Claude Code: Auto-Memory, /copy Command & Session Reliability

Auto-Memory

Claude Code v2.1.59 introduces auto-memory, a feature that allows Claude to automatically save useful context to persistent memory as it works. This means Claude can remember patterns, preferences, and project-specific knowledge across sessions without requiring manual setup. Users can manage auto-memory through the /memory command, which provides control over what Claude retains.

/copy Command

A new /copy command has been added that shows an interactive picker when code blocks are present in Claude's response. Users can select individual code blocks to copy to their clipboard or copy the full response, making it significantly easier to extract specific code snippets from conversations.

Smart Bash Command Prefixes

The "always allow" prefix suggestions for compound bash commands have been improved. Previously, compound commands like cd /tmp && git fetch && git push were treated as a single unit for prefix matching. Now Claude Code computes smarter per-subcommand prefixes, resulting in more granular and accurate permission suggestions.

Memory Efficiency in Multi-Agent Sessions

Memory usage in multi-agent sessions has been improved by releasing completed subagent task state once it is no longer needed. This reduces the memory footprint of long-running sessions that spawn multiple subagents.

Bug Fixes

• Fixed an MCP OAuth token refresh race condition when running multiple Claude Code instances simultaneously, which could cause authentication failures.
• Fixed shell commands not showing a clear error message when the working directory has been deleted.
• Improved ordering of short task lists.