# Agent Rules

## Purpose

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

## General Principle

Agents may help clarify, structure, draft, refactor, and prepare work. They must not silently turn options into commitments.

## Allowed Agent Activities

Agents may:

- draft documents;
- improve structure and wording;
- create task breakdowns;
- propose repository structures;
- classify items by activation level;
- summarize open decisions;
- propose next actions;
- generate templates;
- prepare implementation prompts for other repositories.

## Requires Human Approval

Agents must request approval before:

- marking a topic as critical;
- deleting significant content;
- changing project activation levels;
- converting options into commitments;
- changing legal, tax, financial, or company-related conclusions;
- publishing external-facing claims;
- creating production deployment instructions;
- touching secrets, credentials, payment systems, or legal documents.

## Forbidden

Agents must not:

- create artificial urgency;
- overload the task list;
- treat all ideas as work;
- optimize for output volume over personal sustainability;
- make legal, tax, or financial decisions;
- act on behalf of the company externally without explicit instruction.

## Preferred Output Style

Agent outputs should be:

- concise;
- structured;
- explicit about assumptions;
- clear about next actions;
- careful to separate facts, options, and recommendations.