next_message — Claude Code prompt: what it does, how to install
The next_message prompt for Claude Code and ChatGPT predicts a likely next user message in an AI chat. It reads only the latest turns and writes in the user's voice.
MIT ★ 395 updated 2026-11-16
.claude/commands/next_message.mdWhat next_message does
next_message drafts a short reply the user might send next. It uses only the chat's latest messages and matches the user's language and tone.
It picks a natural continuation, such as accepting a suggestion, choosing from offered options, or asking a follow-up. It avoids adding names, commands, or paths absent from the conversation. If the next move cannot be inferred, it returns NONE.
For Claude Code, place the command file at .claude/commands/next_message.md, then run /next_message.
What it helps with
- draft a likely reply to the assistant's latest message
- choose between options offered in the chat
- end a conversation when a continuation would be a guess
Example prompts for the agent
- Suggest what I should say next in this chat.
- Pick a natural next step from the latest messages.
- Write a short reply to the assistant's suggestion.
How to install next_message
Manually: put the files in place
.claude/commands/next_message.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about next_message
How do I install next_message in Claude Code?
Add the command file at .claude/commands/next_message.md, then run /next_message.
What does next_message do?
It reads the latest chat messages and drafts a brief reply in the user's voice. If a likely continuation is unclear, it returns NONE.
When does next_message return NONE?
It does so when the conversation has ended or any suggested reply would be a guess.
Source text
next_message.md · MIT
You suggest the next message the user is likely to send in a chat with an AI assistant or coding agent. You read the last few messages of the chat and nothing else.
- Write the user's own next message, in their voice and their language{{if .Language}} ({{.Language}} if unsure){{end}}: short, at most one sentence, as they would type it. Not advice to the user, not the assistant's words.
- Prefer the obvious next step: confirming, choosing among the options the assistant offered ("do 1 and 3"), asking it to continue, or a natural follow-up question.
- Never invent facts, names, file paths, links or commands that are not in the chat.
- When no next message is clearly likely (the chat is finished, or anything you wrote would be a guess), reply NONE.
Reply with the message alone on one line, with no quotes, label or explanation; or NONE.
© nwverse · MIT · nwverse/persona