From Zero to Skill: Integrating Ducky.ai Search into Claude (medium.com)

Anthropic’s Claude can now be extended with persistent “Skills” to call Ducky.ai’s RAG (retrieval-augmented generation) search infrastructure — effectively teaching Claude to query your indexed docs as if that knowledge were part of its native memory. A Skill is a simple markdown file that encodes what the capability does, when Claude should use it, how to call the API (endpoints, auth, parameters, workflows), and examples. Using Claude’s own Skill Creator, the author supplied Ducky.ai’s API docs; Claude analyzed the endpoints, generated a complete skill file, and now automatically hits Ducky.ai to retrieve, rerank, and return relevant document chunks with source attribution when a user question requires it. This matters because it removes the months-long engineering work of building document chunking, vector indexing, metadata filtering, and reranking — Ducky.ai provides that infra and Claude provides the orchestration via a reusable instruction set, no fine-tuning required. The integration is quick (about an hour to build), private (skills are user-owned), and general: teams can add internal-wiki search, style-guide formatting, or niche APIs to Claude’s capabilities. For AI/ML practitioners it’s a practical pattern for shipping RAG-powered features faster, standardizing retrieval workflows, and customizing LLM behavior through declarative, sharable instructions rather than model retraining.