I have 52 slash commands in my terminal that I use every day. They're not aliases or shell scripts — they're Claude Code agent skills: structured instruction sets that tell an AI model exactly how to approach a specific type of work.
/reputrack scaffolds a full Next.js SaaS with auth, Stripe, and Supabase in one command. /hunt-xss runs a methodical cross-site scripting audit against a target. /dawit applies my personal monochrome design system to any project. /ethiopic builds bilingual Amharic–English interfaces.
This is what I mean when I say Claude Code is a second brain.
The skill system
Claude Code's slash command system lets you define reusable agent behaviors in Markdown. A skill file is a plain text document — structured instructions, context, patterns, examples — that gets loaded into the model's context when you invoke it.
The key insight: the bottleneck in AI-assisted development isn't the model's capability — it's context. A generic "help me build a SaaS" prompt gives you generic output. A skill that explains your specific stack, your design system, your conventions, your past mistakes, gives you output that fits.
What I've learned about prompt engineering
Writing 52 skills has taught me that good prompts aren't about being clever. They're about being specific and honest about what you want.
The best skills I've written have three things:
- A clear trigger — when exactly this skill applies
- Grounding in reality — specific file paths, library versions, patterns that actually exist
- Explicit anti-patterns — what NOT to do is often more valuable than what to do
The open source angle
I published all 52 skills on GitHub because the value compounds with sharing. Someone who builds on one of my security research skills and makes it better feeds that back into the ecosystem.
If you use Claude Code, clone the repo and try /hunt-idor on your next bug bounty target. It will change how you work.