Claude Code now auto-resumes when your usage limit resets
Version 2.1.234 also cuts the claude-api skill's context cost 8x, hardens file reads against an NTLM leak, and opens /permissions mid-turn.
Copy markdown
No more restarting when the limit resets
Claude Code now continues your session automatically the moment a claude.ai usage limit resets — no manual re-run. If you'd rather it wait, switch it off in /config under “Continue automatically at usage limit.”
The claude-api skill got 8x lighter
Loading the built-in claude-api skill dropped from ~200k+ tokens to ~25k by fetching its reference docs on demand. That's a large slice of context handed back to your actual work every time you use it.
A quiet NTLM security patch
Remote file reads, session restore, CLAUDE.md includes, workflow scripts and uploads now reject Windows NT-namespace (\??\) paths, closing a credential-leak vector. Nothing to configure — it's live once you update.
Open /permissions and /add-dir mid-turn
You can now change permission rules or add a directory while Claude is working, and they apply to the rest of the current turn. /goal also self-clears on unrecoverable errors and checks in when background tasks stall past 30 minutes.
Small quality-of-life wins pile up
Your own prompts now render markdown — highlighted code, inline code, lists — in the transcript, auto-generated session titles read as short names like “Login button bug,” and Esc in fullscreen no longer wipes a text selection.