In this article
caveman
| Field | Value |
|---|---|
| Kind | skill |
| Source | .github/skills/experimental/caveman |
| Invocation | Invoked directly as /caveman; model invocation is disabled, so agents do not load it |
| Interactive | No |
What it does
Ultra-compressed response style that reduces output token count while preserving technical accuracy, with intensity levels and auto-clarity safety rules
When to use it
Invoke Caveman explicitly when you want compressed assistant prose across the
current conversation. lite keeps full sentences, full permits fragments,
ultra is telegraphic, and wenyan adds a Classical Chinese register. A generic
"be brief" request does not activate the mode, and agents do not select it on
their own.
This changes output style, not code or reasoning-token usage. Security warnings, destructive-action confirmations, quoted tool output, and ambiguous instructions temporarily return to normal clarity.
Example usage
Enter /caveman lite, then ask for a summary of a supplied test failure. Expect a
short explanation with the exact test name, error text, and command arguments
preserved. Switch with /caveman ultra if fragments remain clear, or use
/caveman off to restore normal responses.
Success is less prose without lost technical meaning. Generated code, commit messages, PR bodies, and release notes stay in normal style. The latest visible activation or exit directive governs the mode; it has no state file and resets when the relevant conversation context is gone. If you ask for clarification, clarity takes precedence over compression.