Tip: Turn On Focus View (Ctrl+Alt+F) for Any Run Longer Than ~20 Tool Calls
Releasebot·medium signal
Focus view in Claude Code 2.1.221 collapses tool activity into an expandable summary, leaving the model's actual decisions legible in the transcript. For long agent sessions the bottleneck is human review bandwidth, not model capability — you cannot catch a bad decision you scrolled past. Toggle it on for autonomous runs and expand only the tool calls adjacent to a step you want to verify.