What changed
Claude Code 2.1.183 shipped on June 19, 2026. Auto mode gains safety guards that block destructive commands you didn’t ask for, alongside new settings for model deprecation warnings, attribution.sessionUrl, and /config --help. On top of that, a batch of bugs across subagents, teammates, MCP, and terminal rendering gets cleaned up.
New features
-
Warning for deprecated or auto-updated models
There was no notice when the requested model was deprecated or automatically updated to a newer one. A warning is now shown on stderr in print mode (
-p) and now also covers models set in agent frontmatter. -
attribution.sessionUrlsettingCommits and PRs from web and Remote Control sessions always carried the claude.ai session link. The new
attribution.sessionUrlsetting lets you omit that link. -
/config --helpIt was hard to see which shorthand keys
/config key=valueaccepts./config --helplists all available shorthand keys.
Key improvements
-
Improved auto mode safety
Auto mode could run operations you hadn’t asked for, risking the loss of local work. Destructive git commands (
git reset --hard,git checkout -- .,git clean -fd,git stash drop) are now blocked when you didn’t ask to discard local work,git commit --amendis blocked when the commit wasn’t made by the agent this session, andterraform destroy/pulumi destroy/cdk destroyare blocked unless you asked for the specific stack. -
Changed
/configtoggle behaviorThe keys for changing a setting in
/configweren’t intuitive. Enter and Space both change the selected setting now, and Esc saves and closes instead of reverting. -
Removed the startup “setup issues” line
The “setup issues” line under the logo always showed and cluttered startup. It’s gone — run
/doctorto see configuration issues or use--debug.
Bug fixes
Sessions & output
thinking.disabled.display400 errors on subagent spawns — affected configurations failed subagent spawns and session-title generation withthinking.disabled.display: Extra inputs are not permitted. Fixed.- Turns silently completing with no visible output — when the model returned only a thinking block. Claude now re-prompts once.
- WebSearch returning empty results in subagents — Fixed.
Subagents & teammates
- tmux teammate panes failing to launch — when the shell has slow rc-file initialization, and keystrokes typed during agent spawn leaking into the new tmux pane instead of the leader prompt. Fixed.
- Background tasks started by a teammate being killed — when the teammate finishes a turn. Fixed.
- Scheduled task and webhook trigger deliveries being treated as keyboard input — they now classify as task notifications and can no longer approve a pending action or set the session title in auto mode.
MCP
- MCP servers requiring authentication exposing auth-stub tools to the model — in headless/SDK mode. Fixed.
Display & rendering
- Terminal cursor stranded above the prompt — after navigating history in vim mode with the native cursor enabled. Fixed.
- Fullscreen TUI corruption in Windows Terminal — statusline mid-screen, duplicated spinner rows, merged text under heavy nested-subagent load. Fixed.
- User-level skills appearing multiple times in slash-command autocomplete — when multiple plugins are enabled. Fixed.
- Focus mode showing “Ran N PostToolUse hooks” timing lines — under each response. Fixed.
Notes
- Auto mode blocks destructive commands — destructive git commands and
terraform destroy/pulumi destroy/cdk destroyyou didn’t explicitly ask for are blocked in auto mode.git commit --amendis allowed only for commits the agent made this session. - Model warnings show on stderr in print mode — deprecation and auto-update warnings appear on stderr in print mode (
-p), and now cover models set in agent frontmatter. - Omit the session link from commits and PRs — the
attribution.sessionUrlsetting drops the claude.ai session link from commits and PRs in web and Remote Control sessions. - Check
/configshorthand keys — run/config --helpto list all shorthand keys for/config key=value.