Agents That Write Code You Can Live With
A quality harness for AI coding agents
0 of 12 pages done0%
Getting started
-
Welcome aboard, done
3 min
Who the course is for, how it's built, and the small print.
-
Pick your path, done
3 min
In your browser with nothing to install, or on your own machine. Your code or FridgeLister. A short quiz helps you choose.
-
Path A: No install, done
10 min
Run FridgeLister and Claude Code on a ready-made Linux machine in your browser. Nothing to install.
-
Path B: Your own machine, done
20 min
Check what you have, install what's missing (PHP, Composer, Git, a coding agent), and get the code.
-
Meet FridgeLister, done
10 min
What's cooking in the practice codebase: the story, the demo users, two ground rules, and small tasks to give your agent.
-
Git: commit to it, done
10 min
The one habit and handful of Git commands the course uses. You make your course branch and practise undo on it.
-
Ready, set, agent, done
5 min
Three checks that your code, your branch and your agent work together. Then the taster.
Free taster: the one-hour quality harness
-
The one-hour quality harness, done
10 min
Why code written by AI agents rots, and the one-hour fix that makes the biggest difference.
-
The seven layers of a quality harness, done
15 min
The seven layers that decide whether an agent's code stays maintainable.
-
Exercise: a starter AGENTS.md and one check command, done
30 min
Give your agent your conventions and a single command it must pass before saying done.
-
Self-assessment: how strong is your harness?, done
10 min
Score your codebase on the seven layers, see your gaps in the order to fix them, and keep a copy to score again later.
-
Where to go from here, done
5 min
What the full course covers, and how to keep going on your own.
Part 1: Habits Coming soon
-
Module 1: Context engineering Give the agent the knowledge it can't get from the code: your conventions, your decisions and your words. 1 h 30 min
-
Module 2: Tests as the specification Tell the agent what done means: examples and tests before code, and the edge cases agents miss. 1 h 30 min
-
Module 3: Shaping the work Small tasks, a plan before the code, and diffs small enough to review properly. 1 h 30 min
-
Module 4: Feedback loops the agent can run One fast, decisive check the agent runs itself, wherever it works, and the same check in CI. 1 h 15 min
Part 2: Measure and guard Coming soon
-
Module 5: Baseline Measure where your codebase stands, so you know where to put guardrails and can tell later whether the harness helps. 1 h 10 min
-
Module 6: Guardrails Turn your most expensive "never" rules into hooks and permissions, and guard the boundary between your machine and production. 1 h 30 min
Part 3: Going deeper Coming soon
-
Module 7: Complexity the agent can see Cyclomatic complexity: what it counts, why 10 is a good limit for agent-written code, and how to add it to your check. 1 h 15 min
-
Module 8: Architecture the agent can't erode Coupling, cohesion and layers, and how to turn rules about structure into architecture tests that fail. 1 h 15 min
-
Module 9: Do the tests protect anything? Measure whether the tests protect the code that matters, with CRAP and mutation testing, and review agent diffs with a checklist. 1 h 50 min
-
Module 10: Keeping it alive, done
25 min
Refactoring routines, a budget for technical debt, and context files that stay true as the code and the agents change.