One great tool to load into your AI agent, every day

Turns coding agents into structured security auditors that map attack surfaces, hunt by coverage, challenge candidate findings, and produce verified reports.

Security reviews fail when one agent both finds and validates its own work. Cloudflare’s skill separates reconnaissance, coverage-led hunting, adversarial verification, and reporting, while keeping unresolved leads distinct from confirmed vulnerabilities. It is the single-repository starting point that evolved into Cloudflare’s fleet-wide vulnerability discovery harness.

Teaches coding agents to use Jev for fast, typed decisions such as routing, scoring, verification, and guardrails.

Not every decision inside an AI workflow needs another slow, expensive language-model call. Jev gives up free-form text to return typed choices, scores, and probabilities in parallel, while the TypeSafe skill teaches agents when to use it and how to integrate it safely. TypeSafe reports 70–500 ms latency and up to 193.6× faster, 444.6× cheaper performance on its own workflow evaluations.

Helps coding agents discover source-reviewed game projects and reusable systems, then turn selected ingredients into a grounded build plan.

Most AI game demos begin from a blank prompt and forget where their mechanics came from. HeadStart gives Codex a researched catalog, source evidence, reuse caveats, and twelve game-development skills so a builder can start from real projects instead of vibes alone. It was built end to end with GPT-6 Astra for OpenAI H@.

Turns Claude Code's waiting time into language practice with quick vocabulary cards, spaced repetition, and one-key answers above the prompt.

Claude already makes you wait while it thinks. ClaudeLingo turns those loose seconds into Spanish, French, Italian, or another language you generate on demand, then disappears when Claude needs you back. Your progress stays local, the quiz never enters the conversation context, and no separate API key is required. We gave Claude Code a minor in Spanish.

Gives Claude Code and Codex a coordinated team of skills for building playable, polished Three.js browser games with gameplay, graphics, UI, debugging, and release QA.

Flappy Bird was a tiny 2D game built by one developer, and at its peak Dong Nguyen told The Verge it was generating about $50,000 a day from ads. This pack gives one vibe coder a game director plus specialists for gameplay, AAA-style graphics, UI, debugging, browser testing, mobile checks, and optional 3D, image, and audio generation. The demos range from hover racing and championship snooker to starship combat. Better tools do not guarantee the next viral hit, but the cost of trying has collapsed. Newzoo estimated games generated $188.8 billion in consumer revenue in 2025, compared with roughly $33.5 billion at the global theatrical box office. The next breakout studio may begin with one obsessive builder, one strange idea, and Claude or Codex connected to this skill pack.

Lets an agent search and query Bureau of Labor Statistics data for prices, employment, wages, productivity, and labor markets.

BLS has unusually rich public data hidden behind cryptic series IDs. We used this MCP to combine grocery prices with earnings data and build The Minute Aisle, where every item is priced in minutes of work.

Lets your agent search, compare, and run image, video, audio, and other AI models through Replicate.

Instead of learning a separate interface for every model, your agent can discover and run them from the same chat workflow. One MCP connection exposes Replicate’s API while keeping model choice flexible.

Keeps a coding agent working through a backlog while a separate reviewer decides whether each issue is actually finished.

A builder should not grade its own homework. This pattern separates implementation from acceptance, makes the reviewer test the running product in a browser, and loops rejected work back with a concrete reason.

Connects AI agents to structured stock-market data, SEC filings, financial statements, prices, ownership, and earnings.

Finance agents are only as reliable as the numbers they can reach. Financial Datasets gives them a direct, structured data source instead of making them scrape websites or guess, with coverage across 27,000+ US tickers and more than 30 years of fundamentals.

Pulls current, version-correct docs for whatever library you are using straight into the model.

Models are confident about library APIs that changed six months ago. Context7 fetches the real docs for the version you actually have installed, so you stop debugging code written against a function that no longer exists.

Gives AI agents social research, creator analysis, private analytics, content scheduling, and publishing in one MCP connection.

Most content agents can write, but they cannot see what is actually working. Eden grounds them in real high-performing posts, creator baselines, your saved research, and your own analytics, then carries the same workflow through scheduling and publishing.

Lets your AI agent create, modify, light, animate, and render editable 3D scenes directly inside Blender.

Image generators give you a finished picture. Blender MCP gives your agent control of the actual scene, including its objects, materials, lighting, camera, and geometry. The result is not just generated. It is a world you can keep editing.

Lets your coding agent open Chrome, watch the console and network tab, and profile the page it just built.

Your agent writes a fix, then checks its own work in a live browser. It sees the actual console error, the actual slow request, the actual layout. We started the series here because it is the fastest jump from "the AI guessed" to "the AI checked."