How do you approve Claude Code permission prompts from your phone?

Why Claude Code stops and waits

Claude Code asks before it runs shell commands, edits files or reaches the network, depending on the permission mode. In Manual mode it asks before most of these. When you're not at the terminal, that means the task sits there until you come back. Anthropic's docs put it simply: "If Claude hits a permission prompt while you're away from the terminal, the session pauses until you respond."

You have two ways out: answer the prompt from your phone, or reduce how often Claude Code asks. This guide covers both, and why turning approvals off entirely is a bad trade on your main computer.

How the 🔐 Allow / Deny flow works in TextMyWorker

1. Claude Code asks for permission. A Claude Code session on your computer wants to do something that needs your OK, like running git push or deleting a file.
2. The request comes to WhatsApp. If you've been away from the keyboard and mouse for a few minutes, or the task came from WhatsApp, TextMyWorker sends you a message saying what Claude wants to run and where, with Allow and Deny buttons. If it's part of a task you texted, your message also gets the 🔐 reaction.
3. You tap Allow or Deny. Only a tap counts, or a reply to that exact message. Typing "yes" in the chat doesn't approve anything, and neither does an emoji reaction.
4. No answer means no. If you don't answer within 5 minutes, nothing is approved. For a session you started at your desk, the request goes back to the normal dialog on your computer.
5. Back at your desk, the dialog moves back. As soon as you touch the keyboard or mouse, a pending request returns to the normal Claude Code dialog on your screen.

This works for every Claude Code session on your computer, not only the tasks you text. It's the same for Mac and Windows.

What counts as risky

In the worker's own session (the one you text), these always wait for your tap:

• Deleting files
• git push and rewriting history
• Deploys and publishing
• Sending email or messages
• Uploads
• System commands

In your other Claude Code sessions, you get a button for whatever Claude Code itself would ask you about, based on your own permission settings. Honest limit: risky commands are recognized by pattern, so a deliberately disguised command could slip past the button. On a Mac, at the one-folder level, it would still run inside Claude Code's sandbox. Windows has no system sandbox. More in What your worker can use.

Ways to answer permission prompts from your phone

Sources: Remote Control, Desktop and Dispatch, Channels reference, Happy. For the full comparison of phone options, see How to use Claude Code from your phone.

Alternatives: get fewer prompts in the first place

• Permission rules. Pre-approve commands you trust and block ones you never want, in your Claude Code settings. Allow rules cut prompts; ask rules (for example Bash(git push *)) make sure you're always asked.
• Accept edits mode. File edits and common filesystem commands run without asking; other commands still prompt.
• Auto mode. A second model, a classifier, reviews actions instead of you, so you see far fewer prompts. Recent Claude Code versions start terminal sessions in it. Anthropic's own warning: it "reduces permission prompts but does not guarantee safety."
• Sandboxing. Run commands in Claude Code's sandbox so a mistake can't reach the rest of the system.
• Bypass permissions (--dangerously-skip-permissions). Skips the checks. Anthropic says to use it only in isolated environments like containers or VMs, not on the computer that holds your client work and keys.

These combine well with approving from your phone: let safe things run on their own, and keep the few risky ones on a button. Source: Choose a permission mode.

FAQ