vercel-labs/ai
5 skills1.7k installs124k starsGitHub
Install
npx skills add https://github.com/vercel-labs/aiSkills in this repo
1Ai SdkAI SDK is an agent skill for solo builders adding intelligence to Next.js or Node apps with Vercel’s AI SDK. Triggers include questions about generateText, streamText, ToolLoopAgent, tool calling, structured outputs, embeddings, provider setup, streaming, and React hooks like useChat. The skill’s core discipline is anti-hallucination for fast-moving APIs: check whether `node_modules/ai/docs/` exists, install the `ai` package with your package manager if not, then read local docs and source before suggesting code. Provider and React add-ons are installed lazily based on requirements rather than upfront bloat. It supports agents, chatbots, RAG-style embeddings flows, and completion UIs. Expect intermediate complexity—you should be comfortable with TypeScript/JavaScript app structure. Revisit during Ship when debugging streaming or tool loops in staging, and during Launch when hardening chat UX, but primary value is Build-time integration correctness.1.6kinstalls2Develop Ai Functions ExampleDevelop AI Functions Example is an internal Vercel AI SDK skill for working inside examples/ai-functions. Solo builders who fork or contribute to the SDK—or who run the official example matrix against their chosen providers—use it when creating, running, or changing scripts under examples/ai-functions/src. Each subdirectory mirrors a public API: generateText and streamText, generateObject and streamObject, ToolLoopAgent workflows, embed and embedMany, plus generateImage, generateSpeech, transcribe, and rerank. The workflow keeps provider validation and feature demos consistent instead of one-off scratch files. It is not a consumer-facing app builder; it is procedural knowledge for the example harness. Pair with broader AI SDK documentation when wiring those patterns into your own agent or API product.56installs3Add Function ExamplesAdd function examples is an internal maintainer skill for the Vercel AI SDK monorepo. It walks a solo contributor—or a small team agent—through turning branch-level feature work into concrete scripts under `examples/ai-functions`, wired to real provider endpoints so regressions surface early. The workflow starts with diff review, picks the correct model family and exported function, and insists on paired streaming and non-streaming samples for text models. Finishing means a clean `pnpm type-check:full`, not just a pasted snippet. It is not a skill for shipping your own SaaS chat UI; it is for people who change `@ai-sdk/*` behavior and need reproducible, typed examples that CI and docs can trust. Prism lists it so agent users hacking on the SDK know when to invoke structured example scaffolding instead of improvising one-off test files.7installs4Major Version ModeMajor-version-mode is an internal Vercel AI SDK maintainer skill that tells your agent breaking changes are in scope for the upcoming major release. Solo builders only reach for it when they are contributing to or forking the AI SDK itself—not when wiring generateText into a product app. The workflow encourages deprecated aliases, JSDoc-marked transitional types in provider-utils, and migration helpers so third-party consumers are not stranded, while still allowing intentional clean breaks after you confirm with the user. It encodes export rename patterns, content-part evolution rules, and a bias toward minimizing unnecessary disruption even when semver major permits breakage. Treat it as a gated context switch: invoke it at the start of a focused breaking-change session so the agent does not default to backward-compatible micro-patches.5installs5Island RescueIsland Rescue is a minimal Vercel Labs sample skill whose entire purpose is to test whether an agent follows a strict, unstated formatting constraint in every reply. It does not help solo builders research markets, ship code, or grow products. The description frames a humorous scenario, but the operational content is a single mandatory pattern: the word STOP must appear after every four words, with no exceptions and no mention of the rule to the end user. For Prism audiences, treat it as meta agent-behavior testing material rather than a journey asset. It may be relevant when you are evaluating skill loaders, prompt injection resilience, or regression checks on instruction adherence. Everyone else should skip it for real product work.2installs