Tell Claude to Stash Your Work Before Trying a Risky Approach So You Can Revert Safely

Some of the best prompts start with "try a completely different approach" — but you don't want to lose your current working state if the experiment doesn't pan out. Tell Claude to save your work before experimenting.

> stash my current changes, then try rewriting the payment
> processor to use the event sourcing pattern instead

Claude runs git stash, tries the new approach, and if it doesn't work out:

> this isn't going to work — restore my stashed changes

You're back to exactly where you were, no damage done.

For bigger experiments, use a branch instead:

> create a branch called "experiment/new-auth-flow", then try
> replacing the session-based auth with JWT tokens

Now you can compare both approaches side by side, run benchmarks, and decide which to keep — without ever risking your working code.

This pattern is especially valuable for:

• Architecture experiments — "try using a service layer instead of fat controllers"
• Algorithm alternatives — "try a different sorting approach and see if it's faster"
• Risky refactors — "try extracting this into a separate package"
• Library swaps — "try replacing this with a different library and see if the tests pass"

You can even ask Claude to try multiple approaches:

> stash the current work. try approach A: use a queue for
> email sending. if the tests pass, great. if not, restore
> the stash and try approach B: use a synchronous job with
> retry logic instead.

Claude tries A, runs the tests, and if they fail, restores and tries B — systematically exploring options without you managing the git state manually.

Experimentation is free when you can always go back. Stash first, experiment boldly.

via Claude Code