6 parts
Scope the role, define success criteria, and stand up the CLAUDE.md contract and persistent Memory your workspace reads on every invocation.
Turn the role's standards into four Rules files the workspace enforces on every turn — voice, process, tools, and safety — written once and reused on every interaction.
Build two Skills — a brief-to-artifact Generator and a rules-backed Checker — as reusable, scoped prompt packs your workspace runs the same way every time.
Configure two specialist Agents — one for planning, one for production — that own a point of view, delegate sub-tasks, and review their outputs before saving them.
Wire your Skills and Agents into a multi-stage pipeline with explicit handoffs, a human in the loop at every boundary, and persisted artifacts at every stage.
Put recurring work on Scheduled Tasks, audit the full system, and prove the approach generalizes by extending it into an adjacent role.
A quick reference for the plain-English meaning of every term used across the cowork guide.