Skip to content
CHICKENPIE.Start a project →
Tool Tuesday

Building Claude Skills: Our First Iteration

Week 1 of building a Claude Code skills library. From concept to first commit - here's how we're creating reusable AI-powered workflows.

Read the paper ↓
Code editor showing Claude skills configuration with terminal output
A PAPER FROM THE DESK
THE DETAILS
At a glance
CategoryTool Tuesday
PublishedJan 23, 2026
Read time2 min
Covered here
  • Productivity
  • AI Tools

This week marked a major milestone: we shipped the first iteration of our Claude Code skills library. If you've ever wished you could teach Claude to do repetitive tasks your way, this is for you.

What Are Claude Skills?

Claude Skills are reusable instruction sets that extend Claude Code's capabilities. Think of them as recipes that tell Claude exactly how to handle specific tasks - from content management to code deployment.

Diagram showing Claude skills architecture - skills folder structure with markdown files and configuration
The skills architecture: Simple markdown files that pack a punch

The First Iteration

Our initial release includes three core skill categories:

  1. Content Management Skills - Sanity CMS integration, document creation, scheduling workflows
  2. Scheduling Tools - Calendar integration, task management, deadline tracking
  3. Development Workflows - Git operations, deployment pipelines, code review automation
Screenshot of terminal showing Claude Code executing a skill command with colorful output
Running a skill from the command line - it just works

How It Works

Each skill is a markdown file with structured instructions. Here's the basic anatomy:

1. Skill metadata (name, description, triggers)

2. Context requirements (what Claude needs to know)

3. Step-by-step instructions (the actual workflow)

4. Output specifications (what to produce)

Code editor split view showing a skill markdown file on the left and Claude executing it on the right
Writing skills in markdown makes them easy to version control and share

The GitHub Repository

We've open-sourced everything. The repository includes:

  • Ready-to-use skill templates
  • Documentation and examples
  • Integration guides for Sanity, GitHub, and more
  • A content agent example (what powers this very blog!)
GitHub repository page showing the claude-skills project with folder structure and README visible
The full repository - fork it, star it, make it yours

Lessons Learned

Building this first iteration taught us a lot:

"The best AI tools don't replace your workflow - they adapt to it."

The key insight: Claude works best when you give it clear, structured instructions that match how you actually think about tasks. Skills are just a way to codify that knowledge.

What's Next

Next week we're focusing on:

  • Adding more MCP server integrations
  • Building a skill discovery/sharing platform
  • Creating video walkthroughs

If you're building with Claude Code, give the skills library a try. And if you create something cool, we'd love to feature it!

Share
MORE FROM THIS LITTLE DESK

A little worth keeping.

Stories, experiments, and things I find along the way. Occasionally, in your inbox.

Get the letter ↗
/feed/building-claude-skills-first-iteration

Have a project in mind?

Tell us what you are making and what you need. An unfinished idea is welcome too.

Start a project →

Another paper for you.

Mr. Chicken sitting at a small desk coding on a laptop with an offline cloud icon on screen, in a brick-walled room

BiNeuron promises a local AI coding assistant with no cloud bill. Here's what "local" actually costs you.

BiNeuron is a free, open-source, fully-offline AI coding assistant. Here's what the whole-file-replacement design, the Tkinter GUI, and the 2-star repo status actually mean before you install it.

Open paper
CHKN flat-vector illustration: Mr. Chicken in a navy hoodie at a home-office desk searching a laptop media library, hard drives and notebook beside him

Clipto just raised $15M to be your computer's memory. It's good. It's not there yet.

Clipto's local AI media search is real and useful. The pricing page contradicting itself and a shaky Windows beta keep it from a full recommendation.

Open paper
Mr. Chicken at a desk pointing at a laptop screen showing a colorful AI-generated image, with reference photos stacked nearby

6 AI Art Tool Tips That Stop Your Marketing Images From Screaming "AI"

6 quick tips for using AI art tools like Nano Banana and Gemini 3 to make marketing visuals that don't scream "AI" — prompting, texture, consistency, and a pre-publish check.

Open paper
Chickenpie mascot Mr. Chicken pointing at a glowing screen showing chaotic color swatches organizing into labeled design system tokens

Stop Hardcoding Colors: Build a Real Design System with Figma Variables

A hands-on 9-step tutorial: wire a Primitives, Semantic, and Brand variable system in Figma so your whole file updates from one place, light mode and dark mode included.

Open paper
Mr. Chicken at a cluttered home studio desk, working on an AI image generation interface with an uncanny AI-slop draft on one side and a clean composited image on the other

How to Art-Direct AI Images So They Stop Looking Like AI

Nine steps to art-direct AI image generation like a photographer, not a wishlist — light, texture, junk tokens, and the two spots (hands, signage) that always give it away.

Open paper
Mr. Chicken at a nighttime desk poking a laptop screen showing a bouncy Lottie squeeze-toy animation

I Tried to Build a Clickable Lottie Squeeze Toy. It Took Three Tries to Feel Right.

Watch three failed attempts to build an interactive Lottie squeeze toy — from a dead mousemove hack to a broken freeze-frame click, to the segment-based fix that finally feels tactile.

Open paper

You might also like