Not a headline release, but one of those versions that makes daily work measurably nicer: Claude Code 2.1.203 focuses mostly on long-running background sessions — exactly where the agent feature hurts most in practice when something goes wrong.
Login about to expire? Now you get a warning
The most annoying background-session scenario: you kick off a long job, walk away — and midway through, your login expires and the session dies. There’s now a warning before that happens. You can re-authenticate in time instead of discovering afterward that half your work is gone.
The mode is finally visible
Ever since manual permission mode became the default a few days ago, it wasn’t always obvious which mode you were in. New: a grey pause badge in the footer whenever manual mode is active. Small, but it ends the “wait, why is Claude asking me at every step?” guessing game.
And the rest
There are also clearer agent status indicators and a batch of fixes for macOS, Windows, worktrees, and the daemon. Background sessions are more robust — long commands now survive more reliably when the process is stopped, restarted, or updated. Anthropic also tuned streaming responsiveness and trimmed startup memory.
My take
This is maintenance work in the best sense. None of it makes headlines, but these are precisely the details that decide whether you’ll trust an agent with a multi-hour job or feel you’d better sit next to it. A login warning sounds trivial — until a dropped four-hour session has ruined your afternoon once.
The through-line of the recent releases is clear: away from the “impressive demo,” toward the “reliable tool that runs in the background without surprising you.” For anyone using Claude Code in production, that’s the right direction.
Sources: