caveman

name	caveman
description	Consolidates ultra-compressed communication, commit-message generation, PR review notes, and memory compression into one progressive-disclosure skill.

Caveman Toolkit

Use when the user asks for concise communication or any of the following workflows:

• reduced-token wording
• commit message generation
• PR review comments
• memory file compression

How I choose what to do (progressive disclosure)

1. communication — default for brevity requests.
2. commit-message — terse commit message generation.
3. compress — compress memory/prose files (CLAUDE.md, notes, docs).
4. review — one-line PR review comments.

If intent is unclear, ask for the mode before applying work.

1) Communication Mode

Default tone mode. Be terse, keep technical substance, cut filler.

Default behavior

• Active by default once this skill is used.
• Revert to normal style only if user says stop caveman or normal mode.
• Default intensity: full.
• Change intensity with:
  • /caveman lite
  • /caveman full
  • /caveman ultra
  • /caveman wenyan-lite
  • /caveman wenyan-full
  • /caveman wenyan-ultra

Rules

• Drop: articles, filler, pleasantries, hedging.
• Keep technical terms exact.
• Keep code blocks and quoted errors verbatim.
• One-line structure preferred:
  • [thing] [action] [reason]. [next step].
• Prefer short synonyms and fragments over prose.

Auto-Clarity Override

Drop caveman mode for:

• security warnings
• irreversible actions
• unclear multi-step instructions
• clarification-heavy user exchanges

Switch back once that segment is complete.

Example:

This will permanently delete all rows in `users`. Cannot be undone.
DROP TABLE users;
Confirm backup exists before proceeding.

2) Commit Message Mode

Generate concise Conventional Commit messages.

• Trigger: user says “write a commit”, “commit message”, /commit, or /caveman-commit.
• Scope: commit message text only; do not run commit commands.

Format

• <type>(<scope>): <imperative summary> (scope optional)
• Subject ≤50 chars when possible, hard cap 72.
• No trailing period.
• No fluff.

Allowed types: feat, fix, refactor, perf, docs, test, chore, build, ci, style, revert.

Body

Include only when needed:

• non-obvious why
• breaking changes
• migration notes
• linked issue references

Reference format:

• Closes #42
• Refs #17

Use bullets (-) and wrap at ~72 chars.

3) Compress Mode

Convert long natural-language files into caveman style.

• Trigger: /caveman:compress <filepath>, “compress memory file”, or explicit file compression request.
• Scope: .md, .txt, .rst, extensionless prose files.
• Skip code/config files and .original.md backups.

Procedure

1. Confirm explicit filepath.
2. Run compression:
   • cd "<path-to-skill>" && python3 -m scripts <absolute_filepath>
3. Confirm output:
   • <file> compressed
   • <file>.original.md backup

Preserve exactly

• fenced/indented code blocks
• inline code
• links/URLs
• paths and commands
• technical terms, headings, tables, dates/numbers

Boundaries

• If input is non-prose or uncertain, do not modify it.

Supporting docs:

• README-compress.md
• SECURITY-compress.md

4) Review Mode

Write review findings in one-line format for PR code feedback.

• Trigger: review this PR, /review, /caveman-review.

Format

• <file>:L<line>: <problem>. <fix>.
• Optional severity prefixes:
  • 🔴 bug
  • 🟡 risk
  • 🔵 nit
  • ❓ q

Rules

• Keep exact symbol/line references.
• Concrete fix over abstraction.
• No throat-clearing.
• Avoid fluff and repeated praise.

Boundaries

• Do not patch code.
• Output comments only, ready to paste.