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

# Starting, Results and Stopping Tasks

> Preview, fixed goals, step sequence, canary, errors, and safe repetition.

The task is the specific execution of the script on the selected nodes. It stores the version, steps, composition of goals and results. Closing the page does not cancel the task: the dashboard and agents continue to work regardless of the browser tab.

## Pre-execution

At launch, Telepúrte checks the user, targets, agent status and resolution of each step, preview captures the exact version of the script and node, lasts ten minutes, its confirmation checks the current state.

If you change your version, access, agent, or policy, you get a new preview: a new machine that appears in the group after confirmation is not added to the task you've already created, and a repeat confirmation of the same preview by the author returns the previous task, rather than creating another one.

One run supports up to 100 nodes. For a large group, divide the work into meaningful sets and check the result between launches.

## What happens after confirmation?

The task gets in line. The agent gets the work for his node, takes the step and reports the result. The next step on this machine opens after the successful previous one.

Without an explicitly specified rollout, normal goals are independent, output is limited to sequential mode. A single node error is not a promise of automatic cancellation of all others. A separate rollout rule is needed to stop by canary and error threshold.

Before issuing a step, the rights and policy are checked again, revoking access between steps prohibits the next step, but the action already taken by the agent may have time to complete.

## Task card

Open **Playbooks and Schedules → Executing Tasks** or **Launch History**. The list is updated automatically, you can update it manually. In the card, see not only the general status, but also each goal.

| Field | Why watch | |
| - | - | - |
| Node and | where the action | |
| Current move by | Where did the | script stop? |
| Status and number of attempts | Did the agent get the job and how many times was the | read operation issued? |
| Step Result | Service, Package, File, Diagnostics or Script Code Status | |
| `changed` / `unchanged` | Whether a change was necessary to achieve the result of | |
| Error Code | Cause of Failure, Not Just the Error Results of | |
| Output trimming | Whether saved stdout/stderr | |

A common "completed" does not replace application verification. A service can be active but return an error to users; a script can return a zero without verifying its business result.

## Canary and parties via API

The preview API supports `rollout`. This is an extended mode; the current conventional run wizard does not offer separate canary and threshold fields.

```json theme={null}
{
  "rollout": {
    "canary_host_id": "hst_TEST_NODE",
    "batch_size": 5,
    "max_failures": 1
  }
}
```

It's a preview request fragment, not a stand-alone command, and the canary must be part of the target selection, batch size 1 to 20, error threshold 1 to number of targets.

First, the canary is executed. When it fails, further goals are stopped. After success, games are opened; the next begins after the previous one is completed. Reaching the error threshold stops the work that has not yet been issued. The actions already performed can be completed and save the result.

## Stoppage.

In the active task card, request a stop. Telepúrte stops issuing new actions, waiting for the goal to be canceled. The action already transmitted to the agent does not turn into "not completed": it can be completed after pressing a button.

For a job that hasn't started completely, you can cancel it before you get your first job. For someone who's already working, use the stop and wait for the results of what you're doing, and closing the tab doesn't replace that.

## Lost result and repetition

For a reading operation, it is permissible to reissue after a loss of response. For a change, Telepúrte does not assume that no response means no effect: the result becomes `execution_uncertain`.

With this result:

1. Stop spreading the change further if it is still in progress.
2. Check a specific node with an independent state reading.
3. Determine which steps have actually been completed.
4. Fix the link or the policy.
5. Create a new launch just for the job you need.

| Code | Next step | |
| - | - | - |
| `policy_changed` | Review current agent permissions and create a new | preview |
| `policy_denied` | Check the exact action and resource in the local policy | |
| `access_denied` | Check the current rights of the author | |
| `operation_blocked` | Find a valid ban action or resource | |
| `execution_uncertain` | First check the actual state of the | node |
| `stopped_after_failure` | Disassemble Canary Error or Reach the | Threshold |

## Verification

Compare the number of goals, the final states of all the steps, and the expected results on the servers. For partial success, make a separate list of machines that still need work; do not restart the entire fleet automatically.

## Rollback

Stopping stops future output, but doesn't return server status. Reverse changes are done either in a separate script or manually. Save the original task ID to match the nodes and results when you restore.

Automation capabilities depend on the agent’s version and local policies.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.