Claude Code 2.1.108: Prompt Caching, Recap, Slash Commands

Claude Code 2.1.108 adds 1-hour prompt caching, session recap feature, and slash command discovery. Major fixes for paste, diacritical marks, and session resumption.

Claude Code 2.1.108: Prompt Caching, Recap, Slash Commands

TL;DR

  • Prompt caching now supports 1-hour TTL via ENABLE_PROMPT_CACHING_1H env var across all providers
  • New recap feature auto-summarizes context when you return to a session
  • Model can now discover and invoke slash commands like /init, /review, /security-review via Skill tool
  • Significant bug fixes for paste in login, diacritical marks, and session resumption

New

  • 1-hour prompt cache TTL — Set ENABLE_PROMPT_CACHING_1H to extend cache lifetime on API key, Bedrock, Vertex, and Foundry; ENABLE_PROMPT_CACHING_1H_BEDROCK is deprecated but still honored
  • Recap feature — Automatically provides context when returning to a session; configure in /config or invoke manually with /recap; force with CLAUDE_CODE_ENABLE_AWAY_SUMMARY if telemetry disabled
  • Slash command discovery — Model can now discover and invoke built-in slash commands like /init, /review, and /security-review via the Skill tool
  • /undo alias/undo now works as an alias for /rewind
  • Improved /model switching — Warns before switching models mid-conversation since the next response re-reads the full history uncached
  • Better /resume picker — Defaults to sessions from the current directory; press Ctrl+A to show all projects
  • Smarter error messages — Server rate limits now distinguished from plan usage limits; 5xx/529 errors link to status.claude.com; unknown slash commands suggest the closest match
  • On-demand grammar loading — Reduced memory footprint for file reads, edits, and syntax highlighting by loading language grammars on demand
  • Verbose indicator — Added indicator when viewing detailed transcript with Ctrl+O
  • Prompt caching warning — Startup warning when prompt caching is disabled via DISABLE_PROMPT_CACHING* environment variables

Fixed

  • Paste not working in /login code prompt (regression in 2.1.105)
  • Subscribers with DISABLE_TELEMETRY falling back to 5-minute cache TTL instead of 1 hour
  • Agent tool prompting for permission in auto mode when safety classifier's transcript exceeded context window
  • Bash tool producing no output when CLAUDE_ENV_FILE ends with a # comment line
  • claude --resume <session-id> losing the session's custom name and color set via /rename
  • Session titles showing placeholder example text when first message is a short greeting
  • Terminal escape codes appearing as garbage text in prompt input after --teleport
  • /feedback retry: pressing Enter to resubmit after failure now works without editing the description
  • --teleport and --resume <id> precondition errors exiting silently instead of showing error message
  • Remote Control session titles set in web UI being overwritten by auto-generated titles after third message
  • --resume truncating sessions when transcript contained a self-referencing message
  • Transcript write failures (e.g., disk full) being silently dropped instead of logged
  • Diacritical marks (accents, umlauts, cedillas) being dropped from responses when language setting is configured
  • Policy-managed plugins never auto-updating when running from a different project than where they were first installed

Update Claude Code: npm install -g @anthropic-ai/claude-code or check your IDE extension marketplace for the latest version.

Source: Claude Code