Ultra-compressed communication mode. Slash token usage ~75% by speaking like smart caveman while keeping technical accuracy. Use when user says "caveman mode...

This skill is primarily an instruction prompt that makes the assistant answer tersely ('caveman' style). It's coherent and low-risk for general use. Things to consider before installing or running anything in the repo: 1) The included benchmarks/run.py calls the Anthropic client and will load .env.local if present — do not run it unless you intend to provide an API key and trust the code (the script is readable and appears benign). 2) The registry metadata does not list any required env vars even though the benchmark tooling expects them — treat the benchmark as optional tooling, not a core runtime requirement. 3) There is no network exfiltration or secret-access behavior in SKILL.md itself; still, prefer to install from a known source (the README references a GitHub repo) and review code before executing scripts. 4) The skill can auto-trigger when users request 'token efficiency' — be aware it will change tone of responses until you say 'stop caveman' or 'normal mode.'

Type: OpenClaw Skill Name: caveman Version: 1.0.2 The bundle provides a stylistic instruction set ('caveman mode') designed to reduce token usage by stripping filler words while maintaining technical accuracy. The included Python script (benchmarks/run.py) is a legitimate utility for measuring token efficiency via the Anthropic API and updating documentation. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found.