Claude Code Ships Four Updates in Four Days: Concise Mode, Default Model Env Var, and GitLab Badges

Anthropic shipped four Claude Code releases between August 17 and August 20, each adding something useful without much fanfare. The biggest addition lands in 2.1.237: a built-in Concise output style that cuts preamble without cutting corners. ## Concise output style (v2.1.237, August 20) The new Concise mode changes how Claude Code formats responses. When enabled, Claude leads with the result, skips the narration and preamble, and still does the full work underneath. It does not reduce what gets done; it reduces what gets said about what's being done. You turn it on via `/config` under "Output style." It sits alongside the existing Default and Verbose options. The same release fixes prompt caching for sessions that route through an LLM gateway or a custom base URL, which had been broken. ## ANTHROPIC_DEFAULT_MODEL and idle notifications (v2.1.236, August 19) Version 2.1.236 adds the `ANTHROPIC_DEFAULT_MODEL` environment variable. Unlike `ANTHROPIC_MODEL` (which locks the model), this one sets the starting model for new sessions only. A `/model` pick inside a session still overrides it and persists across restarts. It is useful for teams or CI setups that want a consistent default without preventing per-session overrides. The same version introduces `notify_when_idle` for cross-session `SendMessage` on macOS and Linux. You can ask another local Claude Code session to send one notification when it next goes idle. It is opt-in and one-shot, not a polling loop. macOS sandboxing also gets tighter: wildcard read-deny rules like `**/.env` now take priority inside allowed read regions, cover matched directories' contents, and cannot be bypassed by renaming the denied file. Other fixes in this version include a long-standing bug where removing a directory a session had `cd`'d into would break clipboard copy, background sessions, and local MCP logs (present since 2.1.229). The fullscreen renderer no longer exits permanently after a single failed start; it falls back to the classic renderer. The `/model` picker no longer renders taller than the terminal window. ## Spellcheck in the prompt input (v2.1.235, August 18) Version 2.1.235 adds optional spellcheck to the prompt text field. When enabled, it underlines misspelled words as you type using your locally installed `aspell`, `hunspell`, or `ispell`. It is off by default. ## GitLab MR badge and auto-continue at usage limits (v2.1.234, August 17) Version 2.1.234 adds a GitLab merge request badge to the footer and statusline. Repos with a GitLab remote and an authenticated `glab` CLI now show the current MR number with draft, pending, or green status. The feature mirrors the GitHub PR badge that has been in Claude Code for a while. The same release adds auto-continue behavior when a claude.ai usage limit resets. When a session is paused at a limit, Claude Code will automatically pick back up once the reset happens. You can disable this in `/config` ("Continue automatically at usage limit"). There is also a new `CLAUDE_CODE_PROJECT_DIR_NAME` environment variable for hosts that give each session its own config directory. Setting it to a short name changes what the per-project transcript directory is called. A `selection:clear` keybinding action is new, making it possible to bind a key to clear an in-progress text selection in the main input or the agents view. --- The changelog for this stretch is at [code.claude.com/docs/en/changelog](https://code.claude.com/docs/en/changelog). The DevelopersIO writeup on 2.1.236-237 has additional context at [dev.classmethod.jp](https://dev.classmethod.jp/en/articles/20260820-cc-updates-v2-1-237/).