# AI agent permission approvals, explained

> How permission approvals work in Claude Code, Codex, Cursor and Antigravity: why agents ask, what to check before you allow, and how to answer quickly.

Canonical: https://herdbell.com/blog/ai-agent-permission-approvals-explained
Author: Herdbell team
Published: 2026-11-01
Last updated: 2026-11-01

A coding agent can edit files and run commands on your machine. A permission approval is the moment it stops and asks before doing something that could change things. It is your main safety check, and also the main thing that slows an agent down.

## How an approval works

1. **The agent wants to act,** for example by running a command or using an MCP tool.
2. **It checks its rules.** Anything you pre-approved runs straight away.
3. **Otherwise it stops and asks.** Work pauses until someone answers.
4. **You allow or deny.** Allow runs that action. Deny stops it, and the agent has to find another way or ask you.

## What to check before you allow

For more, see the [AI coding agent safety checklist](https://herdbell.com/blog/ai-coding-agent-safety-checklist).

- What the command does, especially anything that deletes, overwrites or installs.
- Which project it runs in.
- Whether it reaches outside the project or the internet.

## Approvals in each agent

| Agent | Answer from Herdbell? |
|---|---|
| Claude Code | Yes |
| Codex | Yes |
| Cursor | Yes (shell commands and MCP tools) |
| Antigravity | Shown in Herdbell, approve in Antigravity |

## Answering without switching windows

The prompt normally sits in the agent's own window, so it can wait unnoticed. With Herdbell your pointer glows amber and a card shows the command at your cursor. Press **Return** to allow or **D** to deny. Risky commands ask twice, and if you do not answer in time the agent falls back to its own prompt, so nothing is approved by accident.

## Common questions

### What is a permission approval in an AI coding agent?
It is the agent stopping to ask before an action that could change something, such as running a command. It only continues once you allow it.

### What happens if I do not answer?
The agent keeps waiting. In Herdbell, after the wait you set (5 minutes by default) the request goes back to the agent's own prompt, so nothing is allowed on your behalf.

### Should I turn approvals off?
Only somewhere you can throw away or restore. Without approvals, commands run without you seeing them. See [how to cut prompts safely](https://herdbell.com/blog/stop-ai-coding-agent-asking-permission).

## Related
- [Herdbell: pricing](https://herdbell.com/#pricing)
- [All posts](https://herdbell.com/blog)
- [How to approve Claude Code permissions without switching windows](https://herdbell.com/blog/approve-claude-code-permissions)
- [How to stop Claude Code and Codex asking for permission every time (safely)](https://herdbell.com/blog/stop-ai-coding-agent-asking-permission)
- [AI coding agent safety checklist: 10 rules before you let it run commands](https://herdbell.com/blog/ai-coding-agent-safety-checklist)