Claude Skill Hub

Scaffold a pipeline where editing one line in a GitHub file triggers a scheduled Claude run that writes a post and publishes it to Hugo + GitHub Pages.

Scaffold a daily collect-and-triage pipeline — pull from RSS, Reddit, HN and GitHub, have Claude rate each item act-now / backlog / learn / ignore, then ship it to a site.

On-demand comparison posts — add one line like "REST vs GraphQL" to a file, and the push triggers Claude to generate an SVG diagram plus comparison table and deploy it. No cron, no fallback content.

Turn pasted class transcripts into structured English-study posts — idioms with examples, vocabulary, spoken-mistake corrections, fill-in-the-blank quizzes and a mini diary — published on push.

A Claude Code harness — 14 agents and 15 skills — that takes a topic to a finished EPUB: research (web/paper/community) → plan → chapter drafting → editing → continuity and fact-check → cover → EPUB build, with review gates between stages. Driven by the book-writing-orchestrator skill.

Write a business-idiom section as collapsible blocks — one bold English line and a single key bullet per expression, nothing else.

Write Cisco CCNP/CCIE knowledge-base docs in a fixed structure — definition, characteristics, components, packet flow, comparison tables, config and exam points — with Mermaid diagrams.

Author CISA (IS audit) best-practice guides across the six ISACA domains — a standard 10-section layout with Mermaid diagrams and HTML audit checklists.

Compare and choose between modern data platform architecture styles — Lakehouse, Medallion, Data Mesh, Data Fabric, Lambda/Kappa, Streaming Lakehouse, Modern Data Stack and Zero-ETL — then write the decision up with the constraints that drove it.

Explain any topic, code, concept or error to a named audience — a five-year-old, a manager, a parent, a new engineer — matching vocabulary, analogies and depth to who is actually listening. Triggers on "explain like I'm…", "dumb it down", "how do I tell my boss…".

Audit and optimise a site across five lanes — SEO, AEO, GEO, LLMO and NEO (Naver) — so it becomes a primary source that search, answer engines, generative AI and Naver's AI briefing cite. Diagnose → implement → measure, with a crawler's-eye check (what the HTML shows without JS), a no-black-hat rule, and a blog/content pipeline. Bilingual references.

Rewrite Korean text that reads as AI-written — translationese, over-quoted English, mechanical parallelism, passive overuse, uniform rhythm, emoji and bullet excess — across 10 categories of tells, changing style and rhythm without touching the content. Routes to 1, 2 or 3+ passes by severity.

Write internal communications in house formats — status reports, leadership updates, newsletters, FAQs, incident reports and project updates.

Write framework and methodology content in the Korean 기술사 exam format — a 3-section structure with block diagrams, Mermaid visualizations and comparison tables.

Write up an ICT incident, error log or troubleshooting session as a structured problem note for a Quartz v5 second brain — plus concept notes, MOC pages and a masking check before anything goes public.

Add expression-practice pages to a Docusaurus English-learning site — thematic collapsible groups, bold expressions with key/source/similar bullets, tip admonitions and sidebar wiring.

Generative art with p5.js — seeded randomness, flow fields, particle systems, and an interactive parameter-exploration loop.

Build a web animation from scratch by making the decisions in the order that settles whether it feels right — whether to animate at all, which tool, which properties, which curve, how it interrupts and exits — then write the implementation.

The same animation decision order for React Native and Expo — which thread it runs on, spring or timing, how the gesture hands off, how it degrades — implemented with Reanimated, Gesture Handler, Expo Router and expo-haptics.

Reverse-lookup glossary that turns a vague description of a motion effect into its exact term — "the bouncy thing when a popover opens" becomes Pop in, "the iOS rubber-band scroll" becomes Rubber-banding.

Stop coding agents from shipping generic UI — ground interfaces in UIZZE's library of 800k+ real web and iOS screens, define a design contract, cover required states, and run a hard pre-ship finish gate. Works in Claude Code, Cursor, Copilot, Codex.

Apple's approach to interface design and fluid, physical motion translated for the web — gesture-driven UI, spring and interruptible transitions, translucent materials and depth, optical typography, reduced-motion.

A curated collection of 67 DESIGN.md / SKILL.md design systems — minimalism, brutalism, glassmorphism, luxury, dashboards and more — each with WCAG 2.2 AA typography, colour and spacing so agents ship magazine-grade layouts.

Banners for social, ads, website heroes and print — multiple art directions per brief across Facebook, X, LinkedIn, YouTube, Instagram and Google Display sizes.

Apply Anthropic's official brand colors and typography to any artifact that benefits from a consistent house look-and-feel.

Premium brand-guideline boards, logo systems and identity decks — minimalist, cinematic, editorial, dark-tech and luxury systems with art-directed mockups.

Brand voice, visual identity, messaging frameworks and asset management, with compliance checks for branded content.

Design posters, art and static pieces as .png or .pdf using a real design philosophy rather than default template output.

Art direction and visual design decisions for product work, backed by the Pro Max reference data.

Design mobile app screens through the hosted sleek.design API — create a project, send the intent as one message with an authored style direction, then poll the run and pull screenshots or HTML. Needs a SLEEK_API_KEY; sustained use is a paid plan.

Three-layer token architecture (primitive → semantic → component), CSS variables, spacing and type scales, and component specs.

Anti-slop frontend for landing pages, portfolios and redesigns — read the brief, infer the right direction, and ship interfaces that don't look templated.

Emil Kowalski's philosophy on UI polish, component design, animation decisions and the invisible details that make software feel finished.

Search a codebase or UI for the places that don't animate but should, and reject everything that shouldn't. Read-only — it proposes motion with exact values rather than implementing it.

Design like a high-end agency — the exact fonts, spacing, shadows, card structures and animations that make a site feel expensive, with the cheap-looking defaults blocked.

Turn one reference image into an animation-ready, procedural Three.js model built entirely in code — staged, quality-gated generation for object rebuilds and stylized characters, rendered live in the browser and orbitable by mouse. No 3D modeller needed.

A design language that pulls agent UI out of default-SaaS territory (Inter, purple-blue gradients) — 23 named commands like /critique and /colorize plus automatic checks, so you steer toward a real brand look with explicit instructions instead of "make it nicer".

Survey a codebase's motion code as a senior motion advisor and produce a prioritised audit plus self-contained implementation plans other agents can execute. Read-only on source — it plans the fixes, it doesn't apply them.

Raw mechanical interfaces fusing Swiss typographic print with military terminal aesthetics — rigid grids, extreme type contrast, analog degradation.

Clean editorial interfaces — warm monochrome palette, typographic contrast, flat bento grids, muted pastels. No gradients, no heavy shadows.

Generate premium app-native mobile screen concepts and flows — clean hierarchy, multi-screen consistency, controlled palettes, framed in a subtle phone mockup. Images only, no code.

Build several genuinely different versions of a UI piece behind a visual picker, so you can flip through them live and promote the one that feels right. Explicit invocation only.

Upgrade an existing site or app to premium quality — audit the current design, name the generic AI patterns, and raise the bar without breaking functionality.

Review motion code against a high craft bar drawn from Emil Kowalski's design engineering standards. Defaults to flagging — approval is earned.

Generate agent-friendly DESIGN.md files for Google Stitch — strict typography, calibrated color, asymmetric layouts, perpetual micro-motion and GPU-friendly performance rules.

Build HTML presentations with Chart.js, design tokens, responsive layouts and copywriting formulas chosen per slide's job.

Style any artifact — slides, docs, reports, landing pages — with one of 10 preset color/font themes, or generate a new theme on the fly.

Build accessible interfaces with shadcn/ui on Radix + Tailwind — themes, dark mode, responsive layout and consistent styling patterns.

Design intelligence for web, mobile and desktop interfaces — a searchable local corpus of 79 styles, 192 palettes, 74 font pairings, 119 UX guidelines, GSAP presets and per-stack implementation notes.

Generate premium landing-page design references — one horizontal image per section, enforced composition variety, a narrative concept spine and one consistent palette throughout.

Behavioural guidelines that head off the coding mistakes Karpathy flags in LLM output — overcomplication, sprawling edits, unstated assumptions, and success criteria nobody can verify.

Architecture, sequence, data-flow and state diagrams as explorable standalone HTML with inline SVG — light/dark themes, optional trace motion, and PNG/SVG/WebM export. Takes plain language or pasted Mermaid, and can read the repo when the diagram must match real code.

Build elaborate multi-component claude.ai HTML artifacts with React, Tailwind and shadcn/ui — for artifacts that need state, routing or a component library.

Working guide to Sonner, the React toast library — wiring the Toaster, picking the right toast() call, promise and loading toasts, styling and theming, and the failures people actually hit: toasts that never appear, appear twice, lose Tailwind classes or sit behind a modal.

Turn a book, document or folder of sources into a skill of its own — frameworks, principles, techniques and anti-patterns extracted into SKILL.md plus on-demand chapter files, instead of pasting the whole book into context.

Explore intent, requirements and design before any implementation, so creative work starts from a settled brief rather than a guess.

Drive the user's real, logged-in Chromium from an agent — work happens in a separate Agent Window, user tabs must be explicitly borrowed, and sessions you are already signed into just work. Needs the bsk CLI and browser extension installed first.

A no-mercy, hyper-critical senior-engineer code review. Paste a snippet, file, PR or diff and ask for a roast — it tears the code apart with engineering-grade scrutiny. Triggers on "be harsh", "no sugarcoating", "what sucks".

Unvarnished technical criticism — Linus Torvalds' precision, Gordon Ramsay's standards, James Bach's BS-detection. Surgical, no sugar-coating: what is broken in your code, tests or technical decisions, and why. Ships assessment scripts and rubrics.

A curated library of 78 offensive-security skills for authorized red-team and pentest work — web (SQLi, SSRF, XSS, SSTI, deserialization…), cloud, containers/K8s, AD, wireless, exploit development, EDR evasion and more, each a SKILL.md priming Claude with methodology for one attack surface.

Every's official Compound Engineering plugin — 35 skills that turn each solved problem into durable, reusable repo knowledge: capture non-obvious reasoning (ce-compound), plus review, debug, planning, commit/PR and brainstorm workflows built around it.

Draw concepts, architecture, flows and whiteboards in three output modes — clean SVG for explanatory and process diagrams, architecture SVG for software and cloud infrastructure, and Excalidraw for a file someone can keep editing.

Split 2+ tasks that share no state or ordering across parallel agents instead of running them in sequence.

Work through a written implementation plan in a separate session, with review checkpoints between steps.

Vercel's official skill-discovery helper — when you ask "is there a skill for X" or "how do I do Y", it finds and installs matching agent skills via the open `npx skills` tool.

Decide how to integrate finished work once implementation is complete and the tests pass.

Override default truncation behavior — enforce complete code generation, ban placeholder patterns, and handle token-limit splits cleanly.

Find code by asking what it does — the `jg` CLI uses the Jev relevance model to walk folders, files and declarations and returns a ranked file list plus verbatim source excerpts, so a coding agent starts from the right files. Upstream measured ~26% lower total agent cost at the same solve rate on 10 SWE-bench tasks. Needs Node 22+ and a Vercel AI Gateway, TypeSafe, OpenRouter or OpenCode Zen key; searched source is sent to the provider.

Drive microsoft/markitdown to turn PDF, Word, PowerPoint, Excel, images, audio, HTML, CSV/JSON/XML, ZIP, EPUB and YouTube URLs into LLM-ready Markdown — CLI, batch folders, Python API, and the Azure Document Intelligence and Content Understanding paths when local conversion is not enough.

Build high-quality MCP servers that expose external APIs to LLMs through well-designed tools — in Python (FastMCP) or Node/TypeScript (MCP SDK).

Run Alibaba's `ocr` CLI over a diff, commit, branch comparison or whole file and get line-level review comments — a deterministic pipeline picks files and matches rules, then an LLM agent reads the surrounding code. Needs its own model endpoint configured.

The same review engine with no LLM endpoint to configure — `ocr` does only the deterministic part, file selection and rule resolution, and hands the reviewing to the agent already in the session. The way to use Alibaba's rulesets without a second API key.

Pick the right frontend library for the task from a curated, opinionated list — numbers, OTP inputs, charts, command menus, virtualization, drag and drop, toasts, state and styling. Explicit invocation only.

Handle review feedback with technical rigor — verify each suggestion instead of agreeing performatively or implementing blindly.

Ask for review at task completion, major features and pre-merge, so work is checked against its requirements.

Create and update Claude skills — the guide for packaging specialized knowledge, workflows and tool integrations into a skill.

Execute a plan's independent tasks through subagents in the current session, with task briefs and a review package.

Work a bug, test failure or surprise back to its root cause before proposing any fix.

Write the failing test before the implementation, for every feature and bugfix.

Get an isolated workspace before feature work or plan execution, via native tooling or a git worktree fallback.

Entry point for the Superpowers framework — how to find and invoke the right skill before answering, including clarifying questions.

Run the verification commands and read the output before claiming anything is done, fixed or passing — evidence before assertions.

Drive and test local web apps with Playwright — verify frontend behavior, debug the UI, capture screenshots and read browser logs.

How to write modern Swift well — value types, Swift 6 data-race safety and approachable concurrency, protocols and generics, API design, ARC and performance, Swift Testing and macros, including the language features agents don't know yet.

Turn a spec or set of requirements into a written multi-step implementation plan before touching code.

Create, edit and verify Claude skills before deploying them.

Render a cinematic chaptered explainer video for any CLI or dev tool — Remotion "Terminal Noir": code that types itself, streaming terminal output, growing dependency graphs, continuous BGM.

Drive MoneyPrinterTurbo headlessly — turn a topic into a finished short with an LLM script, stock footage, TTS voiceover and burnt-in subtitles, through its CLI and batch manifest instead of the Streamlit WebUI.

Synthesise speech in 600+ languages with OmniVoice — clone a voice from a reference clip, design one from a written description, or let the model pick — driven through its CLI, JSONL batch runner or Python API.

Turn a product URL into a Korean-market shopping short — crawl the product, profile the target persona, recommend script × preset, approve the scene map, then generate via Higgsfield.

Build animated GIFs sized for Slack — composable animation primitives plus validators for Slack's size constraints.

Turn a frontend project or webpage into a cinematic product video — Remotion with real page screenshots, 2.5D camera moves, beat-synced cuts and sound design, driven by shot recipe cards against a validated template.

A Claude Code harness — 27 agents and 6 skills — that produces one webtoon episode end to end: trend research → scenario → panel breakdown → panel render → a finished vertical-scroll HTML viewer, with a quality-review agent gating each stage. Driven by the webtoon-orchestrator skill.

A Claude Code research harness: installs a /hyperresearch entry skill plus 16 step skills, then runs decompose → width sweep → depth → synthesis, where 4 parallel adversarial critics only file findings and a patcher applies them as surgical edits instead of regenerating the draft.

Run an 8-agent research pipeline — Orchestrator, 3 Researchers, 2 Fact-Checkers, Writer, Gatekeeper — applying MECE and the Pyramid Principle to produce a sourced consulting-style report.

Wire Algolia DocSearch into a Docusaurus v3 site on GitHub Pages — automates every config file, then walks you through the manual Algolia account steps.

Stand up a Docusaurus documentation site with GitHub Pages deployment, grouped navigation, per-topic sidebars, Korean locale and Mermaid support.

Add a GitHub Sponsors funding button to any static site — Hugo, Jekyll, Docusaurus or plain HTML — including the FUNDING.yml wiring.

Initialize a technical knowledge base on Hugo's Hextra theme with Mermaid, LaTeX and search, wired for GitHub Pages deployment.

Scaffold a topic learning-roadmap site on Hugo + Hextra — hero home, numbered roadmap docs by category, Labs, Tools, a Blog with RSS and a termbase-backed glossary.

Build a Hugo blog or docs site with GitHub Actions deployment — PaperMod or Hextra — with every known pitfall (TOML ordering, submodules, go.mod conflicts) handled up front.

Stand up a Hugo blog on the Stack theme (v4) via Hugo Modules on 0.157+, wired end to end for GitHub Actions deployment to GitHub Pages.

Build a documentation or knowledge-base site on Hugo's Lotus Docs theme (installed as a Hugo Module) and deploy it to GitHub Pages, custom domain included. Topic-agnostic.