Every skill starts with the same cycle: you describes the workflow in natural language, the agent generates the SKILL.md in the official format, it becomes a slash command (e.g., /draft-email), and from there you reuse — preferably with a clean context, so each run is predictable.

Illustrative diagram — the blue box is the agent writing the SKILL.md. Build once; reuse forever. The dotted loop is reuse with /clear between runs.

What it is

Building a skill starts with a good request. You describes the workflow — what the inputs, what’s the behavior expected, and where to save the result — and asks the agent to turn it into a slash command following the official skill format. It's the same mindset as vibe coding: you say the "what," the agent writes the SKILL.md (the “how”).

A good request is specific in all three dimensions. Too vague ("make an email skill") forces the agent to guess; too specific about technical steps ("create a .md file with YAML frontmatter containing…") throws away the benefit of letting it choose. The sweet spot: describe the inputs, behavior, and destination — and leave the rest to it.

Entre em Plan Mode.

Quero uma skill que .
- Entradas: 
- Comportamento: 
- Onde salvar: 

Crie isso como um slash command seguindo o formato
oficial de skill (SKILL.md com nome, descrição e
instruções). Me mostre o plano antes de gravar.

Why learn

A request in the official format ensures the skill is created for reuse — with a name, description, and instructions the agent will reread every time you invoke it. Asking in Plan Mode gives you a review point: you can correct the scope before it becomes a file.

Key concepts

What it is

Let's build the first real skill: /draft-email. It receives three inputs—the bullets of the content, the recipient e o tone — and writes a polished, concise email without rambling (fluff). The draft is saved in a temporary folder, and the entire procedure becomes the command /draft-email.

---
name: draft-email
description: Rascunha um e-mail polido a partir de
  bullets, destinatário e tom. Use quando o usuário
  pedir para escrever ou rascunhar um e-mail.
---

# Rascunhar e-mail

## Entradas
- destinatário: para quem é o e-mail
- tom: 
- bullets: os pontos a cobrir

Se faltar alguma entrada, PERGUNTE antes de escrever.

## Comportamento
1. Escreva um e-mail conciso e polido, sem fluff.
2. Respeite o tom pedido; assunto curto e claro.
3. Salve o rascunho em /email.md.

Why learn

Email is the perfect example of a repetitive, standardizable task: you write dozens a week, and they all follow the same structure. Encapsulating this in a skill means that the “how to write well” lives in the SKILL.md, and you just provide the content. Build it once, save time forever.

Key concepts

What it is

The robustness test for a good skill: call /draft-email without sending any data. Instead of running blind and making up a recipient and tone, the skill should ask for the missing details—the recipient, tone, and bullets—before writing a single line. That instruction of SKILL.md ("if any input is missing, ASK") is what ensures that.

Illustrative diagram — invoking it without data isn't an error: it triggers data collection. The skill only writes after it has all three inputs.

Why learn

Skills that ask before running prevent automation’s worst nightmare: a plausible but wrong result based on guesses. When the skill collects what’s missing, you control the input—and never get an email addressed to the wrong person in the wrong tone.

Você:
/draft-email

Skill (esperado):
Para escrever o e-mail, preciso de:
1) Para quem é? 
2) Qual o tom? 
3) Quais pontos cobrir? 

(a skill NÃO escreve nada até você responder)

Key concepts

What it is

The real payoff from building a skill comes from reuse. To reuse it well, give it /clear (clear the conversation) and call /draft-email again with new entries. The result: an output fresh and consistent, following the same rules as the SKILL.md, without the previous conversation’s clutter contaminating the new email.

Why learn

Without clean context, the third email of the day might inherit the tone of the first or mention the wrong recipient. Clearing it before each reuse makes every run predictable — the skill always applies the same rules, starting from the same point.

/clear

/draft-email
destinatário: 
tom: 
bullets:
- 
- 
- 

Key concepts

What it is

A more ambitious skill, but built the same way: document → slides. It reads a document and extracts the key points and create a professional presentation using Python tools. The input document is in temporary/resources and the generated presentation comes out in temporary/outputs — organized folders, with input and output kept separate.

Crie uma skill chamada  que:
- Lê um documento de temporary/resources/
- Extrai os pontos-chave do conteúdo
- Cria uma apresentação profissional usando
  ferramentas Python
- Salva o resultado em temporary/outputs/

Faça no formato oficial (slash command). Se faltar
o arquivo de entrada, PERGUNTE qual usar.

Why learn

This skill shows that the same pattern scales: clear inputs, described behavior, defined destination—now with the agent writing Python code under the hood. You don't need to know Python; you need to describe the result. And the separation resources/outputs keeps the project clean and the skill reusable with any document.

Key concepts

What it is

Skills are composed. Instead of describing every visual detail of your presentation, you stacks an existing design skill layered on top of the skill doc-to-slides. The design skill provides the typography, color, and spacing rules; your skill handles the content. The visual quality improves without you having to become a designer.

Illustrative diagram — adding a design skill doesn't replace yours; each covers a different responsibility, and together they improve the result.

Why learn

Composition is what makes your skill library worth more than the sum of its parts. A ready-made design skill becomes a "theme" you can plug into any skill that generates a visual artifact. You reuse design knowledge that isn't even yours—and the result looks like it was made by an expert.

Key concepts

What it is

Now that you know how to build, test, invoke, and reuse, the final step is to accumulate: build your own skills library. Each skill is one of your repetitive tasks, encapsulated once. Below is a catalog of ready-made ideas—not to copy, but to serve as starting point from your own collection.

Catalog of starting points

The life cycle of a skill

Why learn

A skills library is capital that earns interest: every skill you build saves time on every future run, and good skills combine into larger workflows. Start small — one or two skills for your most repetitive tasks — and let the collection grow with use.

Key concepts