Claude Skills
The open format is called Agent Skills and works in Claude Code, Codex, Cursor and other agents — most people know it as Claude Skills.
Every Agent Skill we could find on GitHub, deduplicated by content. 79 600 files from 1 763 authors, of which 61 947 are unique — the rest is the same skill repackaged into someone else's repository. For each one: what it weighs in tokens, whether it ships runnable scripts, and which MCP servers it needs.
Load when a frontend task needs Apple-platform design guidance or physical, gesture-driven, velocity-aware, interruptible interaction; skip generic visual styling and use review-animations for report-only motion review.
Load when a user describes a Web animation or motion effect without knowing its name and needs the precise term or close alternatives; use only for naming, not design, implementation, or review.
Load when the user explicitly asks to clone, reverse-engineer, or rebuild a website they own or are authorized to reproduce; do not load for generic redesign, inspiration, scraping, phishing, impersonation, or copying third-party brands.
Load when a task needs Claude API or Anthropic SDK build, debug, migration, or tuning work; skip provider-neutral, OpenAI, or non-Anthropic model work.
Load when the user asks for private agent interaction audits from local Codex/Claude Code traces, task profiles, tool-call decisions, collaboration bottlenecks, prompt/rule or automation-log audits, closeout retrospectives, or evidence-backed Skill/project-rule/Eval/SOP/OKF improvements.
Load when a task needs agent run, tool loop, context drift, or recoverable harness/tool failure debugging; use diagnose for product/runtime bugs.
Load when a change needs module-boundary, dependency-direction, interface-depth, or architecture simplification decisions; do not use for routine style or formatting rules.
Load when a task needs a deep, report-only review across independent Standards and Spec axes, using bounded read-only native Subagents when available and adding risk lenses only when the diff warrants them.
Load when a user wants a source-backed interactive HTML course that explains how a codebase works; do not load for chat tutoring, repository review, articles, slide decks, or general browser artifacts.
Load when a task needs anti-template frontend visual direction, design-read calibration, or pre-flight critique for landing pages, portfolios, marketing pages, or redesigns; do not load for dashboards, data tables, multi-step product UI, routine frontend logic, HTML reports, slide decks, or generic code explanation.
Load when a task needs hard bugs and performance regressions, failing commands, hard-to-reproduce behavior, or unknown root causes diagnosed; use agent-introspection-debugging instead for agent/tool harness failures.
Load when Codex 即将要求用户在 2-3 个互斥方案中做真实决定、准备询问阻塞性澄清问题或需要确认高影响取舍；将选择转换为宿主原生可点击 UI，并在工具不可用时诚实文本回退。不要用于低风险实现细节、只有一个合理答案的问题、普通信息采集或不需要用户决策的完成汇报。
Load when a task needs to collaboratively draft, restructure, or reader-test docs, PRDs, RFCs, proposals, specs, or decision records; skip implementation itself.
Load when a task needs current library, framework, SDK, API, CLI, or cloud-service documentation; fetch docs instead of relying on training data or ordinary repo evidence.
Load when a task needs durable Playwright E2E suites, Page Object Models, fixtures, CI browser tests, or flaky-test strategy; use webapp-testing for one-off local inspection.
Load when agent pauses to report relatively complex information needing Chinese-first clear complex communication, alignment, multi-option choice, status/incident, long-task fact ledgers, implementation plans, reviews, maps, explainers, evidence, risks, validation, handoff; choose plain text/Markdown/visual Markdown/HTML by decision cost; do not load merely because answer is long; skip trivial chat/bundled apps.
Load when a task needs polished product UI, web pages, landing pages, dashboards, React components, HTML/CSS layouts, or styling; skip routine frontend logic and reports.
Load when a repository mutation creates or materially changes durable project contracts, OKF knowledge, specs, ADRs, or architecture/API/design documents; run a source-backed alignment pass that reuses grill-me's dependency-frontier batch protocol.
Load when beginning every repository mutation task to run one dependency-layered batch interview; a fully aligned task takes the zero-question path, while durable documentation work uses grill-with-docs to reuse the same decision graph.
Load when a task needs React or Next.js frontend logic, component patterns, state, forms, routing, accessibility, or responsive behavior; use frontend-design for visual direction.
Load when a task needs a browser-based HTML presentation, talk or pitch deck, existing-deck enhancement, or PPT/PPTX-to-web conversion; do not load for reports, product UI, or interactive courses.
Load when a task needs to hand off current work, compact a conversation for another agent/session, or create restart notes; do not load for visual HTML repo reports.
Load when a task needs internal communications such as status reports, leadership updates, 3P updates, newsletters, FAQs, incident reports, or project updates; skip public marketing copy.
Load when a task needs MCP server design, build, review, testing, tool schema, resource, or prompt work; skip ordinary REST clients and non-MCP integrations.
Load when a task needs a coding behavior baseline for writing, reviewing, or refactoring code: surface assumptions, avoid overcomplication, keep changes surgical, and define verifiable success criteria; do not use as the top-level owner or for trivial one-line edits.
Load when a task needs to implement or continue an existing OpenSpec change; use the repository's normal development contract for non-OpenSpec implementation.
Load when a task needs an explicit OpenSpec proposal with design, spec deltas, and tasks; use product-capability for ordinary implementation contracts.
Load when a task needs to archive a completed OpenSpec change after implementation and spec sync decisions are resolved.
Load when a task needs explicit OpenSpec exploration, discovery, or requirement clarification; use the repository's normal development contract for default change alignment.
Load when tracking newly published package or model-package releases across package registries and release feeds for AI/developer-tool monitoring; do not load for ordinary docs lookup, broad GitHub trend scanning, or implementing package clients.
Load when coding work should be deliberately minimal: YAGNI, reuse existing code, prefer stdlib/native/already-installed dependencies, shortest correct diff, no unrequested abstractions, or over-engineering review; do not load for non-coding concise reports, general summaries, or communication artifact density.
Load when an existing Web product, especially a dashboard, dense-data view, or multi-step app, needs a deep evidence-backed experience review; report findings only, and use design-taste-frontend for marketing-page direction, code-review for source diffs, web-design-guidelines for checklist compliance, or frontend-design for implementation.
Load when accepted product intent must become a concise, implementation-ready local capability specification with explicit constraints, non-goals, test seams, and unresolved decisions.
Load when existing animation or motion code, a motion diff, or a rendered interaction needs an evidence-backed craft review; report findings only and use code-review for general source review.
Load when a task needs an explicit source-backed learning project, study plan, syllabus, book/technical topic coaching, teach-back, quizzes, or staged review; do not load for routine implementation, one-off facts, or article writing.
Load when a task needs a throwaway prototype, state-model sanity check, UI variant, image-assisted visual mockup, mock interaction, or playable design; do not use for production feature work.
Load when a task needs security-sensitive code, auth, user input, secrets, API endpoints, payments, injection risk, or unsafe IO reviewed; use code-review for broader review.
Load when creating, updating, adapting, or evaluating a standard agent skill; do not load for ordinary documentation edits or one-off prompt advice.
Load when a task needs to create, optimize, or validate an animated GIF for Slack emoji or messages; skip static images, video editing, and non-Slack animation.
Load when turning an external article, blog, release note, interview, or report into a source-backed public post with Chinese fidelity, media references, an effective-interact summary layer, and project-iteration review; do not load for ordinary summaries, full copyrighted reposts, or production site design.
Load when the user asks to draft, edit, or review English prose for AI-writing tells; do not load for code explanation, Chinese output, technical specs, or routine status reports.
Load when implementing a confirmed behavior change through red-green-refactor; diagnose first when the root cause is unknown, and use prototype for throwaway exploration.
Load when a task needs a coherent visual theme for slides, docs, reports, HTML artifacts, or landing pages; skip product UI implementation and accessibility review.
Load when an accepted specification must be decomposed into dependency-aware, project-owned implementation slices without creating a new tracker or publishing remote issues by default.
Load when claiming completion requires the repository's actual deterministic build, type, lint, test, diff, smoke, and artifact gates; use diagnose while a failure's root cause is unknown.
Load when a task needs complex standalone React/Tailwind/shadcn browser artifacts with state, routing, or bundled components; use effective-interact for reports and frontend-design for production UI.
Load when a task needs UI, UX, accessibility, or web interface guideline compliance review; use frontend-design when creating or changing production UI.
Load when a task needs one-off local web app inspection with Playwright, screenshots, console logs, or UI issue reproduction against a dev server; use e2e-testing for durable suites.
Use when planning or writing server-side code that uses `@supabase/server` — Edge Functions, Hono apps, webhook handlers, or any backend that creates Supabase clients or validates inbound auth. Trigger **before** writing or modifying any file that imports from `@supabase/server` (or sub-paths like `@supabase/server/core`); calls `withSupabase`, `createSupabaseContext`, `createAdminClient`, `createContextClient`, `verifyAuth`, `verifyCredentials`, or `extractCredentials`; configures an `auth:` mode (`'none'` | `'publishable'` | `'secret'` | `'user'`, or keyed variants like `'secret:*'`); or lives under `supabase/functions/` and authenticates an inbound request. Also trigger during planning — if a plan mentions any of the above, load the skill before drafting code; do not extrapolate `auth:` values or auth modes from neighboring functions. Also trigger when you see legacy patterns to migrate to this package — `Deno.serve`, `createClient(Deno.env.get('SUPABASE_URL'))`, imports from `esm.sh/@supabase` or `deno.land/std`, usage of `SUPABASE_ANON_KEY` / `SUPABASE_SERVICE_ROLE_KEY`, or the deprecated `allow:` config option / removed `'always'` / `'public'` mode values / removed `authType` field.
Lightweight, script-driven variant of data2motion for turning data into a smooth, on-brand animated chart (one self-contained HTML) — built for a TEXT-ONLY model that cannot see its own output. You never write HTML/CSS/SVG/animation; you extract the data, pick a chart, fill a small JSON spec, and run one build script that owns 100% of the look and the motion, so the result cannot drift from the house template style or lose its smoothness. Use to turn a number/stat/table/CSV/paragraph into a quick animated chart, data reveal, or KPI when you want guaranteed template fidelity with minimal reasoning and no image feedback.
Use when the user wants to turn a static scene, character images, aerial map, drawn path, or route-control image into an immersive first-person FPV AI video prompt, especially Seedance/Kling/Runway/Veo style one-shot videos with numbered stop markers, red-line path control, world-map flythroughs, camera route planning, variable character counts, non-human POVs such as drones, pets, robot vacuums, character references, timed interactions, dialogue, spatial audio, and negative constraints.
Create or update a complete repository skill from a user's idea, including the workflow instructions, references, scripts or assets, agent metadata, skill-card artwork, cinematic banner artwork, README links, discovery metadata, and validation. Use when the user asks to create a new skill, add a skill to this collection, turn a workflow into a reusable skill, or make a skill's documentation and artwork consistent with the repository.
Transform saved links, papers, articles, posts, videos, and reference collections into approachable AI teaching artifacts for later study. Use when a user wants to queue learning material, create a readable explanation from a source, teach a paper or post step by step, or run an interactive tutor that validates understanding over multiple sessions.
> Discovery-scale research harness. A cheap scout maps the topic, the Lead designs topic-specific parallel researcher assignments from the scout's map (drawing on a source-class tactics library — academic, repos, production patterns, web, experts), then verifies claims against sources and writes a decision-oriented report. Use when brainstorming a project or feature, choosing a technology, or asked to "research X", "state of the art", "deep research". For narrow slice-level fact checks inside the build loop, /lead handles those inline.
>- Apply a disciplined engineering workflow to any code change. Use whenever implementing a feature, fixing a bug, or refactoring — before writing code, not after. Walks orient → baseline → smallest change → test → verify → self-review, and enforces language-agnostic hard gates (don't mass-reformat, keep the linter and type-checker clean, keep the build and tests green, make interface changes additive, protect security invariants).
>- The Indie Maker Blueprint. Use when a user is starting or sizing an app idea, choosing what to build next, building a minimal product, preparing a launch, diagnosing stalled growth, deciding pricing or monetization, automating operations, maintaining a per-app MAKE.md tracker, or evaluating whether to sell a product.
Create truthful, human-centered marketing campaigns for an app or product, including positioning, channel copy, original artwork, editable layouts, README banners, and selective website integration. Use when asked to make launch materials, promotional artwork, social assets, campaign kits, ads, or marketing content from an existing product; to adapt a visual reference without copying it; or to add approved campaign art to product surfaces. Do not invent claims, fake UI, publish, deploy, or replace product proof without explicit evidence and authorization.
Inspect an unfamiliar repository, turn a focused Markdown behavior scenario into a deterministic test in the repository's native test stack, run it, and preserve traceability between intent and code. Use when asked to add scenario tests, compile acceptance criteria or Given/When/Then Markdown into executable tests, reproduce a user-visible regression, or convert a narrow workflow specification into stable web, API, CLI, desktop, or mobile interaction coverage. Do not use for broad exploratory journeys or agent-judged smoke tests.
> goal into a spec-approved GitHub issue plan, freeze acceptance checks, dispatch parallel builder jobs, review completed jobs, answer blockers, and finish a run with a single PR. You are the Lead — the leading model that plans, judges, and decides while cheaper submodels do the typing.
Use when an agent is asked to draft, rewrite, edit, review, polish, copyedit, simplify, humanize, or create written prose, including creative writing, essays, posts, scripts, speeches, emails, documentation, product copy, and other style-sensitive text. Apply George Orwell's six rules and ASD-STE100 Simplified Technical English as a plain-English discipline while preserving the user's intended meaning, audience, tone, and explicit constraints.
Claude Skills — questions
Answers built from the skills we actually parsed.