Plugin manifest

.claude-plugin/plugin.json is what Claude Code reads when you run claude plugins add d2c-ai/d2c. It names the plugin and describes what it does; the skills themselves live under skills/ and are discovered by convention.

Fields

Two fields today — both required.

Field	Type	Meaning
`name`	string	Plugin identifier shown in the marketplace list. Must match the repo name / install target; today `d2c`.
`description`	string	Long-form summary. This is the text Claude Code shows on the plugin detail screen, and it’s what ChatGPT-like semantic search indexes against, so the wording matters for discoverability.

If upstream ever adds fields (version pinning, dependency declarations, permission scopes), this page will need a prose update — the raw block below stays authoritative either way.

Raw manifest

The file exactly as served from upstream, pretty-printed:

{
  "name": "d2c",
  "description": "Design-aware iterative code generation from Figma to production-ready frontend code. Supports React/Next.js, Vue/Nuxt, Svelte/SvelteKit, Angular, Solid.js/SolidStart, and Astro. Generates design tokens from your codebase, pulls Figma designs via MCP, generates code matching your project conventions, and visually verifies output via Playwright screenshot comparison — iterating until the result matches the design."
}

Install paths

The manifest is consumed by all three install paths documented on Install:

Plugin — claude plugins add reads this file to register the plugin, then pulls skills/ alongside.
npm skills — the skills CLI reads this file for discovery but installs the individual skill definitions directly.
Manual— the install script copies the entire repo; the manifest comes along for free but isn’t actively used by Claude Code in this mode.