# Agent Rules

## Purpose

This document defines how AI coding and writing agents may assist within `whynot-control`.

## General Principle

Agents may help clarify, structure, draft, compare, and analyze prototype ideas. They must not silently turn experiments into product commitments.

## Allowed Agent Activities

Agents may:

- draft prototype cards;
- classify ideas by lifecycle stage;
- propose smallest useful tests;
- summarize feedback;
- draft signal records;
- compare prototype candidates;
- propose beta plans;
- identify promotion targets;
- prepare structured tasks for Helix or Coulomb;
- improve wording and structure.

## Requires Human Approval

Agents must request approval before:

- promoting a prototype to productization;
- marking an idea as commercially validated;
- creating public-facing claims;
- proposing paid beta or investment mechanics;
- contacting external users or communities;
- changing activation level;
- creating new implementation repositories;
- adding payment, legal, or investment language.

## Forbidden

Agents must not:

- create artificial urgency;
- treat all prototype ideas as products;
- infer willingness to pay without evidence;
- present weak signals as strong validation;
- create legal, financial, or investment commitments;
- publish external beta announcements without approval;
- act on behalf of Binky, Plenitude, or any company entity externally.

## Preferred Output Style

Agent outputs should be concise, evidence-oriented, explicit about uncertainty, and careful to separate idea, hypothesis, signal, and decision.