New

What's new in Claude Code, August and September 2026: the useful recap

/design, a live /diff panel, /skill-doctor, cache diagnostics in /cost, fork mode, the Concise style, the end of Ultraplan: the Claude Code changes from v2.1.220 to v2.1.269 worth your attention.

  • Reference
  • Productivity
Published

Quick answer

Between early August and mid-September 2026, Claude Code went from v2.1.220 to v2.1.269. That's a lot of versions, and the full changelog is hard to follow. We sorted it: here's what actually changes how you work, grouped by use rather than by date.

Some topics get their own article: cross-session messaging, claude plugin eval and auto mode by default.

First, update:

claude update
claude --version

Better visibility into what Claude does

/diff becomes a live panel

In fullscreen rendering, /diff now opens a panel beside the conversation. It lists changed files with added and removed line counts, and refreshes each time Claude edits a file or runs a shell command. You can select lines with the mouse to attach them to your next prompt. It needs a git repo and a terminal at least 110 columns wide.

It's the difference between getting a report when the job's done and having a window onto the site while work progresses.

The Concise output style

A new built-in style: Claude leads with the result and skips preamble and narration, while working as thoroughly as with the default style. Error reports, security warnings and confirmations for destructive actions stay complete. 

{
"outputStyle": "Concise"
}

Turn it on under Output style in /config, or in ~/.claude/settings.json, then run /clear or start a new session.

Keeping costs in check

/cost explains your cache misses

/cost now shows a Prompt cache (main) line: the share of input tokens served from cache, the cache misses, whether the cache is warm, and a likely cause for the last miss when Claude Code can name one. Status line scripts get a matching prompt_cache object.

With Opus 5.5, where cache reads cost $0.20 per million tokens versus $4 for regular input, every cache miss is expensive. So this line is the first thing to check if a session feels costly.

/skill-doctor finds unused skills

Every listed skill adds text to your context on every turn, whether or not Claude uses it. /skill-doctor shows what each skill costs in context and how often it gets used, so you can turn off the useless ones. It requires v2.1.252 or later.

Cap effort with maxEffortLevel

The maxEffortLevel setting, at the top level or per model under modelSettings, caps the effort level on every provider, including Bedrock, Google Agent Platform and Foundry. Any higher level runs at the cap. Useful for a team that wants to avoid max sessions out of habit.

Another practical detail: /effort and the /model picker now save a separate effort level for each model.

Working in parallel

Fork mode on by default

Since v2.1.232, in interactive sessions, Claude can spawn a fork subagent that inherits the full conversation and prompt cache instead of starting fresh. No more re-explaining the context for a side task. 

> /subtask draft unit tests for the parser changes so far

To turn it off: CLAUDE_CODE_FORK_SUBAGENT=0.

GitLab joins GitHub

Plugin marketplaces accept gitlab.com URLs, and since v2.1.233, --worktree accepts a GitLab merge request URL to branch from it. Claude Code also redacts GitLab tokens such as glpat- in its output. 

claude --worktree https://gitlab.com/group/project/-/merge_requests/42

Design and interface

/design, in research preview

The /design skill brings Claude Design's artboard workflow into the CLI and Claude Code Desktop. You describe what you want, Claude publishes a canvas of editable artboards, you pick one, and Claude implements it. Available on Pro, Max, Team and Enterprise, from v2.1.234. 

> /design redesign the composer based on what people actually use it for

On Desktop

  • Pop-out windows: any pane (diff, terminal) can move into its own window, on a second screen for instance.
  • /resume in Desktop: pick up in the app a session started from the terminal, with the full conversation.
  • Background computer use: on macOS, in beta for Pro and Max, Claude acts in the apps you approved while you keep working.

From your phone

Any machine running claude remote-control shows up as a device card in the Code tab of the Claude app. One tap starts a session on it. Remote Control is also out of research preview.

For teams and CI

  • Restricted mode: claude --restricted starts without the tools that run commands or code, removes WebFetch, confines file tools to the working directories and refuses bypassPermissions. Built for evaluation harnesses on a shared machine.
  • Self-hosted environments: Claude Code cloud sessions can run on your organization's infrastructure, in public beta on Team and Enterprise.
  • No more 200-subagent cap per session: concurrency and depth limits still apply. 
claude --restricted -p "review src/ for SQL injection risks"

What's gone: Ultraplan

The Ultraplan research preview was removed during the week of August 3 to 7, including the /ultraplan command and the ultraplan keyword. Anthropic recommends plan mode or Claude Code on the web instead.

And the models?

Two new models arrived in Claude Code during the period: Fable 5.1 (with the fable alias, from v2.1.257) and Opus 5.5 (v2.1.280 minimum), which becomes the default on most plans. Details are in our dedicated articles.

Next steps