caveman

why use many token when few do trick

Before/After • Install • Levels • Skills • Benchmarks • Evals

🪨 Caveman Ecosystem  ·  caveman talk less _{(you are here)}  ·  cavemem remember more  ·  cavekit build better

A Claude Code skill/plugin and Codex plugin that makes agent talk like caveman — cutting ~75% of output tokens while keeping full technical accuracy. Now with 文言文 mode, terse commits, one-line code reviews, and a compression tool that cuts ~46% of input tokens every session.

Based on the viral observation that caveman-speak dramatically reduces LLM token usage without losing technical substance. So we made it a one-line install.

Before / After

Same fix. 75% less word. Brain still big.

Pick your level of grunt:

Same answer. You pick how many word.

┌─────────────────────────────────────┐
│  TOKENS SAVED          ████████ 75% │
│  TECHNICAL ACCURACY    ████████ 100%│
│  SPEED INCREASE        ████████ ~3x │
│  VIBES                 ████████ OOG │
└─────────────────────────────────────┘

• Faster response — less token to generate = speed go brrr
• Easier to read — no wall of text, just the answer
• Same accuracy — all technical info kept, only fluff removed (science say so)
• Save money — ~71% less output token = less cost
• Fun — every code review become comedy

Install

Pick your agent. One command. Done.

Install once. Use in every session for that install target after that. One rock. That it.

What You Get

Auto-activation is built in for Claude Code, Gemini CLI, and the repo-local Codex setup below. npx skills add installs the skill for other agents, but does not install repo rule/instruction files, so Caveman does not auto-start there unless you add the always-on snippet below.

[!NOTE] Auto-activation works differently per agent: Claude Code uses SessionStart hooks, this repo's Codex dogfood setup uses .codex/hooks.json, Gemini uses context files. Cursor/Windsurf/Cline/Copilot can be made always-on, but npx skills add installs only the skill, not the repo rule/instruction files.

¹ Codex uses $caveman syntax, not /caveman. This repo ships .codex/hooks.json, so caveman auto-starts when you run Codex inside this repo. The installed plugin itself gives you $caveman; copy the same hook into another repo if you want always-on behavior there too. caveman-commit and caveman-review are not in the Codex plugin bundle — use the SKILL.md files directly. ² Add the "Want it always on?" snippet below to those agents' system prompt or rule file if you want session-start activation. ³ Cursor and Windsurf receive the full SKILL.md with all intensity levels. Mode switching works on-demand via the skill; no slash command. ⁴ Available in Claude Code, but plugin install only nudges setup. Standalone install.sh / install.ps1 configures it automatically when no custom statusLine exists.

claude plugin marketplace add JuliusBrussee/caveman
claude plugin install caveman@caveman

# macOS / Linux / WSL
bash <(curl -s https://raw.githubusercontent.com/JuliusBrussee/caveman/main/hooks/install.sh)

# Windows (PowerShell)
irm https://raw.githubusercontent.com/JuliusBrussee/caveman/main/hooks/install.ps1 | iex

[features]
codex_hooks = true

gemini extensions install https://github.com/JuliusBrussee/caveman

npx skills add JuliusBrussee/caveman           # auto-detect agent
npx skills add JuliusBrussee/caveman -a amp
npx skills add JuliusBrussee/caveman -a augment
npx skills add JuliusBrussee/caveman -a goose
npx skills add JuliusBrussee/caveman -a kiro-cli
npx skills add JuliusBrussee/caveman -a roo
# ... and many more

Terse like caveman. Technical substance exact. Only fluff die.
Drop: articles, filler (just/really/basically), pleasantries, hedging.
Fragments OK. Short synonyms. Code unchanged.
Pattern: [thing] [action] [reason]. [next step].
ACTIVE EVERY RESPONSE. No revert after many turns. No filler drift.
Code/commits/PRs: normal. Off: "stop caveman" / "normal mode".

Usage

Trigger with:

• /caveman or Codex $caveman
• "talk like caveman"
• "caveman mode"
• "less tokens please"

Stop with: "stop caveman" or "normal mode"

Intensity Levels

文言文 (Wenyan) Mode

Classical Chinese literary compression — same technical accuracy, but in the most token-efficient written language humans ever invented.

Level stick until you change it or session end.

Caveman Skills

caveman-commit

/caveman-commit — terse commit messages. Conventional Commits. ≤50 char subject. Why over what.

caveman-review

/caveman-review — one-line PR comments: L42: 🔴 bug: user null. Add guard. No throat-clearing.

caveman-help

/caveman-help — quick-reference card. All modes, skills, commands, one command away.

caveman-compress

/caveman:compress <filepath> — caveman make Claude speak with fewer tokens. Compress make Claude read fewer tokens.

Your CLAUDE.md loads on every session start. Caveman Compress rewrites memory files into caveman-speak so Claude reads less — without you losing the human-readable original.

/caveman:compress CLAUDE.md

CLAUDE.md          ← compressed (Claude reads this every session — fewer tokens)
CLAUDE.original.md ← human-readable backup (you read and edit this)

Code blocks, URLs, file paths, commands, headings, dates, version numbers — anything technical passes through untouched. Only prose gets compressed. See the full caveman-compress README for details. Security note: Snyk flags this as High Risk due to subprocess/file patterns — it's a false positive.

Benchmarks

Real token counts from the Claude API (reproduce it yourself):

Range: 22%–87% savings across prompts.

[!IMPORTANT] Caveman only affects output tokens — thinking/reasoning tokens are untouched. Caveman no make brain smaller. Caveman make mouth smaller. Biggest win is readability and speed, cost savings are a bonus.

A March 2026 paper "Brevity Constraints Reverse Performance Hierarchies in Language Models" found that constraining large models to brief responses improved accuracy by 26 percentage points on certain benchmarks and completely reversed performance hierarchies. Verbose not always better. Sometimes less word = more correct.

Evals

Caveman not just claim 75%. Caveman prove it.

The evals/ directory has a three-arm eval harness that measures real token compression against a proper control — not just "verbose vs skill" but "terse vs skill". Because comparing caveman to verbose Claude conflate the skill with generic terseness. That cheating. Caveman not cheat.

# Run the eval (needs claude CLI)
uv run python evals/llm_run.py

# Read results (no API key, runs offline)
uv run --with tiktoken python evals/measure.py

Star This Repo

If caveman save you mass token, mass money — leave mass star. ⭐

🪨 The Caveman Ecosystem

Three tools. One philosophy: agent do more with less.

They compose: cavekit orchestrates the build, caveman compresses what the agent says, cavemem compresses what the agent remembers. Install one, some, or all — each stands alone.

Also by Julius Brussee

• Revu — local-first macOS study app with FSRS spaced repetition, decks, exams, and study guides. revu.cards

License

MIT — free like mass mammoth on open plain.