English

← Back to list

Claude's voiceproof plug-in.

itadol5j

♪ Caveman mode: Let AI speak only human words ♪

Recently, one called "Caveman." The Claude Code plugin exploded on Hacker News. The core idea of this plugin is simple and harsh:To force Ai to return to the time of the caveman.I don't know.

The blogger says:"Enough with AI bullshit!"

How bad is AI's disease?

People who used Claude and ChatGPT know that these AIs are sick--Too much talk.I don't know. Ask it a simple technical question, it can write you a paper:

  • I'll explain to you the history of the concept.
  • Analysis of strengths and weaknesses of programmes
  • Then give a detailed description of the steps.
  • And finally, a summary and best practices.

Results? You just want to know how to change the code, and it gives you 2,000 words of background.

Core logic of the Caveman plugin

This plugin's prompt design is a very simplistic example:

Systems Prompt:

You speak in short, given events.

Example:

How do I enter a div?

You: Use flexbox. display: flex; justy-content: center; align-ites: center;

No, "There are special ways..." Just the answer.

✨ Effect comparison

scene Normal Mode Caveman Mode
Q: How to change this bug? "The problem involves... first we need to understand... there are several solutions... The first is... the second is... a combination, I suggest... "Leave a semicolon in line 3."
Q: What does this function do? "This function is a tool function used primarily to process... It is designed to describe... parameters... returns values... using examples... (800 words) "Computation of user points."
Question: Recommended a library "In accordance with your needs, I recommend the following reservoirs... the advantage of the A library... is that the B library is characterized by... the C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C-C "with lodash."

How to use Claude Code

Method I: directly modify system program

Add in Claude Code configuration file:

{ "systemPrompt": "You are a caveman. Speak in short, grunted sentences. No fluff. Just the answer." }

Method II: Use of shortcut instructions

Create /caveman Command, switch mode at any time:

/caveman on  - 开启穴居人模式/caveman off - 关闭穴居人模式

Methodology III: Switching for different tasks

  • Write code: In Caveman mode, do not explain the code
  • Writing documents: In normal mode, need details
  • Debug: In Caveman mode, just cause of error
  • Learning new concepts: using normal patterns, requiring background knowledge

I'm sorry.

1. Customity of rhetoric

It's not just a switch, it's a setup:

  • Level 1 – Caveman: only keywords
  • Level 2 – Robot: Short sentences, uncompensated
  • Level 3 – Engineer: technical details, no bullshit
  • Level 4 – Normal: normal dialogue
  • Level 5 – Professor: detailed explanation, examples

2. Scenario configuration

There are different scenarios in different formats:

# 代码相关 -> 穴居人模式# 文档相关 -> 正常模式# 调试相关 -> 穴居人模式# 架构设计 -> 工程师模式

3. Reverse operation: speech mode

Sometimes you need an AI to say more? Here. /professor Mode:

“Please explain in detail the historical origin of this concept, its theoretical basis, the rationale for its realization, the context of its application, the comparison of strengths and weaknesses, best practices, common problems...”

Attention

  • Not for beginners.If you're a beginner, you need to explain it.
  • The context will be lost.: Too concise may lead to missing important information
  • Questions need to be precise.: A simple answer presupposes that the question itself is clear
  • Some scenarios don't apply.: Write a document, e-mail or do you need a full expression

🎭 Brilliant comments from readers

"Finally, I don't need to read the little book I wrote."

"That's what I want, not what I want."

"Recommends that OpenAI and Anthropic internalize this model"

"It's time to cure al-Ai's disease."

"The Caveman Model x Society Fear of Gospels"

At the end.

The explosion of the Caveman plugin reflects a deep need:What we need is not a more talkable AI, but a better AI.I don't know.

Know when to explain in detail, when to answer briefly — that's real intelligence.

So next time you're drowning in AI's bullshit, say to it:

"Be a Caveman."