Skip to main content
Use the selected task or run as the execution record. Chat tells you what was requested; the task state, steps, and output tell you what is happening now and what actually finished.

​
Reopen the exact run

You can reach the same work from several places:
  • Select a task under Currently Running on Home.
  • Select a task row in the sidebar Recents list.
  • Open Chats and tasks, choose Tasks, and select the task.
  • Open Activity and select its Agent work card.
  • Open the Agents sidebench in the originating conversation and select the specialist or nested run.
  • Follow a task link from Notifications.
Before you act, match the task title, specialist, original conversation, attached context, and relative time. Selecting a task in Chats and tasks opens its Activity run detail; selecting a chat opens the complete conversation instead.

​
Read the lifecycle on the current surface

Compact task rows use these native labels: Activity can also present family-level states such as Queued, Working, Waiting, Completed, or Partially completed when it combines a parent with nested work. Read the selected child before deciding which part needs attention. The selected detail can show Working for time, Worked for time, Failed after time, Canceled after time, or Waiting. Elapsed time describes the run; it is not a deadline or completion estimate.

​
Resume waiting work

1

Open the waiting interaction

Use the question or approval mounted inside the same run. If you started from a compact task row, open its detail first.
2

Resolve it once

Submit the missing answer or use the displayed approval controls. Keep the response within the original objective.
3

Wait for the state transition

The interface can show Sending your answer… or Sending…. Keep the run open until it returns to Running or Working.
4

Follow the next active step

A resumed run can encounter another question or approval. Read the new interaction instead of assuming the first decision released every branch.
5

Review the terminal result

Open the output only after the task is Done, Completed, Failed, Partially completed, or Canceled.
If delivery fails and the interaction remains available, use its retry path. If Marketer says delivery may already be in progress and tells you not to retry, wait for the run state to reconcile in Activity.

​
Stop active or waiting work

Stop is available in the selected run detail while the run is non-terminal. It can remain available while the run is waiting for input, which lets you end work you no longer want to resume.
  1. Confirm that the selected task is the one you intend to stop.
  2. Select Stop.
  3. Keep the detail open while the control shows Stopping….
  4. Wait for Canceled or another terminal state.
  5. Review any output already produced before the stop.
Stopping a chat response through Stop generating is different. It stops the current conversational turn. Use Stop in the task or Agents detail for a background run. If Marketer reports that the run did not acknowledge the stop and is still active, do not treat it as canceled. Keep the same run open and try again as the message instructs. If only part of the work stopped, inspect the remaining active child before you start replacement work.

​
Recover by terminal state

​
Failed

  1. Open the failed run instead of immediately resubmitting the prompt.
  2. Read the last successful step and the displayed error.
  3. Check the required connected account, workspace record, permission, file, or destination.
  4. Open Outputs and preserve any usable result.
  5. Correct the prerequisite, then start a revised run for the missing work.

​
Partially completed

Activity can show Partially completed when a run family produced usable output but one branch did not finish. Open the parent and child rows, identify which output is supported, and request only the missing branch. Do not discard valid output or rerun finished generation by default.

​
Canceled

A canceled run is terminal. It does not resume from a new chat message. Return to the originating conversation, revise the objective or constraint, and start a new proposal or task when the deliverable is still needed.

​
Done but incomplete

A successful execution state means the run ended normally, not that the deliverable satisfies your request. Compare the actual result with the requested sections, sources, date window, and completion criteria. Ask for the missing output as a new, bounded request.

​
Still running after a quiet period

Do not infer failure from a quiet conversation. Reopen the task through Chats and tasks or Activity, read its current status, and inspect the active specialist in Agents. Start a replacement only after the original run is terminal or you intentionally stop it.

​
Use Activity detail for recovery evidence

The Activity run detail is the durable place to reopen a task when the conversation is not in view. It can include:
  • The current status and elapsed work
  • The run transcript or progress body
  • The active question or action approval
  • A Stop control while work is non-terminal
  • Outputs links for artifacts produced by the run
  • The specialist family and nested work in the Agents view
Open an output from Activity, then verify it in the surface that owns the object. Activity proves that work was recorded; it is not the current campaign, brief, creative, workflow, or provider state.

​
Verify before closing the task

Use the matching final check:
Done, an approval receipt, and a notification are three different records. None replaces verification of the current object.

​
Next step

Chats and tasks

Reopen a conversation or recent task from the combined history page.

Tasks and activity

Understand task progress and workspace-wide Activity.

Questions and approvals

Resolve the interaction that placed a run in Waiting.

Inspect specialists

Trace the parent task into nested specialist work, sources, and outputs.
Last modified on August 9, 2026

Related topics

Notifications and billingTasks and activityTroubleshoot integrationsSecurity and troubleshootingApprovals, guardrails, and autonomy