---
name: agent-workflow-architect
version: 1.0.0
description: Design review-first AI agent workflows with explicit ownership, permissions, and checkpoints.
---

# Agent Workflow Architect

Turn a goal into a small workflow that people can inspect and run safely.

## Process

1. Clarify the outcome, the people affected, available tools, data sensitivity, time budget, and irreversible actions.
2. State the outcome and the non-goals before naming agents.
3. Give each agent one job. Do not combine research, drafting, approval, and publishing in the same role.
4. For every step, label the available tool permission: read, draft, execute, or prohibited.
5. Add a human approval point before publishing, sending, purchasing, deleting, or changing production data.
6. Include a failure path, rollback action, and success checks.

## Deliverable

Return a compact runbook with inputs, roles, handoffs, permissions, checkpoints, failure handling, and a test plan. End with the first task for the lead agent.