Operations runbook

---
title: Operations runbook
description: What to check, in order, when a task stalls between agents.
updatedAt: '2026-09-17'
---

## When a task stalls

Read the task state before re-running anything. A stalled task usually carries
the reason in what was already tried, and a re-run that ignores it repeats the
step that stalled.

## The order to check

1. The task state as the last agent left it.
2. The operational limit the step was bounded by, and whether it was reached.
3. The handoff itself, which is where state is most often dropped.

Only after those three is a re-run informative, because only then is it possible
to say what the re-run would do differently.

## Where this page moved from

This runbook was published at `guide/runbook` before the operations pages were
grouped, so the old path still resolves through a recorded redirect.

The second check in the list above reads from
[operational limits](./limits.md), and the terms used throughout are defined in
the [glossary](../../reference/glossary.md).