Claude Code v2.1.116 adds performance improvements and safety guardrails for large sessions

Anthropic released Claude Code version 2.1.116 on April 20, introducing performance and stability improvements across session handling, terminal interaction, and agent operations. The most significant change addresses resumption speed: sessions larger than 40MB now resume up to 67% faster, with optimized processing of sessions containing many dead-fork entries. This directly improves user experience when working with long conversation histories.

The release enhances user interface responsiveness in multiple areas. Terminal scrolling is now smoother across VS Code, Cursor, and Windsurf implementations, with a revised /terminal-setup command that calibrates the editor's scroll sensitivity. The thinking spinner—which displays Claude's reasoning process—now shows granular progress states (still thinking, thinking more, almost done thinking) inline rather than as a separate row.

A security improvement tightens the sandbox auto-allow feature to prevent circumventing dangerous-path safety checks when using rm or rmdir commands targeting critical system directories including root, home, and other protected locations. This prevents agents from accidentally or maliciously deleting essential files.

The update includes numerous bug fixes affecting keyboard navigation, character rendering, and API handling. Fixed issues include broken column alignment in Devanagari and other Indic scripts, keyboard protocol incompatibilities in terminals like WezTerm and Ghostty, and an intermittent API 400 error related to cache control TTL ordering during parallel requests.