Free preview4 min

Map Your Own Repo

Reviewed by Human · Updated September 22, 2026

Open a project you actually use an agent on. How many of the five subsystems already exist as real files? Guess, then tap.

How Do You Map a Repo to the Five Subsystems?

Mapping is a two-column audit: list what exists, then mark what's missing. **Subsystem map**: a one-line-per-subsystem inventory of which file (if any) currently does that job in your repo. For each subsystem, name the file — or write "none." A "none" isn't a verdict on your model; it's a to-do. This is the exact audit a minimal harness closes first, and it's the same lens that separates a repo an agent drifts in from one it works reliably in.

The blanks in your map are your backlog. Every "none" predicts a class of failure.

Your turn

Do the audit on a real repo. Fill the file name that does each job today, or write "none."

Reflect

Count your "none"s. That number is roughly how surprised you should be when the agent misbehaves.

Challenge

Close one blank

Pick the single "none" from your audit that would remove the most pain, and draft the smallest file that fills it. Not a perfect file — a real one that changes agent behavior next session.

You add a detailed AGENTS.md but the agent still 'finishes' broken features. Which subsystem is most likely still empty?

The whole point, in one line

The harness doesn't make the model smarter — it makes its output reliable. You're not upgrading the pilot. You're building the instruments that keep a capable pilot from flying into the ground.

📚

Go deeper

end of module

You've finished this module.

Mark it complete to earn your XP and keep your streak alive.

Progress saved locally · Sign up to earn XP

←Who Does What: Model vs. Harness