Two versions on, and this one is less about security than about flow. 2.1.250 clears up a few spots where Claude Code used to stall.
The change that helps me most: when a session runs into your usage limit, it now picks up on its own as soon as the limit resets. No cut-off mid-thought, no restart, no “where was I again?” If you like kicking off long runs in the evening the way I do, you know the annoyance. From now on it’s gone.
Second: you can address another running session by name with @. That sounds like a toy, but it’s the logical next step after the session-to-session messaging from recent releases. Instead of two agents finding each other blindly through ListAgents, you point straight at the one working on the API layer and hand it your decision.
It fits that subagent forking is now on by default. Agents you start interactively keep running in the background rather than blocking your foreground. For anyone juggling several threads at once, that shifts the rhythm of the work noticeably — you wait less and delegate more.
One thing that might trip you up: the todo and task tools (TaskCreate, TaskUpdate, TodoWrite and friends) are no longer available on the newer models — Opus 4.8, Sonnet 5, Fable 5, Mythos 5. Those models plan for themselves now, so the explicit tracking tool was more baggage than help. If you still want it, bring it back with CLAUDE_CODE_ENABLE_TODO_TOOLS=1.
The rest is the usual quiet upkeep. Remote Control syncing got expanded so local sessions carry across devices more cleanly. GitLab finally gets MR badges. And in the terminal, Devanagari and other Indic scripts that used to wreck column alignment are fixed — a detail most people never see, but for those affected it’s the difference between usable and unreadable.
Not a spectacular release, but automatic continuation at the limit is one I’ll take right away. Exactly the kind of small friction you only miss once it’s gone.
Sources: Claude Code Changelog · Releasebot: Claude Code