B-TIER technique
Claude 'caveman' mode — token-saving terse-output skill
100k ft — What is this?
Claude 'caveman' mode — token-saving terse-output skill
50k ft — Why care?
@sourcetms made a 'caveman' Claude skill (from a Reddit post) that forces terse caveman-speak responses to save tokens. Meme-y but a legit compression technique for long sessions.
10k ft — Key takeaways
(no takeaways yet)
5k ft — Tools & techniques
Anthropic, Claude, Claude Code
AI agents
1k ft — Deep-dive / try it
Redditor make a post. Me read. Me try. Tokens saved. Me make video. Create skill to make more Claude caveman. Share with you.
Comment “caveman” to save tokens too.
eddynthings
13w
This for many reasons is why anthropic hates openclaw. 😂 you can see in the leaked code that they literally burn tokens just to burn tokens.
Transcript excerpt
I don't know why I find this so funny, but this guy has created a way that Claude talks to him like a caveman to save on his token usage. Because that's what I do on this account, let me show you how to use it. By the way, if you comment caveman, I'll just give you the skill directly to install it. Okay, so I've updated the communication style and you can either do that directly inside of your MD file or what you can do is you can ask Claude to update your system prompt through the agent. Then you can give it an example command just to test it out. So here Claude is thinking, got it, pattern understood, describe change, paste content, me apply directly. So this super short precise form of communication from your AI is actually what you want because if you've created some of the most intelligent systems in the world, ideally they can also communicate to you at the level of a caveman.