> ## 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.

# One-off teams and profiles

> Interactive SSH, fixed agent commands and ad hoc shell: rights, preview, output and stop.

In Telepúrte, there are several ways to execute a command, and you can choose the task and the access model you want: an interactive terminal, a repeatable playbook, an authorized agent profile, or a one-time shell.

| Method | Suitable for | Features |
| - | - | - |
| Web SSH / SSH client | Interactive diagnostics | Terminal, system accounting, session record |
| Script playbook | Repetitive work and schedule | Script version, input files, environment |
| Fixed Agent Profile | Narrow Pre-Resolved Command | Exact Binary, Arguments, and Term |
| Ad hoc Agent | One-time random shell command | Needs special resolution and profile `adhoc` |
| SSH command via MCP | Non-interactive call through bastion | OAuth rights and existing SSH access |

## Fixed profile

The node administrator specifies the profile locally with the agent. Example configuration fragment of a separate Telepúrte Agent:

```json theme={null}
{
  "command_profiles": [
    {
      "id": "status",
      "executable": "/usr/bin/uptime",
      "args": [],
      "timeout_seconds": 5
    }
  ]
}
```

This is an addition to an existing file, not a replacement for the entire configuration. After the change, check the configuration, restart the agent and wait for its state update in the dashboard.

The operator selects the profile, but does not substitute arbitrary arguments. For all mass launch purposes, the profile with this ID must have the same binary, arguments and deadline. The executable file must exist and not be accessible to the recording by unauthorized persons.

The profile of the interpreter or universal wrapper actually extends the possibilities much more than the profile of `uptime`. Do not use this resolution as a “state only”.

## Rights and launch

The operator requires permission for a specific profile and node; a regular SSH grant and the right to a typed operation do not replace it.

The managed command contract is available through the `/api/v1/commands` API: first `previews`, then `runs` with preview ID. The preview shows the exact profile, parameters, goals and batches. It does not execute the command. Mass launch supports the canary, batch size and error threshold.

If your profile, policy, or rights have changed, form a new preview: Save your launch ID and check the results of each target, rather than repeat the request with a new ID when the network fails.

## Agent's arbitrary shell

For ad hoc modes on the node, a `adhoc` profile with `/bin/sh` and empty arguments is explicitly allowed. It executes the transmitted text through the shell; the mere appearance of a profile does not give the user the right.

This mode is not interactive: stdin is closed, the working directory begins with `/`, the environment and the current directory are not saved between launches. Profile time is limited to 30 seconds, command text is 1 MiB. For long repetitive work, use the [script playbook](https://docs.axottle.com/en/telepurte/scripts.md), for interactive - terminal.

## Conclusion and Uncertain Results

The agent saves up to 4,096 bytes of the total stdout/stderr. As a result, there's a completion code, a checksum, and a crop sign. If the output is cropped, the rest won't appear when you read the same result again.

Losing the change-start confirmation gives you `execution_uncertain`; there is no automatic repeat command. Check the actual state of the node. Reading the saved output does not repeat the command and is recorded in the audit.

## Prohibition and suspension

The administrator can disable the profile on one node or throughout the installation, the prohibition is on top of the permissions, including when issuing a pending task to an agent, the command already issued can run out.

Stopping a startup stops future output; it doesn't equal sending a `Ctrl-C` to an already running process. Prohibiting a profile doesn't filter commands inside an interactive SSH or automatically block other execution methods.

## Verification

Start with a `status` reader profile on one node. Check the exact preview profile, completion code, output and audit. For multiple nodes, compare profile definitions and their pre-launch policies.

## Rollback

Withdraw the grant, remove an unnecessary profile from local policy or impose a ban, then wait for the agent's status update. Changes made by the team are returned separately; disabling the profile does not cancel them.

Extended parameters, designated as APIs, may not be present in the current interface form.


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