Claude Code v2.1.183: Destructive Git Blocked, TUI Fixed

Claude Code v2.1.183 ships with critical auto-mode safety improvements, several terminal UI fixes, and new configuration options. If you're using Claude Code in auto mode or on Windows Terminal, this release directly affects your workflow.

Auto Mode Safety Enhancements

Destructive git commands are now blocked unless you explicitly ask to discard local work. The following commands are intercepted:

git reset --hard
git checkout -- .
git clean -fd
git stash drop

git commit --amend is blocked if the commit wasn't made by the agent in the current session. Infrastructure destroy commands are also guarded: terraform destroy, pulumi destroy, and cdk destroy are blocked unless you explicitly ask for the specific stack.

Model Deprecation Warnings

A warning is now printed on stderr when the requested model is deprecated or automatically updated to a newer model. In print mode (-p), this warning appears on stderr. It also covers models set in agent frontmatter.

New Config Options and Behavior Changes

Added attribution.sessionUrl setting to omit the claude.ai session link from commits and PRs in web and Remote Control sessions. Use /config --help to list all available shorthand keys for /config key=value. The /config toggle behavior changed: Enter and Space both change the selected setting, and Esc now saves and closes instead of reverting.

The startup "setup issues" line under the logo has been removed. Run /doctor to see configuration issues or use --debug.

Bug Fixes

Subagent spawns: Fixed thinking.disabled.display: Extra inputs are not permitted 400 errors on subagent spawns and session-title generation.
WebSearch: Fixed returning empty results in subagents.
Terminal cursor: Fixed cursor being stranded above the prompt after navigating history in vim mode with native cursor enabled.
Windows Terminal TUI: Fixed fullscreen TUI corruption (statusline mid-screen, duplicated spinner rows, merged text) under heavy nested-subagent load.
Silent turns: Fixed turns silently completing with no visible output when the model returned only a thinking block; Claude now re-prompts once.
Skills autocomplete: Fixed user-level skills appearing multiple times in slash-command autocomplete when multiple plugins are enabled.
MCP servers: Fixed servers requiring authentication exposing auth-stub tools to the model in headless/SDK mode.
tmux teammates: Fixed pane launch failure with slow rc-file initialization, and keystrokes typed during spawn leaking into the new pane.
Background tasks: Fixed tasks started by a teammate being killed when the teammate finishes a turn.
Scheduled tasks/webhooks: Fixed deliveries being treated as keyboard input; they now classify as task notifications and can no longer approve pending actions or set session titles in auto mode.
Focus mode: Fixed showing "Ran N PostToolUse hooks" timing lines under each response.