Skip to main content

Monitoring runs

A run is one execution of an exact published workflow with captured input. Open the workflow's run history or the cross-workflow run list, narrow the results and select a run. Check its version and triggering record or schedule before interpreting its results.

Understand the stateโ€‹

Run stateMeaning
QueuedAdmitted and waiting to start execution.
RunningExecution is active, including durable timers and human waits.
SucceededThe selected path finished normally, including an If/Else path connected to End.
FailedAn execution error prevented normal completion.
Timed outA timeout was observed for the execution.
CancelledExecution cancellation was observed.
TerminatedExecution was externally terminated.

A waiting step does not imply a stuck run. Inspect its deadline or open question. Skipped steps may simply be on an unselected branch.

Inspect the evidenceโ€‹

Select a step to compare its resolved inputs with its outputs and outcome. Trace an unexpected value back to the preceding source step or captured trigger input. Editing the current record does not rewrite the run's captured values.

Look for created Task links, retained record results, branch decisions, wait times, human answers, Script logs and accepted Agent output. Agent usage distinguishes confirmed consumption from pending or unknown amounts. Following a Task or record link still requires access to that resource.

History can arrive laterโ€‹

Displayed history is synchronized from execution. A terminal run may briefly have incomplete step history. Refresh and check freshness before concluding that an effect did not happen. Missing history is not permission to repeat an operation.

Agent conversational activity may be retained, partial or expired. Accepted final outputs, validated source links, committed business effects and usage remain separate from conversational retention. A proposed tool call alone does not establish that it ran.

Cancel deliberatelyโ€‹

Cancellation prevents new effects and closes pending question Tasks, then reaches execution asynchronously. A cancellation request and a cancelled status are different observations; concurrent completion can win. Existing Tasks and record changes are not rolled back.

Automatic recovery handles supported interruptions within the same run. There is no general user-facing retry-from-failed-step operation. Starting another manual run is a new execution and can repeat effects. Inspect the first run before doing so.

Use Troubleshooting for next steps. The run API reference documents filters, freshness and retained history.