antfu
antfu/skills · Agent Skill
Opinione di Anthony Fu sull'strumentazione e le convenzioni per i progetti JavaScript/TypeScript.
Vai all'installazione ↓Panoramica
| Autore | |
|---|---|
| Repository | antfu/skills ↗ |
| Cartella dello skill | antfu |
| Stelle | 6011 |
| Fork | 338 |
| File inclusi | 6 |
| Licenza | MIT |
| Aggiornato | 23 ottobre 2026 |
| Link | Sorgente dello skill ↗ · Repository ↗ |
Verifica Mozgi
Cosa abbiamo trovato nei file dello skill. I dettagli con l’abbonamento.
Installazione
Uno skill è una cartella con un file SKILL.md. Scegli il tuo agente ed esegui i comandi in un terminale (macOS, Linux o Git Bash su Windows) dalla radice del progetto.
Incolla il prompt in Claude, Codex o un altro assistente: controllerà lo skill e lo installerà.
Install the “antfu” skill from mozgi.io (https://mozgi.io/skills/antfu-antfu/) for me. 1. Download the archive: https://mozgi.io/skills/antfu-antfu/download If it fails, take the skill folder from the source: https://github.com/antfu/skills/tree/main/skills/antfu 2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me. 3. Put the “antfu” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.). 4. Make sure <skills folder>/antfu/SKILL.md exists. 5. Briefly tell me what this skill does and offer to try it on my task. If you cannot do a step yourself, tell me what to click or type.
git clone --depth 1 --branch main https://github.com/antfu/skills.git /tmp/skills mkdir -p .claude/skills cp -r /tmp/skills/skills/antfu .claude/skills/antfu
Riavvia Claude Code: lo skill compare nell'elenco; per richiamarlo a mano usa /antfu. Per tutti i progetti: ~/.claude/skills.
git clone --depth 1 --branch main https://github.com/antfu/skills.git /tmp/skills mkdir -p .agents/skills cp -r /tmp/skills/skills/antfu .agents/skills/antfu
Codex cerca gli skill in .agents/skills dalla cartella corrente fino alla radice del repo; richiamalo con /skills o $antfu. Per tutti i progetti: ~/.agents/skills.
git clone --depth 1 --branch main https://github.com/antfu/skills.git /tmp/skills mkdir -p .cursor/skills cp -r /tmp/skills/skills/antfu .cursor/skills/antfu
Cursor carica lo skill all'avvio; richiamalo con “/” nella chat Agent. Cursor legge anche .claude/skills e .agents/skills. Per tutti i progetti: ~/.cursor/skills.
git clone --depth 1 --branch main https://github.com/antfu/skills.git /tmp/skills mkdir -p .gemini/skills cp -r /tmp/skills/skills/antfu .gemini/skills/antfu
Gemini CLI propone di attivare lo skill quando il compito corrisponde: tu confermi. Per tutti i progetti: ~/.gemini/skills.
Leggi SKILL.md e i file correlati prima di installare. Clicca un file per l'anteprima.
Contenuto di SKILL.md
Testo originale dell'autore · sola lettura> Reference template: antfu/starter-ts ↗. When scaffolding a new TypeScript project, mirror its package.json scripts, eslint.config.js, knip.json, tsdown.config.ts and workflows rather than inventing a new layout.
Coding Practices
Code Organization
- Single responsibility: Each source file should have a clear, focused scope/purpose
- Split large files: Break files when they become large or handle too many concerns
- Type separation: Always separate types and interfaces into
types.tsortypes/*.ts - Constants extraction: Move constants to a dedicated
constants.tsfile
Runtime Environment
- Prefer isomorphic code: Write runtime-agnostic code that works in Node, browser, and workers whenever possible
- Clear runtime indicators: When code is environment-specific, add a comment at the top of the file:
// @env node
// @env browser
TypeScript
- Explicit return types: Declare return types explicitly when possible
- Avoid complex inline types: Extract complex types into dedicated
typeorinterfacedeclarations
Explicitness
Favor explicit, traceable code over implicit "magic". A reader (human or agent) should be able to follow where every name comes from without running tooling.
- Explicit imports: Prefer explicit
importstatements. Avoid auto-imports — when a framework provides them (e.g. Nuxt/Nitro), turn them off for new projects (see app-development ↗). - No path aliases by default: Use relative imports (
./foo,../bar). Only use path aliases (@/,~/,#imports, etc.) when they are *already* configured in the project; don't introduce new ones for greenfield code.
Comments
- Avoid unnecessary comments: Code should be self-explanatory
- Explain "why" not "how": Comments should describe the reasoning or intent, not what the code does
Testing (Vitest)
- Test files:
foo.ts→foo.test.ts(same directory) - Use
describe/itAPI (nottest) - Use
toMatchSnapshotfor complex outputs - Use
toMatchFileSnapshotwith explicit path for language-specific snapshots
---
Tooling Choices
@antfu/ni Commands
| Command | Description |
|---------|-------------|
| ni | Install dependencies |
| ni <pkg> / ni -D <pkg> | Add dependency / dev dependency |
| nr <script> | Run script |
| nu | Upgrade dependencies |
| nun <pkg> | Uninstall dependency |
| nci | Clean install (pnpm i --frozen-lockfile) |
| nlx <pkg> | Execute package (npx) |
Checking npm Package Versions
Use fast-npm-meta ↗ to look up the latest version of a package — it queries a small metadata endpoint instead of downloading the full registry payload (which can be megabytes per package).
nlx fast-npm-meta version vite # 7.3.1
nlx fast-npm-meta version "nuxt@^3.5" # 3.5.22 — range-aware
nlx fast-npm-meta version vite nuxt vue # multiple at once
nlx fast-npm-meta version vite --json # JSON for scripting
nlx fast-npm-meta full vite # full version list + dist-tags
Prefer this over npm view <pkg> version when you only need the latest version, and over reading package.json from the registry directly.
TypeScript Config
{
"compilerOptions": {
"target": "ESNext",
"module": "ESNext",
"moduleResolution": "bundler",
"strict": true,
"esModuleInterop": true,
"skipLibCheck": true,
"resolveJsonModule": true,
"isolatedModules": true,
"noEmit": true
}
}
ESLint Setup
// eslint.config.js
import antfu from '@antfu/eslint-config'
export default antfu({
type: 'lib', // or 'app' (default)
pnpm: true, // enforce pnpm catalogs in package.json
antislop: true, // flag redundant/duplicated AI-style code, ban explicit `any`
})
Always enable antislop: true. It requires eslint-plugin-slop and eslint-plugin-sonarjs as dev dependencies.
For detailed configuration options: antfu-eslint-config ↗
Knip
Use Knip ↗ to catch unused files, exports and dependencies — the linter only sees one file at a time.
// knip.json
{
"$schema": "https://unpkg.com/knip@6/schema.json",
"project": ["src/**/*.ts"],
"ignoreDependencies": []
}
Fix what Knip reports by deleting; only add to ignoreDependencies for tools invoked outside package.json scripts (e.g. taze).
Scripts and the ci Gate
Every project exposes a ci script that runs all checks in one command:
{
"scripts": {
"build": "tsdown",
"lint": "eslint --cache",
"typecheck": "tsc",
"knip": "knip",
"test": "pnpm run build && vitest",
"ci": "pnpm run lint && pnpm run typecheck && pnpm run knip && pnpm run test --run"
}
}
Before every commit, run nr lint --fix to format, then nr ci and make sure it passes. Do not commit with a failing ci; fix the root cause instead of silencing rules or adding ignores.
Git Hooks
simple-git-hooks + nano-staged. The pre-commit hook runs the full ci gate; prepare also installs agent skills via skills-npm:
{
"scripts": {
"prepare": "git config core.hooksPath .githooks && simple-git-hooks && skills-npm"
},
"simple-git-hooks": {
"pre-commit": "pnpm i --frozen-lockfile --ignore-scripts --offline && pnpm run ci && pnpx nano-staged"
},
"nano-staged": {
"*": "eslint --fix --no-warn-ignored"
}
}
Add .githooks to .gitignore and simple-git-hooks: true under allowBuilds in pnpm-workspace.yaml.
skills-npm
skills-npm ↗ symlinks agent skills shipped inside installed npm packages (and fetches those declared in a skills field) into the agent's skills directory on every install. Install as a dev dependency and wire it into prepare as above; add skills/npm-* to .gitignore. Commit the generated skills-npm-lock.json.
Publishing
Prefer npm Trusted Publishing (OIDC) over NPM_TOKEN secrets: releases run on CI from a v* tag via release.yml with id-token: write, and nr release (bumpp) only bumps, tags and pushes. Details: setting-up ↗.
pnpm Catalogs
Use named catalogs in pnpm-workspace.yaml for version management:
| Catalog | Purpose |
|---------|---------|
| prod | Production dependencies |
| inlined | Bundler-inlined dependencies |
| dev | Dev tools (linter, bundler, testing) |
| frontend | Frontend libraries |
Avoid the default catalog. Catalog names can be adjusted per project needs.
---
References
| Topic | Description | Reference |
|-------|-------------|-----------|
| ESLint Config | Framework support, antislop, formatters, rule overrides, VS Code settings | antfu-eslint-config ↗ |
| Project Setup | .gitignore, GitHub Actions, OIDC trusted publishing, VS Code extensions | setting-up ↗ |
| App Development | Vue/Nuxt/UnoCSS conventions, auto-import control, Storybook component testing | app-development ↗ |
| Library Development | tsdown bundling, pure ESM publishing, publint, API snapshots | library-development ↗ |
| Monorepo | pnpm workspaces, centralized alias, Turborepo | monorepo ↗ |
Domande frequenti
Che cosa fa lo skill antfu?
Opinione di Anthony Fu sull'strumentazione e le convenzioni per i progetti JavaScript/TypeScript.
Come si installa antfu?
Apri il blocco “Installazione” in questa pagina e scegli il tuo agente — Claude Code, Codex, Cursor o Gemini CLI — per i comandi pronti che copiano la cartella dello skill nel posto giusto. Oppure copia il prompt dalla scheda “Con l'assistente” e lascia che sia l'assistente a controllare e installare lo skill.
Con quali modelli AI funziona antfu?
Gli skill sono semplici istruzioni in Markdown (formato Agent Skills), quindi antfu funziona con qualsiasi agente che legge SKILL.md: Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode e altri. Il modello carica le istruzioni solo quando il compito corrisponde alla descrizione dello skill.
antfu è gratuito?
Lo skill è pubblicato apertamente da antfu con licenza MIT. Lo skill in sé non costa nulla: paghi solo il tuo fornitore AI per i token usati.
È sicuro installare gli skill?
Uno skill è un insieme di istruzioni che il tuo agente seguirà: leggi SKILL.md e i file correlati prima di installarlo. Gli script nella cartella dello skill vengono eseguiti solo se li esegue l'agente: controlla cosa fa e non concedere permessi superflui.
In cosa differisce uno skill da un server MCP o da un plugin?
Un server MCP o un plugin danno al modello nuovi strumenti: codice che viene eseguito e restituisce un risultato. Uno skill fornisce conoscenza e metodo: come affrontare il compito, quali passi seguire e com'è un buon risultato. Non servono server né chiavi API.