Expert-written
Every SKILL.md prompt is hand-authored by domain experts — not auto-generated boilerplate.
Skills for AI agents
Specialized agent skills that streamline common developer workflows and reduce token usage.
/plugin marketplace add allonsy-studio/agent-skills
/plugin install design-system@agent-skills
Agent skills are focused, self-contained capability
packs you can drop into an AI coding agent. Each skill bundles a
SKILL.md (the prompt and trigger description), a
Node.js implementation (where applicable), and a robust test suite — all under one directory.
Claude Code installs them natively via the plugin marketplace. Other
agent harnesses — Cursor, OpenCode, Aider, or anything built on the
Claude Agent SDK — can vendor any skill directory with one
npx command. No vendor lock-in.
Design Reference Tested
Build, name, document, and audit UI components with expert ARIA, accessibility, design token, typography, and i18n guidance written by a Staff engineer with experience building design systems at scale.
These aren't prompt snippets pasted from a gist. Every skill is engineered, tested, and maintained like production software.
Every SKILL.md prompt is hand-authored by domain experts — not auto-generated boilerplate.
Each skill ships a node --test suite with c8 coverage, so behavior is verified, not assumed.
ESLint and Prettier run on every commit for consistent, reviewable code.
LLM eval suites check that each skill triggers and behaves correctly on real prompts.
Tests, linting, and evals run in GitHub Actions before anything merges to main.
Plain SKILL.md directories — install natively in Claude Code or vendor into any harness.
All skills currently published in this marketplace.
Design Reference Tested
Build, name, document, and audit UI components with expert ARIA, accessibility, design token, typography, and i18n guidance written by a Staff engineer with experience building design systems at scale.
Games Implementation Tested Live preview
Cooking, gathering, and inventory management for Dreamlight Valley.
GitHub Implementation Tested
Review, summarize, and manage GitHub notifications via an interactive local dashboard. Unsubscribe from noisy threads, mark individual notifications or your whole inbox as done, all from a single prompt.
SEO Implementation Tested
Inject or generate schema.org structured data (HTML Microdata or JSON-LD) into web markup, then validate it. Picks the right format per project, injects at build/server-render time only, and emits a reviewable diff...
Pick the card for your agent harness. Every install uses the same skill payload — there's no Claude-specific wiring inside SKILL.md.
Recommended
Register this repo as a Claude Code marketplace, then install skills via the /plugin UI.
Drop any skill into your agent's SKILL.md directory. No Claude-specific wiring.
Cursor reads skill folders from your project. Vendor the skill, then point Cursor at it.
Aider, OpenCode, or anything that reads SKILL.md directories. Choose your target folder.
For Node-runtime skills with dependencies, run
npm install in the vendored directory once.
Tool builders can enumerate skills dynamically via the exported getSkills() helper.
import { getSkills } from "@allons-y/agent-skills";
const skills = getSkills();
// returns =>
// [
// {
// name: "design-system",
// path: "/path/to/skills/design-system",
// description: "Build, name, document, and audit UI components...",
// mdPath: "/path/to/skills/design-system/SKILL.md",
// },
// ...
// ]
New skills, improvements, bug fixes, and documentation are all welcome. See CONTRIBUTING.md for environment setup, the test/lint/eval workflow, and the PR checklist.
Some ideas if you're looking for a starting point:
design-system)