Claude Code v2.1.182 to v2.1.183 Major Updates

This is Ishikawa from the Cloud Business Division. Here is a summary of the Claude Code updates for v2.1.182 ~ v2.1.183 (2026-06-18 ~ 2026-06-19).

Update Summary

This covers 2 versions (v2.1.182 ~ v2.1.183, 2026-06-18 ~ 2026-06-19). v2.1.182 has no independent entry in the CHANGELOG (though it has been published to npm), and all documented changes are included in v2.1.183. The highlight of this release is enhanced safety in auto mode, with the addition of blocking for destructive git commands and infrastructure destruction commands, along with bug fixes centered around sub-agents, teammates, MCP, and TUI display.

I can't wait for Claude Fable 5 to resume.

https://x.com/wsj/status/2067276454803423467

Notable Updates

• New Feature: The attribution.sessionUrl setting has been added, allowing you to omit the claude.ai session link from commits and PRs (v2.1.183).
• Bug Fix: A bug where a turn would silently complete with no output when the model returned only a thinking block has been fixed (v2.1.183).
• Security: In auto mode, destructive git commands such as git reset --hard and infrastructure destruction commands such as terraform destroy will now be blocked unless explicitly requested (v2.1.183).

Target Versions and Period

Note that while v2.1.182 has been published on npm, there are no independent change entries in the CHANGELOG. It is considered a release/build-only version, and all changes introduced below are documented as part of v2.1.183.

New Features

• Added the attribution.sessionUrl setting. This allows you to omit the claude.ai session link attached to commits and PRs in Web sessions and Remote Control sessions (v2.1.183).
• Added /config --help. You can now view a list of shorthand keys available with /config key=value (v2.1.183).

Safety & Security

Safety in auto mode has been enhanced. These are important changes to prevent unintentional loss of your work.

• In auto mode, destructive git commands (git reset --hard, git checkout -- ., git clean -fd, git stash drop) will now be blocked if you have not requested the discarding of local work. Additionally, git commit --amend on commits not created by the agent in the current session will be blocked, and terraform destroy / pulumi destroy / cdk destroy will also be blocked unless the target stack has been explicitly requested (v2.1.183).
• Fixed behavior where scheduled task and webhook trigger deliveries were being treated as keyboard input. These are now classified as task notifications and can no longer approve pending actions or set session titles in auto mode (v2.1.183).

Improvements

• A warning is now displayed when the requested model is deprecated or has been automatically updated to a newer model. In print mode (-p), this is shown on stderr, and models specified in agent frontmatter are also included (v2.1.183).
• Changed the toggle behavior of /config. Both Enter and Space can now be used to change the selected setting, and Esc now "saves and closes" instead of the previous "discard changes" behavior (v2.1.183).
• Removed the "setup issues" line displayed below the logo at startup. Configuration issues can now be checked by running /doctor or using --debug (v2.1.183).

Fixes (Major)

Here is a selection of fixes related to stability and usability.

• Fixed a bug where turns silently completed with no output: Fixed an issue where a turn would end with no output when the model returned only a thinking block, causing Claude to re-prompt once (v2.1.183).
• Fixed 400 errors on sub-agent startup: Fixed thinking.disabled.display: Extra inputs are not permitted 400 errors that occurred during sub-agent startup and session title generation in certain configurations (v2.1.183).
• Fixed empty results from WebSearch in sub-agents: Fixed a bug where WebSearch returned empty results inside sub-agents (v2.1.183).
• Fixed fullscreen TUI corruption in Windows Terminal: Fixed display corruption that occurred when running many nested sub-agents, including status lines appearing mid-screen, duplicate spinner lines, and mixed text (v2.1.183).
• Fixed model exposure of MCP authentication stub tools: Fixed a bug where MCP servers requiring authentication exposed authentication stub tools to the model in headless/SDK mode (v2.1.183).
• Fixed tmux teammate pane startup failures: Fixed a bug where tmux teammate panes failed to start in environments with slow shell rc file initialization, and a bug where keys typed during agent startup leaked into new tmux panes (v2.1.183).
• Fixed forced termination of background tasks started by teammates: Fixed a bug where background tasks started by a teammate were forcibly terminated when that teammate's turn ended (v2.1.183).
• Additionally, numerous minor bugs have been fixed, including vim mode cursor position, duplicate user skill display in slash command completion, and hook timing line display in focus mode.

Closing

v2.1.182 ~ v2.1.183 was an update centered on enhanced safety through the blocking of destructive commands in auto mode. It also includes a wide range of bug fixes for sub-agents, teammates, MCP, and TUI display. This release is particularly impactful for those actively using auto mode or the sub-agent and teammate features. If any of the changes catch your interest, please update and check them out.

References

https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

https://code.claude.com/docs/en/changelog