Claude Code: /btw Side-Chain Chats for Mid-Task Questions
Claude CodeClaude Code v2.1.72 introduces the /btw (By The Way) command, enabling ephemeral side-chain conversations while a main task is running. The feature spawns a lightweight read-only agent that reuses the parent conversation's prompt cache, letting developers ask quick questions, get clarifications, or verify facts without polluting the main context or interrupting in-progress work. Anthropic claims up to 50% token savings for sessions where frequent mid-task questions are common.
Sources & Mentions
6 external resources covering this update
The simple /btw command that cuts Claude Code costs in half
Link
I Tried (New) Claude Code /btw Command (Now You Can Chat As You Code)
Link
Claude Code Adds /btw Side-Chain Chats: Latest Productivity Boost for Developers
Link
How to Use the Claude Code /btw Command to Save Tokens and Ask Questions Mid-Task
Link
Claude Code /btw by the way slash command
Link
Claude Code has introduced the /btw command
Link
/btw Side-Chain Chats
Claude Code v2.1.72 introduces /btw, a new slash command that lets developers ask side questions while Claude is actively working on a task. Typing /btw [question] spawns an ephemeral agent that answers the query in a temporary overlay, without appending the exchange to the main conversation history.
How It Works
The /btw command creates an isolated side-chain conversation that has full visibility into the current session context. It operates in read-only mode — it cannot edit files, run commands, or trigger tools. Answers appear in an overlay panel and disappear when closed.
Critically, the ephemeral agent reuses the parent conversation's prompt cache, so the marginal cost of a /btw query is minimal compared to a regular message. Since the exchange is excluded from the main thread, subsequent messages don't pay the input-token cost of carrying those Q&A pairs forward.
Token and Cost Savings
For sessions where 40–50% of exchanges are lightweight mid-task questions (factual lookups, style preferences, quick confirmations), Anthropic estimates up to 50% reduction in cumulative token costs. The savings scale with session length and model tier.
Best Use Cases
- Quick factual lookups and verifications
- Clarifying function behavior or return types
- Confirming code style or output format preferences
- Injecting constraints or infrastructure context
- Sanity-checking assumptions without derailing the main task
Limitations
- Read-only: no file editing, no command execution, no tool access
- Limited to existing session context
- Answers do not persist in conversation history
- Not suited for multi-step debugging or consequential decisions that need future reference
- Should not replace
CLAUDE.mdfor persistent cross-session context