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

# Composer: mentions and commands

> Point the box at one of your questions, rivals, pages, missions or teammates with @, run a / command such as /counter or /win, and read what agents do on their own, right beside the box. Every command maps to one MCP verb.

The box where you start a [chat](https://docs.attensira.com/agent/chats.md), above [Today](https://docs.attensira.com/agent/today.md), does three more things:

* **`@` mentions** attach one of your objects to the message, so the agent reads that object instead of guessing from a name.
* **`/` commands** are shorthands for the work people ask for most, and each one is the same request an assistant makes over [MCP](https://docs.attensira.com/mcp/tools.md) or a key makes over the [REST API](https://docs.attensira.com/api/overview.md).
* **The autonomy chip**, on the right of the box, shows what agents do on their own. It only shows information and changes nothing.

## `@` mentions

Type `@` and then a few words. The menu lists your context ([Topics](https://docs.attensira.com/tracking/topics.md) and connected integrations) and, under **Mention**, the objects whose name contains every word you typed:

| What | Matched on | Sent as |
| - | - | - |
| A **question** you track | Its wording | `attensira://{workspace}/question/{id}` |
| A **competitor** | Its name or domain | `attensira://{workspace}/source/{domain}`: the rival's domain as a cited source |
| A **page** of yours that answers cite | Its title or address | `attensira://{workspace}/page/{id}` |
| A **mission** | Its name or `#slug` | `attensira://{workspace}/mission/{id}` |
| A **teammate** | Their name or `@handle` | `attensira://{workspace}/member/{user_id}` |

A query can include spaces: `@best GEO` finds "Best GEO tools for SaaS". Pick a row with Enter, Tab or a click. It becomes a chip in the box.

When you send, each mention is added under your words as one line with its address:

```text theme={null}
How do we win this?

Mentioned:
- question “Best GEO tools for SaaS”: attensira://acme-1/question/q_geo
```

The agent reads each object from its address with `get`, the same way an assistant does over MCP. That is why a mention is exact and a typed name is not. Mentions are sent with a new chat and with a reply inside one.

Mentioning a teammate here tells the agent which person you mean, and tells the teammate too: they get one email that links to the chat, under their own [notification settings](https://docs.attensira.com/account/notifications.md). Naming yourself tells nobody. See [Mentions in a chat](https://docs.attensira.com/team/members.md#mentions-in-a-chat).

If nothing matches, the menu closes and your `@` stays in the text as you typed it. If your questions, competitors, pages or teammates can't be read, the **Mention** section is empty and the message still sends as typed.

## `/` commands

Type `/` at the start of a new chat. The menu lists the [modes](https://docs.attensira.com/agent/chats.md#modes), your skills and your projects. The commands come last. Picking a command writes it into the box, for example `/counter `, and you type what it needs after it.

| Command | What it does | MCP verb | REST |
| - | - | - | - |
| `/plan <what you want>` | A Plan chat: the agent says what it would do. Nothing changes until you agree | `ask` with `mode: "plan"` | `POST /v1/threads` |
| `/report` | A report on what shipped this month and what moved, or on what you write after it | `ask` with `mode: "report"` | `POST /v1/threads` |
| `/counter <rival>` | A chat that works: it finds the questions where answer engines name the rival and not you, and goes after them. The rival is attached when it is one of your competitors | `ask` with `mode: "do"` | `POST /v1/threads` |
| `/win "<question>"` | Proposes a [mission](https://docs.attensira.com/agent/projects.md) to win one question: a goal, the lanes and a budget. Nothing runs until you accept it. The question is attached when you track it | `mission` | `POST /v1/missions` |
| `/audit <url>` | A chat that works: it reads the page the way answer engines do and fixes what stops it from being cited. The page is attached | `ask` with `mode: "do"` | `POST /v1/threads` |
| `/explain` | Why your numbers moved this week, or whatever you write after it. It reads your data and changes nothing | `ask` with `mode: "plan"` | `POST /v1/threads` |
| `/share` | Copies a read-only link to your [receipt](https://docs.attensira.com/account/receipt.md) | `share` | `GET /v1/projects/{id}/receipt/share` |

In the app a thread is a chat. `ask` with `mode: "do"` starts an ordinary (Ask) chat, and `mission` starts the project-scoping chat that ends in a plan for you to accept.

A command that is missing what it needs, such as `/counter` with no rival or `/audit` with something that isn't a web address, sends nothing. One line says what to add, and your text stays in the box. A command you picked a mode chip for is sent as plain text in that mode. Skills you have already set up still appear in the `/` menu.

Every command is also in the command palette: press **⌘K** (Ctrl-K on Windows and Linux) on any page. The palette lists them under **Commands**. Type a command with what it needs, with or without the `/`, for example `counter Peec`, and press Enter: the command goes into the box above Today, ready for you to send. The palette never sends it for you. The palette also has the [queue's commands](https://docs.attensira.com/app/queue.md#from-the-command-palette): approve next, decline next, stop, undo and switch brand.

No command can raise an autonomy level, raise a budget or change a setting. Those stay on your own screens, where only a signed-in person can change them.

## The autonomy chip

The chip on the right of the box shows the level agents run at, across your lanes.

* When every lane is at one level, the chip shows what that level does and then the level: **Waits for your yes · L2**, the default. On a phone it shows **L2**.
* When the lanes differ, it shows the range, for example **L1–L3**.

The chip shows the level each lane actually runs at. That can be lower than the level you set while a playbook's quality checks have not passed yet; see [Approval modes](https://docs.attensira.com/agent/approval.md).

Click the chip to see each lane's level. A lane held below the level you set says so, with the reason. The list is read-only. To change a level, follow **Change levels in Settings**. Raising a level there takes a signed-in person, as it does on every door. No assistant or key can raise one.

If the levels can't be read, the chip shows **—** instead of a level. It reads them again the next time you move to it or click it.

The same levels are `get(attensira://{workspace}/autonomy)` over MCP and `GET /v1/autonomy` over REST.

The chip shows no spend. What you have spent this month and your budget live on the Money page alone, so there is one place to read them.


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