caveman-output-style-claude-code
Health Uyari
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 8 GitHub stars
Code Uyari
- Code scan incomplete — No supported source files were scanned during light audit
Permissions Gecti
- Permissions — No dangerous permissions requested
This tool is a custom output style for Claude Code. It provides a prompt template designed to make Claude's responses more concise and terse, which reduces token usage and removes conversational filler.
Security Assessment
Overall risk: Low. The installation process simply copies a single Markdown file (`caveman.md`) into your local Claude configuration directory. It does not request dangerous permissions, does not execute shell commands, and makes no network requests. The automated code scan was unable to detect standard source code files to analyze, but given that the tool's core function is just providing a text prompt, the attack surface is virtually non-existent. No hardcoded secrets or malicious payloads were found.
Quality Assessment
The project is actively maintained, with its last push occurring today. It uses the permissive MIT license. However, community trust and visibility are currently very low, evidenced by only 8 GitHub stars. Despite the low visibility, the concept is well-documented and provides a clean, easy-to-revert method for users to save on API token costs. The claims are even backed by links to relevant research regarding language model brevity.
Verdict
Safe to use.
Caveman output style for Claude Code: 40% fewer output tokens, always-on formatting
Caveman Output Style for Claude Code
Better consistency and adherence than CLAUDE.md or skills. Always-on formatting. Less fluff, same technical signal.
quick start
One command. Done.
# macOS / Linux
mkdir -p ~/.claude/output-styles && cp ./.claude/output-styles/caveman.md ~/.claude/output-styles/caveman.md
# Windows
New-Item -ItemType Directory $HOME\.claude\output-styles -Force | Out-Null; Copy-Item .\.claude\output-styles\caveman.md $HOME\.claude\output-styles\caveman.md -Force
Then: /config → Output Style → Caveman.
Want it always on? Add this to ~/.claude/settings.json:
{ "outputStyle": "caveman" }
Why use caveman?
| prompt | normal answer | caveman answer | rough token cut |
|---|---|---|---|
| why build fail? | Build is failing because TypeScript cannot resolve the @/auth import after the alias change. Update the path mapping or import path. |
Build fails: TypeScript cannot resolve @/auth. Fix alias or import path. |
~42% |
| what changed in auth? | I updated the middleware to reject missing bearer tokens before decoding the JWT, which prevents the null token crash. | Auth middleware now rejects missing bearer token before JWT decode. Crash gone. | ~45% |
| how set default style? | Open your Claude settings file and add the outputStyle field with the value caveman, then start a new session. |
Add "outputStyle": "caveman" to settings. Start new session. |
~47% |
Real token count will vary. Table shows rough reduction only. Pattern is the same: less fluff out -> less output spend.
Same answer, less bla. Shorter replies every turn, lower token bill and less energy wasted on filler.
Use when:
- You keep typing "be concise"
- You want the same terse format on every turn
- You want less output spend without losing technical substance
Backed by research: Brevity Constraints Reverse Performance Hierarchies in Language Models. Shorter answer keeps quality, and sometimes even improves it.
why output style, not other things
Claude Code official documentation show output styles change how Claude responds, not what Claude knows.
Output styles are the perfect matching tool for achieving "talk short on every turn" goal:
| tool | good for | why worse? |
|---|---|---|
CLAUDE.md |
project rules, conventions, codebase memory | adds user message after default prompt. not purpose-built for response style, directive lost on long threads |
| skills | reusable workflows, task prompts | skills load when invoked or relevant. Claude Code may decide against using it. Not always-on formatting layer |
--append-system-prompt |
one-off system prompt append | appended at end of system prompt. lower priority. busts cache if value changes |
All methods add roughly the same amount of input tokens. Output styles win on adherence and consistency, not input cost.
how it works
Output styles inject a dedicated section into Claude Code's system prompt:
# Output Style: Caveman
[style text here]
This section is:
- Dedicated — purpose-built for response formatting
- System prompt level — not buried in user messages like CLAUDE.md
- Always on — no model discretion, no invocation needed
- Survives long threads — style persists without dilution
available styles
Caveman (default)
Standard terse mode. Drop articles, filler, pleasantries, hedging. Fragments OK.
Caveman Ultra
Maximum terseness. Telegraphic fragments. Symbols over words. Abbreviated prose.
Use when you want the most aggressive compression.
# macOS / Linux
cp ./.claude/output-styles/caveman-ultra.md ~/.claude/output-styles/caveman-ultra.md
# Windows
Copy-Item .\.claude\output-styles\caveman-ultra.md $HOME\.claude\output-styles\caveman-ultra.md -Force
Then: /config → Output Style → Caveman Ultra.
Tokenizer-safe terseness
Caveman (default) avoids aggressive abbreviations that may hurt clarity or tokenization on non-Anthropic models. Caveman Ultra goes further with symbols and prose abbreviations — use it when you trust your model's tokenizer.
FAQ
Does caveman make Claude dumber?Not at all. Same knowledge, same reasoning, just shorter output. Research shows output brevity can even improve answer quality.
Will it break code generation?No. keep-coding-instructions: true keeps all coding behavior intact. Only the voice changes.
Sure! /options → pick different style. Or remove it from settings.json.
Yes. Output style is the base voice; skills and agents still work normally.
Why not just tell Claude "be concise"?You can, but you have to say it every. single. turn. Output style is always on, set once and forget.
Does it save money?Yes. Fewer output tokens = lower cost, especially on long sessions.
Can I customize it?You are more than welcome. Edit caveman.md in your ~/.claude/output-styles/ folder and make it your own.
license
MIT. Do what you wish with it.
Like this? Star the repo. Share with your team.
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi