PowerClaude
oliverstraus246-beep/powerclaude
Claude Code com memória, skills automáticas e roteamento de modelos — pronto em 2min
Ver no GitHubSistema completo de configuração para Claude Code com vault de memória, roteamento inteligente de modelos, automação por hooks e janelas especializadas. Transforma Claude Code de ferramenta básica em ambiente profissional de desenvolvimento em 2 minutos.
Instalação
npx convez-ai oliverstraus246-beep/powerclaude1 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
Casos de uso
- Manter contexto persistente entre sessões de codificação
- Automatizar tarefas repetitivas com skills e hooks
- Otimizar custos roteando tarefas para modelos apropriados (Flash/Sonnet/Opus)
- Gerenciar múltiplos projetos com janelas especializadas
- Centralizar chaves API e configurações de stack
Veredito

PowerClaude é um wrapper sólido que resolve limites reais do Claude Code — memória perdida, re-explicações constantes e desperdício de tokens. A instalação é trivial, mas o valor está na versão paga (personalization engine). Recomendado para devs que querem Claude Code como IDE principal, não apenas como assistente pontual.
README
A tradução para o português deste README ainda está na fila.
You're running Claude Code at 20% capacity.
The other 80% is memory, skill routing, multi-window specialization, hook automation, and model cost control. None of it ships by default. This repo wires it all up.
Before vs. After
| Stock Claude Code | PowerClaude |
|---|---|
| No memory between sessions | Vault loads your projects, decisions, and patterns at every session start |
| Re-explaining your stack every conversation | CLAUDE.md carries your full context -- stack, rules, key paths, active projects |
| Claude improvises every task | 120+ skills auto-invoke based on what you're doing -- no slash commands, no prompting |
| One window doing everything | 11 specialized windows: planning, debugging, code review, UI, infra, each expert at one job |
| Every task runs on Sonnet | Model routing: summarize a file -> Gemini Flash. Real code changes -> Sonnet. Architecture -> Opus. |
| Raw code ships as-written | Hook system: format on save, type check on edit, vault logging on completion, build verify on stop |
| You copy-paste API keys from memory | Centralized key store -- every project pulls from one place |
What You Get -- Free
Install the foundation in under 2 minutes.
- Setup script (PowerShell + bash) -- creates vault structure, installs CLAUDE.md, wires hooks
- CLAUDE.md starter template -- fully documented, every section explained, ready to customize
- Window prompt templates -- planning window, debug window, code review window, general window
- Hook boilerplate -- format on save, incremental type check, vault logging patterns
- Model routing guide -- decision matrix: which task goes to which model, why, and how to wire it
- API key management pattern -- single JSON store, referenced by every project
What You Get -- Paid ($25)
The free tier is a skeleton. The paid tier is the conversation that builds your actual system.
Most setups fail because they're generic. A CLAUDE.md written for someone else's stack doesn't help you. The paid tier includes a Personalization Engine -- a structured prompt you paste into Claude Code. It asks 13 questions about your projects, stack, workflows, preferences, and working style. In return, it outputs a complete, configured environment built around how you actually work.
- Personalization Engine -- answer 13 questions, get a fully generated CLAUDE.md, vault structure, and hook config built around how you actually work
- Full advanced CLAUDE.md -- complete proactive behaviors, full model routing logic, the entire skill auto-invoke library (every trigger condition documented)
- Complete window prompt library -- all 11 specialized windows, production-grade: Planning, Debug, Code Review, UI/Design, Infrastructure, Research, Prompt Engineering, Refactor, Briefing, Security, Handoff
- Full hook library -- pre/post hooks for TypeScript, Python, React, Node; vault-logging hooks that capture decisions and patterns automatically
- Graphify + GitNexus integration guide -- semantic codebase navigation: query your codebase like a database instead of reading files
- Lifetime updates -- every improvement to the system lands in your repo
Install
Windows:
irm https://raw.githubusercontent.com/oliverstraus246-beep/powerclaude/main/install.ps1 | iex
Mac / Linux:
curl -sSL https://raw.githubusercontent.com/oliverstraus246-beep/powerclaude/main/install.sh | bash
The script:
- Creates the vault directory structure under your home folder
- Installs
CLAUDE.md.templateinto~/.claude/ - Wires the baseline hooks into
~/.claude/settings.json
Then open Claude Code. The vault loads automatically on every session start.
After installing, run the validator to confirm everything is wired correctly:
node ~/.claude/validate.js
Personalize in 5 Minutes (Free Tier)
After install, three files need your actual info:
1. ~/.claude/CLAUDE.md (copy from the template)
Copy-Item ~/.claude/CLAUDE.md.template ~/.claude/CLAUDE.md
Open it and fill in every [PLACEHOLDER] section. The template has inline instructions for each one.
2. ~/.claude/hooks/user-prompt-submit.js
Fill in your ACTIVE_PROJECTS array -- name, path, and stack for each project you're actively working on. This injects your project context at the start of every conversation.
3. ~/.claude/api-keys.json
Add your Gemini API key (free at aistudio.google.com) to unlock cheap model routing. Without it, everything runs on Sonnet.
That's it. Everything else auto-loads from your vault.
Why This Exists
Claude Code out of the box is a very capable tool that starts from scratch every single session.
No memory of your stack, your projects, your preferences, or what you worked on yesterday. No automatic behavior — every skill, every model routing decision, every workflow is something you have to ask for explicitly. No specialization — the same Claude handles planning, debugging, code review, and UI work, switching between them mid-session with no focused context.
The gap between that and a tuned setup is not small. It is the difference between a smart assistant you have to constantly manage and a system that works the way you think.
I built this over months of running Claude Code as my primary development environment. Everything in this repo came from fixing something that was actually slowing me down. This is what I use every day.
Get the Full Setup
$25 one-time. Lifetime updates.
Text or CashApp to get access:
- CashApp: $oliverstraus
- Text: 303-946-4224
Send $25 and your GitHub username. I'll add you to the private repo within a few hours.
Not sure if it's worth it for your setup? Text me first -- 303-946-4224. I'll tell you straight.
FAQ
Do I need to know how skills work?
No. The whole point is they fire automatically. You write useEffect and the React skill activates. You see an error and the debugging skill kicks in. You don't manage it.
Does this work on Windows? Yes. The install script, vault system, and all hook patterns are tested on Windows (PowerShell) and Mac/Linux (bash).
What if I already have a CLAUDE.md?
The installer backs it up to CLAUDE.md.backup before installing the template. Nothing gets overwritten without a safety copy.
How does the paid tier work? You pay once ($25), send your GitHub username, and get added as a collaborator on the private repo. No subscription. Pull anytime for updates.
What's the difference between the free template and the paid advanced CLAUDE.md? The free template has every section with instructions and examples -- a solid starting point you fill in yourself. The paid version is the full, production-grade file: every auto-invoke rule, every skill trigger, the complete model routing matrix, and vault conventions. The exact architecture running daily in a production setup.
Do I need Node.js? Yes, for the hooks. Node.js 18+ is required. The installer checks for it and warns if it's missing.
What is the vault?
A folder of Markdown files -- your persistent memory. Claude reads Home.md on every session start. Over time it accumulates your project notes, decisions, and patterns. Think of it as a second brain that loads automatically.
File Structure
powerclaude/
install.ps1 # Windows installer (irm | iex)
install.sh # Mac/Linux installer (curl | bash)
free/
CLAUDE.md.template # Starter CLAUDE.md with instructions
api-keys.json.example # API key store template
hooks/
session-start.js # Injects vault Home.md at session start
user-prompt-submit.js # Injects project context at conversation start
settings.json.example # Hook wiring for ~/.claude/settings.json
vault-templates/ # Seeded vault files (Home, Projects, etc.)
window-prompts/ # Specialized session prompts (general, planning, debug, review)
docs/
vault-setup.md # Vault philosophy and structure guide
model-routing.md # Model routing decision matrix and setup
graphify-setup.md # Semantic codebase querying guide
mcp-guide.md # MCP server setup guide
Troubleshooting
Vault not loading? Hooks not firing? Model routing broken? See TROUBLESHOOTING.md for solutions to every common setup issue.
Contributing
Issues and PRs welcome. If you've built a hook, skill trigger pattern, or window prompt that works well -- open a PR. The goal is the best possible default setup, not a showcase of one person's config.
Built on Claude Code by Anthropic. MIT License -- free tier. Paid tier files are proprietary.