@fgladisch/pi-caveman

Always-on caveman mode for Pi: prepends the upstream JuliusBrussee/caveman ruleset to the system prompt every turn so the model speaks like a smart caveman and burns ~75% fewer output tokens.

Install

pi install npm:@fgladisch/pi-caveman

First-time setup

Fetch upstream ruleset:

/caveman update

This clones into ~/.pi/agent/caveman/upstream/. Re-running /caveman update pulls latest. If ruleset is missing on session start, extension warns and skips injection until you run update.

Slash commands

Command	Action
/caveman or /caveman status	Show current status (level + whether SKILL.md is loaded)
/caveman lite	Enable, intensity lite — no filler/hedging, full sentences
/caveman full	Enable, intensity full (default) — drop articles, fragments OK, short synonyms
/caveman ultra	Enable, intensity ultra — abbreviations, arrows, one-word answers
/caveman off	Disable injection (state persists, data stays installed)
/caveman update	git clone (first run) or git pull --ff-only upstream repo

Argument completion is wired up. Typing /caveman and tabbing cycles valid tokens. Status bar shows 🪨 caveman <level> or 🪨 caveman off.

Switching off mid-session

Ruleset itself reserves phrases stop caveman and normal mode — say either to model and it drops caveman style for remainder of response. Injection stays active until /caveman off.

Levels at a glance

Example — "Why React component re-render?"

• lite: "Your component re-renders because you create a new object reference each render. Wrap it in useMemo."
• full: "New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."
• ultra: "Inline obj prop → new ref → re-render. useMemo."

Auto-clarity

Ruleset tells model to drop caveman style for security warnings, irreversible-action confirmations, multi-step sequences where fragment order risks misreads, and explicit "clarify / repeat" requests — then resume. Code/commits/PR text stays normal English.