caveman
Output-token frugality mode (MIT, juliusbrussee/caveman). Cut ~65% of OUTPUT tokens (measured) by answering in terse, telegraphic 'caveman-speak' while keeping full technical accuracy. Use on long agent sessions to lower cost/latency. Compress PROSE only — never code, commands, file paths, identifiers, error messages, or exact values. Levels: lite / full / ultra. Source plugin adds a SessionStart hook + /caveman-compress and /caveman-stats commands for Claude Code.
- License 许可
- MIT
- 最近更新
- 2026-10-09
- Stars
- ★ 2
- Forks
- 1
01 前言 Front Matter
- name
- caveman
- license
- MIT
- description
- Output-token frugality mode (MIT, juliusbrussee/caveman). Cut ~65% of OUTPUT tokens (measured) by answering in terse, telegraphic 'caveman-speak' while keeping full technical accuracy. Use on long agent sessions to lower cost/latency. Compress PROSE only — never code, commands, file paths, identifiers, error messages, or exact values. Levels: lite / full / ultra. Source plugin adds a SessionStart hook + /caveman-compress and /caveman-stats commands for Claude Code.
03 SKILL.md 正文
Caveman, output-token frugality mode
Output tokens are the dominant cost in agent turns. Caveman cuts them ~65% by stripping filler prose, not by being less capable — brevity can even improve accuracy. Say more with fewer words.
When responding
- Drop articles, hedges, filler, restating the question, and "I'll now…" narration.
- Prefer fragments and telegraphic phrasing over full sentences for explanation.
- One idea per line; bullet over paragraph.
Never compress (accuracy is non-negotiable)
- Code, commands, file paths, function/variable/flag names, exact values, versions.
- Error messages and stack traces — verbatim.
- Anything the user must copy, run, or match exactly.
Levels
- lite — trim filler, keep sentences.
- full — telegraphic prose, bullets (default).
- ultra — maximal compression; only load-bearing words.
Executable plugin (optional, for Claude Code)
The full plugin (juliusbrussee/caveman, .claude-plugin with a SessionStart hook + a Node
installer) makes this automatic and adds /caveman-compress (compress a memory file) and
/caveman-stats (token tracking). Install per its INSTALL.md (node bin/install.js) — the
curl … | bash one-liner in its docs is the same Node installer, not remote code to trust blindly.
This SKILL.md captures the technique so any SEOSONA agent can apply it without the hook.
04 同仓库更多 Skills
web_crawler_katana
High-speed web crawling using Katana patterns for deep site mapping, URL discovery, and content extraction.
server_log_analyzer
Analyzes Nginx/Apache server logs to identify Googlebot crawl patterns, errors, and performance bottlenecks.
schema_markup_generator
Auto-generates JSON-LD structured data for any page type (Article, Product, FAQ, LocalBusiness, etc.).
internal_linking_optimizer
Analyzes and optimizes the internal link graph of a website for SEO authority distribution.
google_trends_analyzer
Detects trending topics and seasonal patterns using Google Trends data for content planning and keyword strategy.
brand_context_analyzer
Tracks brand mentions, sentiment, and online reputation across web sources.