@felixgeelhaar/glossa-cli

Build-time tooling for Glossa. Walks a source tree, extracts translation keys, syncs them with the API, and pulls bundles to disk for build-time baking. ~620 LOC.

pnpm add -D @felixgeelhaar/glossa-cli
# or
pnpm dlx @felixgeelhaar/glossa-cli init

Commands

| Command | Purpose | |---|---| | glossa init | Scaffolds a glossa.config.json in the current directory. | | glossa scan | Walks source files, extracts <glossa-text key="..."> / <glossa-rich> / <glossa-plural> / <glossa-select> keys + JS/TS string-literal keys, POSTs them to /keys:scan so they exist before translators see them. | | glossa pull | Fetches every locale bundle to disk (default ./glossa/bundles/<locale>.json). Use these as static fallbacks the app loads when Glossa is offline. | | glossa push | Pushes a translated bundle back (translator workflow — for users editing JSON in their fork before submitting back). |

Configuration

glossa.config.json:

{
  "apiUrl": "https://glossa.example.com/api/v1",
  "apiKey": "glossa_...",
  "project": "brotwerk-site",
  "scan": {
    "include": ["src/**/*.{ts,tsx,vue,astro,html}"],
    "exclude": ["src/**/*.test.*"]
  },
  "pull": {
    "output": "./glossa/bundles"
  }
}

Include/exclude globs are composed with the repo's .gitignore via the ignore package, so generated / vendored directories stay out without extra config.

Extraction model

Regex-driven, not AST-driven. The patterns that matter (key="..." attributes, t("...") calls, string literals passed to known SDK methods) are stable enough that a regex pass is both fast and accurate. Each extracted key includes its source file + line number so missing-key diagnostics point straight at the call site.

Honest scope

The CLI is build-time only. Runtime updates come via SSE in <glossa-provider> from @felixgeelhaar/glossa-elements — no polling, no CLI involvement.

License

MIT