Part 3: Instruct with Intent
The Problem
The gap between “meh” AI output and “that's exactly what I needed” isn't the AI — it's the instruction.
Claude Code can do incredible things, but only if you learn to tell it what you actually want. This lesson teaches you the patterns that work.
The Instruction Spectrum
The quality of your instruction determines the quality of the result. Here's how to move from bad to great:
Vague
Claude Code guesses. Results are generic.
Directional
Better. But “improve” means different things to different people.
Specific
Now Claude Code knows exactly what to do.
Contextual
Best. Uses project context, gives constraints, specifies what NOT to change.
Remember This
The best instructions tell Claude Code three things: what to do, what to keep, and where to look.
Essential Commands
These slash commands are your shortcuts. Memorize these four and you'll be ahead of most users:
See all available commands
Your safety net. When you're not sure what to do, start here. Shows every command Claude Code supports.
Summarize the conversation to free up context
Use when Claude Code starts forgetting things you told it earlier. It compresses the conversation so the AI can keep working with full context.
Start fresh
When you want a clean slate. Use this when switching to a completely different task and the previous conversation is no longer relevant.
Switch between different AI models
Opus for complex, nuanced work. Sonnet for speed when you need quick changes. Think of it like choosing between a specialist and a generalist.
The @ Symbol — Your Secret Weapon
The @ symbol lets you point Claude Code at a specific file. This is how you make sure it's looking at the right thing:
“Look at this specific file”
“Check our dependencies”
Use this when Claude Code is looking at the wrong file or needs to focus on something specific.
With @ reference (specific)
Without @ reference (vague)
Claude Code might look at the wrong file or create a new component instead of editing the existing one.
Prompting Patterns That Work
Five proven patterns you can use immediately. Each one fits a different situation:
The Scout
Understand what's there
Use when: You just joined a project or need to understand what's there before making changes.
The Surgeon
Make a precise, surgical change
Use when: You need one specific thing changed and nothing else touched.
The Builder
Create something new from scratch
Use when: You need something new built and want it to match what's already there.
The Reviewer
Understand recent changes
Use when: You want to understand what your team has been doing without reading code yourself.
The Debugger
Fix what's broken
Use when: Something is broken and you need it fixed without understanding the technical details yourself.
What You Just Learned
- You know the difference between vague and effective instructions
- You can use slash commands (/help, /compact, /clear, /model) confidently
- You understand how @ references help Claude Code focus on the right files
- You have 5 prompting patterns you can use immediately
Next: Now that you can instruct effectively, let's learn how to automate repetitive work.