Agent Runs is the inspection surface for agent work. Use it when Command Center reports a failed or stuck run, or when an admin needs to confirm that scheduled agent work is moving through the system.
What to inspect first
- Status - queued, running, completed, failed, retried, cancelled, or deferred.
- Events - the step-by-step history of what the run attempted.
- Error - the operator-visible failure reason, when available.
- Policy decisions - whether a policy allowed, blocked, or limited an action.
- Checkpoints - saved state that can help explain progress before a failure.
- Artifacts and output - drafts, proposals, summaries, or other results produced by the agent.
- Cost - spend and latency tied to the run.
Retry or cancel
Retry a run when the failure was temporary: a worker issue, dispatch failure, transient provider problem, or a dependency that is now fixed.
Cancel a run when the work is no longer needed, the inputs are wrong, the same failure repeats, or policy/budget state says the run should not continue.
Do not retry blindly if the run failed because of permissions, policy, missing required data, or malformed input. Fix the underlying condition first.
Escalation signals
Escalate to an admin when:
- the run is repeatedly stuck in queued or running,
- policy blocks an action that operators expected to be allowed,
- budget status is over or repeatedly approaching,
- checkpoints stop before the same step every time,
- the run affects customer-facing output.