# Check what the worker did, then decide what runs next.

Reviewed: 2026-09-12

Canonical: https://clankercloud.ai/learn/review-recurring-worker-results

Open the Loop or Schedule history and inspect the run’s progress, conversation, files, and outcome. Verify the result at its destination before retrying uncertain work. Pause recurrence while resolving a failure that could repeatedly produce incorrect outputs or duplicate an action.

## What you are working toward

A reviewed worker run with a clear decision to continue, repair, pause, or stop.

## Start with

- An existing Loop or Schedule and a known expected result

- The source list, output destination, and notification rule

- Access to inspect any destination the worker changes

## Know what a successful run contains

A recurring worker needs a definition of completion beyond “finished without a visible error.” For a daily source monitor, that might mean checking three named sources, recording the observation date, comparing against the prior reviewed result, and saving a brief. A run missing one source has a different conclusion from a run finding no changes.

Keep the expected output and notification rule in the goal. A worker that should notify only on actionable changes still needs a way for you to inspect its history. Silence by itself does not prove success or failure. Review the worker’s recorded progress and files when you need to establish what happened.

## Inspect the run and its actual output

Open the worker’s conversation, progress, files, and run history. Identify the specific run under review instead of relying on the most recent file with a similar name. Compare the source dates, input scope, and saved artifact with the goal. Open the artifact rather than accepting a summary that says it was created.

For computer jobs, inspect the reported job result and whether the requested output is available. Allow computer jobs and an Agent permitted to work unattended are prerequisites for such runs. Answer only, Research, and Always ask Agents cannot run unattended computer jobs. A configuration mismatch requires a setup decision, not repeated run-now attempts.

## Classify a fictional monitoring run

Suppose a fictional worker checks three public supplier-status pages every weekday and saves a brief. One source is unavailable, one has changed, and one is unchanged. The correct report describes partial coverage; it cannot conclude that all three suppliers are operating normally.

The decisions below are examples for an owner reviewing that run. The worker does not gain new access or permission merely because its instructions ask for a complete report. A narrower, explicit outcome is easier to repair than an “all clear” message that hides the missing source.

| Observed evidence | Conclusion you can support | Next decision |
| --- | --- | --- |
| Two of three sources were checked. | Coverage is incomplete. | Record the missing source and decide on a later check. |
| A brief opens in Project Files. | That artifact was saved. | Review its dates, claims, and links. |
| An external update has no confirmed outcome. | Completion is uncertain. | Inspect the destination before another attempt. |
| The same source fails every run. | The task repeatedly lacks an input. | Pause and repair the source or revise the goal. |

## Reconcile uncertain actions before a retry

Separate execution, saving, and downstream actions. A worker can compute a result but fail to save it, or save it before another step fails. If it writes to an external system, inspect that system or available history to determine whether the change already happened. Retrying the entire goal can duplicate completed work.

Make the repair specific to the unfinished step. Regenerating a missing draft from known inputs is different from sending a second message or submitting a form again. When you cannot confirm an external outcome, retain the uncertainty and choose an appropriate manual check instead of interpreting a timeout as a definitive failure.

## Pause a recurring failure and test the repair once

Use pause or stop when the recurring setup would repeat the same error or consume resources without useful results. Fix the identified cause: an obsolete source, ambiguous instruction, unavailable computer, or missing prerequisite. Keep the repair within the intended task rather than broadening the worker’s access simply to make a run finish.

Test one corrected run and inspect its saved output before resuming recurrence. Check the notification rule as part of the test. A task intended to report meaningful changes should not suddenly produce repetitive updates because a repair changed its wording. Record what changed so a later reviewer can interpret the difference in outputs.

## Review timing and active environments together

A Loop repeats after the previous run and a delay; a Schedule runs at set times. Confirm the timing and timezone appropriate to the task. Hosted workers can run in the cloud, while work on a connected computer still depends on that machine being online and available.

Review actual usage after a representative run and after changing the goal or frequency. Loops use 10× model tokens, and computer usage is charged separately. Pause unused workers and stop active environments when finished; closing a viewer may leave compute running. Check current pricing and the account balance rather than estimating from an unrelated task.

## Try this prompt

Copy into a Workspace Session. Opening Workspace does not send it automatically.

```text
Review the latest run of this worker using its history and saved results. Identify the intended inputs, what was accessed, which outputs are confirmed, and what remains uncertain. Check whether the result meets the agreed criteria. Do not repeat external actions while their outcome is uncertain. Recommend a narrow repair and explain whether recurrence should remain paused during it.
```

## Before using the result

- The reviewed run and its input coverage are identified.

- Saved files open and match the requested period.

- Uncertain external outcomes are reconciled before retry.

- A repair passes one inspected run before recurrence resumes.

- Timing, notifications, and active compute match the intended workload.

## Common questions

### Should I use a Loop or a Schedule?

Use a Loop for work that repeats after the previous run and a delay. Use a Schedule for work at set times. Test the task in a Session first, then choose its Project, Agent, goal, timing, and notification rule.

[Read more in the FAQ](https://clankercloud.ai/faq#loops-schedules)

### Can recurring work run scripts while I am away?

Allow computer jobs lets a Loop or Schedule run scripts and commands while your browser is closed. Its Agent must permit unattended work; Answer only, Research, and Always ask Agents cannot run unattended computer jobs. Review progress and job results in the worker history.

[Read more in the FAQ](https://clankercloud.ai/faq#unattended)

### Can it work while my laptop is closed?

Hosted Loops and Schedules can run recurring work in the cloud. Tasks using a connected computer still need that computer to be online and available. Simply starting a browser conversation does not turn it into a recurring background task.

[Read more in the FAQ](https://clankercloud.ai/faq#background-work)

### Does closing a viewer stop its computer?

Not necessarily. Active browsers, desktops, and cloud computers can continue using credits. Use their stop or shutdown controls when you finish. Availability and controls are shown in Workspace.

[Read more in the FAQ](https://clankercloud.ai/faq#stop-compute)

## Sources

- [Workspace guide: worker history and controls](https://workspace.clankercloud.ai/#page=faq&tab=articles&article=loops-and-schedules)

- [Workspace guide: active environments](https://workspace.clankercloud.ai/#page=faq&tab=articles&article=where-work-runs)

- [Current usage rates](https://clankercloud.ai/pricing)

## Related

- [Test it once. Give it a rhythm.](https://clankercloud.ai/learn/loops-and-schedules)

- [Choose approvals around the actions you intend.](https://clankercloud.ai/learn/choose-approval-mode)

- [Start small. Know what keeps running.](https://clankercloud.ai/learn/workspace-credits-and-costs)

[Open Workspace](https://workspace.clankercloud.ai)

[Current pricing and usage](https://clankercloud.ai/pricing)