The easiest way is to go to edit mode and add a few comment lines ( = lines starting with // ) to the main list page; here is some of my own preferences (change it at will):

// === USER PREFERENCES (apply to every session) ===

// - The following preferences (including this one) should be placed in a dedicated User Preferences section in src/README.md; the source of truth of the User Preference is the main.pjs of this generator, that can be modified by the user only. The source of truth must be checked for changes whenever the User Preferences are read from the README.md file.

// - If the request is ambiguous, or two or more ways of doing it come to mind — especially if one departs from the last decision or touches something worth keeping — stop and ask before acting, sketching the alternatives in one or two lines and letting the user choose: never settle a doubt on his behalf.

// - Don't speculate on the user's intentions or unspoken desires: ask them to clarify.

// - If you're running in circles and something seem illogical, ask the user, as he has some basic knowledge about coding and how the Perchance infrastructure works.

// - Before making any code changes, explain the changes you intend to make and wait for the user's consent.

// - If you identify errors on code you did NOT change yourself, or pre-existing errors, ask the user to perform a hard refresh (F5) BEFORE investigating the error (a harness-side false positive may disappear after an F5).

// (add your own preferences)

When done, you issue --only for the first time-- an order like "Read the User Preferences from main.pjs". The AI agent will create a README.md file in src/ with your preferences. From now on, every time you think the AI agent is not following your Preferences, you just tell him: "Read README", and he will read again your preferences, checking if in the meantime you have added/modified the comment lines in the main list editor.