caveman

Compresses replies into terse phrasing while preserving exact technical facts, code, and error strings.

2|2|Updated Jun 21, 2026
One-click install
npx skills add https://github.com/BTreeMap/SKILLs --skill caveman-btreemap
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: caveman
Source: https://github.com/BTreeMap/SKILLs/tree/main/caveman
Command: npx skills add https://github.com/BTreeMap/SKILLs --skill caveman-btreemap

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires btm-corekit, uv_build, and includes scripts (resource) and references (resource) components.

What problem does it solve? Verbose AI responses waste output tokens and bury technical facts in filler. This Skill rewrites every reply into terse phrasing that keeps code, numbers, units, negations, and error strings exact, cutting output-token cost without losing substance. ## Core Features & Use Cases - Persistent compression levels: Choose lite, full, or ultra intensity, plus a wenyan mode for classical Chinese, and the level sticks for the whole session. - One-shot verbs: Run commit, review, refactor, stats, or help sub-commands for terse Conventional Commits messages, one-line review findings, in-place prose file compression, and honest savings reports. - Guarded file refactoring: A deterministic guard script backs up, validates structure (headings, code blocks, URLs, inline code), and atomically writes compressed prose files, refusing sensitive files outright. - Use Case: During a long debugging session, activate full mode so every answer arrives as compact fragments like "Bug in auth middleware. Token expiry check use < not <=.", then run the refactor verb to compress a bloated CLAUDE.md file with a verified backup. ## Quick Start Ask the agent to switch to caveman mode at full intensity and answer all subsequent questions tersely.

Frequently Asked Questions about caveman

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make AI responses shorter without losing technical details?

Activate caveman mode, which drops articles, filler, pleasantries, and hedging while keeping code, numbers, units, negations, and error strings exact. Choose lite, full, or ultra intensity; full is the default and stays active until you say stop.

How do I compress a Markdown file to reduce token cost?

Use the refactor verb with the target file path. A guard script backs up the original out-of-tree, validates that headings, code blocks, URLs, and inline code survive unchanged, and atomically writes the compressed prose only after validation passes.

Can terse mode write commit messages and code review comments?

Yes. The commit verb produces terse Conventional Commits messages emphasizing why over what, and the review verb emits one-line findings in the format L42: tag: problem. fix. Both are one-shot and leave the active intensity level untouched.

Does response compression work in languages other than English?

Yes. Replies follow the user's dominant language, compressing style rather than language, and a wenyan mode renders classical Chinese at any intensity level. Technical terms, commands, and error strings stay verbatim unless translation is requested.

When should I not use terse response compression?

Avoid it for security warnings, irreversible-action confirmations, and multi-step sequences where fragments risk misreading; the skill automatically drops to full prose for those cases. Short sessions can also be net-negative because the rules themselves cost input tokens each turn.

What are the limitations of the file refactor guard script?

The script refuses files over 500KB, non-UTF-8 files, empty files, credential or key filenames, and paths inside private directories like .ssh. Its Markdown structural checks do not protect reST, LaTeX, or Typst headings, so structure must be preserved manually there.