[Release] anthropics/claude-code v2.1.187: v2.1.187

Claude Code v2.1.187 introduces security enhancements with sandbox credential blocking and org-configured model restrictions, while fixing critical issues with structured output handling, MCP tool timeouts, remote sessions, and UI responsiveness. The release improves user experience with mouse click support in select menus, better navigation controls, and enhanced GitHub App installation workflow. Multiple stability fixes address agent management, channel connections, and terminal text handling across different platforms.

Key Points

• •Added `sandbox.credentials` setting to prevent sandboxed commands from accessing credential files and secret environment variables
• •Implemented org-configured model restrictions with visual indicators in model picker and CLI flags (`--model`, `/model`, `ANTHROPIC_MODEL`)
• •Fixed structured output infinite loop: model can no longer re-call `StructuredOutput` after successful completion, ensuring reliable follow-up turns
• •Resolved remote MCP tool calls hanging indefinitely by implementing 5-minute timeout with error abort (configurable via `CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT`)
• •Fixed `--resume` failures when original `-p` run produced no model turns, improving session recovery
• •Added mouse click support for select menus in fullscreen mode, enhancing UI interaction across permission prompts and configuration dialogs
• •Improved `/install-github-app` workflow by making GitHub Actions setup optional while keeping GitHub App installation mandatory
• •Fixed remote session startup delay (~2.7s) caused by agent proxy CA system-trust installation
• •Resolved channel connection drops after navigating agents view, using `/bg`, `/tui`, or `/update` commands
• •Fixed subagent depth tracking for resumed and forked subagents, ensuring proper spawn depth restoration and depth cap enforcement

Workflow Diagram

Concepts