# Approve what agents can do

> Decides which write operations on your connections agents can call, who enables each one, how to stop one, and when a person decides single actions in Slack.

**Skill for your coding agent:** [`mindset-script/SKILL.md`](https://docs.mindset.ai/skills/mindset-script/SKILL.md)

**Skill for your coding agent:** [`mindset-connect/SKILL.md`](https://docs.mindset.ai/skills/mindset-connect/SKILL.md)

Builders work from a coding agent and operators from an AI client, both through the MCP server, or in the Studio. In Mindset, an agent is a prompt and a model, held to an order by a script when needed. It uses connections to outside systems, operations on them, functions and widgets, and loads Skills when a request calls for them. Every change is a version, and every execution is a run. Every piece sits on one security and compliance layer: the org sets its posture once, and every agent works under it, with the audit trail and, once an admin turns it on, personal data redaction. The platform runs in your own cloud (AWS, GCP or Azure) or is hosted by Mindset. Your app reaches agents through the SDK, your server through the API, and people through the Hub and schedules. Highlighted: Connection, Operation.

Agents reach your systems only through the [operations](https://docs.mindset.ai/docs/connections-and-operations) on each connection, and each operation is a read or a write. A write changes something in that system, such as registering a supplier query, so it's the kind you approve.

## Ask your AI client

```
List every write operation the invoice exceptions agent can call in Production, and any that are still waiting for a person's approval.
```

## What you get back

Each write operation, the connection it's on, and whether it's enabled or waiting for a person. With your org's setting on, which is the default, your AI client enables a new write itself. With it off, a write added from then on waits for a person to enable it in the Studio, and no AI client or agent can. Writes already enabled stay enabled. How consent works is on [Security and approval](https://docs.mindset.ai/docs/security-model).

## How approval works

A write needs consent once per operation, never per call. The org setting **Enable actions that change other systems automatically** is on by default, so your AI client can enable a new write operation as soon as it's added.

> **Warning:** Leave it on and nobody approves new writes. Turn it off and a person enables each new write operation once before anything can call it. Writes already enabled stay enabled.

If a write's definition changes, it needs consent again.

| Button             | On                                   | What it does                                                                                           |
| ------------------ | ------------------------------------ | ------------------------------------------------------------------------------------------------------ |
| **Enable**         | A new write, when the setting is off | Lets agents call it. Recorded with your name                                                           |
| **Lock operation** | Any operation                        | Stops every agent, script and function calling it. Only a person can enable it again, on the same page |
| **Enabled** switch | A whole connection                   | Turned off, refuses every call to any of its operations                                                |

Each takes effect on the next call, with no new version. To stop something right now, this is the fastest way.

## Decide one action during a run

When a person must decide a specific action, the script holds an `$ask` step. It posts one message to Slack with buttons, and the run waits for the answer or a timeout. See [Make it follow steps](https://docs.mindset.ai/docs/write-a-script), and [Post to Slack or send email](https://docs.mindset.ai/docs/post-to-slack-or-send-email) to set up the Slack app.

- **Put it late.** Gather, compare and assess first, then ask before the query is registered.
- **Ask once for the batch.** One message with every disputed line gets read. Fifty get waved through.
- **Put enough in it to judge:** the invoice, the PO, each line that differs and the contract clause.
- **Expect a slow answer.** Don't build anything that goes stale in ten minutes.

## What you decide

- Whether new writes turn on automatically, and who approves them if not. Finance approves finance writes.
- Which actions need a person each time, as an `$ask` in the script.

> **Info:** Write approval is one of your org's [compliance rules](https://docs.mindset.ai/docs/compliance): one setting, applied to every agent and connection.

## Done when

- You can say, for every write the agent uses, whether the org setting or a named person granted it.
- You know how to stop one operation (**Lock operation**) and a whole connection (its **Enabled** switch).

## In the Studio

The setting is on **Settings → Systems → Connections that change other systems**. A connection's operations are listed under **Operations** (**Tools** for an MCP server) on its **Overview**. Open one to see its state and who enabled it. A write waiting for a person shows **Disabled** in the list. Open it and press **Enable**. The **Enabled** switch for the whole connection is at the top of its **Overview**.
