2 min read AI-generated

Claude Code 2.1.234: It Keeps Going When Your Limit Resets

Copy article as Markdown

The new version resumes on its own after a usage limit, GitLab merge requests get a badge in the footer, and the built-in 'claude-api' skill now costs an eighth of the context it used to. Plus a security follow-up on Windows.

Featured image for "Claude Code 2.1.234: It Keeps Going When Your Limit Resets"

2.1.234 is a fat release – over sixty lines in the changelog. Most of them are fixes. But a handful of things genuinely change how the terminal feels day to day.

Keep working when the limit resets

The one you’ll notice right away: Claude Code now waits for your claude.ai usage limit to reset and then picks up on its own. No more nudging it by hand when you hit the wall mid-session. Don’t want that? Turn it off in /config under “Continue automatically at usage limit.”

Sounds small, but it’s exactly the kind of friction that adds up. You kick off something bigger, go grab a coffee, come back – and the session has been hanging for an hour because the limit freed up at 2pm and nobody was there to hit Enter.

GitLab gets a badge

GitLab keeps catching up. The footer and statusline now show a merge request badge: if your repo has a GitLab remote and the glab CLI is signed in, you see MR !N with its state – draft, pending, or green. GitHub used to be the clear favorite here. Step by step, GitLab is becoming a first-class case in Claude Code.

A security follow-up on Windows

2.1.233 already closed an NTLM leak via NT paths (\??\). Now Anthropic tightens the screw further: remote file reads, session restore, CLAUDE.md includes, workflow scripts, and file uploads all reject those paths too. The last spots where a crafted path could still slip through are sealed.

Small stuff that matters

Your own prompts now render as markdown in the transcript – code blocks, inline code, and lists look the way replies do. Auto-generated session titles are shorter and more to the point: “Login button bug” instead of a full sentence restating your request. And the built-in claude-api skill now loads its reference docs on demand – the context surcharge drops from over 200,000 to around 25,000 tokens.

On top of that: /permissions and /add-dir can be opened while Claude is working. Rule changes then apply to the rest of the turn.

My take

Auto-continue at the limit is the kind of feature nobody celebrates in a keynote and that still makes the biggest difference. It removes a wait that was purely technical and served no one.

The rest is solid craft. The 200k-to-25k change on the claude-api skill is also a signal: context is money, and Anthropic is starting to budget its own. If you work on Windows, same as with 2.1.233 – grab it, don’t let it sit.


Sources: