/
Context Engineering in Claude Code

Context Engineering in Claude Code

A practical introduction to context engineering for anyone looking to get more consistent, reliable results from Claude Code.

Pre-Requisites:

  • Claude Code installed — install via npm install -g @anthropic-ai/claude-code or follow the Claude Code quickstart

  • VS Code with a terminal open — go to Terminal → New Terminal, then type claude to start a session

  • A codebase to practice with — any project works; you'll use it for the exercises

  • Claude running and responding — start Claude in your terminal and try: "What is the name of the codebase I have open?"

Duration: ~1 hour

 

Context engineering is how you take an AI that has been trained on everything and focus it on your specific project, your specific stack, and your specific way of doing things. Without context, Claude is incredibly capable but generic. With the right context, it can behave like a teammate who already knows the codebase.

There's a useful way to visualize this. Think of an LLM as a bridge between all the world's knowledge and your particular system. Claude has seen every programming language, every framework, every document type. Context is what tells it: here's the universe you're operating in. Here's how we like to do things.

Context engineering answers two core questions:

  1. How do I get Claude to work in my project?

  2. Why is Claude failing?

If Claude isn't doing what you want, it's almost always a context problem. The model is smart enough — it just needs better instructions. That's what we're building today.

In this workshop we'll cover three things:

  • Skills — reusable instruction sets that Claude (or you) can invoke when needed

  • Instructions files — always-on context that gets attached to every single request

  • Context engineering basics — how the context window works and why it matters

Workshop Sections

Work through each section in order and complete the exercises as you go!

CC - Agent Skills

CC - Instructions Files

CC - Context Engineering Theory

Conclusion

You've now got the full context engineering toolkit for Claude Code.

The core insight is this: Claude isn't failing because it's not smart enough. It's failing because it doesn't have enough of the right information. Skills, instructions files, and a well-managed context window are how you fix that.

Skills encode your workflows. Instructions files encode your environment. Together, they let Claude act less like a generic AI and more like a teammate who knows your codebase.

The best next step is to look at the places where Claude currently gets things wrong in your project — and ask yourself: what information would fix this? Is it a skill that defines the right process? Is it a line in CLAUDE.md that clarifies a convention? Usually it's one of those two things.

Start there, iterate, and let the context do the work.