A small set of Claude Code skills I actually use, packaged to drop into any project. Curated, not exhaustive — each one earns its place or it gets cut.

A skill is a folder with a SKILL.md: YAML frontmatter plus instructions. Claude Code reads the description to decide when to reach for it; the body guides the run. Heavier material loads on demand from a references/ folder, so the trigger stays cheap. They're small, composable, and model-agnostic — fork them, rename them, make them yours.

1. Install with the skills.sh CLI:

   npx skills@latest add othavi0/skills

2. Pick the skills and the agents you want them on. Add --global to install for every project, or -s dev-up for just one.

3. Run /reload-skills (or restart Claude Code).

4. Invoke by slash command, or let the model-invoked ones trigger on their own.

Each one came out of a real friction, not a "wouldn't it be neat" idea.

Dev servers that trip over each other. Run a few projects at once and the terminals blur together — which tab is which port, who threw that error. dev-up pins one browser tab to a chosen port, arms an error watcher, and stays out of the way of everything else already running.

CLAUDE.md that bloats into noise. Project memory grows by accretion: paths that moved, commands that changed, advice the code already enforces. claude-md-prune runs the Boris Cherny / Anthropic filter — "would removing this cause Claude to make mistakes? If not, cut it" — and flags the lines that drifted out of sync with the code.

Prose that reads like a machine wrote it. Portuguese text from an LLM carries tells: inflated vocabulary, negative parallelism, sycophancy, em-dashes everywhere. humanize-pt-br strips 30+ verified patterns — Wikipedia's Signs of AI writing adapted to PT-BR, plus Strunk.

Skills split on one axis: who can invoke them. User-invoked skills run only when you type them (e.g. /dev-up) — they orchestrate. Model-invoked skills can be called by you or reached for automatically when the task fits — they hold the reusable discipline.

User-invoked

• dev-up — /dev-up <port> — start this folder's dev server on a port, open one pinned browser tab, arm an error watcher, hand control back. Built for running several servers and tabs in parallel without disturbing them. First run on a machine bootstraps the claude-in-chrome connection itself.

Model-invoked

• claude-md-prune — subtractive audit of CLAUDE.md: cut derivable content, flag drift (paths, commands, ADRs no longer matching the code). Triggers when you mention trimming or auditing CLAUDE.md.

Model-invoked

• humanize-pt-br — remove AI tells from Brazilian-Portuguese prose: 30+ verified patterns (inflated vocabulary, impersonal passive, negative parallelism, sycophancy). Scoped to Read / Write / Edit, and Portuguese-only by design.

skills/
  engineering/
    claude-md-prune/   SKILL.md + references/
    dev-up/            SKILL.md + references/
  writing/
    humanize-pt-br/    SKILL.md + references/

• dev-up needs the claude-in-chrome browser connection — on first run it walks through the one-time setup itself (references/setup.md), no separate command.
• Prefer to install by hand? Clone and copy a skill folder into ~/.claude/skills (global) or .claude/skills (one project), then /reload-skills.

Part of the noctua toolset, alongside agent-bar. More at othavio.com.