Categories
🛡️ Setup
Foundation and first contact—install, validate, connect, and teach the tone.
Install Claude Code
When to use: first installation on a new machine.
Hi! I want you to install Claude Code on my computer and also install any other dependencies needed to run it here. Let me know when you’re done.
File creation test
When to use: verify that Claude can create folders/files on your system.
Create a folder called "cafe-gelado" on my desktop and write two files inside: one called "por-que-cafe-e-otimo" and another called "por-que-cafe-nao-e-otimo".
Connect/reconnect GitHub via CLI
When to use: when the connection to GitHub dropped or was never established.
I want you to be able to access my GitHub repositories. Connect to GitHub through the CLI and confirm. If you're not connected, update or reestablish the connection, please.
Communication style (instructions for Claude)
When to use: configure how Claude talks to you (goes in CLAUDE.md / system).
Tom direto e informal — nada de enrolação, nada de "corporativês". Sem bullet points a menos que eu peça. Sem travessões (entrega de texto de IA). Nunca use palavras de preenchimento como "certamente", "com certeza", "ótima pergunta" ou "fico feliz em ajudar". Mantenha o texto direto e focado em resultado, não em features. Antecipe problemas (olhe além das curvas), desafie meu raciocínio, seja NÃO-bajulador (não concorde com tudo que eu disser) e seja máximo em busca da verdade. Eu sou: [INSERIR 2 FRASES sobre você/seu trabalho].
🧠 Memory
Organize context into buckets, use a CLAUDE.md for each project, and keep long-term memory.
Organize life into “buckets”
When to use: structure short-term memory from everything you work on.
Based on everything you know about me and all the work I do, I want you to organize my life into six to eight buckets/categories.
Turn the buckets into folders with CLAUDE.md
When to use: turn the buckets into a real folder structure.
Create these six to eight folders for me on my desktop, each with a CLAUDE.md file inside.
Fill in the project CLAUDE.md (brief)
When to use: document the full context of each folder/project.
Adicione um CLAUDE.md em cada uma dessas pastas e depois me faça perguntas sobre cada projeto, para eu te dar o briefing completo. Para cada projeto capture: o que é a pasta; qual é o objetivo (ex.: "maximizar o valor que cada pessoa recebe, crescer a base"); por que ele existe; o stack; as decisões (ex.: "nada de posts gerados por IA"); um mapa de memória; e referências.
“Save memory” rule in CLAUDE.md
When to use: define where to save conversations when you ask to "save/wrap up".
Quando eu pedir explicitamente para salvar, guardar, "fechar" (wrap up) ou lembrar a conversa, salve aqui: [INSERIR LOCAL DA PASTA/WIKI]. Haverá uma subpasta relevante para o tema — escolha a subpasta correta e adicione essa informação nela. (Adicione isto na seção de memória do CLAUDE.md.)
Wrap-up skill for your second brain
When to use: at the end of every conversation, save the essence to the right wiki.
Write me a "wrap-up" skill: whenever I'm having any conversation with any model on my computer, it should capture the essence of that conversation and save it to the relevant wiki in my notes vault. Then rename the skill to "wrap-up".
"Ask" skill (wiki lookup)
When to use: answer questions by consulting your local second brain.
I have a vault folder on my desktop with notes about everything I work on. Create a skill called "ask". Whenever I ask you a question related to anything I do, consult this knowledge base if relevant to give a specific answer, and follow all the instructions in it.
Long-term memory with Pinecone
When to use: vectorize the wiki and have vector-based wrap-up/ask skills.
Build a skill for me with Pinecone. I’ll give you a Pinecone API key. Vectorize everything in my [INSERT] folder and store it in Pinecone (create an index/vault if one doesn’t exist). Then build a wrap-up skill: when I trigger it, capture the essence of the conversation and save it in a specific folder in Pinecone. Add a second skill, "ask Pinecone": when I ask, query that database and bring the information back to me.
⚡ Power features
Skills, MCPs, and subagents — package processes and work in parallel.
Create a skill from an API
When to use: package a repeatable process (e.g., a URL shortener) as a skill.
Quero criar uma skill com você. O processo é: eu te dou [INSERIR ENTRADA, ex.: uma URL] e você [INSERIR AÇÃO, ex.: converte em um link curto via Bitly]. (Se já existir uma skill parecida e você quiser um build limpo, apague-a primeiro.) Me avise quando estiver pronto para a minha API key. Em seguida, com essa API key, construa a skill.
Inspect a marketplace skill
When to use: evaluate a skill before adopting it.
Tell me about the skill [INSERT NAME, e.g., brand guidelines]—I want to understand what it does, then test it and find out whether it works.
Export skills for portability
When to use: take your skills to another environment/model.
Save all my skills to a folder called "Claude skills" on my desktop, with instructions and every detail for each skill, so I can take them to another environment.
Create a local MCP from an API
When to use: expose an API (e.g., YouTube Data) as Claude tools.
I want to create an MCP for [INSERT SERVICE, e.g.: YouTube]. You already have an API key for [INSERT SERVICE] on this system. First, give me a list of the available features. Then build a local MCP for me — the most useful one for me based on everything you know about me. Next, go ahead and build it.
Three subagents with condensed verdicts
When to use: parallel multi-perspective research with a single synthesis.
Quero que você pesquise sobre [INSERIR TEMA] e crie três subagentes. O subagente 1 olha do ponto de vista de [INSERIR ÂNGULO 1]. O agente 2, da perspectiva de [INSERIR ÂNGULO 2]. O agente 3, de um ângulo holístico/todos-os-lados [INSERIR ÂNGULO 3]. Seu trabalho é olhar todos e me dar uma visão condensada, rodando o máximo de iterações possível.
Email drafting routine
When to use: automate inbox responses.
I want you to check all the emails in my inbox and, for any unread ones that don't have a reply or draft, write a reply draft.
Client inquiry workflow (API-triggered)
When to use: respond to leads coming in through the website.
You'll receive information from a client who requested my services through my website. Get their email address and draft a reply. Then schedule a time in a convenient slot between 9 a.m. and 5 p.m. in my time zone.
Critique loop with an external model
When to use: critique and refinement with a second opinion.
Create a critic agent and review the work yourself. When you're done, bring in [INSERT EXTERNAL MODEL] for a fresh perspective and a check.
🌐 Website
From competitor research to SEO, UI/UX, and critical website review.
Competitor research with a judgment matrix
When to use: map and rank the best competitors before building.
Quero criar um site para [INSERIR NEGÓCIO], que [INSERIR DESCRIÇÃO]. Use a conexão do Firecrawl e encontre os melhores/mais bem-sucedidos [INSERIR CATEGORIA]. Vou operar em [INSERIR GEOGRAFIA] (mas pode ser internacional). Identifique uma lista grande e depois o top 5 — pode avaliar do jeito que quiser (Google reviews, ranqueamento de SEO, etc.). Desenvolva uma matriz de julgamento para avaliar isso. Entregue um relatório dos cinco e uma análise profunda do que os cinco melhores têm em comum que os cinco piores não têm, para usarmos esses princípios. Me faça as perguntas que precisar para clarear o entendimento.
Internalize the brief before building
When to use: after saving the research in a melhores-principios.md in the project.
Familiarize yourself with the best-practices document on the left. Let's build a [INSERT TYPE] website together. Let me know when you understand the brief, and then we'll get started.
Build the site inspired by a repo/HTML
When to use: use a design system or downloaded HTML as visual inspiration.
Construa um site para [INSERIR TIPO] usando os melhores-princípios à esquerda, mas com a inspiração de design deste repo do GitHub: [INSERIR URL DO REPO]. (Alternativa: use como inspiração o HTML que eu acabei de baixar no computador.)
Parallel versions in numbered folders
When to use: generate 2–3 variations of the same site and choose the best one.
Crie uma pasta à esquerda chamada "[INSERIR NÚMERO]" e coloque seus designs dentro. (Repita com outros agentes/terminais para ter 2-3 sites em pastas diferentes.) Quando terminar: abra para mim em um localhost.
Save the API key and generate images for the section
When to use: store a key securely and generate images (hero, etc.).
Save this API key securely on my computer so you can access it in the future. Then generate beautiful images of [INSERT TYPE]—or images you think are relevant—for this section you’re implementing here.
SEO optimization from a repo
When to use: research keywords and apply SEO while balancing attention and conversion.
I want to improve this page's SEO. First, understand which keywords I want this page to rank for — strike a balance between capturing attention, converting that attention, and SEO rankings. Use all the insights in this GitHub repo to make that happen: [INSERT SEO REPO URL]. When you're done, give me a detailed list with green check marks of what you did, explaining each improvement; then publish to GitHub and Vercel.
Apply UI/UX principles in depth
When to use: refine an existing website using a design checklist (in parallel with SEO).
Quero usar este repo para melhorar este site — é sobre UI e UX. Olhe todos os melhores princípios e aplique-os sem dó ao meu site real. Vou ter um agente separado cuidando do SEO, então parte do texto pode mudar no caminho.
Critical/adversarial review of the site
When to use: in a clean window, find inconsistencies in copy, design, and UX.
I want you to act as a highly critical reviewer. Go through the site and find inconsistencies, nonsensical text, and poor design, then return with a list of ideas ranked as critical, high, medium, and low. You don't need to invent critical bugs where none exist, but take a real look to help me improve — to make this site as effective, honest, and well-designed as possible. Include any GitHub repo you think is necessary.
📦 Apps
Authenticated backend, payments, rate limiting, and conversion copy.
Lead capture form
When to use: add a lead magnet with questions and save it to the database.
Adicione um formulário de captura de e-mail. Peça o nome, o e-mail e mais uma pergunta — algo como "quanto dinheiro você está faturando", com cinco opções de R$0 até R$1 milhão+. Depois adicione mais uma: "qual é o maior problema que você tem agora?". Quando estiver pronto, salve todas essas informações no Supabase.
Sign-up, login, and authenticated dashboard
When to use: build an authenticated backend with a value dashboard.
Agora quero uma seção de cadastro e login, toda autenticada e gerenciada no Supabase. O usuário cria conta e entra. Ao entrar, quero ver o site dele em um dashboard lindo, no mesmo estilo do site. Mostre métricas diferentes, raspe alguns dados do site dele com um script em background, rode análise de SEO (procure repos no GitHub que façam isso) e deixe essa página o mais valiosa humanamente possível — tudo autenticado e conectado ao Supabase.
Stripe payment integration
When to use: add paid plans (standard/premium) to the app.
Quero integrar o Stripe. No dashboard quero uma seção com cores: standard e premium. O premium é [INSERIR PREÇO] por mês. Me guie no setup do Stripe da forma mais sucinta possível, conferindo a documentação mais recente, com instruções claras e nada mais longas que o necessário. Eu te dou os arquivos necessários. Seu objetivo é exigir o mínimo de mim e, ao mesmo tempo, deixar tudo o mais seguro possível.
Model routing + rate limit (OpenRouter)
When to use: control the cost of the app's AI features.
Para as features de IA, use a API key do OpenRouter. Use o modelo Sonnet até o usuário atingir no máximo 10 mensagens por dia; a partir daí, caia para um modelo mais barato. Implemente rate limits para não dar abuso e fique de olho no gasto total de créditos.
Conversion copy via subagents
When to use: in-depth conversion copy research, critically synthesized.
Spin up a few subagents. Research the best copywriting practices for this type of site in depth, with the goal of converting visitors. See which sites are excelling in this segment, what we can learn, and how to improve our copy to make it as effective as possible. Use multiple subagents; critically analyze their input and synthesize it into a strong result. Be thorough and detailed, but don’t overanalyze.
🧩 Build Anything
Lead scraping, data dashboards, and meta-prompts to get unstuck.
Lead scraping (Apify)
When to use: build a prospect list for outreach.
I want to target [INSERT NICHE] in [INSERT LOCATION]. Use Apify — find the best actor/example you can — and bring me this information for each one: name, email, social media accounts you find, and an interesting fact. Bring me [INSERT NUMBER, e.g.: 10] of them so I can talk to them. Put everything in a polished HTML page for me to view.
"Outlier" content dashboard
When to use: discover viral/outlier content in a niche using metrics.
I want a dashboard that does the following. The outcome I want is to understand which items are outliers — the content that's going viral right now — and, based on that, find out how many views, likes, and comments they got, their engagement, duration, and when they were posted. I want to be able to sort by outlier. And I want to know which creators to follow — actually, suggest a few for me.
"Bottleneck" meta-prompt to get unstuck
When to use: when you don't know the solution and want to think through the problem together.
This is what I'm trying to achieve: [INSERT OBJECTIVE]. Ask me questions to clarify my intentions. And this is the desired result: [INSERT RESULT]. How should I think about the problem more clearly? Then let's think through the solution together.
Validate/enrich emails via API
When to use: clean up a lead list before outreach.
Use a API key do [INSERIR VERIFICADOR DE E-MAIL] e valide todos os e-mails desta lista para mim. API key: [INSERIR].
🎨 Design Systems
Design as code: clone a framework, generate decks in a brand’s style.
Design framework setup (clone and learn)
When to use: start a design-as-code system for slides/sites.
I want to create beautiful slides. Use the repo below: install it, clone it, learn everything about it, and then give me some examples of brands I could use to build beautiful slides. [INSERT DESIGN SYSTEM REPO URL]
Deck in a brand style + images
When to use: generate a presentation aligned with the reference brand.
Let's build a [INSERT N] page deck about [INSERT BRAND]. Go to the brand's site to get some assets and use my [INSERT IMAGE GENERATOR] API key to generate some images that match their brand style perfectly, as you see fit.
Run the brand extraction skill
When to use: encode a brand’s colors, fonts, spacing, and layout.
Baixe isto, descompacte e siga as instruções. (A skill vai te perguntar qual é a marca/site e do que é o seu negócio, e então gerar um design blueprint capturando espaçamento, cores, fontes e princípios de layout.)
🤖 Personal Agent
The "operating system" for agents: install, plan goals, and back up.
Install the agent operating system
When to use: upload the personal agent package (zip) and configure it.
Acabei de baixar este sistema operacional. Abra-o e siga as instruções, e fique à disposição para me ajudar a configurar se for preciso. (Tudo que ele precisa para se configurar e se conectar a este computador está no CLAUDE.md dentro do pacote.)
Mission control: plan goals
When to use: break a medium-term goal down into actionable steps.
This is "mission control" — planning for a medium- or long-term goal. My medium-term goal is: [INSERT GOAL, e.g., "reach 500 subscribers" or "launch an email list with 1,000 people"]. Talk everything through with me, then break this goal into about 4 to 10 individual, sequential prompts — mark which ones are for you (the AI) to execute and which ones are for me — and put together a beautiful roadmap I can follow step by step.
Daily agent backup on GitHub
When to use: protect all of the agent's configuration/memory.
Connect to GitHub and make a daily backup of this entire system, every day, so that—if something goes wrong with my computer or I make a mistake—we can go back in time and restore it.
🔐 Compliance
Security, API costs, and health checks to keep the system running.
Offensive security red team (mega prompt)
When to use: audit the security of a codebase/project before production.
Você é um engenheiro sênior de segurança ofensiva contratado para fazer red-team neste codebase. Seu trabalho é invadir, exfiltrar dados e quebrar o negócio. Seja paranoico e minucioso — não assuma nada. Método: percorra o repositório inteiro de forma sistemática. Para cada arquivo que ler, pergunte: "Se eu fosse um atacante que já tem acesso de leitura, o que eu faria com isto?". Superfícies de ataque a cobrir: segredos e credenciais; segurança de banco (ex.: Supabase) e autorização; LLM / prompt injection; webhooks; dependências e cadeia de suprimentos; logging e exposição. Saída: achados ordenados por severidade, o caminho do exploit e a correção concreta de cada um. Não escreva um scanner — recomende poucos scanners open-source consagrados (ex.: Gitleaks, Semgrep, Trivy), diga quais três adicionar ao meu pre-commit, e assuma que isso barra ~80% dos problemas.
API key via terminal (outside the chat)
When to use: avoid pasting an API key into Claude’s log.
Crie um comando de terminal para eu atualizar a minha API key de [INSERIR SERVIÇO]. (Depois rode esse comando você mesmo no terminal, de modo que o Claude nunca veja a chave em si.)
Fixed API spending cap
When to use: avoid losses if a key leaks.
For any API with variable costs, set a fixed, strict spending cap. Configure the API key with a total cost you’re comfortable with and set the renewal frequency. Limit spending before limiting risk.
System health-check map
When to use: ensure the system keeps working over time.
Pense no meu sistema inteiro e em toda a jornada do usuário, e defina quais devem ser os health checks — o que um sistema funcional teria. Escreva isso como testes para eu saber que continua funcionando. Inclua a checagem de logs de erro e do lado financeiro/pagamentos (ex.: Stripe).
🖥️ Dashboard / OS
Extend the agents' operating system with new tools and sources.
Add an unfamiliar tool to the OS
When to use: when one of your tools wasn't detected by the personal agent.
I use [INSERT TOOL/SYSTEM] and don't see it in the operating system. Can you add it here for me?
Query a knowledge base through an agent
When to use: the agent finds or creates notebooks in an external tool (e.g., NotebookLM).
You have the NotebookLM skill in the "pantheon"—keep it fully up to date. Go to my NotebookLM, find the latest infographic I generated, and then create a new notebook about [INSERT TOPIC].
⌘ Slash commands & trivial commands
They aren’t prompts, but they’re part of the workflow.
/init(high effort) — getting Claude to understand the entire codebase./clear— reset the conversation (avoid "context rot")./context— inspect context window usage./compact— compress the context while preserving what matters./model— switch between Opus / Sonnet / Haiku./cost— see the session’s token usage./agents— manage subagents./memory— open/edit the memory files (CLAUDE.md)./<nome-da-skill>— invoke a skill./newbot(Telegram’s BotFather) — create a bot for your personal agent.