Claude Code v2.1.282: maxProseWidth Setting, Telemetry Settings Changes, and Reserved Skill Namespaces

Claude Code v2.1.282 is a large release whose main themes are settings and how they are trusted: a new maxProseWidth setting for wide terminals, stricter handling of telemetry variables in project settings and of managed settings, and reserved names for skills synced from claude.ai. It also fixes more cases of resumed sessions losing earlier reasoning, a batch of vim mode issues, and items across VS Code, cloud sessions and Claude Tag.

What’s New

`maxProseWidth` for wide terminals

The new maxProseWidth setting caps the width of Claude’s prose in wide terminals, while tables and code blocks keep the full width.

Project settings can no longer turn on telemetry export

Project and local settings now ignore OpenTelemetry variables that turn on export, set its endpoint or capture content, such as CLAUDE_CODE_ENABLE_TELEMETRY and OTEL_LOG_*. A new startup notice, along with new entries in /status and claude doctor, lists telemetry variables in a project’s settings files that were ignored or that turned telemetry off.

Server-side auto mode classifier when telemetry is off

On a direct Anthropic API connection with telemetry off, auto mode now uses the server-side classifier by default; CLAUDE_CODE_AUTO_MODE_SERVER=0 opts out.

Reserved `anthropic-skills` and `claude-ai` names

Skill(anthropic-skills:*) and Skill(claude-ai:*) allow rules now cover only skills synced from claude.ai, not plugins or other skills that merely use such a name. Skill folders, command files and workflow commands in the anthropic-skills or claude-ai namespace no longer load (a plugin with such a name still loads, but synced skills win name ties). MCP servers configured under the name anthropic-skills or claude-ai now list no skills or prompts, although their tools still work; rename the server in your MCP configuration to list them again.

Admin and managed settings

The new allowClaudeInChromeWithManagedMcp managed setting lets claude --chrome run alongside an exclusive managed-mcp.json, and the error shown when Chrome is blocked now names it. sandbox.excludedCommands now ignores project and local settings entries when managed settings or --settings set allowUnsandboxedCommands: false, or managed settings set allowManagedDomainsOnly: true. On Windows/WSL, an admin policy that is present but invalid or unreadable (HKLM, managed-settings.json) now keeps user-writable HKCU and WSL /etc/claude-code settings from applying.

Claude apps gateway readiness grace period

The Claude apps gateway adds store.readiness_grace_seconds, so /readyz can stay ready through a short Postgres outage such as a database failover.

Cloud sessions: GitHub App status and more repository options

Settings › Connectors › GitHub now shows whether the Claude GitHub App is installed and reachable for your account, with steps to connect, install or reconnect. You can attach a repository from a different GitHub owner, such as a fork’s upstream, to a running cloud session that already has one (including sessions started from Slack), and sessions on Git servers other than GitHub get “Open repository” and “Open compare page” links.

Fixes and Improvements

API errors: Fixed every request failing with a 400 error in conversations whose history holds web search results the API cannot decrypt, for example from a turn answered through a third-party gateway.
API errors: Fixed sessions failing on every turn with an “Invalid data in redacted_thinking block” API error; Claude Code now drops the conversation’s thinking blocks and retries once.
Compaction: Compaction no longer fails when the summarization request is refused; it now retries on a fallback model.
Resume: More cases of --continue and --resume re-sending earlier messages in a changed form are fixed, and earlier extended thinking is no longer dropped when /model, /rename, /artifacts or another immediate slash command is used while Claude is working. Very large sessions, including ones that were never compacted, also resume faster.
Models: Fixed a failed turn (“Effort ‘xhigh’ isn’t available with thinking turned off”) after a safety-related model switch in sessions with thinking off and effort above high.
Fable usage credits: An unanswered Fable usage-credits prompt no longer switches models in SDK-hosted sessions such as Claude Desktop (the turn ends instead), and when the plan needs usage credits that aren’t turned on yet, /model with a full Fable model id now opens the usage-credits prompt instead of stopping at an API error.
Login: Requests no longer fail for up to a minute with an “another Claude Code process is refreshing it” login error after that other process was closed or killed mid-refresh.
Permissions: Bash permission rules with a mid-pattern :* were skipped in settings files while --allowedTools honored them; they now work from every source, with a startup warning on how they match.
Managed settings: A mistyped value for a boolean lock key such as disableClaudeAiConnectors or allowManagedPermissionRulesOnly now still applies the lock, and startup names the key. One invalid nested value no longer causes managed permissions, autoMode, worktree or attribution settings to be ignored entirely.
Managed settings: Under managed allowManagedPermissionRulesOnly, repository, user and --add-dir skills, commands and skills-directory plugin manifests can no longer pre-approve their own tools via allowed-tools.
macOS: CLAUDE.md and rules are no longer read at startup through a repository symlink that reaches macOS’s /Network via .. or a /.vol-style kernel path.
Shell tools: Bash and PowerShell no longer hide a full disk quota behind “Exit code 1” or leave large output files in temp.
Input: Pasted multi-line text is no longer submitted line by line after the terminal’s bracketed paste mode was reset mid-session, and the send-now hint shows ctrl+x ctrl+s on terminals that send ctrl+enter as a newline (Windows Terminal before 1.25).
Vim mode: >> no longer indents empty lines, r with a count longer than the line no longer changes text, 2J no longer joins one line too many, the cursor lands on the first non-blank after dd, dj, dG or a whole-line p/P, and a count typed before . is now honored.
Plugins: Plugin uninstall no longer reports success and deletes the plugin’s saved options when its settings file still enables it or cannot be read, and it keeps saved options and secrets when the installed-plugins list cannot be read after the removal.
Vertex AI: Web search is now offered for models Claude Code doesn’t recognize yet, such as newly released ones.
VS Code: Long replies no longer fall behind the stream, because the panel no longer re-parses the whole reply on every update.
Claude Tag: Slack threads whose model has been retired now move to a working model instead of falling back on every reply, and newly connected Slack workspaces follow the current default model instead of keeping the default from when they were connected.

This release contains 86 changes in total; see the full changelog for the complete list.

How to Update

Check your installed version: claude --version
Update in place: claude update
If you configured an MCP server named anthropic-skills or claude-ai, rename it in your MCP configuration to list its skills and prompts again. Telemetry variables in project or local settings that turn on export are now ignored; the startup notice, /status and claude doctor list any that were.