A cookbook with a kitchen diary bound in the back. The course teaches the basic recipes, one chapter at a time. The build log is what happened when I cooked them for real, burns included. Every part is listed here, in order.
Each section opens one at a time. Plain line on top; links and dates underneath.
New to Claude Code? Read the course front to back. Every chapter is usable after its first two pages, and chapters 01 to 04 each end with one fifteen-minute exercise.
Already working with it? Go to the build log and read whichever title catches you. Each part stands on its own.
The course: this site, chapters 00 to 05, written for interns at The Software Dojo. The build log: blog posts on scripthammer.com plus deep-dive pages in chapter 05, oldest first. Every chapter, page and post links back here and to its neighbours.
Six chapters, each answering a question you'll have in your first week. Open one to see what it teaches.
Install Claude Code, check it runs, and have a first useful conversation. Read chapter 00 →
15 min · 00-start-here/
Start a Docker-first repo, either from the starter or from scratch. Read chapter 01 →
15 min · 01-bootstrap-a-repo/ · exercise: clone the starter and ship a change
Write a CLAUDE.md, add knowledge bases, and prime each session so Claude knows your project. Read chapter 02 →
30 min · 02-give-it-context/
Install a starter kit of slash commands and learn which one to reach for when. Read chapter 03 →
30 min · 03-slash-commands/ · starter kit in 03-slash-commands/starter-kit/
Take one feature from a written spec to shipped code, one step at a time. Read chapter 04 →
60 min · 04-the-speckit-loop/ · /speckit.specify → .clarify → .plan → .tasks → .implement
The tmux assembly line: terminal roles, an operator, a task queue and votes on changes. The build log's deep-dive pages live in this chapter too. Read chapter 05 →
2+ hours · Level 3 · nothing in chapters 00 to 04 depends on it
One rule per file, each with the incident that earned it. Read the principles →
docker-first · spec-before-code · context-hygiene · secrets-never-committed · five-file-components
Seven parts so far, oldest first. Each is a real job, mistakes left in. Open a part to see what it teaches and where to read it.
An agent with real credentials can file your paperwork and migrate your database. It can also quietly store a setting that does nothing. Five rules for handing over access. Read the post →
2026-07-05 · blog post · scripthammer.com/blog/cursor-github-identity/
A client's email hides the real request inside chatty prose and screenshots. A mail-reading tool that drafts GitHub issues pins it down before any code gets written. Read the post →
2026-07-07 · blog post · an MCP server that reads mail and drafts tickets
When a collaborator's AI agent sends a flawed pull request, block it with a review and a paste-ready fix. Never commit onto their branch. Read the post →
2026-07-07 · blog post · gh pr review --request-changes + a tracking issue
Context doesn't have to be a codebase. Turn a data export into a knowledge graph you can query, and find the work you forgot to answer. Read the post →
2026-08-05 · blog post · graphify over a Facebook export, 31,030 records back to 2011
One expensive model plans and reviews. Cheaper ones make the edits, each in its own copy of the repo, and tests check every change. The first run cost about 41% less. Read the post →
2026-09-30, updated 2026-10-01 · three ways in: blog post: scripthammer.com/blog/head-chef-line-cooks/ deep dive: 05-advanced-orchestration/head-chef-line-cooks.html the map: ai-kitchen-map.html (PDF)
Set the whole thing up yourself, in the order I did it, including the gotchas that cost a day. Read the tutorial →
2026-09-30, updated 2026-10-01 · tiered helpers, the director workflow, a cheap yes/no check, a free review panel behind a privacy gate, a per-client usage ledger
Which AI does what, which machine it runs on, and how they voted on their own name. Meet the team →
2026-10-01 · nickname: "the fleet"
Paste what you're about to post and see every phrase that makes it read like AI. Runs in your browser. Check something →
2026-10-05 · latest part · the same rules as Claude Code's pre-send check, for agents with only a browser
New parts land here first. If you found the series through a post, a share or a stream, this page is the map.
This page: tortoisewolfe.github.io/AI_Workflow/series.html The blog: scripthammer.com/blog/ The course source, issues welcome: github.com/TortoiseWolfe/AI_Workflow