The problem it solves: your agent is deep in a multi-file refactor, and you realize you forgot to mention a constraint. Before /btw, your only options were to interrupt and restart, or hope it figures it out. Now: /btw your constraint, and Claude incorporates it while continuing.
How It Works
Type /btw followed by your message. Claude reads it as a side note — not a new instruction set that overrides the task, but additional context to factor in.
Examples:
- "/btw we're on PostgreSQL not SQLite"
- "/btw the API rate limit is 100 calls/minute"
- "/btw the client changed this requirement: [update]"
The agent absorbs the aside and adjusts without starting over.
Why This Matters More Than It Sounds
Long agent sessions have always had an awkward problem: you can start them and stop them, but you can't gracefully course-correct them mid-flight. /btw adds a middle state.
For complex tasks — multi-file migrations, large document generation, extended research tasks — this means fewer false starts. You can launch with imperfect context and improve on the fly.
It's a small UX feature with a large practical impact. The fact that Anthropic shipped it signals they're thinking seriously about the daily experience of working with agents, not just benchmark scores.