Makes the agent answer in terse caveman speech, dropping fluff but keeping every technical fact.
JuliusBrussee/caveman
Cuts token use for coding agents by answering in compressed caveman-style language while keeping every technical fact.
- 22 skills
- 115k stars
- Reached #10 on GitHub's daily trending
The 5 most installed
Ordered by installs on skills.sh- caveman641k installs
- caveman-compress445k installs
Compresses a memory file like CLAUDE.md into caveman format to save input tokens, keeping a readable backup.
- caveman-commit444k installs
Writes a short Conventional Commits message that captures only the intent of your change.
- caveman-review440k installs
Reviews a diff or pull request in compressed form, one line per finding with location, problem and fix.
- caveman-help430k installs
Shows a quick-reference card of the caveman skills and their commands.
The other 17 skills
- caveman-stats389k installs
Shows token usage and mode attribution for the current Claude Code session, or finds the host's usage report.
- cavecrew386k installs
Delegates code locating, small edits and diff reviews to compressed sub-agents so your main context lasts longer.
- caveman-explore134k installs
Explores the repository read-only to locate code across files, returning only path and line citations.
- caveman-learn133k installs
Reviews a Caveman learn report, applies cost-lowering fixes with your consent and reports the savings.
- safe-refactor133k installs
Restructures code while preserving behavior, verifying before and after each structural edit.
- caveman-optimize132k installs
Turns a Caveman optimization observation into a candidate with a baseline comparison, needing your approval.
- investigate-first132k installs
Diagnoses unclear or intermittent failures and ranks hypotheses by evidence before any code is edited.
- verify-and-stop131k installs
Checks that existing work meets acceptance conditions without expanding scope or adding new work.
- surgical-patch131k installs
Fixes bugs and small behavior changes at the narrowest layer, with regression proof and relevant tests.
- lean-build131k installs
Builds new features with strict scope, reuse of existing code and an explicit stop condition.
- caveman-discover131k installs
Finds and labels every LLM workflow in your repository so Caveman Cloud can group spend by workflow.
- caveman-evidence-review131k installs
Reviews Caveman Cloud evidence read-only, covering cost, scores, workflows, traces, latency, errors and savings.
- caveman-manage130k installs
Inspects Caveman Cloud experiment lifecycle and blocks unsafe runs, letting you start, approve or roll back experiments.
- migration130k installs
Implements reversible migrations for schemas, data, APIs, protocols, config or dependencies with rollback proof.
- caveman-setup130k installs
Routes a repository through the Caveman Cloud gateway so every LLM request is measured without changing behavior.
- ultracave53.6k installs
Makes the agent answer at maximum compression, using fragments and stating each fact once.
Runs only when you type /ultracave
- megacave52k installs
Makes the agent answer in Classical Chinese, using far fewer characters while keeping technical terms verbatim.
Runs only when you type /megacave