What is an agent skill?

An agent skill is a packaged set of instructions and references that teaches an AI coding agent—like Claude Code—how to perform a specific task. Each skill loads on demand, giving the agent focused expertise for frontend, backend, writing, design, and more.

How do I install every skill in this catalog?

Run npx skills add xingkaixin/skills in your terminal. The command downloads the full catalog and installs each skill into your AI coding tool, making all of them available to your agent at once. No extra configuration is required.

How do I install a single skill?

Run npx skills add xingkaixin/skills --skill react-state-management-decisions, swapping react-state-management-decisions for any skill slug. Only that skill is installed, which keeps your agent's context lean when you need just one capability.

Which AI tools are supported?

These skills target Claude Code, Codex and other agents that follow the open Agent Skills format. Any coding tool able to read a SKILL.md instruction file can load them, so the catalog stays portable across compatible AI assistants.

Agent skills catalog

Agent skills for coding, writing and design.

32 installable skills for Claude Code and Codex, covering code review, frontend development, writing, design and releases. Choose a task and copy its install command.

Install the whole catalog

npx skills add xingkaixin/skills

or add one: --skill <slug>

32 skills

designself

Layout, typography and color

visual-design-principles

Apply grouping, alignment, repetition, contrast, typography, and color to interfaces, posters, slides, and documents, then inspect and refine the actual result.

updated 2026-09-14 writingupstream

Natural prose editing

ai-avoid

Edit prose to remove inflated vocabulary, formulaic contrasts, false suspense, and repetitive sentence rhythms while preserving meaning.

updated 2026-09-11 writingadapted

Article illustration prompts

article-illustrator

Find where an article needs illustrations and write image prompts by information type, visual style, and optional palette. Includes educational diagrams and black-ink visual notes.

updated 2026-09-11 writingadapted

Article cover image prompts

cover-image

Turn an article into a cover-image prompt with configurable type, palette, rendering, text, and mood. Choose from 11 palettes and 8 rendering styles, with color instructions kept out of visible image text.

updated 2026-09-11 backendself

CLI design for AI agents

agent-friendly-cli-design

Design a CLI that AI agents can actually drive: resource and verb layering, non-interactive defaults, token-lean output, and errors that teach the agent how to retry.

updated 2026-07-18 ciself

Bilingual release notes

bilingual-release-changelog

Ship a release with Chinese and English changelogs in sync — commit-range analysis, user-facing wording, version bumps per stack, and recovery when a release half-fails.

updated 2026-07-18 deployself

Cloudflare deployment and migration

cloudflare-deploy-migration

Move a site onto Cloudflare — static to Pages, Next.js via vinext, full-stack to Workers with D1 and KV — with the wrangler config splits and pitfalls mapped out.

updated 2026-07-18 workflowself

Code review checklist

code-review-checklist

Review a diff in layers — design first, then call sites, then naming, then actually run it — and know what to skip and how to phrase the hard feedback.

updated 2026-07-18 workflowself

Codebase audit workflow

codebase-review-loop

Audit a codebase through numbered improvement proposals: evidence-backed findings, a read-only recheck before fixing, one fix at a time, status written back.

updated 2026-07-18 backendself

Coding agent session formats

coding-agent-session-formats

Parse local session files from Codex, Claude Code, Kimi, OpenCode, and Pi — storage paths, message schemas, a unified export shape, and cache-aware cost math.

updated 2026-07-18 workflowself

Implementation planning

decision-complete-plan

Write an implementation plan another agent can execute without asking anything: explore before you ask, separate facts from preferences, state every assumption.

updated 2026-07-18 designself

UI from a design system

design-system-as-input-ui

Stop AI-generated screens from each inventing their own blue. Feed the agent a design system as a token contract instead of describing pixels every time.

updated 2026-07-18 desktopself

Electron app releases

electron-desktop-release

Package, sign, and notarize an Electron app for distribution outside the store, and keep its secrets somewhere that is actually a secret store.

updated 2026-07-18 writingself

English to Chinese literary translation

en-zh-literary-translation

Translate English prose into Chinese that reads as if it were written in Chinese, using the faithful/fluent/elegant framework and concrete rules for killing translationese.

updated 2026-07-18 designself

Actionable error messages

error-message-writing

Write error messages that answer what happened, why, and how to fix it — with four anti-patterns to avoid and a nine-point check for each message.

updated 2026-07-18 frontendself

Frontend performance for large datasets

frontend-large-data-performance

Make a frontend survive 50-300MB files and streaming data, with benchmark-driven fixes for quadratic batching, full re-derivation, and virtual-list conflicts.

updated 2026-07-18 ciself

Git commit messages

git-commit-message

Write commit messages that carry the why a diff cannot: the seven-rule convention, with a good and a bad example for each.

updated 2026-07-18 iosself

On-device vision and ML for iOS

ios-on-device-vision-ml

Build on-device visual recognition on iOS: viewfinder crop geometry, Vision foreground segmentation, classification and color, and evaluating models on synthetic data.

updated 2026-07-18 iosself

iOS audio recording integration

ios-recording-system-integration

Wire iOS recording into Live Activities, the Action Button, and background audio, starting from the four hard platform limits that shape the whole design.

updated 2026-07-18 iosself

iOS widgets and App Store releases

ios-widget-appstore-shipping

Ship a WidgetKit extension and get the app through review: App Group data sharing, refresh pitfalls, encryption declarations, and real rejection post-mortems.

updated 2026-07-18 workflowself

Issue-driven development

issue-driven-dev-loop

Work a backlog end to end: pick up an issue, settle the design, implement, open the PR, update status after merge, move to the next.

updated 2026-07-18 workflowself

Markdown issue tracking

issue-tracker

Track issues in Markdown inside the repo itself — milestones, status lanes, PR links, and an HTML board — with no Linear or GitHub Issues involved.

updated 2026-07-18 backendself

Monorepo architecture

monorepo-architecture-principles

Organize a monorepo that stays organized: a one-way dependency graph over apps and packages, ten recurring principles, and the anti-patterns worth naming.

updated 2026-07-18 frontendself

React state management

react-state-management-decisions

Decide where each piece of React state belongs — useState, useRef, useMemo, Context, Zustand, or React Query — from a decision tree instead of habit.

updated 2026-07-18 desktopself

Tauri menu bar agent apps

tauri-menubar-agent-app

Build a macOS menu-bar app in Tauri v2 that watches external coding agents: native window layering through objc2, Unix-socket hook bridges, event normalization.

updated 2026-07-18 backendself

Project stack scaffolding

tech-stack-scaffolding

Bootstrap a new project on house conventions across four stacks — web, browser extension, Python CLI, Go CLI — with drop-in configs and script naming.

updated 2026-07-18 writingself

Structured deep reading

triple-mind-deep-reading

Read a long article or talk through three deliberate lenses — Challenger, Builder, Practitioner — instead of flattening it into a summary.

updated 2026-07-18 workflowself

Writing agent skills

writing-agent-skills

Build an agent skill that actually triggers and delivers: pick the type, write directive instructions, engineer the description as a routing contract, then eval it.

updated 2026-07-18 writingself

Save the Cat story structure

save-the-cat-writing

A Save the Cat checklist for tech writing: sharpen the topic, test the headline, rescue a flat opening, and cut hard before publishing.

updated 2026-04-24 writingself

Chinese article writing

klip-writing

Write a KLIP: a structured technical design document that carries one decision from background and motivation all the way to acceptance criteria.

updated 2026-04-21 backendself

Modern Go development

use-modern-go

Write Go that matches the version the project actually targets, reaching for slices, maps, and cmp instead of the legacy patterns they replaced.

updated 2026-03-12 financeself

Stock research reports

stock-report

Generate a fixed-format HTML report for a single A-share stock through this repository's ds CLI data pipeline.

updated 2026-03-11

FAQ

What is an agent skill?: An agent skill is a packaged set of instructions and references that teaches an AI coding agent—like Claude Code—how to perform a specific task. Each skill loads on demand, giving the agent focused expertise for frontend, backend, writing, design, and more.
How do I install every skill in this catalog?: Run npx skills add xingkaixin/skills in your terminal. The command downloads the full catalog and installs each skill into your AI coding tool, making all of them available to your agent at once. No extra configuration is required.
How do I install a single skill?: Run npx skills add xingkaixin/skills --skill react-state-management-decisions, swapping react-state-management-decisions for any skill slug. Only that skill is installed, which keeps your agent's context lean when you need just one capability.
Which AI tools are supported?: These skills target Claude Code, Codex and other agents that follow the open Agent Skills format. Any coding tool able to read a SKILL.md instruction file can load them, so the catalog stays portable across compatible AI assistants.