Claude Code adds /btw for side questions without interrupting the main task

• Anthropic added a /btw command to Claude Code that opens a side-channel prompt while the main coding task keeps running, according to the launch post.
• The new side question path is deliberately limited: the technical details say it is "a single turn of output," "cannot do any tool calls," and still has "the entire context" of the current conversation.
• Practitioner testing described by an early user frames /btw as a way to ask "a quick question about the current session" without interrupting the main task, with the follow-up adding that it is read-only and does not persist in history.
• Anthropic also pointed users to the Claude Code docs, which document /btw under interactive mode as part of the product rather than a one-off demo, as shown in the docs pointer.

Claude Code now supports /btw for "side chain conversations while Claude is working," as Anthropic described in the announcement. The attached demo Side question demo shows the main pane continuing to generate code while a separate side exchange stays visible, which makes this a workflow feature for in-flight tasks rather than a new model mode.

Anthropic's interactive mode docs place the feature inside Claude Code's existing terminal workflow. The practical change is that users can inspect or clarify the current session mid-run instead of waiting for the active task to finish or breaking the thread with a new top-level prompt.

The implementation is narrow by design. In the technical thread, Anthropic says /btw "cannot do any tool calls" and returns only "a single turn of output," but it still has access to the full session context. That means it is better understood as a read-only inspection path than as a second concurrent agent.

Early users are already describing the behavior in the same terms. One follow-up post says /btw is "read-only," has "no tool access," and "doesn't get added to conversation history"; dismissing it removes it entirely. That combination matters for engineers because it reduces two common side effects of mid-task prompting: spawning extra tool work and polluting the main transcript with transient questions.

The result is small but concrete workflow polish. As the practitioner summary put it, you can ask a quick question about the current session "without interrupting the main task," which is exactly the kind of interaction detail that makes coding agents easier to keep in flow during longer runs.