Skip to content

v1.0.0: First public release

Choose a tag to compare

View all tags
@roadhero roadhero released this 28 Jun 13:14
v1.0.0
47d7733

First public release of my actual Claude Code setup. The same configuration I run day to day, and the one described in the book 42: The AI Builder's Stack.

Most people publish a single CLAUDE.md and call it a setup. This is the whole system: a global engineering spine, rules that auto-activate per stack, a roster of scoped agents, and hooks that stop bad commits before they happen.

What's in here

  • The spine — CLAUDE.md, a stack-agnostic engineering baseline. Workflow, git rules, coding guidelines, secrets handling, anti-patterns.
  • 50 scoped subagents across four stacks — 15 web/default, 7 Android, 7 iOS, 21 compute. Each loads only the context it needs, so they're faster and drift less than one mega-prompt.
  • Auto-activating rule packs — rules/{web,android,ios,compute}.md, loaded automatically based on what's in your repo.
  • Two safety hooks — guard-commit.sh blocks force-pushes, non-human committers, AI attribution, and staged secrets. format.sh auto-formats by extension across every stack.
  • A repo scaffolder — the new-repo skill spins up a new project with the right CLAUDE.md, quality gate, and release workflow.
  • Templates + a worked example — blank project templates plus one filled-in example so you can see what "done" looks like.

How to use it

Install steps are in STRUCTURE.md. Two tiers: a global config you set once, and a short per-project CLAUDE.md you fill in per repo. Strip what you don't need.

Honest caveats

Opinionated, built for how my team works. Git rules assume PR flow with protected branches. Hooks assume formatters are installed (they no-op cleanly if not). The compute stack is C++/CUDA/Python and is overkill if you only ship web apps.

Free, MIT licensed, no email gate. Take what's useful.

The book: 42: The AI Builder's Stack — sample chapters are at eltexsoft.com. The full book is on Amazon.

Assets 2
Loading