caveman is a response style: when enabled, the entire conversation leads with the conclusion followed by reasons and next steps, without pleasantries, restatements, or phrases like "hope this helps." Each sentence covers only one point, using ASD-STE100 simplified English as the baseline. Code blocks, commands, paths, and error messages are preserved exactly as they are; negations are never omitted. When clarity conflicts with compression, clarity takes precedence. It compresses style, not language: if the user asks in Chinese, the response is in Chinese. There are also more extreme ultra and wenyan (classical Chinese) modes.
Why it matters Output tokens are billed by volume and read in the terminal; eliminating pleasantries and restatements directly saves money and time, while code, commands, and error messages are preserved exactly as they are.
skills.sh installs 626207 (+2101 / 24h)
Security Audit: Risk warnings present
- Initial screening only: Only SKILL.md was scanned; no full repository audit was performed, so no installation command is provided yet.
- Audit scope: GitHub juliusbrussee/caveman/plugins/caveman/skills/caveman/SKILL.md (only the SKILL.md text was scanned)
Reasons
- skills.sh installs 2101 in 24 hours (indicating attention only)
- Pure text rules; does not execute commands or connect to the internet
Usage
- Enter /caveman or say "caveman mode" to enable it; say "stop caveman" or "normal mode" to revert. Use /caveman status to check the current mode.
Risks
- The scan only examined SKILL.md; there are 20 additional companion skills and hooks in the same repository that have not been audited, so no installation command is displayed yet.