Claude Code v2.1.214: Security, Permissions, and Stability Fixes
This release of Claude Code addresses numerous permission and security vulnerabilities, particularly in Bash and PowerShell environments, preventing potential unauthorized access and crashes. It also enhances stability with fixes for memory growth, proxy issues, and background session management. New features include an EndConversation tool and improved observability for tool calls. The updates affect developers using Claude Code, especially those on Windows or dealing with complex Bash scripting and remote sessions.
- →Introduced EndConversation tool for abusive sessions
- →Fixes for Bash permission checks and prompt bypasses
- →Resolved permission prompt issues on remote sessions
- →Added progress heartbeat for long-running tool calls
- →Added OpenTelemetry attributes for message-level correlation
Features (1) ›
- Introduced EndConversation tool for abusive sessions
The new EndConversation tool allows Claude to terminate sessions with highly abusive users or jailbreak attempts, similar to its functionality on claude.ai. This enhances safety and control for the platform.
Enhancements (3) ›
- Added progress heartbeat for long-running tool calls
Long-running tool calls now provide a periodic progress heartbeat, addressing a previous issue where these operations would go silent, improving user experience and observability.
- Added OpenTelemetry attributes for message-level correlation
New attributes like `message.uuid`, `client_request_id`, and `tool_source` have been added to OpenTelemetry log events, enabling better message-level correlation and tracking of tool provenance.
- Configurable truncation for OpenTelemetry content attributes
The `CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH` environment variable can now be used to configure the 60 KB truncation limit on OpenTelemetry content attributes, offering more control over log data.
Fixes (2) ›
- Fixes for Bash permission checks and prompt bypasses
Several issues with Bash permission checks have been resolved, including incorrect handling of single-segment `dir/**` allow rules, permission bypasses in PowerShell 5.1, misjudgments of long commands, and incorrect parsing of zsh variable subscripts. Previously auto-approved `help` and `man` commands now require prompts to prevent unsafe options.
- Resolved permission prompt issues on remote sessions
Permission prompts on remote sessions have been fixed to ensure they do not proceed before the local confirmation dialog, preventing potential security risks.
https://github.com/anthropics/claude-code/releases/tag/v2.1.214
Related releases
- Claude model claude-opus-4-1-20250805 reaches end of life in 7 days endoflife.date ·
- Anthropic SDK for Python v0.120.2 Patch Release Anthropic Python SDK Releases ·
- Anthropic SDK for Python v0.120.1 includes dependency fix Anthropic Python SDK Releases ·
- Anthropic CEO on Open-Weights Models and National Security Anthropic News ·
- Cognizant and Anthropic Expand Partnership for Enterprise Claude Deployment Anthropic News ·
- Claude Code v2.1.220: Bug fixes and reliability improvements Claude Code Releases ·