I've been working with agentic coding for over two years now, and I've genuinely never had more fun building software. But it's easy to lose your footing — the field is moving fast, and the moment you start, the buzzwords come at you: subagents, harnesses, agent teams. There is no one correct way to do things, no golden standard. The field has changed, but in essence it also stays the same.
In the last two years I tried most of the new things — from agent teams to full-blown agentic command-line harnesses like Spec Kit — and Skills are what stuck. For me they're the simplest way to capture how my own mind works — my way of working and thinking — and let an LLM use it. The point of this repo is not to hand you my setup as gospel. It's to show how I captured the way I like to work in an agentic coding setup. I highly recommend not copying someone's setup, tool, or workflow wholesale — instead, find your own, and make it your AI-supported brain.
Which is the whole point: it's a tool. A remarkable one, but a tool. Linus Torvalds put it better than I can:
"AI is a great new tool, but it's a tool, and when I see people saying, 'Hey, 99% of our code is written by AI,' I literally get angry, because those same people — I can pretty much guarantee — that 100% of their code is written by compilers. But they never say that." — Linus Torvalds
We don't brag that the compiler wrote our machine code. AI is the next layer up. Learn to wield it well, stay responsible for what it produces, and get back to having fun.
| # | Page | What it covers |
|---|---|---|
| 1 | The journey | From a simple store to "skills stuck" — and what it means for our job |
| 2 | Project setup | The standard Claude Code layout, plus my shared dev-environment layer |
| 3 | The workflow | My spec pipeline: how an idea becomes shipped code, artifact by artifact |
| 4 | Capture | The knowledge loop — offloading links & follow-ups into Obsidian |
| 5 | Glossary | Every term in this take-along, defined simply — and linked to the deep-dives |
Top to bottom tells a story (page 1 → 5). But each page stands alone — skim the diagrams and tables first; the prose is there to connect them, not to be studied.
The one idea to take home: don't copy a workflow. Notice how you like to work, then ask where an LLM can carry the parts you don't want to. That's the whole game.