🟡 Skills, Plugins & Memory

🟡 What It Is

One Skill is a reusable set of instructions that teaches Claude to perform a specific workflow predictably. It’s a SKILL.md with a frontmatter (name + description) that lives in .claude/skills/.

💡 Why learn

The secret to Skills: they only load when needed. Unlike CLAUDE.md (always in memory), a Skill saves context and appears only at the right time — Claude reads the description and decides to invoke.

🔑 CLAUDE.md vs. Skill

Rule of thumb: if Claude should always know → CLAUDE.md. If it’s for a specific task → Skill.

🟡 What It Is

You don’t need to write the Skill from scratch: the skill skill-creator interviews you and generates the SKILL.md. The key is description well-written — that’s what lets Claude know when use.

💡 Why learn

A well-crafted Skill eliminates repetition: you write the “how to” once, and it reuses it forever. Anthropic’s official guide organizes this into 6 chapters, from the fundamentals to distribution.

🔑 Minimum Skill Structure

---
name: pr-review
description: Use quando pedir revisão de PR, diff ou código alterado.
---
Ao revisar um PR:
1. Leia o diff completo
2. Aponte riscos de segurança
3. Verifique cobertura de testes
4. Classifique os achados por severidade

🟡 What It Is

Plugins bundle skills, commands, agents, and hooks into an installable package. You add a marketplace (a GitHub repository) and install plugins from there with /plugin.

💡 Why learn

Plugins are the fastest way to supercharge Claude Code with community work. E.g., Claude Mem provides memory across sessions; the Vision plugin gives it eyes to read images/screens.

🔑 Installing a Plugin

# 1. adicione o marketplace (repo)
/plugin marketplace add thedotmack/claude-mem
# 2. instale o plugin
/plugin install claude-mem

🟡 What It Is

Besides CLAUDE.md, Claude Code has persistent memory in .claude/projects/<projeto>/memory/, managed by the /memory. O Memory 2.0 / Auto Dream consolidates these memories automatically.

💡 Why learn

Memory across sessions = you stop re-explaining the project every time. The agent remembers yesterday's decisions, corrections, and conventions. That's what turns Claude into a colleague who knows your code.

🔑 How "dream" Works

🟡 What It Is

The CLAUDE.md contains hierarchy: your rules (global), project rules, and local rules work together. A good CLAUDE.md is short, specific, and in bullet points—not a novel. The idea behind a “universal prompt” is to have a reusable skeleton across projects.

💡 Why learn

Bloated CLAUDE.md uses up context in every session. The advanced part is knowing what no put in (what's in the code), and what’s worth gold (commands, conventions, “never do X”).

🔑 Hierarchy + Skeleton

# Comandos
- Dev: `pnpm dev` · Test: `pnpm test` · Build: `pnpm build`
# Convenções
- TypeScript estrito, sem `any`
- Componentes em PascalCase
# Regras
- NUNCA commitar na main; sempre branch + PR

🟡 What It Is

Design skills teach Claude to create truly beautiful UI. The frontend-design (Anthropic) is already included in the kit; the Impeccable is an enriched fork—with 7 domain references always loaded and an “AI slop” detector.

💡 Why learn

The difference between a “generic AI” page and a premium page lies in these skills. Impeccable brings a iteration vocabulary (23 verbal commands) to refine, not just "build a page."

🔑 Impeccable in Brief

When to use Anthropic’s frontend-design: quick “one-shot” UI, without installing anything extra. Impeccable: when you’re going to iterate and want a real design system.