TSD AI Workflow

A teaching curriculum for interns at The Software Dojo. Learn to work with Claude Code the way TurtleWolfe works with Claude Code — Docker-first, spec-first, and terminal-orchestrated.

This repo does not teach you Claude Code itself. It teaches you a workflow for building real software with an AI pair programmer, using conventions that have survived 20+ production repos.

• New interns at TSD who have Claude Code installed but don't know what to do with it on Day 1
• Developers who've used Claude Code for scripts and want to adopt it for real feature work
• Mentors who want a single link to hand new hires instead of repeating the same onboarding conversation

Every intern shows up with the same three questions. This repo answers them in order:

1. How do I start a new repo with AI? → Chapter 01
2. How do I give Claude context so it's actually useful? → Chapter 02
3. What slash commands do I use, and when? → Chapter 03

Everything else is advanced material you grow into.

Each chapter follows the same shape:

1. Day-1 minimum — the first two pages. Read this, do the exercise, you're productive.
2. Dig deeper — the rest of the chapter. Come back to it when you have questions.
3. Exercise — a 15-minute hands-on task. Success = a real artifact sitting in your repo.

Chapters are independent. You can read 03 before 02 if you're in a hurry. But the exercises build on each other.

Plus principles/ — the why behind every rule, one file per principle, one war story each.

• The AI Kitchen map (PDF): the whole current setup on one page. An Opus "head chef" hands work to cheap Haiku/Sonnet cooks in separate git worktrees, tests decide pass or fail, and a blind Opus review tastes before anything ships.
• Head chef, line cooks: why it's built this way. What the "let cheap agents do the grunt work" advice gets right, why local models don't fit an 8 GB card, and how the first real run went.
• Build your AI kitchen: the setup tutorial. How to wire up the tiered helpers, the director workflow, the cheap Jev yes/no check, notes with a second assistant, and the gotchas that cost us a day.
• The Action Figure Automata: the team behind it all, nicknamed "the fleet". Who each AI member is, which machine it runs on, and how they voted on their own name.

Chapter 01's exercise has you clone hello-world-ai (a companion repo) as a starting point instead of bootstrapping from zero. It's a minimal Next.js + Docker setup with the 10-command starter kit already installed at .claude/commands/. First commit: under 15 minutes from git clone to docker compose up to a shipped change.

Note: hello-world-ai lives at ~/repos/hello-world-ai on the machine that produced this curriculum. Ask your mentor for the current clone URL.

• Not a Claude Code tutorial. It assumes you've run /help at least once. If you haven't, read Anthropic's docs first.
• Not a generic SpecKit guide. SpecKit has its own docs. This repo teaches how the TSD workflow uses SpecKit inside a larger loop.
• Not a replica of Claude_Commandz. That repo is TurtleWolfe's full 65-command disaster-recovery backup. The Chapter 03 starter kit here is the minimal 10 commands an intern actually uses in week one.
• Not a code library. No generators, no scaffolds. This is documentation plus exercises.

• Claude Code installed and working (claude --version returns something)
• Docker Desktop installed (required — Chapter 01 fails without it)
• Git installed
• A terminal you're comfortable in
• Optional but helpful: a GitHub account

If you hit a question the repo didn't answer, or an exercise that didn't work, open an issue. The /principles directory exists precisely because every rule in here started as a war story. Your question is the next war story.

Ready? Start with Chapter 00.