💡 Claude Code v2.1.185: Reading the New Stream-Stall Hints Correctly
Claude Code v2.1.185 rewrites the messaging that appears when a response stream stalls mid-turn — the moment your terminal goes quiet after tokens were flowing a second ago. Previously the hint text was generic enough that it was hard to tell a genuine network hiccup from a long tool call still running in the background. The updated copy now distinguishes the two cases explicitly, so you know whether the right move is to wait or to check your connection.
Why this matters for long sessions
- A stall during a large multi-file edit or a long-running bash command is normal and self-resolves — the new hint says so, rather than looking like an error.
- A stall caused by an actual dropped connection now surfaces distinctly, so you're not left watching a frozen cursor for minutes before realising you need to retry.
So what should you actually do?
If you're running long agentic sessions unattended (overnight refactors, big test-suite runs), this small wording fix has an outsized effect: you can now grep your terminal log for the specific stall-type text afterwards to tell whether a session actually failed or just had a slow tool call. Worth updating any log-monitoring scripts that previously matched on the old generic stall string.
Claude Code
v2.1.185
stream stall
changelog
long-running sessions
debugging
💡 /code-review Now Runs as a Background Subagent — Your Conversation Stays Clear
The /code-review command no longer runs inline in your main conversation. It now launches as a background subagent, with your currently stacked slash commands as its review target. Practically: you can kick off a review and keep working — asking follow-up questions, editing other files, or running a second command — while the review happens out of view, then surfaces its findings when ready rather than filling your active thread with review commentary you have to scroll past.
What changes in practice
- Your main conversation context is no longer consumed by review output — useful on long sessions where context budget is already tight.
- You can stack a review behind other in-flight work instead of blocking on it.
- Findings still land as a structured report when the subagent finishes, same as before — only the execution model changed.
Good habit: review before you commit, not after
Because the review no longer costs you conversation space, there's less reason to skip it on smaller changes. A sensible default: fire /code-review as soon as a diff feels "done," keep working on the next thing, and treat its findings as a pre-commit gate rather than a post-hoc audit.
/code-review
background subagent
Claude Code
context budget
code review workflow
