What it does
Claude Skills, titled Emdash Skills in its README, is a repository of skill documents, agent definitions and slash commands for AI coding tools. A short prompt, even just a domain name, is routed to the smallest useful subset of its 163 reference docs, which cover product briefs, planning, architecture, building, quality checks, deployment, brand and design. The skills are grouped into 19 numbered categories, from a core operating policy through architecture, the build loop, verification, deploy and runtime, content, design system, motion, media, growth and observability, to MCP authoring. Named agents such as architect, security-reviewer, visual-qa and test-writer are assigned to Opus, Sonnet or Haiku model tiers. The README says every deploy has to clear hard gates, including Playwright end-to-end tests at six breakpoints, a visual QA score of at least 8 out of 10, a Lighthouse accessibility score of at least 95 and zero axe-core violations. The documented stack targets Cloudflare Workers with Hono, Drizzle and Bun. It can be installed as a Claude Code plugin, as an npm or JSR package, or by cloning the repository for Codex. The repository also generates rule files in the native format of tools such as Cursor, Windsurf, Copilot, Cline, Aider, Gemini CLI and Zed.
Capabilities
- 163 reference docs in 19 skill categories, loaded selectively by a skill router
- Named agents mapped to Opus, Sonnet and Haiku model tiers
- 20+ slash commands such as /saas, /run-evals and /audit-mcp-fleet
- Hard deploy gates for end-to-end tests, visual QA, Lighthouse and axe-core
- Rule-file variants for 32 AI coding tools
- MCP authoring templates, including an OpenAPI to MCP forger
When to use it
- You want an AI coding tool to take a one-line prompt through planning, build, QA and deploy
- You deploy to Cloudflare Workers with Hono and want matching conventions
- You use several AI coding tools and want one shared set of rules across them