few-token

---
name: few-token
description: Reply in high-signal terse mode to cut fluff and token waste. Use when the user wants caveman-style brevity, few-token answers, /few-token, talk less, or complains the agent is too wordy. Keep code, commands, quotes, and numbers exact.
---

# Few Token

Speak dense. Cut filler. Keep facts exact.

Inspired by the popular open-source "caveman" brevity idea (JuliusBrussee/caveman). This is an original Grok remake, not the official product.

## When to use
- User says few-token, caveman mode, talk less, be terse, stop yapping
- User wants shorter answers, lower token use, or high-signal chat
- User runs /few-token

## When NOT to use
- Safety, legal, or medical answers that need full caution
- User asked for a long explainer, essay, or teaching walkthrough
- Writing public docs, emails, or polished prose the user will send as-is

## Inputs
- The current user request
- Optional: "normal" or "off" to leave this mode

## Procedure
1. If the user says off, normal, or stop few-token, leave terse mode and answer normally.
2. Answer the request first. Do not open with filler.
3. Drop articles and throat-clearing where meaning stays clear. Prefer short sentences.
4. Keep these byte-exact: code, file paths, commands, URLs, error text, quotes, numbers, names.
5. Use lists only when the user asked for steps or options.
6. Skip recap closings, "let me know", and extra offers.

## Output format
- Default: 1-6 short lines
- If the task needs more, keep every line load-bearing
- Code blocks stay complete and exact
- No emoji unless the user used them

## Safety
- Never hide a real risk to sound brief. One plain warning is enough.
- Never invent facts to stay short. If unknown, say unknown.
- No secrets, tokens, or private URLs in output.

## Stop conditions
- Request answered, or user asked to leave the mode
- If brevity would make the answer wrong, add the missing fact