v2.1.152
Claude CodeSummary
This release introduces powerful new capabilities for code review, allowing fixes to be applied directly to the working tree, and enhances skill management with the ability to restrict tools and reload skills dynamically. Several quality-of-life improvements were made, including better handling of model fallbacks, clearer thinking indicators, and significant fixes to terminal rendering and plugin interaction stability.
New Features
- Added the ability for the `/code-review --fix` command to apply review findings directly to the working tree, offering reuse, simplification, and efficiency suggestions.
- Enabled skills and slash commands to specify `disallowed-tools` in frontmatter to restrict available tools during skill execution.
- Introduced the `/reload-skills` command to re-scan skill directories without requiring a session restart.
- Enabled `SessionStart` hooks to return `reloadSkills: true` to make newly installed skills available within the same session.
- Allowed `SessionStart` hooks to set the session title using `hookSpecificOutput.sessionTitle` upon startup or resume.
- Added a `MessageDisplay` hook event allowing hooks to transform or hide assistant message text before it is displayed to the user.
- Added the managed setting `pluginSuggestionMarketplaces`, allowing admins to allowlist specific organization marketplaces for context-aware plugin suggestions.
- Enabled the `/usage` breakdown to include large session files, utilizing streaming reads to maintain flat memory usage during scanning.
- Vim mode now opens reverse history search (`Ctrl+R` equivalent) when pressing `/` in NORMAL mode.
Bug Fixes
- Fixed terminal styling degradation in very long sessions by recycling the renderer's style pool.
- Fixed the sandbox-enabled warning failing to appear in condensed startup mode; it now shows in all layouts.
- Fixed the loading spinner incorrectly showing "still thinking"/"almost done thinking" while a tool was running, and now resets the thinking status after each tool execution.
- Fixed focus mode displaying a spurious "N messages hidden" count when no hidden activity occurred.
- Fixed clicking a link inside an expanded tool result incorrectly collapsing the section instead of opening the link.
- Fixed various markdown rendering issues including table cell borders inheriting inline code color, wrapped continuation lines losing style, and empty header cells showing an incorrect label in narrow-terminal stacked layout.
- Fixed plugin MCP servers with identical commands but different environment variables being incorrectly deduplicated.
- Fixed `/doctor` reporting marketplace or plugin not found errors due to stale `enabledPlugins` entries referencing removed resources.
- Fixed plugins tracking a git branch silently failing to receive updates after the plugin registry was rebuilt.
- Fixed remote MCP servers failing to connect in Claude Code Remote sessions when the egress proxy was enabled.
- Fixed the effort-change confirmation dialog appearing unnecessarily when the conversation was empty or when switching between equivalent effort levels.
- Fixed the Agent tool description incorrectly referencing an unavailable agent list when running with `--bare` or attachments disabled.
- Fixed a background worker crash in `claude agents` when accepting a stale permission prompt after a subagent cancellation.
- Fixed `cache_creation_input_tokens` reporting as 0 in usage statistics when cache writes were reported only via the nested `cache_creation` breakdown.
- Fixed the PushNotification tool incorrectly reporting failure when Remote Control was enabled in SDK-hosted sessions.
- Fixed sessions getting stuck after model or login switches by proactively stripping stale thinking-block signatures from history.
Improvements
- The `/simplify` command now invokes `/code-review --fix`.
- Claude Code now defaults to the configured `--fallback-model` for the remainder of the session if the primary model is unavailable, preventing request failures.
- Auto mode no longer requires explicit opt-in consent.
- Thinking summaries in collapsed groups now remain readable for a minimum of 3 seconds, render as markdown, and are capped at 10 lines.
- In fullscreen mode, the thinking indicator now counts up live and retains its final value if the user interrupts the thought process.
- Simplified the Workflow tool's inline progress display; live agent counts are now only shown in the persistent status row below the prompt.
- The post-response timer now displays a cumulative time once background agents/workflows finish when they are still running.
- Added the session entrypoint as an OpenTelemetry metric attribute (opt-in).