← all tracks
T2

Context & workflow

The habits that make an agent go from frustrating to fast: a context file it reads, working from a spec instead of a wish, and building your own commands so your workflow repeats.

Claude CodeCodex

Most of the gap between “AI is useless” and “AI is a force multiplier” isn’t the model — it’s how you set up the work. This track is the setup.

What you’ll learn

  • The context fileCLAUDE.md / AGENTS.md. A short file the agent reads every time so it knows your stack, conventions, and the commands to run. The single highest-leverage thing you can add to a project. (guide)
  • Spec-first building — write a one-page spec before code. A good spec is the biggest lever on output quality. (Addy Osmani on specs, GitHub Spec Kit)
  • The explore → plan → code → commit workflow — the loop, formalized, so you stop letting the agent sprint before it understands.
  • Your own commands & subagents — automate the tasks you do over and over (a /review command, a test-writing subagent).
  • Parallel work — git worktrees and multiple sessions, so you’re not stuck watching one agent think.

The build

You’ll turn a repetitive part of your own workflow into a custom slash command or subagent — project 08. Done means it does a real, repeatable task for you on demand.

Full lessons land after Track 1. Join the newsletter to get them first.

Build these in this track
08 Intermediate

Custom slash command + subagent

done → A /command or subagent that does a repeatable task for you.

Free · the newsletter

Get the build logs + the Mirror System

One email a week: a real build broken down, plus a working reference agent you can clone. No fluff, unsubscribe anytime.

No spam. Unsubscribe in one click.