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

# Proposed actions

> How an AI finding can propose a next step — post to a channel, file a ticket, attach a runbook, or call a webhook — that a human approves before it runs.

A **proposed action** is a next step an [AI finding](/concepts/ai-investigations) suggests during an incident — *"file an issue for this"*, *"post a summary to `#major-incidents`"*. scaling.cloud never runs a proposed action on its own. It waits in the incident's **Proposed actions** panel until a responder clicks **Approve**, and only then does it execute.

This is the deliberate guardrail of the AI layer: the AI can diagnose and suggest, but a human is always in the loop before anything outside scaling.cloud happens.

## Action types

Version one keeps the blast radius small — every action type is **reversible and low-risk**:

| Action              | What it does                                           |
| ------------------- | ------------------------------------------------------ |
| **Post to channel** | Posts a message to the incident's chat channel.        |
| **Create ticket**   | Files a linked issue in [Linear](#external-ticketing). |
| **Attach runbook**  | Links a runbook URL to the incident.                   |
| **Call webhook**    | Sends a signed `POST` to an external system.           |

<Note>
  Infrastructure-mutating actions — rollback, scale, restart, flag toggles —
  are deliberately **not** part of this set. They carry real blast radius and
  are out of scope for proposed actions today.
</Note>

## The approval flow

1. A finding proposes an action. It appears in the **Proposed actions** panel on the incident, with an `ai` origin and a status of `proposed`.
2. A responder reviews it and clicks **Approve** (or **Dismiss**).
3. On approval, scaling.cloud executes the action and records the outcome on the incident.

Execution is **durable and idempotent** — an approved action that hits a transient failure is retried without firing twice, so approving "create a ticket" never leaves you with two tickets.

## External ticketing

The **Create ticket** action files into your issue tracker. Connect it the same OAuth way as every other integration — from **Settings → Integrations**, connect [**Linear**](/integrations/linear). Once connected, an approved create-ticket action files a linked issue and links it back to the incident.

<Warning>
  A create-ticket action can only run if **Linear** is connected. If it isn't,
  approve a different action or connect Linear first.
</Warning>

## Entitlement

Proposed actions are part of the AI layer and require a plan that includes **AI Investigations** (**Pro** or **Enterprise**). For human-configured automation that does not involve AI, see [Automation rules](/concepts/automation-rules) — available on every plan.

<CardGroup cols={2}>
  <Card title="AI Investigations" icon="wand-magic-sparkles" href="/concepts/ai-investigations">
    Where findings — and the actions they propose — come from.
  </Card>

  <Card title="Automation rules" icon="diagram-project" href="/concepts/automation-rules">
    Declarative when → then automation, no AI and no approval step.
  </Card>
</CardGroup>