convez.ai
SkillProdutividadeIntermediário
raushan1107

Compact Mode — Skill de Modo Compacto para Claude

raushan1107/compact-mode-claude-skill

Reduz tokens 60–75% mantendo precisão técnica com 3 níveis ajustáveis

Ver no GitHub
InicialPluginScore · de 0 a 100

Reduz o consumo de tokens do Claude em 60–75% com respostas compactas e sem perda de precisão técnica. Oferece três níveis de compactação (lite, dev, ultra) e detecta automaticamente conteúdo didático para preservar explicações completas quando necessário.

Instalação

$npx convez-ai raushan1107/compact-mode-claude-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Otimizar custos em conversas longas de debug e troubleshooting
  • Desenvolver APIs e backends com respostas ultra-compactas
  • Alternar entre modo pedagógico e modo operacional automaticamente

Veredito

Skill bem implementado que de fato economiza tokens sem sacrificar clareza técnica. Ideal para quem trabalha com APIs, cloud e debugging — a detecção automática de conteúdo didático é um diferencial honesto.

README

A tradução para o português deste README ainda está na fila.

🗜️ compact-mode — Claude Skill

Cut Claude's token usage by 60–75%. Full signal. Zero fluff.

RR Skillverse GitHub License: MIT Stars

Built by Raushan Ranjan · Senior Trainer & Developer · RR Skillverse


⭐ If this skill saves you tokens, give it a star — it helps others find it!

📢 Share it with your team, in your Discord, or on LinkedIn. The more people use it, the better it gets.


What It Does

Switches Claude into a terse, high-signal response mode — dropping filler, hedging, and pleasantries while keeping every bit of technical substance.

Unlike a generic "be brief" prompt, this skill is context-aware:

PrincipleBehaviour
📚 Learner-facing contentAuto-exits compact mode — course labs, blog posts, training material always get full prose
🧑‍💻 Dev / debug tasksUltra-compact by default
⚡ Why-before-HowLevel-dependent — always on lite, smart on dev, off on ultra
☁️ Cloud / API / backendStack abbreviations pre-mapped

Intensity Levels

Switch anytime with /compact <level>. Default is dev.

LevelWhy-before-HowStyle
lite✅ AlwaysNo filler or hedging. Full sentences. 1-line reason before every answer.
dev⚡ Only if non-obviousDrop articles. Fragments OK. Short synonyms. Stack abbreviations.
ultra❌ OffMax abbreviation. Arrows for causality. Pure output — zero explanation overhead.

Quick Example

Prompt: "Why is my Azure OpenAI call returning 429?"

LevelResponse
lite"TPM quota on your deployment is exhausted — Azure throttles when you exceed it. Implement exponential backoff or increase your TPM quota in Azure AI Foundry."
dev"TPM limit hit. Add exponential backoff or raise TPM quota in AI Foundry portal."
ultra"TPM quota hit → 429. Backoff or ↑ TPM in AI Foundry."

lite includes the reason why. dev skips it if obvious. ultra drops it entirely — pure fix, minimum tokens.


Installation

Option A — ZIP Upload via Claude.ai ⭐ (easiest — no Git needed)

  1. On this repo page, click Code → Download ZIP
  2. Go to claude.ai → Customize → Skills → Add Skill
  3. Upload the downloaded ZIP file
  4. Done — /compact is now available across all your Claude conversations

No terminal. No Git. Works on any OS — Windows, Mac, Linux.


Option B — Claude Code (for developers)

# Windows (PowerShell)
git clone https://github.com/raushan1107/compact-mode-claude-skill.git
cp -r compact-mode-claude-skill\compact-mode $env:USERPROFILE\.claude\skills\compact-mode
# Mac / Linux
git clone https://github.com/raushan1107/compact-mode-claude-skill.git
cp -r compact-mode-claude-skill/compact-mode ~/.claude/skills/compact-mode

Verify by opening Claude Code and typing:

claude skills

You should see /compact listed in the skills table.


Option C — Claude Project Instructions (no Skills feature needed)

  1. Go to claude.ai → Projects → New Project
  2. Click "Set project instructions"
  3. Paste the full contents of compact-mode/SKILL.md
  4. Every conversation in that project now has compact mode active

Option D — API / Custom App

with open("compact-mode/SKILL.md") as f:
    skill = f.read()
# Prepend to your existing system prompt
system_prompt = skill + "\n\n" + your_existing_system_prompt

How to Use — Claude.ai

After installing via ZIP (Option A) or Project Instructions (Option C):

Activate compact mode:

/compact

Or just say any of these naturally in your message:

compact mode      short mode       save tokens
less words        be brief         skip fluff

Switch levels mid-conversation:

/compact lite     ← full sentences, reasons always included
/compact dev      ← default — smart compression
/compact ultra    ← maximum compression, pure output

Turn it off:

stop compact
normal mode
full mode

Example session:

You:    /compact ultra
Claude: [switches to ultra mode]

You:    my react component re-renders on every keystroke
Claude: useCallback or useMemo missing on handler/value passed as prop. Add deps array.

You:    /compact lite
Claude: [switches to lite — reasons back on]

You:    stop compact
Claude: [back to normal]

How to Use — Claude Code

After installing via clone (Option B), open any project in Claude Code:

Activate by slash command:

/compact

Switch levels:

/compact lite
/compact dev
/compact ultra

Check available skills at any time:

claude skills

Auto-trigger — Claude Code will also activate compact mode automatically when it detects you're in a quick fix / debugging flow. No need to type /compact every time.

Example Claude Code session:

You:    /compact ultra

You:    fix the 422 on my POST /transactions route

Claude: Body schema mismatch. amount field expects float, you're sending string.
        Cast before passing: float(request.amount)

You:    /compact lite

You:    why does that cause a 422 specifically?

Claude: FastAPI uses Pydantic for request validation. When a field type doesn't 
        match the model definition, Pydantic raises a ValidationError which 
        FastAPI converts to a 422 Unprocessable Entity response.

Smart Auto-Behaviour

ContextBehaviour
Code debugging / fixultra by default
Cloud platform configdev — exact portal paths kept
API / backend devdev — reasons on non-obvious fixes
Training content / course labsAUTO-EXIT — full prose always
Blog posts / course materialAUTO-EXIT — full prose always
Security warningsAlways full prose, no compression
Destructive actionsAlways full prose, no compression

Session Control Reference

CommandEffect
/compactActivate at default (dev) level
/compact liteActivate at lite level
/compact devActivate at dev level
/compact ultraActivate at ultra level
stop compactDeactivate, back to normal
normal modeSame as stop compact
full modeSame as stop compact

Level persists for the entire session until you change it.


Why This Over Generic "Be Brief"?

Feature"be brief"caveman-skillcompact-mode
Drops filler / hedging✅✅✅
Why-before-How (level-aware)❌❌✅ lite=always · dev=smart · ultra=off
Protects learner-facing content❌❌✅
Stack-specific abbreviations❌❌✅
Context-aware auto-exit❌❌✅
Intensity levels❌✅✅
Works in Claude.ai + Claude Code✅✅✅

Inspired By

caveman-claude-skill by @amanattar


About RR Skillverse

RR Skillverse is a trainer-led learning platform by Raushan Ranjan — Microsoft Certified Trainer with 9+ years in IT training and software development, specialising in Azure, Power Platform, .NET, and AI Agent development.

More tools, courses, and skills at rrskillverse.in.


Support This Project

If compact-mode saves you time and tokens:

  • ⭐ Star this repo — helps others discover it
  • 🔁 Share on LinkedIn / X / Discord — tag @rrskillverse
  • 🐛 Found a bug or have an idea? — open an issue
  • 🤝 Want to contribute? — PRs welcome

Made with 🧠 by Raushan Ranjan  ·  RR Skillverse  ·  GitHub

MIT License · v1.2.0

Ver o original no GitHub

Bom para

Relacionados