Every check in this chapter started as something that went wrong. That is how a harness grows, one mistake at a time, and it brings the chapter back to where it began.
Ratchet
A rule of thumb: when the agent makes a new kind of mistake, add something that catches that kind from then on. On an old codebase, record today's count of a problem and fail only when it goes up. Quality can then rise, and it can't quietly slip back.
Without it you fix the same class of problem again and again. With it, your instructions file becomes a list of real past failures, which is also the best evidence you can show of your own process.
“You have made this mistake twice now. Add a check that fails on the old version of this code, confirm that it does, then fix the code so the check passes. If the old code breaks it in many places, record today's count and fail only when it goes up.”