> ## Documentation Index
> Fetch the complete documentation index at: https://getpullrequest.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Handle task failures

> Recover from errored, interrupted, and waiting-for-machine states.

## Failed (`errored`)

Common causes:

* Verification exhausted (tests kept failing)
* Unrecoverable pipeline error
* Agent spawn / unresponsive failures

**What to do**

1. Read the task activity and session error.
2. Fix the underlying issue (tests, agent auth, missing binary).
3. **Rerun** the task. Rerun resets status to `backlog` and checkpoint to `none`, and re-queues work. The durable ACP session is kept when it is still usable.

## Interrupted

You stopped the run. Checkpoint is preserved.

* Send a follow-up on the linked session or task to continue.
* Or request a full rerun if you want a clean slate.

## Waiting for machine (`waiting_for_agent`)

The daemon went offline mid-flight.

```bash theme={null}
gpr status
gpr reconnect
gpr doctor
```

When the host is back, the task returns to the queue and resumes from the last checkpoint.

## Needs your input

Not a failure. Answer the plan questions on the task.

## Machine-side checks

```bash theme={null}
gpr discover    # agent still ready?
gpr logs -f     # spawn errors, ACP crashes
gpr doctor      # pairing / websocket health
```

Confirm the agent CLI still authenticates (for example `cursor-agent`, `claude`, `codex`).

More: [Troubleshooting](/docs/reference/troubleshooting).