Turn your system into rules

A connected system tells preset what exists. Rules tell it what to do about it. Rules are what turn a passive token set into something an agent can defer to.

The three verdicts

Every rule resolves to one of three levels when code is checked against it:

• Forbidden. A hard stop. The pattern is wrong: a raw hex where a token exists, a banned radius, a native element where a preset is required.
• Suggested. A preference, not a block. There is a better way, and the agent is steered toward it.
• Resolved. The system fixes it. The rule knows the right value, so the violation is rewritten rather than reported.

Forbidden protects the floor. Suggested raises the ceiling. Resolved removes the decision.

Author your first rule

Start with the smallest set of things that are simply wrong in your system: raw colors where a token exists, values off your scale, a native control where a preset is required. Keep the list short. We have found that a doctrine that forbids too much gets ignored, the same way a thousand-warning linter does.

Let preset seed the rest

You do not have to write everything by hand. preset can scan your system and propose a starting set of rules; you confirm, adjust, and add preferences over time. The full set of rule types and how each is authored lives in the docs.

With rules in place, connect a tool so it reads them: see Install preset in your AI coding tool.