Claude Code /recap: Instant Context Refresh When Returning to a Session
Claude CodeClaude Code v2.1.108 introduced a /recap command that generates a summary of what happened in a session, giving developers an immediate context refresh when they return to work they stepped away from. The feature is configurable in /config for automatic recaps on session re-entry, or can be invoked manually at any time with /recap. The same release also added an opt-in ENABLE_PROMPT_CACHING_1H environment variable to restore 1-hour prompt cache TTL for API key, Bedrock, Vertex, and Foundry users.
Sources & Mentions
3 external resources covering this update
/recap: No More Lost Context When Returning to a Session
Claude Code v2.1.108, released on April 14, 2026, introduced a /recap command that solves one of the most common frustrations in long-running development sessions: losing context when returning to a conversation after a break.
Running /recap generates a concise summary of the session — what was worked on, what decisions were made, what files were modified, and what remains in progress. Developers get an instant orientation without having to scroll back through the conversation history.
Manual and Automatic Modes
The feature ships with two usage patterns:
Manual invocation — Typing /recap at any point in a session generates a summary on demand. This is useful when picking up work after a meeting, returning the next morning, or simply wanting a progress checkpoint mid-session.
Automatic recap — In /config, developers can enable automatic recaps that fire whenever a session is re-entered after being idle. With this setting on, every time a developer opens a paused session, Claude Code immediately generates and displays the recap before any further interaction. This mode is particularly valuable for developers who maintain several active sessions across different projects.
ENABLE_PROMPT_CACHING_1H
V2.1.108 also introduced an opt-in environment variable, ENABLE_PROMPT_CACHING_1H, for developers who want to restore the 1-hour prompt cache TTL. This applies to users authenticating via API key, Amazon Bedrock, Google Vertex AI, and Foundry — platforms where the cache TTL had been adjusted in a previous release.
Setting ENABLE_PROMPT_CACHING_1H=true restores the longer TTL, which can materially reduce costs for long sessions where the same context is referenced repeatedly. The variable is opt-in, leaving the default behavior unchanged for users who do not set it.
Additional Changes in v2.1.108
/undo Alias
The release adds /undo as an alias for existing undo functionality, making the command more discoverable for developers familiar with standard text-editor conventions.
Skill Tool Auto-Invoke
Skill tool auto-invocation is now supported, allowing configured skills to trigger automatically when relevant conditions are met rather than requiring explicit invocation. This reduces friction in workflows where a skill should always run in response to a specific pattern.
/model Warning
The /model command now displays a warning when developers attempt to switch to a model that may not be fully compatible with the current session configuration. This prevents silent degradation where a model switch would succeed but produce unexpected behavior.