I Get More Out of Claude Code by Leaving

I just stumbled on a cool Claude Code unlock by telling it I'm leaving.

Most of the time, I work pretty closely with Claude. But sometimes, especially on a personal project, I want it to push forward more aggressively. Yesterday I decided to go for a walk so I just told it:

If there is a big decision you don't have a strong inclination on yourself, defer that until we talk again and tackle the more obvious things first. I'm stepping out for a bit and will be back in a while to regroup.

I came back to eight commits. It had fixed a bug I didn't know was there, added 56 tests to a project that had exactly zero, and broken a 1,900-line file into six. Oh and it did run into a couple questions along the way that it checked in with me about at the end to wrap things up.

I decided to put a skill together because I'm definitely going to be doing that again. I called it stepping-out and have it:

• Work the task most impactful to least, and keep moving instead of stalling on the first fork.
• Make the call on anything low-stakes, reversible, or clearly the right default. Note it so I can veto it later.
• Defer the big, hard-to-reverse, or taste-driven decisions. Don't guess, don't block, just do the obvious things around them and set them aside.
• Never do anything destructive or irreversible on its own. That's always a defer.
• Come back with a regroup: what got done, what defaults it took, and a "your call" list of what actually needed me.

That last part is the real deliverable. The point of the mode isn't that Claude does everything while I'm gone. It's that it does the obvious 80% well, and hands me a clean, trustworthy list of the 20% that's genuinely mine to decide.

I'm still figuring out where the edges are. I wouldn't hand it something load-bearing in production this way yet, and the regroup is only as good as its judgment about what counts as a "big decision." But for a side project on a Saturday afternoon, it was great.