Rewind

Rewind works a lot like loading a previous save point in a video game. When you have Claude Code (Anthropic's command-line coding agent) running a task for hours, there inevitably comes a moment when it veers off in the wrong direction. Trying to fix that by typing more instructions usually just wastes more time. Rewind lets you skip that entirely and jump straight back to the point where things were still going well.

It works like autosave. Every time you enter a prompt, a checkpoint is created capturing the code and conversation at that moment. When something goes wrong, you can pick a point from that list and revert to it. You can choose to roll back just the code, just the conversation history, or both, depending on what you need.

For example, say you spent a while experimenting with environment setup before getting into the actual implementation. Once the real work is done, you might want to clear out all that messy setup back-and-forth and keep only the conversation about the implementation itself — that's exactly the kind of situation Rewind is built for. After rewinding, you can also ask Claude to summarize what came after (or before) the checkpoint and continue from there.

As described in coverage: "Rewind lets you jump straight back to the last checkpoint instead of trying to steer Claude back on track with more prompts when it goes off in the wrong direction." A common misunderstanding is that this isn't just a simple 'undo' button — it's a granular feature that lets you choose whether to roll back the code, the conversation, or both.

In Claude Code, press Esc twice with an empty prompt to open the Rewind menu. Since a checkpoint is created every time you enter a prompt, it's worth opening the menu early in a long session to see what's been saved. After rewinding, try something like this:

Summarize only the conversation after this checkpoint and continue from there. Leave the code as is and just roll back the conversation history.