caveman

You can reduce token usage in multi-turn LLM conversations by enabling an ultra-compressed communication mode that delivers terse, technically precise responses while preserving essential content. This mode persists across turns to help you move conversations faster and cheaper.

The best way to get concise, technically accurate responses is to use a brevity mode that strips unnecessary text while keeping technical accuracy. You can easily activate it by issuing the /caveman command or saying caveman before asking your question.

Yes, you can adjust the intensity of token reduction modes using lite, full, ultra, and wenyan variants. These intensity levels support auto-switching to help you control the exact level of brevity and token compression needed for your specific prompts.

Caveman mode requires no external dependencies, but it does require a YAML frontmatter block with a name and description in SKILL.md. It may also include optional scripts, references, or assets for extended behavior without needing additional packages.

You should not use ultra-compressed communication modes when your LLM responses require detailed explanations, conversational tone, or non-technical context. The mode is designed for terse, technically precise outputs, meaning it will strip away conversational filler and lengthy formatting.