Docs · Getting Started

Quickstart

Get to a working agent in five minutes. We'll scaffold a project, load a skill, and call an LLM.

Following along?

You'll need an OPENAI_API_KEYin your environment. Skillarc itself doesn't require any keys — it's just a skill registry.

Scaffold a project

terminal

mkdir my-agent && cd my-agent
pnpm init
pnpm add @skillarc/sdk openai dotenv

Load a skill

Skills are referenced by slug. The SDK fetches the MDX file, parses the frontmatter, and gives you a typed handle:

agent.ts

import "dotenv/config";
import { loadSkill } from "@skillarc/sdk";
import OpenAI from "openai";

const client = new OpenAI();
const prompt = await loadSkill("rag-retrieval-prompt");

Call an LLM

The skill exposes a render() method that fills frontmatter variables in a type-safe way:

agent.ts

const message = prompt.render({
  retrieved_chunks: [
    "[1] Transformers replaced RNNs in 2017.",
    "[2] Self-attention scales to long sequences.",
  ],
  question: "What replaced RNNs?",
});

const response = await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "system", content: message }],
});

console.log(response.choices[0].message.content);

Compose with tools

Tools follow the same pattern. Convert to the format your runtime expects with toOpenAI(), toAnthropic() or toGemini():

agent.ts

import { loadTool } from "@skillarc/sdk";

const search = await loadTool("web-search-tool");

const response = await client.responses.create({
  model: "gpt-4o",
  input: "What's the latest stable Python release?",
  tools: [search.toOpenAI()],
});