Claude Code Stuck Waiting? How to Answer from Your Phone

You kick off a Claude Code session, head to a meeting, and come back an hour later to find it asked a question two minutes after you left. The entire session was idle for 58 minutes. Sound familiar?

This is the single biggest productivity killer with AI coding agents. The agent does great work autonomously, but the moment it needs human input, everything stops. And you have no way to know it happened unless you are staring at the terminal.

Why Claude Code Blocks

Claude Code is designed to be safe by default. It asks for confirmation before:

• ●Creating or deleting files
• ●Running shell commands
• ●Making large-scale changes
• ●Choosing between multiple approaches

This is good behavior. You want your AI agent to check before doing something destructive. But it means the agent frequently needs a human to say "yes" or "no" or "take option 2."

The Real Cost of Stalled Sessions

When Claude Code is waiting, it is not just idle — it is costing you money. Anthropic charges for active sessions based on time. A session that could have finished in 20 minutes instead takes 2 hours because the agent waited 1 hour and 40 minutes for a response.

Beyond direct costs, there is the opportunity cost. Work that should have been done by lunch is not ready until end of day. Pull requests pile up. Your momentum disappears.

The Solution: Telegram Remote Control

Remocode connects your AI coding sessions to Telegram. When Claude Code asks a question, you get an instant notification on your phone. You can respond right there, without opening a laptop.

Setting It Up

• ●Install Remocode on your Mac
• ●Open Settings and connect your Telegram account
• ●Start Claude Code in a Remocode terminal pane
• ●Name the pane (e.g., "claude")

That is it. Remocode monitors the pane for prompts and sends you Telegram messages when input is needed.

Key Commands

Once connected, you have full control from your phone:

• ●`via claude reply yes` — Send "yes" to Claude Code
• ●`via claude reply 2` — Select option 2 from a numbered menu
• ●`peek` — See the current terminal output across all panes
• ●`status` — Check which agents are running and which are blocked
• ●`submit` — Send a multi-line message to the active pane
• ●`help` — See all available commands

Handling Different Prompt Types

Claude Code uses several types of prompts:

Yes/No approvals. The most common. Claude wants to run a command or create a file. Just reply via claude reply yes or via claude reply no.

Numbered menus. Claude presents options like "1. Continue 2. Skip 3. Abort." Reply via claude reply 2 and Remocode translates that into the correct arrow-key navigation.

Open-ended questions. Claude asks "What should I name this component?" Reply with via claude reply UserProfileCard and Remocode types it into the terminal.

Going Fully Autonomous

For sessions where you trust the agent, you can go beyond manual Telegram responses:

Auto-Yes Mode

Toggle Auto-Yes on a pane and Remocode will automatically approve prompts where the first option is "Yes." This uses simple pattern matching — no AI API calls, zero cost. Perfect for trusted sessions where you just want to keep Claude moving.

AI Supervisor Mode

For smarter automation, enable the AI Supervisor. It uses a configurable AI model (like Claude Haiku or GPT-5 Nano) to evaluate each prompt against your project brief. It can approve safe operations, reject risky ones, and escalate ambiguous decisions to you via Telegram.

Each supervisor decision costs roughly $0.001 to $0.01 depending on the model — far cheaper than the cost of a stalled session.

The Workflow in Practice

Here is what a typical day looks like with Remocode:

• ●Morning: Start three Claude Code sessions for different tasks
• ●Meeting: Get a Telegram notification — Claude needs a file name. Reply in 5 seconds
• ●Lunch: Check status from your phone. Two agents finished. One is still working
• ●Afternoon: Agent hits an error. Remocode catches it with regex monitoring. You see the alert, fix the issue, and tell Claude to retry

No more coming back to stalled sessions. No more wasted hours. Your AI agents keep working while you do everything else.