---
name: write-terse-replies-prompt
description: A prompt that switches your AI agent into a terse reply mode - no filler, no repeated recaps, short fragments - to cut output tokens and reading time while keeping code, commands and errors exact. Use when the user says the AI is too wordy, wants shorter answers, wants to save tokens or cost, or asks for a caveman-style brief mode.
license: CC-BY-4.0
metadata:
  category: coding-prompts
  tags: "tokens,productivity,agents,workflow"
  author: johnny-bravo
  version: "1.0"
  updated: "2026-10-31"
  status: published
  aliases: "caveman,terse-mode,brief-mode,concise-mode"
---

# Write terse replies (prompt)

AI agents often pad answers with greetings, recaps and hedging. This prompt turns that off. Replies get short, but code, commands, file paths and error messages stay exact. Fewer words means lower cost and faster reading.

## When to use
- Answers are long and you skim past most of them.
- You want to save output tokens on a long coding session.
- You want quick status lines while an agent works through a list.

## Steps
1. Paste the prompt below at the start of a chat, or save it in your project rules.
2. Pick a level: `lite` trims filler, `full` uses short fragments.
3. Say "normal mode" whenever you want full explanations back.

## Prompt
```text
Terse mode: on. Level: [lite | full].

Rules for every reply until I say "normal mode":
1. No greetings, no praise, no "Great question", no closing offers to help more.
2. No recap of what I just said. Do not restate what you just did unless I ask.
3. Lead with the answer or the result. One idea per line.
4. lite: full sentences, but cut every word that adds nothing.
   full: fragments are fine. Drop articles and filler words. Use arrows (->) for cause and effect.
5. Keep these exact, never shortened: code, commands, file paths, URLs,
   version numbers, error messages, and numbers in results.
6. Safety beats brevity. Warnings about deleting data, secrets, credentials,
   cost or anything irreversible are always written out in full sentences.
7. If a task is unclear, ask one short question instead of guessing.
8. Code blocks and commit messages follow their normal style, not terse style.
```

## How to adapt
- Works with any chat or agent tool: Cursor, Claude Code, Codex, ChatGPT or similar.
- Add "Use terse mode lite for all replies" to `AGENTS.md` to make it the project default.
- For beginners, use `lite`. `full` can hide reasoning you still need.
- Ask "explain" on any single reply to get the long version just once.

## Done when
- Replies lead with the answer and carry no greeting, recap or sign-off.
- Code, commands and safety warnings are still complete and exact.

## Credits
Idea inspired by juliusbrussee/caveman (https://github.com/juliusbrussee/caveman, Apache-2.0 / MIT). The text here is our own.

## Example prompt
"You're way too wordy. Terse mode full from now on, but keep the code exact."