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

# Approvals

> How Agents ask before consequential actions, and how to answer.

When an Agent wants to do something that needs your permission, the action is **held** before it runs and a card appears in the chat and on Home. Nothing runs until you approve. A denied, expired or abandoned card, a cancelled run and a service restart all count as "no".

## Your choices

| Choice | Effect |
| - | - |
| Approve once | This call runs. |
| Always allow | Also stores a standing rule for that tool on this Agent. |
| Allow for this reply | The same tool is allowed for the rest of this reply (at most 20 calls). |
| Allow for the rest of the day | The same tool is allowed on this Agent until your next local midnight. You can list and revoke these. |
| Deny | The action does not run. |

The "for this reply" and "for the day" options appear only on cards where they are safe. They are never offered for secrets, for browser password entry, uploads and downloads, or for routines and Group turns.

## Spend

Autonomy and approvals are separate from spend. A daily limit (per Agent and for the account) stops runaway cost whatever the autonomy level.

## What a card shows

A card describes the request in words for you. It never contains a secret value.


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