Claude Skills - Coleção Pessoal de Skills Portáteis
JCaraballo113/claude-skills
Skills portáteis para Claude Code com instalação segura e versionamento
Ver no GitHubRepositório com skills reutilizáveis para Claude Code que podem ser instalados globalmente ou por projeto. Inclui ferramentas como geração de artefatos HTML, melhoria de UX e triagem de PRs, com sistema de instalação seguro que não quebra ao mover ou deletar o repo.
Instalação
npx convez-ai JCaraballo113/claude-skills1 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
- Automatizar fluxos de trabalho com skills personalizadas
- Compartilhar skills entre projetos via instalação global
- Gerar visualizações e artefatos HTML interativos
- Melhorar experiência do usuário em aplicações
Veredito

Repositório bem estruturado com sistema de instalação robusto e skills práticas. Ideal para quem quer estender Claude Code com ferramentas reutilizáveis, mas requer familiaridade com linha de comando e conceitos de agents.
README
A tradução para o português deste README ainda está na fila.
claude-skills
Personal collection of Claude Code skills, portable across machines.
Install
Clone anywhere, then run the install script:
git clone git@github.com:JCaraballo113/claude-skills.git ~/Repos/claude-skills
cd ~/Repos/claude-skills
./install.sh # install all skills globally
Each install is a copy, not a symlink. The repo folder and the installed skill are independent — deleting or moving the repo later doesn't break the installed skills.
Global vs project scope
./install.sh --global # default: ~/.claude/skills/ (all projects see it)
./install.sh --project # $(pwd)/.claude/skills/ (just this repo)
Install a single skill
./install.sh triage-pr-comments # global
./install.sh --project triage-pr-comments # project-local
Updating installed skills
When you pull new versions from origin, re-run install to sync:
cd ~/Repos/claude-skills
git pull
./install.sh # re-copies everything
./install.sh triage-pr-comments # re-copy just one
Re-install safely overwrites skills this repo owns (tracked via a .installed-from marker file). It won't clobber skills that came from somewhere else — if there's a name collision and the existing skill isn't ours, install skips with a warning.
Uninstall
./install.sh --uninstall # remove all (global)
./install.sh --uninstall <name> # remove one (global)
./install.sh --uninstall --project # remove all (project)
./install.sh --uninstall --project <name> # remove one (project)
Uninstall only removes skill directories whose .installed-from marker matches this repo — safe to run even if you have unrelated skills in the same dir.
Editing workflow
The installed copy is a snapshot. To change a skill:
- Edit in the repo:
~/Repos/claude-skills/<skill>/SKILL.md - Commit + push
- Re-run
./install.sh <skill>to update the installed copy
Do not edit files inside ~/.claude/skills/<skill>/ directly — those edits will be overwritten on next install and don't get tracked in git.
Skills
| Skill | Description |
|---|---|
| add-agent-skill | Scaffold a new agent skill in this repo. Interviews for name and triggers, writes the frontmatter + outline, installs locally, and suggests a commit. Use when starting a new skill. Named agent-skill to avoid collision with a potential future Anthropic-shipped add-skill. |
| html-artifact | Generate a single self-contained HTML artifact (inline CSS + inline SVG + optional inline JS) that visually explains a system, feature, or codebase area. Output is one file in docs/ that opens offline. Use when the user wants to visualize a system, generate a flow diagram, build a one-pager explainer, or asks to "make an HTML artifact" / "visualize X" / "explain X visually". Good for state machines, planner/decision logic, data flows, and architectural overviews with optional interactive widgets. |
| improve-user-experience | Find "bridging opportunities" — gulfs the user has to cross themselves — using Don Norman's gulf vocabulary (execution / evaluation, signifier, feedback). Walks the flows, presents candidates as a temp-dir HTML report, then grills the chosen one and designs the bridge in Pencil, held to impeccable's bar. Anchored to EXPERIENCE.md (human-owned) + CONTEXT.md. Requires the Pencil MCP and the impeccable skill. Use when the user wants to improve UX, reduce friction, or fix where users get stuck. |
| open-pr | Open a reviewer-oriented pull request from the current branch. Reads diff and commits, interrogates the author one question at a time (intent, type, scope, risk, out-of-scope, ticket), builds a What / Why / Notes-for-reviewers body inline, then pushes and opens via gh as a draft by default. Honors any repo PR template. |
| pr-review-status | Read-only overview of the current branch's PR review comments — groups into addressed / pending / in-discussion / deferred. No edits, no posts. Pair with triage-pr-comments when you want to act on what you see. |
| teach-me | A teaching session that makes sure you deeply understand a specific piece of work in this repo. Grounds itself in the code, git history, and ADRs, then walks you through the problem, the solution, and the broader context one item at a time — restating, drilling the whys, and quizzing — and won't finish until you've demonstrated mastery. Optionally saves durable study notes to Obsidian or Notion. |
| triage-pr-comments | Active triage workflow: classifies each comment into one of five states (valid-fix / partial / invalid / defer / needs-info), asks clarifying questions when ambiguous, implements approved fixes, and gates commit/push/reply on explicit user approval. |
Adding a new skill
- Create a directory at the repo root with the skill name (kebab-case).
- Add a
SKILL.mdfile with frontmatter (name,description) — see existing skills for examples. - Run
./install.sh <name>to install it locally. - Commit and push.
The description field is what Claude uses to decide when to invoke the skill, so make it specific about the triggers (e.g. "use when user says X" / "use when Y condition").