Answering Claude Code Prompts From Your Phone via Telegram

Claude Code's Permission Model

Claude Code is designed to ask before acting. It requests permission to edit files, run shell commands, and perform other potentially impactful operations. This safety-first approach is excellent when you are at your desk. When you step away, it means Claude sits idle waiting for your approval.

Remocode bridges this gap by forwarding Claude Code's prompts to your Telegram chat with interactive response buttons.

What Claude Code Asks

Claude Code typically asks three categories of questions:

File Edit Permissions — Before modifying a file, Claude shows you the proposed changes and asks for confirmation. Remocode forwards this with Allow and Deny buttons.

Shell Command Permissions — Before running commands like npm install, git commit, or build scripts, Claude asks for approval. Again, Allow and Deny buttons appear in your Telegram chat.

Clarification Questions — When your instructions are ambiguous, Claude asks follow-up questions. These arrive as free-text prompts in Telegram, and you reply with your answer.

The Workflow in Practice

Here is what a typical remote session looks like:

• ●You start Claude Code in a Remocode terminal named "webapp" with a task like "refactor the authentication module to use JWT."
• ●You leave your desk and go about your day.
• ●Claude begins working and soon asks: "I need to modify src/auth/middleware.ts. Allow?" Your phone buzzes with the Telegram notification. You tap Allow.
• ●Claude continues. A few minutes later: "Should I also update the existing tests to use the new JWT format?" You see Yes and No buttons. You tap Yes.
• ●Claude finishes the tests and asks: "The refresh token expiry is currently 7 days. What should I change it to?" You type "30 days" and send it.

Each interaction takes seconds on your phone, while Claude keeps making progress in between.

Using Via Mode for Complex Interactions

For simple Allow/Deny decisions, the button interface is fastest. But sometimes Claude presents a long diff or a detailed question that needs context. In these cases, drop into via mode:

via webapp

Now you see the full terminal output, including the complete diff Claude is proposing. You can read through it, then type your response directly. When you are done, exit with !exit.

Handling the Retry System

Claude Code is an Ink-based application, which means Remocode uses smart typing when you send input via via mode. Your text is sent, followed by an Enter keypress after a 1-second delay. If Claude does not produce output within 8 seconds (perhaps because it is processing), Remocode retries at 8, 18, and 28 seconds.

This means you should not re-send your answer if Claude seems slow to respond. The retry system handles it automatically. If you want to verify that your input is sitting in the buffer, use !peek.

Dangerous Command Protection

Claude Code sometimes needs to run commands that Remocode classifies as dangerous: git push --force, rm -rf, sudo, npm publish, and others. Even if Claude has permission to run shell commands, Remocode adds an extra confirmation layer for these. You will see a warning message with confirmation buttons before the command executes.

Catching Up After Being Away

If you have been away and Claude has been working (or waiting), use the reply command to fetch the last AI responses:

reply_5

This retrieves the last 5 AI responses from the terminal buffer, giving you a quick summary of what happened without needing to scroll through the full via mode output.

Tips for Claude Code Specifically

Keep terminal names short since you will type them often on a phone keyboard. Set up your Telegram notifications so that Remocode messages get through even in Do Not Disturb mode. This ensures you never miss a blocking prompt that is keeping Claude idle.