Skip to main content

Agentic Skills — teach AI agents to run SKU through your API

AI agents are getting genuinely useful — but pointing one at a platform with thousands of API endpoints is a slow, error-prone start. The agent has to rediscover which endpoint does what, what fields it needs, and how the pieces fit, every single time. So teams end up rebuilding the same integration glue over and over. Agentic Skills fixes that. It's a new open-source library of ready-made skills — curated playbooks that tell an AI agent exactly how to accomplish a real business task in SKU: which endpoints to call, in what order, with which fields, plus the e-commerce context to get it right. Point your agent at a skill and it just works.

✨ What this means for you:

  • Put your AI agent to work in SKU today — ready-made skills for real tasks like build a product catalog, create a sales order, adjust inventory, and create a saved view.
  • Works with the agent you already use — every skill is published for Claude, OpenAI, and Gemini, so you're not locked into one model.
  • Grounded in the real API — skills call actual SKU endpoints with the right fields and scopes, so your agent stops guessing and stops hallucinating endpoints.
  • Open source and yours to extend — MIT-licensed on GitHub. Use them as-is, or have your agent improve one and open a pull request.

What's a skill?​

A skill is a curated shortcut for an agent. Instead of exploring thousands of endpoints, the agent loads one skill — for example build a product catalog — and follows a proven path: how to read the source data, map it onto SKU's product fields, de-duplicate against what's already there, and create the products cleanly. The hard-won practical details (what's a supplier SKU vs. a real SKU, where the wholesale price goes, how to capture leftover attributes) are baked in, so your agent gets them right the first time.

The Agentic Skills page on the SKU.io developer portal, showing the cross-model skill library and example skills

Write once, run on any model​

Each skill is authored once in a single, model-agnostic format and compiled into each platform's native shape:

  • Claude — an Agent Skill you drop into your project
  • OpenAI — GPT/Assistant instructions plus an importable Action and function-calling tools
  • Gemini — system instructions plus function declarations

Fix or improve the knowledge in one place, and every model benefits.

Use one in minutes​

Grab the skill for the task you want and install it into your agent. For Claude, that's as simple as copying the skill folder into a .claude/skills/ directory; for OpenAI and Gemini, paste the generated instructions and register the tools. Authenticate with a SKU Personal Access Token, and your agent is ready to work.

It gets better every time it's used​

Agentic Skills is built to evolve. If a skill doesn't do everything you need, don't work around it — have your agent improve it and open a pull request. Every skill even ends with a short note explaining exactly how. The library grows from real usage, so the next person's agent benefits from your fix.

Where to find it​

Read the overview at developer.sku.io/agentic-skills, and get the skills on GitHub at github.com/skuio/sku-skills.