nonstop

nonstop

    "Why do you write like you're running out of time?"

    Go to sleep. Wake up to finished work.

  Install ·
  How it works ·
  Compared to

Your overnight Claude session, except it actually finishes. Type /nonstop before you walk away. Claude asks every question now, gets your sign-off on anything dangerous, then works through the night: solving blockers, routing around problems, and knowing when to stop trying.

The Problem

You hand Claude a 4-hour task and go to bed. You wake up and it stopped 3 minutes in:
*"Should I use the existing auth middleware or write a new one?"*

Or it hit a permission error on step 2 of 47 and just... sat there. Waiting. For 8 hours.

nonstop makes Claude do what a good engineer does when the boss leaves: figure it out, work around it, or write it down and move on to the next thing.

Install

Tell Claude (easiest)

Paste this into Claude Code:

Plugin marketplace

Manual

How It Works

Before you leave: the pre-flight

/nonstop triggers a 4-phase sequence. Claude won't start working until all of this is done.

1. Mental simulation. Claude walks through every step of your task in its head and flags anything that would make it stop:

2. Dangerous ops manifest. Anything that touches the outside world gets explicit approval:

You can approve all, deny all, or scope it ("delete in /tmp, not /data"). Your call.

3. Permissions check. Claude checks your permission mode. If you're not on bypass or auto, it'll warn you that permission prompts will block it while you're gone.

4. Confirm and go. Scope, workaround policy, ground rules. Then:

*"Nonstop mode ON. Go rest, I've got this."*

While you're away: the decision framework

A stop hook blocks Claude from halting. When something goes wrong:

What Claude will never…