/resume — Pick up a past session

What it does

/resume picks up a previous Claude Code session without losing context or history. Every session you run—whether you explicitly close it or it times out—is stored locally with its full conversation thread. /resume shows you a list of available sessions, lets you choose one, and restores the full context so you and Claude continue exactly where you left off, including all prior messages, file edits, and decisions made together.

When to use it

Resume shines when you're working on something that takes multiple sittings. Start a feature on Monday, close the terminal, come back Thursday—instead of repeating "I'm working on X, here's the context," run /resume, pick your Monday session, and the model already knows the full history of what you built and why.

It's also cheaper and faster than re-priming. If you spent 30 minutes in a session exploring code, writing a plan, and getting halfway through a feature, closing and starting fresh means the model loses all that context. The next session burns tokens re-reading files and re-explaining the problem. /resume avoids that: the model has the conversation history, sees what was tried, remembers decisions, and jumps straight to the next step.

Resume is particularly useful when your work spans different files or involves architectural decisions. A refactor across five components, a database schema change with multiple migrations, or a complex debugging session—all of these are easier to continue when the full context is loaded.

Try it yourself

Open your terminal and run /resume to see a list of your past sessions with timestamps and brief context. Pick one that looks relevant—it'll show the working directory and a snippet of what you were doing. Once you select it, the full conversation restores and you're back in context. Try interrupting a multi-file change now (commit it if you like), close Claude Code, and /resume it later to confirm the history is there.

Gotchas

Sessions are stored locally on your machine—they don't sync across devices or accounts. If you switch to a different computer, that session list won't be there.

Old sessions stick around forever unless you explicitly clean them up. After months of work, your session list grows long. You can safely delete old sessions if you no longer need them, but once deleted, the history is gone—there's no archive.

The model's context window is still finite. If you're resuming a session that was already long, or you resume multiple times and keep adding to the same thread, you'll eventually hit the context limit. When that happens, the system compresses older messages to keep you in the conversation. The compression is transparent, but very old decisions might get less weight in the model's reasoning.

Sessions timeout if idle for a while (typically 24+ hours), but the conversation history is saved. Resuming after a timeout picks up the history but starts a fresh request to the model, which is fine for most work.