Claude Code /btw — Ask Questions Mid-Task Without Breaking Flow

You are deep in a Claude Code session. Claude is refactoring a big chunk of your codebase — reading files, making changes, things are moving. Then you suddenly remember something. A quick question like: "What was the config file it read earlier?" or "Is this approach compatible with Node 18?"

You have two options. Interrupt (which kills the current task) or wait (and forget your question by the time it's done). Neither is great.

There is a third option, added to Claude Code just recently, and most people don't know it exists yet. It's called /btw.

What /btw Does

/btw stands for "by the way" — and the name explains the behavior perfectly.

Think about real life. Your friend is cooking in the kitchen, focused on a task. You walk in and say "By the way, do you have any salt?" They answer quickly and carry on cooking. The conversation didn't really happen. It was a side note.

That's exactly what /btw does in Claude Code:

You type /btw followed by your question
Claude answers it in a small overlay on your screen
You read it, dismiss it (press Escape, Space, or Enter)
Your main task continues completely uninterrupted

The important part: that question and answer never get added to your conversation history. The main context stays completely clean. You got the answer, and it's like it never happened as far as the session is concerned.

Why This Matters for Token Cost

This isn't just a UX improvement. Every time you send a message, the model reads your entire conversation history from the very beginning — every message, every response. That's how it maintains context.

If your session history is already 20,000 tokens long, every new message costs at least 20,000 input tokens before you even count what you're typing.

Every throwaway question — "what did we name that variable?" or "is this file already modified?" — gets permanently baked into your history. From that point on, every future message carries that extra weight.

Does /btw cost zero tokens? No. It does use tokens. But because it reuses your session's prompt cache, the cost is significantly lower than a normal message. And more importantly, because the exchange never enters your history, it doesn't inflate the cost of every future message in the session.

The saving isn't in the /btw itself. The saving is in every message that comes after it.

The Key Limitation

/btw cannot run commands or read new files. It can only answer from what's already in the current session context.

If you want Claude to go find out something new, that's not what /btw is for.

/btw vs Sub-Agents

The official Claude Code docs describe /btw as the inverse of a sub-agent:

/btw = Full context, no tools
Sub-agents = Full tools, empty context

Use /btw when you want to ask about something Claude already knows from this session. Use a sub-agent when you need Claude to go research something new, outside the current context. They're complementary, not competing.

When to Use It

Quick clarification about earlier decisions in the session
"What was the file path we changed?" type questions
Checking if Claude already considered something
Any throwaway question you don't want cluttering your history

Availability

Added in Claude Code v2.1.72 (released March 10, 2026). If you haven't updated recently, run the update first.

It won't change how you code. But for long sessions with frequent mid-task questions, this is one of those small features that genuinely adds up.

Watch the full demo: Claude Code /btw — Ask Questions Mid-Task Without Breaking Flow