Claude Code v2.1.183: Enhanced Safety and UX Improvements
This release of Claude Code includes significant safety enhancements, particularly around destructive Git and infrastructure commands, and improves the user experience with clearer model warnings and streamlined configuration. It addresses several bugs impacting subagent performance, terminal interaction, and task handling. These updates are primarily for developers and engineers interacting with Claude Code for coding and infrastructure tasks.
- →Improved auto mode safety for destructive commands
- →Added warning for deprecated or updated models
- →New setting to omit session links from commits and PRs
- →Updated `/config` command for easier management
- →Removed startup "setup issues" line, added `/doctor` command
Enhancements (4) ›
- Improved auto mode safety for destructive commands
Destructive Git commands like `git reset --hard`, `git checkout -- .`, and `git clean -fd` are now blocked when not explicitly requested. `git commit --amend` is also restricted when the commit was not made by the agent in the current session. Similarly, infrastructure commands like `terraform destroy` and `pulumi destroy` are blocked unless the specific stack was requested.
- Added warning for deprecated or updated models
A warning is now displayed when the requested model is deprecated or automatically updated to a newer version. This warning appears on stderr in print mode (`-p`) and now also covers models specified in agent frontmatter.
- New setting to omit session links from commits and PRs
The `attribution.sessionUrl` setting has been added, allowing users to omit the claude.ai session link from commits and pull requests in web and Remote Control sessions.
- Updated `/config` command for easier management
The `/config --help` command now lists all available shorthand keys for `/config key=value`. The behavior of Enter and Space keys in `/config` has been changed to modify the selected setting, and the Esc key now saves and closes instead of reverting.
Fixes (1) ›
- Fixed various bugs affecting subagents, terminal, and task handling
Several bugs have been fixed, including `thinking.disabled.display` errors on subagent spawns, empty results from WebSearch in subagents, terminal cursor issues in vim mode, fullscreen TUI corruption in Windows Terminal, silently completing turns with no output, duplicate skills in autocomplete, authentication issues with MCP servers, tmux teammate pane failures, background task termination, and incorrect classification of scheduled task/webhook deliveries.
Maintenance (1) ›
- Removed startup "setup issues" line, added `/doctor` command
The "setup issues" line previously shown under the logo at startup has been removed. Users can now run `/doctor` to view configuration issues or use the `--debug` flag for troubleshooting.
https://github.com/anthropics/claude-code/releases/tag/v2.1.183