12tower - Sistema Operacional para Founder Solo
bioztar/tower
Co-fundador IA com 14 especialidades para gestão de startup solo.
Ver no GitHubAgent multidisciplinar que agrupa 14 domínios especializados (ideação, produto, GTM, vendas, marketing, growth, fundraising, operações, finanças, legal) em skills sob demanda. Funciona como um co-fundador IA que roteia problemas de startup para o expert certo. Inclui 3 sub-agents (pesquisa de mercado, análise financeira, due diligence) que rodam automaticamente quando necessário.
Instalação
npx convez-ai bioztar/tower1 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
- Validar hipóteses de negócio e fazer pivotar/matar ideias
- Estruturar financeiro (unit economics, runway, projeções)
- Preparar pitch decks e materiais para investors
- Planejar launches, definir go-to-market e primeiros 100 clientes
- Desenhar PRDs e arquitetura de produtos
- Análise de competidores e mercado via web search automático
Veredito

Produto robusto que automatiza decisões-chave de startup em conversas simples. Excelente para founders solo que precisam de orientação multidisciplinar rápida, mas requer experiência para interpretar recomendações em contextos complexos. A instalação é limpa e o design de towers é intuitivo.
README
A tradução para o português deste README ainda está na fila.
12tower
One-Person Unicorn Operating System for Claude Code
An AI-native startup operating system: 14 domain towers + 3 specialized sub-agents that cover every function a solo founder needs — from idea validation to fundraising. Built for Claude Code.
What It Does
Each "tower" is a Claude Code skill — a domain expert loaded on demand. Ask about pitch decks and you get a fundraising strategist. Ask about churn and you get a customer success expert. Ask what to work on and the Command Center routes you.
| Tower | Command | Domain |
|---|---|---|
| 0 · Command Center | /tower | Orchestration, routing, sprint planning |
| 1 · Idea Validation | /validate-idea | Mom Test, assumption mapping, GO/PIVOT/KILL |
| 2 · Business Model | /business-model | Unit economics, LTV/CAC, pricing |
| 3 · Product & Engineering | /product | PRDs, architecture, MVP design |
| 4 · Go-to-Market | /gtm | Launch strategy, channels, first 100 customers |
| 5 · Sales | /sales | Founder-led sales, pipeline, objection handling |
| 6 · Marketing & Brand | /marketing | Positioning, content, copy, distribution |
| 7 · Growth & Analytics | /growth | AARRR, cohort analysis, experiment design |
| 8 · Fundraising | /fundraise | Pitch decks, investor prep, SAFEs, term sheets |
| 9 · Operations | /ops | OKRs, hiring, automation, board management |
| 10 · Finance | /finance | Burn rate, runway, projections, accounting |
| 11 · Customer Success | /customer-success | Onboarding, churn, NPS, support |
| 12 · Legal & Compliance | /legal | Entity, equity, privacy, IP, compliance |
| 13 · Competitive Intel | /competitive-intel | Market research, competitor analysis |
Sub-agents (invoked automatically by towers when research is needed):
market-researcher— Deep web research: market sizing, competitors, trendsfinancial-analyst— Financial modeling, projections, unit economicsdue-diligence— Investor-grade scrutiny, risk analysis, stress-testing
Installation
Option A: Project mode (recommended)
Clone into a dedicated workspace directory and run Claude Code from there. All skills load automatically.
git clone https://github.com/bioztar/tower.git ~/12tower
cd ~/12tower
claude
Then start:
/tower diagnose
Your decisions and sprint logs persist in decisions/ inside the cloned repo.
Option B: Global install
Makes all 14 skills available in every Claude Code session, in any directory.
git clone https://github.com/bioztar/tower.git ~/12tower
cd ~/12tower
chmod +x install.sh
./install.sh
The installer creates symlinks from ~/.claude/skills/ to the repo. To update later:
cd ~/12tower && git pull
# symlinks update automatically — no reinstall needed
Option C: Manual (per-project)
Copy just the skills you want into your existing project's .claude/skills/ directory:
# Copy specific skills
cp -r ~/12tower/.claude/skills/00-command-center your-project/.claude/skills/
cp -r ~/12tower/.claude/skills/01-idea-validation your-project/.claude/skills/
# Copy sub-agents
cp ~/12tower/.claude/agents/*.md your-project/.claude/agents/
Requirements
- Claude Code (CLI or IDE extension)
- Claude account (Pro or above recommended for sub-agents)
No other dependencies. No API keys. No configuration files to edit.
Usage
First time — run the diagnostic
/tower diagnose
The Command Center asks 5 questions, determines your startup stage (Exploring → Scaling), and recommends the top 3 towers to focus on this week.
Daily — go directly to the right tower
/validate-idea # Stress-testing a new concept
/product # Writing a PRD or planning architecture
/sales # Preparing for a sales call
/fundraise # Building pitch materials
/competitive-intel # Researching a competitor
/finance # Calculating runway or burn
Weekly — sprint planning
/tower sprint
Reviews what shipped, what blocked you, maps priorities to towers, outputs a focused 5-item plan. Appends to decisions/sprints.md.
Cross-domain decisions
/tower decide
Structured decision framework for choices that span multiple domains (e.g., "should I raise now or stay bootstrapped?"). Logs to decisions/log.md.
Status dashboard
/tower status
One-line health check per tower: GREEN / YELLOW / RED, with next action.
How It Works
Stage-aware routing
The Command Center uses a 7-stage model:
| Stage | Signal | Primary Focus |
|---|---|---|
| 0 · Exploring | No clear idea | Tower 1, Tower 13 |
| 1 · Validating | Has idea, no product | Tower 1, Tower 2 |
| 2 · Building | Building MVP | Tower 3, Tower 4 |
| 3 · Launching | Product exists, <10 users | Tower 4, Tower 5, Tower 11 |
| 4 · Finding PMF | 10-100 users, iterating | Tower 7, Tower 5, Tower 11 |
| 5 · Scaling | PMF found, growing | Tower 8, Tower 9, Tower 10 |
| 6 · Expanding | >$1M ARR | All towers, emphasis on Tower 9, Tower 12 |
Cross-tower feedback loops
Every tower watches for signals that belong elsewhere. When sales hears "users keep asking for X," it routes to Product (Tower 3). When finance sees runway tightening, it flags Fundraising (Tower 8). These handoffs are explicit, not accidental.
Persistent state
All significant decisions are logged to decisions/log.md with rationale, reversibility score, and review date. Sprint plans accumulate in decisions/sprints.md. This is your startup's institutional memory — it persists across Claude sessions.
Sub-agent research
When you need real market data, a tower invokes the market-researcher sub-agent to search the web, triangulate sources, and return structured findings. Financial analysis triggers financial-analyst for model building. Due diligence runs due-diligence for investor-grade scrutiny.
Anti-patterns the system flags
- Tower-hopping — Switching focus across too many domains without depth
- Premature scaling — Building ops/hiring before product-market fit
- Fundraising as progress — Treating raising money as a milestone
- Analysis paralysis — Planning without shipping
- Ignoring unit economics — Building without a business model
- Founder bottleneck — Doing everything manually instead of systemizing
Design Principles
- Stage-aware — Advice adapts to your exact stage, not generic startup advice
- Interconnected — Towers hand off to each other with explicit signals
- Opinionated — Uses tested frameworks (YC, Mom Test, AARRR), not generic tips
- Anti-pattern aware — Flags mistakes before you make them
- Decision-logged — Key decisions persist across sessions
- Research-enabled — Real web research, not hallucinated market data
- Action-oriented — Every interaction ends with a specific next step
Updating
cd ~/12tower
git pull
If you used the global installer (Option B), symlinks update instantly — no reinstall.
Contributing
Each skill is a SKILL.md file in .claude/skills/<tower>/. Contributions welcome:
- Better frameworks within a tower
- New cross-tower connections
- Sub-agent improvements
- Bug fixes in prompts
Open a PR with the change and a brief note on what problem it solves.
License
MIT