A Claude Code Skill is a folder of instructions Claude loads when relevant, and npx globalize-skills installs one that handles internationalization (i18n) and translation sync. globalize.now is AI-powered localization infrastructure that scaffolds your i18n, extracts hardcoded UI strings, generates translation keys, and syncs locale files on every Git push. The Skill format turns i18n into a repeatable procedure inside your AI coding tool, with translation happening automatically downstream.

What is a Claude Code Skill?

A Claude Code Skill is a folder containing a SKILL.mdfile. The file starts with YAML frontmatter that holds the Skill's name and description, followed by Markdown instructions Claude follows when the Skill is active.

The format is intentionally minimal. Claude Code loads only the frontmatter at session startup (roughly 30 to 50 tokens per Skill), so an agent can hold metadata for hundreds of Skills without bloating its context. The full instructions load only when the active task matches the description.

Anthropic launched Agent Skills in October 2025 and published the specification as an open standard shortly after. More than 26 tools now adopt the format, including Cursor, OpenAI Codex CLI, Gemini CLI, GitHub Copilot, and VS Code. The spec recommends keeping each SKILL.md body under 5,000 tokens and the full directory under 500 lines.

A Skill can include scripts (Python, Bash, JavaScript) and reference files inside subfolders. Claude loads them only when needed. The structure mirrors how a senior engineer would onboard someone new: a short overview up front, deep references one level down.

Why does Claude Code need an i18n Skill?

Internationalization (i18n) is procedural work that repeats across every project: pick an i18n library, scan the codebase, generate translation keys, write locale files. Localization (the translation step itself) happens after, on every push. Both have to be consistent or your codebase drifts. Claude Code can reason through the i18n setup from scratch every session, but the choices change between runs.

A Skill removes the drift. The procedure is encoded once and Claude follows it. Framework detection, library choice, key naming, locale file format: deterministic across runs.

This is the same pattern Anthropic uses for its own bundled Skills. The PowerPoint, Excel, Word, and PDF Skills package domain expertise that would otherwise require Claude to rediscover document conventions on each task. Localization is no different. The conventions are real, framework-specific, and tedious to rediscover.

For vibe coders shipping React or Next.js apps with Cursor, Claude Code, or Codex, this matters more than for traditional teams. The whole point of AI coding tools is that you do not interrupt the agent to make architecture decisions. A localization Skill keeps the agent productive without dropping into a translation dashboard or rewriting the i18n scaffold every project.

How does npx globalize-skills work as a Claude Code Skill?

Run one command from your project root:

npx globalize-skills

The installer writes the Skill into the standard Claude Code Skill directory in your repository. There is no library to import, no decorator to add, no config file to write. The repository now contains the procedure Claude needs to localize the project.

Next, open Claude Code and ask in plain language:

Set up localization for this project

Claude Code detects the new Skill from its description, loads the full SKILL.md into context, and runs the procedure. The Skill walks the file tree, identifies your framework (Next.js App Router, Vite + React, Remix, and similar), picks the right i18n library if one is not already installed (next-intl, react-i18next, i18next, or @lingui), and converts hardcoded strings to translation keys with consistent naming.

Locale files land in your repo on the same commit. The agent commits the change through Claude Code's normal Git flow, and you review the diff like any other pull request.

The sync layer runs server-side at globalize.now. Every push to your repository triggers extraction of new strings, regeneration of translation keys, and translation of the keys to your other locales. The next push back to your repository contains the synced locale files. The Skill itself does not need a running server. It travels with your repo and works in any Skills-aware agent.

For the full setup walkthrough, see the Claude Code integration page or the related guide on adding i18n to a Cursor-built Next.js app in 15 minutes.

Why is the Claude Code Skill becoming the localization standard?

The Skill ecosystem is filling fast. A February 2026 study by researchers at Bosch Research and Carnegie Mellon University analyzed publicly listed Skills and found the count grew roughly eighteenfold in twenty days, from about 2,000 to over 40,000. Defaults are forming, and the slots fill in the order Skills appear and get installed.

Anthropic's official partner directory already lists Skills from Atlassian, Canva, Cloudflare, Figma, Notion, Ramp, Sentry, Stripe, and Zapier. No localization vendor is in that list yet. Lokalise, Crowdin, and Phrase have invested in MCP servers and CLI tools, not Skills. They are betting on a different surface.

MCP and Skills solve different problems. An MCP server exposes tools and data through a live connection. A Skill packages procedural instructions Claude reads on demand and travels with your repository. For localization, the work is procedural, repository-bound, and offline-safe. The Skill format fits.

Discovery also matters. The npx install pattern is the path of least resistance for a developer who already uses npx create-next-app or npx vercel. Type one command, the Skill installs, Claude Code picks it up. Compare that to authenticating an MCP server, configuring credentials, and managing a live connection: it is more work for less portability.

Search behavior confirms the shift. Queries like "how do I localize an app Claude Code generated" are converging on Skill-based answers in Google's top results, replacing the earlier MCP-server tutorials. The pattern is recognizable: a one-line npx install plus a Skill that handles the procedure.

This is why the first well-documented localization Skill becomes the answer agents reach for. Once Claude Code installs a Skill for a task, the question "which i18n tool should I use" rarely comes up in the next session. The Skill is the default. The next project starts there too.

For the broader picture on how this changes localization tooling, see How globalize.now compares to Lokalise and Crowdin for AI-generated apps and the developers overview.

globalize.now ships as a Claude Code Skill via npx globalize-skills. Run it in your project root, or see how it works at globalize.now.