@operatorgrade/cli

Scriptable lookup for the Operator-Grade design system, built for AI agents and CI. The published package reads baked JSON only — no component source, no build step, no runtime dependencies.

Install & reliable invocation

npm i -D @operatorgrade/cli

Add a script so agents and CI invoke it deterministically (no PATH guessing):

"scripts": {
  "operator": "node node_modules/@operatorgrade/cli/bin/operator.mjs"
}

Commands

| Command | Behavior | | --- | --- | | operator init [--all] | Write the AGENTS.md + CLAUDE.md agent cheat sheet (non-interactive, idempotent; --all adds theming + page-building pointers). | | operator component [Name] | List components, or print one's props, intents/sizes, import, do/don't. | | operator tokens [group] | List token groups, or print tokens with resolved Day / Night-NVG / Mono values — e.g. operator tokens semantic.color.threat. | | operator docs [topic] | List doc topics or print one (getting-started, theme-as-mode, standards, …). | | operator template [--list \| <name>] | List @operatorgrade/templates page templates, or print one's usage. | | operator check | Run the zero-hardcoded-values token gate; exits non-zero on a raw value. |

For a live agent connection

operator init gives the buyer's agent a file; @operatorgrade/mcp exposes the same catalog/tokens/docs over the Model Context Protocol for a live connection.

Development (this monorepo)

data/*.json is baked from the component source, token JSON, and docs by npm run build:data -w @operatorgrade/cli (also runs on prepublishOnly). npm test -w @operatorgrade/cli includes a drift check: if source changed without a rebake, it fails.