anthropics/skills in Claude Code: what Design & Test really includes

What is the anthropic-design-test-flow listing, actually

anthropic-design-test-flow is Flowy's listing for anthropics/skills (170,431 GitHub stars, by Flowy's own count), Anthropic's own public repository of example Claude skills. The listing's summary leans on design, testing, and skill authoring, and that part is accurate: frontend-design, webapp-testing, and skill-creator are all in here. But the repository itself describes four kinds of skills, not one, and Flowy indexes the full set as 22 components: 19 skills and 3 agents. If you came here expecting only design and test tooling, there is more waiting once you look at the rest.

The four families inside it

Document skills: docx, xlsx, pptx, and pdf read, create, and edit Word, Excel, PowerPoint, and PDF files. The repository's own documentation says these four are the same skills behind Claude's built-in document creation feature, shared here as source rather than as a separate product.

Creative and design skills: frontend-design pushes new UI toward a deliberate look instead of a generic default. canvas-design produces original poster-style art as a PNG or PDF. web-artifacts-builder is for a claude.ai artifact big enough to need React, Tailwind CSS, or shadcn/ui components and real state, not a single HTML file. theme-factory applies one of ten preset visual themes to an artifact, or builds a new one. brand-guidelines applies Anthropic's own colors and type to whatever is being made. algorithmic-art writes p5.js generative art with seeded randomness. slack-gif-creator builds an animated GIF sized for Slack.

Development skills: mcp-builder guides building an MCP server in Python's FastMCP or the Node and TypeScript MCP SDK. webapp-testing drives a local web app with Playwright to check behavior, capture screenshots, and read browser logs. claude-api is a reference for the Claude API itself: model IDs, pricing, streaming, tool use, caching, and token counting. skill-creator builds, edits, and grades a skill, and it is the one that brings in the three agents below.

Enterprise and workflow skills: doc-coauthoring walks through co-authoring a proposal, spec, or decision document. internal-comms writes status reports and leadership updates in a company's preferred format. discernment-nudge fires after Claude hands over a plan, an estimate, or a factual claim, prompting a second look before anyone acts on it. academy-guide points to a matching Claude Academy course when someone asks how to do something with Claude.

The three agents live inside skill-creator, not on their own

analyzer, grader, and comparator are not general-purpose agents for your own project. All three are sub-agents that ship inside skill-creator, for one job: judging whether a new version of a skill actually performs better than the old one. comparator runs a blind comparison between two outputs, without being told which skill produced which. grader checks a transcript and its output against a stated set of expectations. analyzer reads a batch of comparison results, works out why the winner won, and suggests what to change next. Together they are the eval harness behind skill-creator's own claim to benchmark a skill with variance analysis, not a testing product for anything you build yourself.

Who this is actually for

If you write or maintain Claude skills, skill-creator and its three agents are the reason to install this: a working eval loop rather than a description of one. If you are shaping a UI, a claude.ai artifact, or an MCP server, the development and design families cover that ground on their own. And if what you actually wanted was Word, Excel, PowerPoint, or PDF handling, that is here too, filed under document skills, a fourth category the listing's own name does not mention.

How do I install it

Install from the listing page linked at the top of this guide. Flowy indexes what anthropics/skills ships; it does not own, bundle, or host the repository, so the skills and agents above stay on Anthropic's own releases, and updates come from there rather than from Flowy. The repository carries its own disclaimer worth repeating: these skills are shared for demonstration, the behavior Claude gives you directly can differ from what is shown here, and anything you plan to rely on for real work is worth testing first.