The Claude Playbook
The skills, agents, and commands I use every day to keep Claude writing code I would actually put my name on.
I have this small mental problem where I like to build the same thing over and over to see if I can squeak out one more little bit of groovy awesome. For me that thing is a Stripe webhook receiver with digital fulfillment, in TypeScript, on Postgres. Complex enough to exercise some real logic, short enough that it does not eat a weekend.
Every time through, I tune the instructions. Tests written the way I like them, given-when-then, happy path first and then every sad path I can think of, one expectation per spec. Rigorous SOLID. A composition root instead of a pile of singletons hiding in the corners. No repositories anywhere, because I hate repositories. The goal was always the same: swap out a major part of the app without a full rewrite.
I am about 80% of the way to what I want, which is far enough along that it is worth giving away. So here it is, split the way I actually think about it: skills for the stuff that matters, agents for work I want spawned off, and commands for the things I do in a codebase twenty times a week.
“I used your process today to add a feature to a market data C# project. I created a spec.md file as suggested and provided the inputs and the desired outputs, and like in your video, the only human input was my review. I also added that I would like smaller changes so that they are easy to review. In short, I was blown away with the code it produced, and I’m only just getting started.”
I read every reply, so if you tear this apart and find something busted, tell me about it.