Agent Rescue
Resume from facts after a failure.
Inspect an interrupted task before retrying, resuming or handing it over.
Resume from facts after a failure.
After an error, restart or potentially stalled task.
An interrupted task may leave files or external actions in an uncertain state. Rescue gathers the evidence needed for a deliberate continuation.
Review the recorded state, current files and proposed recovery route. Confirm only after the checks explain what can safely continue.
How to get started.
From your first choice to the result. Here is how to use this feature in your own project.
- 01
Review the last known activity and stored task state.
- 02
Check files, subprocesses and external effects that may already have occurred.
- 03
Then choose resume or a suitable handover route.
How it could work for you.
Recover after a failed tool call without blindly repeating a deployment or migration.
How to assess the result
Recovery cannot undo unknown external effects or restore data that was never saved.
Where to find it.
Conversation → Tools → Agent Rescue
Available actions depend on your plan, mode and project state. Features that require a task become available within an appropriate conversation.What is saved?
The recovery check and chosen next action stay linked to the task. Opening it does not itself make a model request.
What you need to know
Not every external effect can be established automatically. A deployment or database change may require manual inspection.