527 skills
Claude Academy course matcher for product questions
Anthropic · Apache-2.0
Anthropic product skill. Recommends academy.claude.com courses — never invent them. Idle when the user wants the task done, not a tutorial.
Seeded p5.js generative-art skill
Frontend · Apache-2.0
Idle unless someone asks for generative art. Original work only; not a copy-an-artist skill.
Anthropic brand colors and type for artifacts
This is Anthropic's look — Poppins, Lora, #d97757 — not “your brand.”
Poster and static visual design in PNG/PDF
Static PNG/PDF posters. Not product UI. Will sit idle unless someone asks for a piece.
Claude API / Anthropic SDK reference before you guess
Vendor + broad-trigger Catch: fires on almost any LLM-shaped task and steers to Anthropic. Skip when another provider is the job.
Structured loop for co-authoring specs and docs
Authoring · Apache-2.0
Anthropic’s three-stage writing loop. Does not trigger if the agent just writes the doc. Distinct from Sentry’s copy in getsentry/skills.
Intentional visual direction so UI is not template-default
Aesthetic direction, not a component library. Idle unless someone is actually choosing a look.
How to build MCP servers in Python or TypeScript
MCP · Apache-2.0
How to *build* an MCP server. Not how to wire one that already exists.
Create, edit, eval, and tune agent skills
Meta-skill for authoring and eval-ing skills. Idle unless you are writing a skill.
Animated GIFs sized and timed for Slack
Other · Apache-2.0
Will sit idle unless someone actually asks for a Slack GIF.
Ten preset themes, or a new one, for artifacts
Ten palettes for artifacts. Idle unless someone wants a theme applied.
Multi-component claude.ai HTML artifacts (React + shadcn)
Locked to claude.ai HTML artifacts (React + Tailwind + shadcn bundle). Not a general web app skill.
Playwright toolkit for local webapp UI and logs
Debugging · Apache-2.0
Playwright against localhost. Needs a running app; a screenshot is not a test plan.
Split independent work across parallel agents
Workflow · MIT
Only when 2+ tasks share no state. Sequential work is a false fire.
Run a written plan in a separate session with checkpoints
Separate session, written plan, checkpoints. Not for writing the plan.
Integrate a finished, tested branch
Git · MIT
Tests already pass. This decides how to land it — not whether it works.
Verify review feedback before you implement it
Verify feedback before implementing. Rubber-stamping is the failure mode.
Ask for review after a real feature or before merge
After a real feature or before merge. Not a substitute for tests.
Execute a plan with independent subagents in-session
In-session subagents. Conflicts with executing-plans (which wants a separate session).
Root-cause the bug before proposing a fix
Debugging · MIT
Evidence lives outside the agent. Reproduce the failure, read the log, then patch. Do not skip verification because the stack trace looked obvious.
Write the failing test before implementation
Red, then green, then code. Agents skip the red step constantly.
Isolate feature work in a git worktree
Isolate via worktree. No-op if you ignore it and stay on the dirty branch.
Run the proof before claiming done
Run the proof before claiming done. Agents say “done” without running the command.
Bite-sized implementation plan after a spec exists
Bite-sized steps after a spec exists. Not for writing the spec (that’s spec-driven-development).
Stable APIs, module boundaries, and public contracts
Broad guidance from addyosmani/agent-skills. Does not trigger if the agent does not load it.
Chrome DevTools MCP for real-browser evidence
Requires the chrome-devtools MCP server. Without it, this is a prompt, not a browser.
Build, test, and deploy pipelines as quality gates
Broad guidance from addyosmani/agent-skills. Not a CI vendor. Does not trigger if the agent does not load it.
Multi-axis review before a change lands
This is a review workflow, not AGENTS.md. Do not treat it as project agent instructions.
Clarify working code without changing behavior
Clarify without changing behavior. Easy to “simplify” into a rewrite.
Write the quality bar so agents cannot quietly lower it
Writes CONSTRAINTS.md. Useless if nobody reads it, or if the agent quietly lowers the bar anyway.
Sunset, migrate, and expand/contract without downtime
Expand/contract, then delete. Not a license to drop a column on Friday.
Record decisions so the next engineer and agent can follow
Authoring · MIT
Record the decision. Agents write ADRs nobody asked for.
Adversarial review of high-stakes decisions before they stand
Adversarial review before the decision stands. Slow by design; skip it and you get confident wrongness.
Production-quality accessible, responsive UI
Frontend · MIT
Production a11y/responsive UI. Overlaps frontend-design; this one is engineering, not vibe.
Commits, branches, PRs, tags, and version bumps
Commits, PRs, tags. Not using-git-worktrees. Does not trigger if the agent does not load it.
Diverge, then converge a vague idea into a sharp concept
Diverge, then converge. Idle unless the idea is still vague.
One-question interviews until the real intent is clear
One question at a time until intent is clear. Annoying if you already know what you want.
Logs, metrics, traces so production is diagnosable
Logs, metrics, traces. Not Sentry-the-product. Does not trigger if the agent does not load it.
Frontend, backend, and query performance with evidence
Needs a metric. Profiling theater without a number is not this skill.
Turn a spec into ordered, estimable tasks
After a spec. Overlaps writing-plans. Does not trigger if the agent does not load it.
Input, auth, data, and supply-chain hardening
Input, auth, data, supply-chain. Not a pentest and not Sentry security-review.
Pre-launch checklist, rollout, and rollback
Checklist, rollout, rollback. Not a deploy button.
Cite official docs instead of remembered APIs
Cite official docs. Agents still answer from memory.
Write the spec first when none exists yet
Write the spec first when none exists. Distinct from writing-plans, which assumes a spec already exists.
Vercel Engineering React and Next.js performance patterns
Vercel · MIT
Vercel Engineering React/Next performance patterns. MIT in SKILL.md frontmatter; vercel-labs/agent-skills has no LICENSE file. Broad React triggers.
React composition that scales past boolean props
Compound components and boolean-prop cleanup, including React 19. Same repo without a declared LICENSE. Not a performance skill.
React Native and Expo performance and native-module patterns
React Native/Expo lists, animations, native modules. Same repo without a declared LICENSE. Idle on web React.
Railway signup, deploy, services, and infrastructure ops
Other · MIT
False-fires on feature flags, MCP, and signup. Drives `railway up` for signup. A free tier exists.
Cloudflare Workers CLI, before you type it
Cloudflare · Apache-2.0
Workers CLI for KV/R2/D1/Queues before you type it. Pulls a non-Cloudflare deploy toward wrangler.
Production Workers anti-patterns
Production Workers anti-patterns (streaming, globals, bindings). Not a generic JavaScript review.
Stateful coordination on Workers
Stateful Durable Objects (SQLite, alarms, WebSockets). False-fires on any “stateful agent” that isn’t a DO.
Bot check via Cloudflare Turnstile
End-to-end Turnstile widget plus siteverify. A generic captcha/bot-check still gets Turnstile.
@cloudflare/sandbox on the default tag
@cloudflare/sandbox on the default tag. Wrong sibling if the app is on @next.
Sandbox SDK 1.0 preview tag
Preview Catch: this is @cloudflare/sandbox@next, not the stable default. Vendor skill for Cloudflare.
Send and route mail from Workers
Workers email send/route and SPF/DKIM. “Add email” on a non-Workers app still steers here.
Stateful agents on Cloudflare Workers
Cloudflare Agents SDK on Workers (state, RPC, Workflows). “MCP/chat agent” triggers overlap the MCP skills.
This writes AGENTS.md, not a skill
Sentry · Apache-2.0
This IS AGENTS.md / CLAUDE.md, not an agent skill. Do not confuse it with SKILL.md. Vendor skill from Sentry.
permissions.json from the repo shape
Writes Claude Code settings.json allowlists from repo shape. Sentry conventions leak; a generic repo still gets Sentry-shaped permissions.
IDOR and tenant isolation in Django
IDOR and tenant isolation in Django views/ORM. Idle off Django; not a generic auth review.
N+1 and queryset audits
N+1 and queryset audit, not a general perf skill. “Optimize Django” false-fires when the bottleneck isn’t the ORM.
Structured pass with the human still in it
Human-in-the-loop doc loop from getsentry/skills. Twin of Anthropic doc-coauthoring; does not trigger if the agent just writes the doc.
Unread review-request notifications
Fetches unread GitHub review-request notifications. Needs gh + a team; an empty inbox looks like “no reviews.”
Pwn-request and workflow injection
Pwn-request and expression-injection in .github/workflows. Exploitation-shaped; not a green-CI checklist.
CI + review loop until the PR is green
Loops CI and review until green, and will not wait for human approval or merge gates. Keeps pushing a draft you wanted parked.
Reviewer-facing title and body
Reviewer-facing title and body only. Will rewrite a PR description you already wrote.
Evals and model-family adapters
Prompt evals and OpenAI/Claude/Gemini adapters. Not product copy; will “optimize” a prompt that wasn’t the job.
Audit a SKILL.md before you install it
Audits a SKILL.md for injection and secrets before install. Scanning is not the same as running the skill.
Bugs on this branch, not the whole repo
Bugs on this branch, not the whole repo. A repo-wide “find bugs” is a false fire.
OWASP-style pass with confidence scores
OWASP-style pass with confidence scores. Overlaps find-bugs and gha-security-review; not Sentry-product config.
Sentry @extend_schema and TypedDicts
Sentry @extend_schema and response TypedDicts. Dead outside getsentry API views.
In-session scan, then patches you apply
In-session scan, then patch files you apply. Claude Code plugin layout; not Sentry security-review.
Hooks, skills, MCP — recommendations only
Recommends hooks, skills, and MCP for Claude Code — it does not install them. Vendor lock.
Claude Code subagent frontmatter
Claude Code subagent frontmatter (when-to-use, tools, colors). “Create an agent” outside plugins is a false fire.
Slash commands for Claude Code plugins
Slash commands and AskUserQuestion inside Claude Code plugins. Not a Unix CLI skill.
PreToolUse and friends, in plugins
PreToolUse/PostToolUse/Stop hooks with ${CLAUDE_PLUGIN_ROOT}. Does not trigger without that plugin layout.
plugin.json at the root, components beside it
Plugin.json scaffold and auto-discovery. First-step only; overlaps the other plugin-dev siblings.
.local.md config for a plugin
Per-project .claude/<plugin>.local.md YAML. “Plugin settings” in an editor is a false fire.
In-chat widgets on an MCP server
In-chat MCP App widgets after the server exists. Not how to build the server itself —
Bundle a local MCP with its runtime
Bundles a local MCP plus runtime as .mcpb. Desktop/filesystem MCP; not a hosted-server tutorial.
Reanimated, Gesture Handler, and haptics on device
Expo · MIT
Reanimated/GH/haptics; wrong for web CSS motion; fights expo-ui (don’t use Reanimated for sheets).
iOS App Clip target beside the parent Expo app
IOS Clip/AASA only; dead on Android/web.
Embed Expo in an existing Swift or Kotlin app
Embed-in-existing-native only; dead on greenfield Expo.
Fetch, React Query, SWR, and Expo Router loaders
“ANY network request” also fires on generic fetch/SWR work outside Expo.
Tokens, variants, and an in-app component library for Expo
Expo token/theme kit; twin of frontend-design if you drop the Expo scope.
Local and TestFlight Expo development clients
Internal dev-client builds; agents mix it with paid eas-app-stores TestFlight.
Web code in a native webview, incrementally
Webview DOM components only; not a full web→native port.
Adapt with-* starters from expo/examples
Not full apps — managed single-screen patterns. catalog.md drifts; confirm live list. Never scaffold on top of an existing app. Distinct from expo-project-structure.
Swift Expo module 1.0 DSL → 2.0 macros
Experiments path, not the main expo plugin. Requires expo ≥ 57.0.7. Not for new modules (expo-module), not SDK upgrades (expo-upgrade), not Android/Kotlin.
New Expo Modules API in Swift, Kotlin, and TypeScript
Native module authors (Swift/Kotlin); does not trigger on JS-only Expo apps.
HIG, semantic color, and SF Symbols on Expo screens
HIG/SF Symbols screens; look vs @expo/ui primitives (expo-ui).
Folder layout for a new Expo Router app
New-app folders only; forbids restructuring existing apps.
File-based routes, Stack, NativeTabs, and modals
Expo Router file routes only; also fires on Next.js App Router and React Navigation work.
@expo/ui SwiftUI and Compose primitives
@expo/ui SwiftUI/Compose controls; not FlashList; not expo-native-ui styling.
Expo SDK upgrades and dependency fixes
Expo SDK bump cookbook; stale after the next SDK; not a generic RN upgrade.
Port Next/Vite/CRA to Expo native, screen by screen
Whole-app Next/Vite/CRA → Expo; not expo-dom.
EAS build and submit to App Store and Play
Eas build/submit to stores. Free-tier build minutes; Apple/Google developer accounts still cost. Not local expo-dev-client.
Expo web and API routes via eas deploy
Expo web/API routes via eas deploy (CF Workers). Authoring routes is free; production deploys use plan allowance.
expo-observe metrics, 10k MAU free cap
Expo-observe metrics. Free plan up to 10k MAU, then paid. Not generic analytics.
Remote iOS/Android sim on EAS cloud
Remote iOS/Android sim on EAS cloud, experimental API. Not local sim.
OTA JS/asset updates via expo-updates
EAS service, costs apply (Free plan has update/bandwidth/storage allowances). Distinct from live eas-update-insights (health/adoption, not publish) and eas-app-stores (native store binaries).