Agent Scopes

Purpose

Agent scopes keep AI worker parallelism safe. Each agent must know its default mode, allowed paths, forbidden paths, required reading, acceptance checks, and escalation rules before it edits anything.

Standard Structure

Every agent scope document uses:

# Agent Name

## Purpose

## Default Mode

## May Edit

## Must Not Edit

## Required Reading

## Typical Tasks

## Acceptance Checks

## Output Required

## Escalation Rules

Global Rules

• AGENTS.md remains the strong rule source.
• A task graph can narrow an agent's scope further.
• Agents must not expand scope silently.
• If two agents need the same file, pause and ask Architect AI to re-plan.
• Integrator Agent is read-only by default.
• Human owner decides merge and deployment.

Agents

• architect-agent.md
• docs-agent.md
• ui-agent.md
• shell-agent.md
• app-agent.md
• registry-agent.md
• api-agent.md
• qa-agent.md
• integrator-agent.md
• business-agent.md
• growth-agent.md
• content-agent.md
• ops-agent.md