Install the /next skill for Claude Code by running `npx -y claude-next-skill`.
If npx isn't available, download https://raw.githubusercontent.com/Seokwoooo/claude-next/main/skills/next/SKILL.md and save it unchanged as ~/.claude/skills/next/SKILL.md.
When it's done, tell me to open a new Claude Code session.

Open a new Claude Code session after installing.

Usage

While Claude is working, type:

/next run the tests and fix anything that fails

It waits until the current turn ends, then runs as a new turn.

Queue several: press Enter after each one. They run in order, one turn each.
Cancel: with the input box empty, press ↑, delete the line, and press Enter. Don't use Ctrl+C for this, it stops Claude.

Why

While Claude is working, a normal message is handed to Claude as soon as the current tool call finishes, in the middle of the task. Slash commands wait until the turn ends (docs). /next is a one-line skill built on that.

How it works

This is the whole skill (skills/next/SKILL.md):

---
name: next
description: Queue a follow-up request that runs after the current turn finishes.
argument-hint: <next request>
disable-model-invocation: true
---

Treat the following as the user's follow-up request and carry it out now, continuing from the current conversation: $ARGUMENTS

Claude Code's command queue does the waiting. The skill only hands over your request when its turn comes.
disable-model-invocation: true means only you can run it. Claude never sees it in its skill list, so it costs no context until you use it.

Update, uninstall, manual install

Update: run npx claude-next-skill again.
Uninstall: npx claude-next-skill uninstall
Without Node: save skills/next/SKILL.md as ~/.claude/skills/next/SKILL.md (on Windows, %USERPROFILE%\.claude\skills\next\SKILL.md).

Tested

tests/verify.sh drives a real Claude Code session in tmux, queues two /next commands during a multi-step task, and reads the session transcript to check what reached Claude and when. It passes on Claude Code 2.1.283 (macOS):

✓ both /next were queued during the first turn (2/2)
✓ neither /next reached Claude during the first turn (leaks: 0)
✓ the second /next didn't reach Claude during the first /next (leaks: 0)
✓ each /next ran as its own turn (2/2)
✓ reply order: FIRST_DONE -> SECOND_DONE -> THIRD_DONE

Run it again after updating Claude Code (needs tmux and jq).

Limits

It waits for the main turn to end, not for background tasks Claude started.
Permission prompts still apply.
It's a personal skill, so it works in local sessions on your machine, not in cloud sessions.