Caveman Micro — Development skill for Claude Code
6-line caveman micro prompt (85 tokens) that outperformed the original 552-token skill.
How to install Caveman Micro
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open kuba-guzik/caveman-micro and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Caveman Micro does
6-line caveman micro prompt (85 tokens) that outperformed the original 552-token skill. Benchmark on Claude Sonnet + Opus included.
Alternatives in Development
- GenericAgent — Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x less token co 6.1k ★
- Claude Engineer — A version of command line interface(CLI) that leverages the power of Claude Opus(3-4.7) 954 ★
- Caveman Claude Skill — Ultra-compressed communication mode for Claude 396 ★
README
caveman-micro
**6 lines. 85 tokens. Outperformed the 552-token original.**
We benchmarked the viral [caveman](https://github.com/JuliusBrussee/caveman) token-saving prompt on real coding tasks. Then we distilled it into 6 lines that beat the original on both Claude Sonnet and Opus.
The micro prompt
Respond like smart caveman. Cut all filler, keep technical substance.
- Drop articles (a, an, the), filler (just, really, basically, actually).
- Drop pleasantries (sure, certainly, happy to).
- No hedging. Fragments fine. Short synonyms.
- Technical terms stay exact. Code blocks unchanged.
- Pattern: [thing] [action] [reason]. [next step].Copy it into your system prompt, custom instructions (ChatGPT), or CLAUDE.md. Works with any LLM.
Benchmark results
Tested on real tasks (incident diagnosis, config extraction) with structured JSON output and quality verification.
Claude Sonnet
| Group | Avg Output Tokens | Quality | Savings |
|---|---|---|---|
| Baseline ("Be concise") | 259 | 100% | -- |
| Caveman full (552 tok) | 225 | 100% | 13% |
| Caveman Micro (85 tok) | 223 | 100% | 14% |
Claude Opus
| Group | Avg Output Tokens | Quality | Savings |
|---|---|---|---|
| Baseline ("Be concise") | 227 | 100% | -- |
| Caveman full (552 tok) | 207 | 100% | 9% |
| Caveman Micro (85 tok) | 180 | 100% | 21% |
Quality: 100% in every run. Zero missing facts. The micro version outperformed the full skill on both models.
Why micro beats full
The model already knows how to be concise. It doesn't need a 552-token tutorial. It needs 6 lines of permission.
A longer instruction set costs tokens to inject and gives the model more noise to process. The short nudge does the same job at one-sixt
Related Skills
Claude Fable 5 Prompt Strategies
Use when composing prompts targeting Claude Opus 5 (Claude 디폴트), Fable 5, Opus 4.8, or Sonnet 5 — 최신 Claude 프롬
JiegeOSBench
Benchmarking a single‑prompt Rust OS kernel implementation, comparing Claude Code Fable, Opus, and Sonnet.
Token Forecaster
Forecasts how many output tokens a Claude reply or agent turn will use, and shows how it is tracking while it
Caveman Skill
save tokens? claude? vibe code? you knocked your head too much, let's go find some bugs to eat
Caveman
caveman — 65% fewer output tokens, same technical accuracy
Ashlr Brief
Toggle response-shortening (caveman-inspired, ashlr-flavored) with three intensity levels. Trims filler from C
Related Agents
Claude Optimizer
Optimizes CLAUDE.md files for maximum effectiveness with Sonnet 4 and Opus 4 models by analyzing structure, co
Escalation Reviewer Opus
Second-level escalation code reviewer (Opus) used ONLY under ORCHESTRATOR section 6, when the Sonnet escalatio
Evm Liquidity Lane
Liquidity and market research lane for EVM token diligence. Use with the one-line pointer prompt printed by br
