How to approve Claude Code permissions without switching windows

What Herdbell does when an agent asks

Risky commands ask twice before they are allowed. If you are away, the agent falls back to its own prompt after the wait you set in Settings (5 minutes by default), so nothing is ever approved by accident.

1. Your pointer glows in the colour of what is waiting (amber for a permission).
2. A small card appears by your cursor naming the agent and the command it wants to run.
3. Press Return to allow, or D to deny. The agent carries on straight away.

Set it up in about two minutes

1. Install Herdbell and enter your license key. Onboarding starts right after.
2. Let Herdbell find your agents and press Connect. It adds small hooks to each agent's own settings, and saves a backup of those settings first.
3. Start a new chat in Claude Code. It shows up in Herdbell by itself, and the first permission request appears as a card.

It works the same for Codex and Cursor

Claude Code, Codex and Cursor all send Herdbell their permission requests through hooks, so Allow and Deny work in Herdbell for all three. Antigravity shows the request in Herdbell too, but you approve it in Antigravity itself.

Good to know

• Herdbell only answers the questions an agent sends it. It cannot see or change what an agent decides on its own.
• If Herdbell is closed, your agents carry on with their own prompts. Nothing breaks.
• Everything stays on your Mac. Herdbell has no account and never calls an AI model.

Common questions