Intro
My AI development setup, versioned. Skills, hooks, slash commands, subagents, MCP configuration, and the workflows around them — installed onto a machine with one script, dropped into a project with another.
The idea
Three kinds of thing, three different lifecycles, and mixing them is what makes a setup rot:
| Lives in | Updated by | |
|---|---|---|
| What I own | this repo, in git | me |
| What others own | ~/.claude/ via the skills CLI | npx skills update |
| What a project owns | the project's own repo | copied in once, then edited there |
The first fifteen skills here arrived by copy-paste and froze at a version two releases behind upstream, with nothing recording that a newer one existed. The split above exists so that cannot happen again.
Start here
- Install — get it onto a machine
- What lands where — the map
- Claude Code fundamentals — models, context, the mechanisms
- Beyond vibecoding — what you still have to know
What is in it
| Area | Contents |
|---|---|
| Skills | 24, in 6 categories, each with recorded provenance |
| Hooks | 9 — secret blocking, dangerous-command blocking, format, lint, test, audit log, sounds |
| Commands & agents | 3 slash commands, 3 subagents |
| MCP | My servers, plus the ones worth adding |
| Workflows | Planning, architecture, coding, debugging, review, devops, observability |
| Stack notes | Frontend, backend, devops, AI — what to use with each |
Official documentation
This site is deliberately thin. Where Anthropic documents something, it links there rather than paraphrasing it into staleness.