Claude Code is busy fixing files, running tests, tracking down bugs — and then a random question pops into your head. What do you do? Until now, you had to hit Esc to stop the task or open another terminal window. Now, a single /btw is all you need.
What Is This?
/btw stands for "by the way." It's a side-question feature in Claude Code that lets you ask a quick question without interrupting the main task. The question and answer appear in a popup overlay, and when you close it, they disappear. They don't get saved to your conversation history at all.
This feature actually had quite a journey. Starting with Claude Code v2.0.73 in December 2025, a hint message reading "Start with btw to ask a quick side question" started appearing, but it didn't actually work. Bug reports were filed on GitHub, and an Anthropic engineer removed the hint, calling it "an experimental feature." About three months later, in March 2026, it came back as an official feature.
Technically speaking, /btw is the exact opposite of a subagent. A subagent can freely use tools (reading files, running commands, etc.) but starts with an empty context. Meanwhile, /btw can see the entire current conversation context but can't use any tools. It can only answer about code Claude has already read, decisions it has made, and things that have been discussed.
Key Analogy
Subagent = Hiring a new employee and sending them off to investigate (tools ✓, context ✗)
/btw = Asking the coworker sitting next to you "Hey, what was that thing earlier?" (tools ✗, context ✓)
It barely costs anything either. Since it reuses the parent conversation's prompt cache, additional token consumption is minimized.
What Changes?
Previously, when a question came up while Claude Code was working, you had two options: stop the task or open another terminal. Either way, your flow gets broken. /btw solves this problem cleanly.
| Old Way | /btw | |
|---|---|---|
| Task Interruption | Must hit Esc to stop | Ask without stopping |
| History Pollution | Piles up in history | Doesn't appear in history |
| Extra Cost | New turn = extra tokens | Cache reuse, minimal cost |
| Context Retention | Context may be lost after interruption | References entire current conversation |
| Follow-up Questions | Freely continue | One-shot (1 question, 1 answer) |
Community reactions have been enthusiastic. One user on Threads said, "I always had to hit Esc and then type my command, glad this is coming," and on Reddit, people responded, "Being able to ask without breaking context seems like it'll be a huge productivity boost." The key takeaway: you can satisfy your curiosity while staying in the flow.
The Essentials: How to Get Started
- Type /btw while Claude is working
When Claude is spinning and writing code, type something like/btw what was the name of that config file earlier?in the prompt. - Check the answer in the overlay
The answer appears as a popup overlay. The main task keeps running in the background. - Close with Space, Enter, or Esc
Once you've read the answer, press any key to close it. No trace is left in your conversation history. - Works between tasks too
You can use it when Claude is idle as well. It's perfect for lightweight lookups that don't clutter your history.
Note: /btw Limitations
No tool access — it can't read files, run commands, or search. It only answers using information already in the current conversation. No follow-up questions either. If you need a conversation, use a regular prompt.
