Skip to content

Writing

I write about

  • 🔧 Building AI Tools
  • 🤖 Using AI Tools

Subscribe to my Newsletter Subscribe to my YouTube

Building Real-World Claude Code Skills: Spreadsheets and PDFs

Agent skills sound great in theory. But what does it actually look like to build one that handles messy, real-world data? For my upcoming LinkedIn Learning course on Claude Code skills, I built two production-grade skills from scratch: an expense categorizer for bank CSV exports and a PDF extraction and transformation toolkit. Both taught me things about skill architecture that I didn't expect.

This post walks through what each skill does, the design decisions that made them reliable, and the patterns that emerged across both.

Why Agent Skills Are The Future of Agents in 2026

In the first month of 2026, something remarkable happened in the AI agent ecosystem: the way we build AI capabilities fundamentally changed. Instead of creating specialized agents for every task, developers discovered they could create modular "skills"—folders containing instructions, scripts, and resources that agents load dynamically when needed.

Within five weeks of launch, over 71,000 skills appeared in community marketplaces. Microsoft, OpenAI, GitHub, Cursor, and 20+ other platforms adopted the same open standard. And researchers demonstrated that capable models like Claude could teach smaller open-source models through generated skills, achieving up to 45% performance improvements.

This isn't just another AI feature. It's a fundamental architectural shift in how we build agent capabilities—and 2026 is the year it reaches maturity.