Claude Code Changelog - Release Notes & Version History

Claude Code v2.1.220 includes 1 change. Use search to find specific features or subscribe to RSS for updates.

Native installations auto-update in the background. To update manually, run claude update. For Homebrew: brew upgrade claude-code. For WinGet: winget upgrade Anthropic.ClaudeCode. npm installation is deprecated — migrate with curl -fsSL https://claude.ai/install.sh | bash.

At github.com/anthropics/claude-code. This site adds search, filters, and RSS.

Subscribe to the RSS feed or watch the GitHub repo.

Claude Code creates temporary files during planning operations. To fix: set CLAUDE_CODE_TMPDIR=/tmp/claude in your environment, or add tmpclaude-* and nul to your .gitignore.

Stop Hooks run when Claude finishes responding and can prevent it from stopping (e.g., requiring tests to pass first). Memory files (CLAUDE.md) provide persistent project context across sessions. Configure hooks in your settings.json file, and view them with /hooks.

Claude Code primarily uses two memory systems: CLAUDE.md files (user-written) and auto memory (Claude-written, stored in ~/.claude/projects/<project>/memory/). You can create and edit CLAUDE.md manually, or Claude will suggest updates after understanding your project. While other .md files might appear, CLAUDE.md is the main memory file. Keep it in your project root and commit it to version control to maintain context across sessions.

Claude Code uses OAuth tokens that expire for security reasons. To minimize re-authentication, ensure you're using the latest version and check if your system clock is accurate.

This is normal behavior - the Claude Code CLI displays all subscription types including Max plans as 'API Usage Billing' in the interface. This doesn't affect your actual plan benefits or usage limits. Your Max plan features remain fully active regardless of this display label in the CLI.

When Claude Code shows 'Resume without any available conversations', try pressing Ctrl+C twice to exit. If that doesn't work, force quit the terminal session. This issue occurs when there's no conversation history to resume. To prevent it, ensure you're properly exiting Claude Code using the /exit command or Ctrl+C in future sessions.

To reduce token usage, use .gitignore patterns to exclude large files and directories from Claude Code's context. Create focused CLAUDE.md files with only essential project information — avoid verbose docs or redundant instructions. Use specific file paths instead of broad commands when possible. Start new sessions for unrelated tasks to keep context small. Using /compact will reduce current session context size.

Claude Code has built-in rate limiting. When you hit a limit, the tool will automatically pause and resume when the limit resets. You don't need to build custom tools - just wait for the automatic retry. If you frequently hit limits, consider changing your plan or adjusting your usage patterns, or using /extra-usage - however this will incur extra costs/charges.

Use CLAUDE.md files to provide project context - place them in your project root, home directory, or .claude/ folder. Define custom skills and slash commands (like /commit, /review-pr) for repetitive tasks. For external tool integration, set up MCP (Model Context Protocol) servers. Use /config to review settings and organize your project configuration for consistent behavior across your team.

You can change the model using /model <alias|name> during a session or claude --model <alias|name> at startup. For effort level, use /effort low|medium|high|max. Effort is supported on Opus 4.6 and Sonnet 4.6; max is Opus-only.

Claude Code's performance can degrade with excessive context usage. To optimize: 1) Check your CLAUDE.md file(s) for overly large project contexts, 2) Disable unused MCP servers or hooks, 3) Clear saved memories in ~/.claude/ if they become bloated. Start with a minimal configuration and add features only as needed.