session/index.ts
index 19de1e5450..8ada7007e2 100644
--- a/apps/web/src/trpc/commands/sandbox-session/index.ts
+++ b/apps/web/src/trpc/commands/sandbox-session/index.ts
@@ -40,6 +40,7 @@ import {
   taskMessages,
   taskRuns,
   tasks,
+  workItems,
   touchTaskActivity,
 } from '@roomote/db/server';
 import { httpBatchLink, TRPCClientError } from '@trpc/client';
@@ -922,31 +923,49 @@ export async function getSandboxSessionByTaskIdCommand(
 
   const shouldCheckHarnessMessages =
     !isExitedRunStatus(taskRun.status) && !isBootingRunStatus(taskRun.status);
+  const shouldResolveLegacyAutomationWorkItem =
+    task.initiatorKind === 'automation' &&
+    (!('agentPromptText' in taskRun.payload) ||
+      typeof taskRun.payload.agentPromptText !== 'string' ||
+      !taskRun.payload.agentPromptText.trim());
 
   // Check whether the task ever produced messages. Used to distinguish boot
   // failures (no messages) from runtime failures (has conversation history).
-  const [bootFailureMessage, firstHarnessMessage] = await Promise.all([
-    shouldCheckBootFailureMessages
-      ? db.query.taskMessages.findFirst({
-          where: eq(taskMessages.runId, taskRun.id),
-          columns: { id: true },
-        })
-      : Promise.resolve(null),
-    shouldCheckHarnessMessages
-      ? db.query.taskMessages.findFirst({
-          where: and(
-            eq(taskMessages.runId, taskRun.id),
-            isNotNull(taskMessages.role),
-            not(eq(taskMessages.role, 'user')),
-          ),
-          columns: { id: true },
-        })
-      : Promise.resolve(null),
-  ]);
+  const [bootFailureMessage, firstHarnessMessage, legacyAutomationWorkItem] =
+    await Promise.all([
+      shouldCheckBootFailureMessages
+        ? db.query.taskMessages.findFirst({
+            where: eq(taskMessages.runId, taskRun.id),
+            columns: { id: true },
+          })
+        : Promise.resolve(null),
+      shouldCheckHarnessMessages
+        ? db.query.taskMessages.findFirst({
+            where: and(
+              eq(taskMessages.runId, taskRun.id),
+              isNotNull(taskMessages.role),
+              not(eq(taskMessages.role, 'user')),
+            ),
+            columns: { id: true },
+          })
+        : Promise.resolve(null),
+      shouldResolveLegacyAutomationWorkItem
+        ? db.query.workItems.findFirst({
+            where: and(
+              eq(workItems.kind, 'auto_fix'),
+              eq(workItems.launchedTaskId, input.taskId),
+            ),
+            columns: { title: true, brief: true },
+          })
+        : Promise.resolve(null),
+    ]);
 
   const hasMessages = !!bootFailureMessage;
   const hasHarnessMessages = !!firstHarnessMessage;
-  const payloadPrompt = getTaskRunVisiblePrompt(taskRun);
+  const payloadPrompt = getTaskRunVisiblePrompt(
+    taskRun,
+    legacyAutomationWorkItem,
+  );
   // Fall back to durable task-level prompt when the run payload has no visible
   // prompt, so failure UIs can still show the original request.
   const prompt: {
diff --git a/packages/types/src/task-runs.ts b/packages/types/src/task-runs.ts
index 90b2221639..a1814a54f3 100644
--- a/packages/types/src/task-runs.ts
+++ b/packages/types/src/task-runs.ts
@@ -258,6 +258,16 @@ export const AUTOMATION_WORK_ITEM_DISPOSITIONS = ['suggest', 'act'] as const;
 export type AutomationWorkItemDisposition =
   (typeof AUTOMATION_WORK_ITEM_DISPOSITIONS)[number];
 
+export function buildAutomationWorkItemDisplayPrompt(item: {
+  title: string;
+  brief?: string | null;
+}): string {
+  const title = item.title.trim();
+  const brief = item.brief?.trim();
+
+  return brief ? `${title}\n\n${brief}` : title;
+}
+
 /**
  * work_items (Stage 4): one table merges the old task_suggestions,
  * automation_work_items, and setup_new_queued_tasks. `kind` selects the flavor