convez.ai
SkillProdutividadeIntermediário
zhitongblog

doaipm — Método Portable de PM para IA

zhitongblog/doaipm-skill

Fale a ideia, IA constrói o produto. Método PM portável para agentes.

Ver no GitHub
InicialPluginScore · de 0 a 100

Uma skill que ensina um agente de IA a converter ideias em produtos reais e funcionais seguindo a metodologia de product manager na era de IA. Inclui cinco fases (Discover → Define → Design → Develop → Validate) e prioriza prototipagem de alta fidelidade com interações reais antes de implementação.

Instalação

$npx convez-ai zhitongblog/doaipm-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
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Prototipagem rápida de aplicações web com Claude Code
  • Disciplinar fluxo de desenvolvimento de IA para evitar detalhes desnecessários
  • Construir produtos shipping com segurança (sem dados reais, sem botões irreversíveis)

Veredito

Skill bem arquitetada que resolve um problema real: como dirigir agentes de IA para construir coisas práticas ao invés de protótipos descartáveis. Prova de conceito sólida com produtos shipping já publicados. Recomendado para quem quer metodologia estruturada ao usar Claude Code e similares.

README

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

doaipm — DO AI PM

Speak it, and AI builds it. 言出法随.

The doaipm method as a portable AI skill — a default working method you can drop into any AI coding tool. It teaches an agent (and you) how to turn an idea into a real, runnable product the way a product manager in the AI era actually works.

🌐 doaipm.com · MIT licensed · works with Claude Code, Codex CLI, Gemini CLI, Cursor, and anything that reads an AGENTS.md.


The method in one breath

Don't ask "what tool should I use?" — ask "how do I describe this?", then build the real, high-fidelity thing.

  • Trust the AI, don't tool-hunt. For almost anything, the model already has an answer.
  • Not knowing how to code is an advantage — focus on what the user needs and say it clearly.
  • High-fidelity first. Skip wireframes; build a real, runnable prototype directly — real content and data, real states (loading / empty / error / success), real interactions.
  • Five phases: Discover → Define → Design → Develop → Validate.
  • Safety net: no real secrets/data in prototypes; the human presses irreversible buttons; don't touch production; ask when unsure.

Full method: SKILL.md.

Install

ToolHow
Claude Codegit clone this repo into ~/.claude/skills/doaipm/ (must contain SKILL.md). Auto-discovered.
Codex CLIClone into ~/.codex/skills/doaipm/, or append doaipm.method.md to ~/.codex/AGENTS.md.
Gemini CLIAppend doaipm.method.md to ~/.gemini/GEMINI.md.
CursorCopy doaipm.mdc into ~/.cursor/rules/ (global) or .cursor/rules/ (per project).
Any other agentDrop doaipm.method.md into the project's AGENTS.md / CLAUDE.md.

One-liner (Claude Code):

git clone https://github.com/zhitongblog/doaipm-skill ~/.claude/skills/doaipm

Try it in 60 seconds

After installing, give your agent a deliberately vague task:

Build me a tip calculator web app.
  • Without doaipm: it dumps a single static HTML file and calls it done.
  • With doaipm: it first asks you 3–5 sharp questions (split the bill? round up? which currency? tax before or after?), then builds a real, runnable high-fidelity prototype — real interactions and real states (empty / typing / error / result), verified by actually opening it in a browser — and it never presses an irreversible button (publish / deploy / pay) without you.

That gap — questions before code, high-fidelity over wireframes, the human keeping judgment — is the whole method. Watch for it on any task.

The proof

This method isn't theory. The doaipm.com site and a row of real, shipping products were all built this way, with Claude Code:

  • SoloMD — Markdown editor
  • Unterm — a terminal AI agents can drive
  • unfetch — download manager for humans & AI
  • StoryAlter — your AI writing alter-ego
  • Unflick — video player for humans & AI

The method explains the work; the work proves the method.

License

MIT © zhitong · doaipm.com

Ver o original no GitHub

Bom para

Relacionados