SkillHub

Skills juliusbrussee caveman

Caveman Bundle of 16 skills

Use it when:
You want shorter agent replies and fewer tokens spent, beyond chat too.
It does:
Makes the agent answer tersely, and adds terse commits, one-line reviews and compressed memory files.
Modes:
caveman's short sentences, ultracave's fragments or megacave's Classical Chinese; the last two start only when you type their command.
Keeps:
Code, commands, errors and numbers exact, and full sentences for warnings and irreversible actions.
Also has:
Six short work patterns, such as investigate-first and verify-and-stop, that keep the agent's work to what the task asks.
Needs:
The Caveman plugin for caveman-stats and for cavecrew's subagents.

TL;DR

Its core is caveman: the agent leads with the answer in short plain sentences, without greetings, hedging or recaps, until you say "stop caveman". Two stricter modes start only when you type them: ultracave cuts replies to fragments, and megacave turns them into Classical Chinese (文言文), for fewer characters. The other skills carry the idea beyond chat. caveman-commit writes terse Conventional Commits messages, caveman-review writes one line per review finding, and caveman-compress rewrites a memory file such as CLAUDE.md in caveman style to save input tokens, backing up the original. To spare the main context, cavecrew hands searches, small edits and reviews to three subagents that report back compressed, and caveman-explore, a read-only code finder, replies with file and line citations only. caveman-stats shows recorded token usage, and caveman-help lists every mode and command. The agent picks up six short work patterns when a task fits: investigate-first diagnoses before editing, lean-build builds only what's asked, surgical-patch fixes at the narrowest layer, safe-refactor keeps behavior unchanged, migration keeps changes reversible and verify-and-stop proves finished work without adding to it. The repo's CLI and proxy, which shrink what the agent reads, and the skills that drive them aren't in this bundle.

Summarizes the repo at 4bf8898.

Any agent, with the skills CLI

npx skills add https://github.com/juliusbrussee/caveman --skill '*'

Claude Code plugin

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

Source on GitHubAll skills from juliusbrussee/caveman

Total Downloads
616,833 of caveman, its most installed skill, on skills.sh, Nov 9, 2026
Repo stars
118,416 juliusbrussee/caveman
First seen
Apr 4, 2026 on skills.sh, its first skill

Its skills, in the order they're used

The ones marked “Only in the bundle” need the rest of it. The others also work alone.

  1. caveman

    617K total downloads

    Switches to ultra-compressed answers that keep technical accuracy.

  2. ultracave

    40.4K total downloads

    Switches the agent to fragment-only replies, one word where one word does and each fact said once, until you say "stop caveman".

  3. megacave

    39.2K total downloads

    Switches the agent to terse replies in Classical Chinese, with code and technical terms kept verbatim, until you say "stop caveman".

  4. caveman-commit

    431K total downloads

    Writes Conventional Commits messages compressed to intent only.

  5. caveman-review

    427K total downloads

    Writes terse code review comments, one line per finding.

  6. caveman-compress

    428K total downloads

    Compresses CLAUDE.md or todo files into caveman format to save input tokens.

  7. caveman-stats

    Only in the bundle 373K total downloads

    Shows token usage and mode attribution for the Claude Code session.

    Prints the token report of hooks that only the Caveman plugin installs, in Claude Code.

  8. caveman-help

    Only in the bundle 417K total downloads

    Shows a quick-reference card for caveman modes.

    A reference card for the bundle's modes and skills, so it only helps once they're installed.

  9. cavecrew

    Only in the bundle 373K total downloads

    Delegates work to terse investigator, builder and reviewer subagents.

    Delegates to three subagents that come with the Caveman plugin, not with the skill.

  10. caveman-explore

    122K total downloads

    Finds where the relevant code lives in a repository, read-only, and replies only with path:line citations for the agent that asked.

  11. investigate-first

    119K total downloads

    Diagnoses ambiguous failures before editing.

  12. lean-build

    118K total downloads

    Builds features with strict scope to avoid overbuilding.

  13. surgical-patch

    119K total downloads

    Fixes bugs at the narrowest layer that owns the wrong behavior.

  14. safe-refactor

    120K total downloads

    Restructures code while preserving behavior, with verification around each edit.

  15. migration

    118K total downloads

    Implements reversible migrations of schemas, data, APIs and dependencies.

  16. verify-and-stop

    119K total downloads

    Proves finished work meets its acceptance conditions with the smallest set of checks, then stops without adding fixes or polish.