You are halfway through a complex refactor and need to remember the name of that config file Claude mentioned 40 messages ago. Previously, asking would derail the current task and add noise to your conversation context. Claude Code now has /btw for exactly this situation.
The command works like a sticky note you pass to Claude while it keeps working. Type /btw what was that config file called? and you get an answer in a dismissible overlay. The question and response never enter your conversation history, so your main context stays clean.
How It Actually Works
The /btw command has access to everything in your current conversation, so it can reference code Claude has already read, decisions made earlier, or any other context from the session. But it deliberately cannot use tools - no file reads, no bash commands, no searches. It only answers from what Claude already knows in the current session.
A few practical details worth knowing:
- Works while Claude is busy. You can fire off a
/btwwhile Claude is mid-response on your main task. The side question runs independently. - Cheap to use. It reuses the parent conversation's prompt cache (the pre-computed token data from your session), so the cost is minimal.
- One-shot only. No follow-ups. If you need a back-and-forth, use a regular prompt.
- Dismiss with Space, Enter, or Escape to get back to your prompt.
Anthropic describes /btw as the inverse of a subagent: a subagent gets full tool access but starts with no context, while /btw gets your full context but has zero tools. That is a clean mental model for deciding which to use. Need Claude to recall something from this session? /btw. Need Claude to go find something new? Subagent.
This is a small quality-of-life addition, but for anyone spending hours in Claude Code sessions, not having to choose between asking a quick question and keeping your context clean removes a real friction point.