Guide · Agent skill
How to use the Caveman Terse Voice agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The caveman skill is an agent skill that changes how an AI agent writes its replies. It removes greetings, hedging and closing summaries, and puts the answer first in short sentences, while keeping code, commands, paths, numbers and error messages exact. This guide explains what an agent skill is, how to install caveman, how to switch it on and off, and when it is the wrong choice.
Caveman Terse VoiceJuliusBrussee/caveman · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the caveman skill ZIP from this page and unzip it on your computer.
- 2Choose a location. Put the caveman folder in ~/.claude/skills/ for all projects, or in .claude/skills/ inside one project.
- 3Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the skill.
- 4Start a session. Type /caveman or ask for caveman mode, such as "talk like caveman" or "be brief".
- 5Ask a normal task. Give the agent a coding or terminal task and read the answer-first reply.
- 6Switch off. Say "stop caveman" or "normal mode" when you want full-sentence replies again.
What an agent skill is and what caveman changes
An agent skill is a folder with a SKILL.md file that gives an AI agent extra instructions it loads when they apply. This one is named caveman, and its description says it is a terse voice: answer first, fluff gone, every technical fact kept.
Once the skill is on, the agent follows the rules for every response in the session. It answers in the pattern [thing] [action] [reason]. [next step]. It drops filler such as "Sure, I'd be happy to help" and words like just, really and basically. Articles are optional when the sentence still reads clearly.
- Negations are never dropped: not, never, no, only and except stay in every sentence.
- Code blocks, commands, paths, API names and errors are copied exactly.
- Numbers and units stay exact.
- Invented abbreviations such as cfg or impl are avoided. Standard acronyms like DB and API are allowed.
- Your reply language stays the same as yours. The style is compressed, not the language.
Install the caveman skill from the ZIP
Download the ZIP for the caveman skill from this page and unzip it. You can then place the folder in one of two locations, depending on whether you want the skill for all projects or one project only.
For a personal install that applies everywhere, put the skill in ~/.claude/skills/. For a single project, put it in the project's .claude/skills/ folder. The skill folder contains agents/openai.yaml and SKILL.md.
- Personal: ~/.claude/skills/ (available across your projects).
- Project only: .claude/skills/ inside the project folder.
- Claude apps: upload the skill under Settings → Capabilities → Skills.
Turn caveman on and off with natural requests
You can start caveman with the command /caveman or with plain requests. The skill's description lists phrases such as "caveman mode", "talk like caveman", "be brief" and "less tokens" as triggers.
It stays on for the whole session until you say "stop caveman" or "normal mode". If you are unsure whether it is still on, the agent is required to keep it on and confirm the switch-off in one line when you ask it to stop.
- Turn on: /caveman, "caveman mode", "talk like caveman", "be brief", "less tokens".
- Turn off: "stop caveman" or "normal mode".
- Check status: /caveman status reports the current mode and changes nothing.
/cavemanSee what caveman replies look like
The source gives one direct comparison. A plain reply starts with a greeting and explains the cause at length. The caveman reply names the bug, the exact fix and stops.
Clipped style does not mean lost detail. Technical content stays intact, so a fix reads the same way it would in a full answer.
- Plain: "Sure! I'd be happy to help. The issue you're experiencing is likely caused by..."
- Caveman: "Bug in auth middleware. Token expiry check use < not <=. Fix:"
Know what the skill needs and what it runs
The listing states that the skill requires nothing in particular, does not run scripts, and does not use the network. It is a set of written instructions, so it adds no extra software to your machine beyond the agent you already use.
Because the skill only changes writing style, the main safety rule is about what the agent still says in plain prose. The skill tells the agent to switch to full sentences for security warnings, irreversible actions, step orders that could be scrambled, and when the agent asks you a question.
- Requires: nothing listed.
- Runs scripts: no.
- Uses network: no.
- Files: agents/openai.yaml and SKILL.md.
- Caveman text is not used in files, commits, docs or tickets that the agent writes. Those stay in normal prose.
Tips for getting the most from caveman
Use caveman when you read replies in a terminal and want quick answers. Ask for a full answer when a topic is new to you, because compressed replies can leave out context.
If the agent is confused or you repeat a question, the skill allows it to fall back to plain prose for that reply. You can also say "normal mode" at any time.
- Read code changes as the changed lines with one or two lines of context, which is how caveman shows them.
- Ask for the whole file if you need it, since the skill shows only changed lines by default.
- Switch to normal mode for beginner explanations or reports to other people.
What people use it for
- Asking an agent in a terminal to diagnose a bug and get the fix line first.
- Cutting output tokens on long coding sessions where you pay per token.
- Getting quick status checks with /caveman status without changing the mode.
- Reading command-heavy answers where paths, numbers and errors must stay exact.
Frequently asked questions
Is the caveman agent skill free?
The source lists the licence as Apache-2.0 for the repository, and this is the licence to check for terms of use. The listing does not state any price, so do not assume a cost beyond what your own agent setup requires.
What does the caveman skill require?
The listing states nothing is required. It does not run scripts and does not use the network. Install it by unzipping it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
How do I turn caveman mode off?
Say "stop caveman" or "normal mode". The skill stays on for the whole session until you say one of these phrases. You can also check the current state with /caveman status, which changes nothing.
Does caveman mode lose important technical details?
The skill is written to keep code, commands, paths, numbers and error messages exact, and never to drop negations such as not or never. Still, the listing warns that compressed replies can omit context, so ask for full sentences when a topic is new to you.
Is caveman good for beginners?
The listing says it is not suited to tasks where the agent must explain nuanced reasoning to a beginner. Use normal mode for learning or formal writing, and caveman for quick, technical answers you can scan.
Does caveman change the language I write in?
No. The skill compresses the style, not the language, and it keeps your reply language unless you explicitly ask for another. Technical terms and error messages stay verbatim.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Caveman suits developers who use an AI agent in a terminal, want shorter answers that are fast to scan, or pay per token and want lower output costs without losing technical detail. It is a poor fit for customer-facing, formal or beginner-focused writing, where the clipped style and possible missing context are a problem.
Download the skill