---
title: "Commands"
description: "Review command history in Alpacon to see who ran each command, on which server and account, and how it ended. Understand rejection reasons and next steps."
---

> Full documentation index: https://docs.alpacax.com/llms.txt

# Commands

Every command sent to a registered server through Alpacon, in or out of a work session, leaves a record here. Use this page to confirm a command ran, see who ran it, and read what came back.

## How to get here

Open the sidebar and go to **Audit → Events → Commands** (or **My events → Commands** if you're a regular user).

## What's in the list

Each row is one command, newest first.

| Column | What it shows |
| --- | --- |
| **Command** | The command line that was sent |
| **Requester** | Who ran the command, as an icon and a name. The icon shows the kind: a person's avatar, a bot for an AI agent, or a key for a personal API token. Hover it for the kind, and for the agent name the client reported, if any |
| **Server** | The server the command targeted |
| **Account** | The system account the command ran as |
| **Result** | How the command ended, or where it is now, as a word. See [What the result means](#what-the-result-means) |
| **Risk** | The risk band assigned to the command, colored from High up. Shows `—` when no band was assigned |
| **Requested** | When the command was sent. The header names the time zone. The only sortable column: click it to switch between newest and oldest first |

A command counts as an agent's when it came through MCP, an API or service token, or an unregistered client, or when the CLI reported that an agent sent it.

On a self-hosted deployment without AI analysis, the Risk column does not appear.

The list updates on its own as commands finish.

## Command details

Click a row to open the command in a side panel. The top line repeats the result and the risk, and links to the [work session](/use/work-sessions/) the command ran in, or says it ran outside a session. Below it:

- The requester, the agent name the client reported, the server, the account and group, the access path (web, CLI, MCP, API token, or service token), and how long the command took.
- **Rejection reason** and its next step, for a rejected command. See [Understanding a rejection](#understanding-a-rejection).
- **Denial reason**, for a command an ACL policy denied.
- **Stated purpose**, when the requester gave one.
- **Output**: the full output, or **No output**.
- **Execution control**: how the command was admitted. See [Verification status](#verification-status).
- **Stages**: when the command was requested, scheduled, delivered, received by the server, and handled, each with its offset from the start.

A scheduled command that hasn't been sent yet has **Cancel scheduled command** at the bottom of the panel. Cancelling removes it from the list, and the activity log keeps a record.

The panel stays in the page URL, so a link opens the list with that command's panel already open.

## Filtering and search

The filter bar narrows the list by **Server**, **Requester**, **Type** (**User** or **Agent**), **Risk**, and **Date**. The `+` button adds **Access path**, **Execution control**, and **Scheduled** (**Scheduled** or **Run now**).

The search box matches the command line, the server name, and the requester's name. The filters and search stay in the page URL, so you can share a filtered list as a link. Links from elsewhere in the console, such as the dashboard, can open the list with extra filters already applied.

## What the result means

The Result column names the furthest point a command has reached, or how it ended.

**Before delivery**

- **Verifying**: the command is being checked before delivery.
- **Awaiting purpose**: the requester is being asked what the command is for. No approver has been notified yet.
- **Awaiting approval**: the command is waiting for a person to approve it before it can be delivered.
- **Rejected**: a check refused the command before it could run. Hover the word for the reason. See [Understanding a rejection](#understanding-a-rejection).
- **Denied**: an ACL policy denied the command. This replaces whatever else the result would have been, and the side panel shows a **Denial reason**.

**On the way to the server**

- **Scheduled**: the command is waiting for the time it was scheduled for.
- **Queued**: the command is ready to be delivered.
- **Delivered**: the command has been sent to the server.
- **Running**: the server is executing the command now.

**Finished**

- **Completed**: the command ran and exited with code 0.
- `Exit N`: the command ran and exited with a non-zero code. It is not colored, because a non-zero exit is not always a failure. Open the row to read the output.
- **Timed out**, **Command not found**, **Not executable**, **Killed by signal**, or **Prior command failed**: the command did not finish on its own, and the word says why.

**Needs a look**

- **Stuck**: no response for over 10 minutes.
- **Error**: an unexpected state; delivery was never confirmed either way.

Waiting states are colored amber, and refusals and failures red.

## Understanding a rejection

For a rejected command, hover the result in the list for the reason, or open the command to read the reason and the suggested next step under **Rejection reason**. The same explanation appears in a work session's command history and live activity when a reason is available.

| Reason | What to do next |
| --- | --- |
| The command access policy did not allow the command | Ask a workspace administrator to review the command's access requirements. |
| The command access policy could not be checked | Ask a workspace administrator to check whether command policy checks are available. |
| The risk assessment blocked the command | Review the command and its purpose with a workspace administrator. |
| The command's risk could not be assessed | Ask a workspace administrator to check whether risk assessment is available. |
| The approval policy did not allow the command to proceed | Ask a workspace administrator to review approval requirements and approver availability. |
| An approver rejected the command | Review the [approval request](/use/approvals/reviewing/) and discuss the work with your team. |
| The approval request expired without a decision | Review the expired request and confirm approval availability before planning the work. |
| The requester canceled the approval request | Check the canceled request and confirm whether the work is still needed. |
| The work session ended while the command waited for approval | Review the session's status and confirm a valid work session before continuing. |
| A workspace rule refused a script body in the command line | Submit the script as a [file execution](/use/work-sessions/commands/file-execution/) instead. |
| The command was cancelled before delivery because its account was linked to another person (shown as the refusal reason) | Run the command again as your own account or another account you may use. If you expected to use that account, ask a Superuser. |
| The file command had no verification signature | Ask a workspace administrator to check file-command signing. |
| The agent could not verify the command's signature | Ask a workspace administrator to check command signing and the agent's verification setup. |

If a recorded reason is not recognized by the console but the blocking stage is known, the explanation names the access policy check, risk assessment, approval, or work session. It does not claim a more specific cause. Ask a workspace administrator to review that stage; for approval or work session blocks, also review the request or session status.

If no reason is recorded or visible to you, only the result appears, even if the stage is known. A recorded reason that the console does not recognize also has no explanation when its stage is unknown. This can affect older records; a missing explanation does not mean signing failed. Commands marked **Denied** keep their separate **Denial reason**.

These explanations apply only when a command was blocked before execution. A rejected judgment on a command that still ran is described under [Verification status](#verification-status).

## Verification status

Open a command to see how it was admitted before it ran, under **Execution control**. This is a separate question from the Result column, which is about what happened once it did run.

| Verification | Meaning |
| --- | --- |
| **Pending** | Still being judged. The command hasn't been delivered yet. |
| **Approved** | Judged and cleared to run. |
| **Rejected** | The judgment refused the command. |
| **Skipped** | Not judged. This covers an agent task, which isn't user-submitted, and a command that ran without a judgment because Alpacon AI was unavailable or the self-hosted deployment runs without AI analysis. |
| **Monitor executed** | The command ran after a rejecting verdict under signing monitor mode. The panel adds **Rejected by verification, but executed**, and the verdict and output remain available for review. |
| **Rejected by verification, but executed via a session-less service token** | The [AI verdict](/use/iam/applications/service-tokens/command-patterns/#what-happens-when-a-tokens-command-fails-ai-verification) rejected a service-token command sent without a work session, but the workspace's token execution policy allowed it to run. This is a separate result from signing monitor mode and from Advisory execution control. |
| **Advisory** | Ran without being gated, because the workspace's [execution control](/use/policies/server-access/) is set to Advisory. The judgment is recorded afterwards, and it changes nothing about the command that already ran. |

An **Advisory** command carries the note **Judged after execution, not gated** under the value, so it isn't mistaken for an approval. The judgment lands a moment after the command, and the Risk column then shows its band. If the judgment can't be produced at all, the command stays **Advisory** with no risk band, which is an honest record that nothing was ever assessed.

Use the **Execution control** filter behind `+` to narrow the list to one verification status. **Monitor executed** includes both signing monitor mode and session-less service-token overrides; open a command to distinguish their labels. Filtering on **Advisory** answers "which commands ran without being gated?" in one query.

### Filtering through the API

The same list is available from [`GET /api/events/commands/`](/reference/api/commands/). Use `override_reason=session_less_service_token` to find only the service-token overrides; the console has no separate override filter. The API also filters by `risk_level` (one or more levels, such as `HIGH,CRITICAL`) and by creation time with `added_at__gte` and `added_at__lt`. For example, rejected High and Critical commands in one week:

```http
GET /api/events/commands/?verification_status=rejected&risk_level=HIGH,CRITICAL&added_at__gte=2026-09-28T00:00:00Z&added_at__lt=2026-10-05T00:00:00Z
```

The filters only narrow the rows you can already see.

## Who sees what

The page is open to all users, but the rows scope to your role.

**Regular users** see only the commands they themselves ran, on the page named **My commands**. There is no requester filter.

**Administrators** see every command across the workspace, on the page named **Commands**.

**On the Free plan**, the list shows only your own commands from the last 7 days, for administrators too, and there is no requester filter.

The narrowing is enforced server-side, and direct URLs follow the same rule.

## Common things you'd check here

- **Operations**: confirm a scheduled command actually ran. Track automated agent tasks and deployment commands.
- **Debugging**: open a failed command to read its output and trace what went wrong.
- **Security**: review privileged or unusual commands, or filter by type to see what AI agents ran. Spot commands no one should have submitted.
- **Personal review**: look up your own past commands to recover an answer or copy the line.

Source: https://docs.alpacax.com/use/audit/events/commands/
